PhDthesisRemcoVerzijlbergh.

PhDthesisRemcoVerzijlbergh.
The Power of Electric Vehicles
Exploring the value of flexible electricity
demand in a multi-actor context
R.A. Verzijlbergh
.
The Power of Electric Vehicles
Exploring the value of flexible electricity
demand in a multi-actor context
Proefschrift
ter verkrijging van de graad van doctor
aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof.ir. K.C.A.M. Luyben
voorzitter van het College van Promoties,
in het openbaar te verdedigen op vrijdag 25 oktober 2013 om 15:00 uur
door
Remco Alexander VERZIJLBERGH
Natuurkundig ingenieur,
geboren te Hellevoetsluis
Dit proefschrift is goedgekeurd door de promotor:
Prof.dr. M.D. Ilić
Copromotor:
Dr.ir. Z. Lukszo
Samenstelling promotiecommissie:
Rector Magnificus
Prof.dr. M.D. Ilić
Dr.ir. Z. Lukszo
Prof.dr.ir. M.P.C. Weijnen
Prof.dr.ir. B. De Schutter
Prof.dr.ir. G. Deconinck
Prof.dr. I.J. Peréz-Arriaga
Dr. P.M.S. Carvalho
Prof.dr.ir. P.M. Herder
voorzitter
Technische Universiteit Delft, promotor
Technische Universiteit Delft, Copromotor
Technische Universiteit Delft
Technische Universiteit Delft
Katholieke Universiteit Leuven
Universidad Pontificia Comillas
Instituto Superior Técnico
Technische Universiteit Delft (reservelid)
Published and distributed by:
Next Generation Infrastructures Foundation
P.O. Box 5015, 2600 GA, Delft, the Netherlands
[email protected], www.nginfra.nl
This research was funded by the Next Generation Infrastructures Foundation
ISBN 978-90-79787-53-1
Keywords: electric vehicles, smart grid, demand response, renewable energy, distribution networks.
Copyright © 2013 by R.A. Verzijlbergh. All rights reserved.
Cover photo by Victor Calado. Electric vehicle in front of a wind-park near Zeewolde,
the Netherlands.
Printed in the Netherlands by Gildeprint Drukkerijen, Enschede
Contents
Acknowledgements
v
1 Introduction
1.1 The changing energy landscape . . . . . . . . . . . . . . . . . . . . .
1.1.1 The growth of renewables and its drivers . . . . . . . . . . .
1.1.2 The advent of electric vehicles . . . . . . . . . . . . . . . . .
1.1.3 The potential synergy between electric vehicles and renewable
energy sources . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.4 Changing roles in future power systems . . . . . . . . . . . .
1.2 This thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Problem description . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 Research objectives . . . . . . . . . . . . . . . . . . . . . . . .
1.2.3 Thesis outline, structure, research methods and scope . . . .
1
1
1
5
2 Electric vehicles in future power systems
2.1 Power systems . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 Technical aspects . . . . . . . . . . . . . . . . . . .
2.1.2 Load and generation profiles . . . . . . . . . . . .
2.1.3 Non-technical aspects: organizational, economical,
latory . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Electric vehicles . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Actor analysis . . . . . . . . . . . . . . . . . . . .
2.2.2 Driving data . . . . . . . . . . . . . . . . . . . . .
2.2.3 EV battery model . . . . . . . . . . . . . . . . . .
2.2.4 Uncontrolled charging . . . . . . . . . . . . . . . .
2.2.5 Electric vehicle charging as optimization problem .
. . . . . .
. . . . . .
. . . . . .
and regu. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
5
6
6
6
7
7
11
11
11
14
17
23
23
27
29
31
32
3 Literature review
3.1 Trends in literature on the role of EVs in smart grids . . . . . . . . .
3.2 Discussion of some important papers per sub-field . . . . . . . . . .
3.3 Relative positioning of this thesis regarding the literature . . . . . .
39
39
42
47
4 Network impacts and cost savings of controlled EV charging
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Research method . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
51
52
i
ii
Contents
4.3
4.4
4.2.1 Distribution networks
4.2.2 New load profiles . . .
4.2.3 Power flow . . . . . .
4.2.4 Energy loss estimation
4.2.5 Costs . . . . . . . . .
Results . . . . . . . . . . . . .
4.3.1 MV/LV Transformers
4.3.2 MV cables . . . . . . .
4.3.3 HV/MV substations .
4.3.4 Economic figures . . .
Conclusions . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 Impacts of controlled EV charging on cross-border electricity flows
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Model formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.1 EV charging model . . . . . . . . . . . . . . . . . . . . . . . .
5.2.2 EV data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.3 Charging scenarios . . . . . . . . . . . . . . . . . . . . . . . .
5.2.4 Typical EV fleet . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.5 One node unit commitment model . . . . . . . . . . . . . . .
5.2.6 Multi node unit commitment model with flexible EV load . .
5.3 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Two node conceptual system . . . . . . . . . . . . . . . . . .
5.3.2 Generator parameters . . . . . . . . . . . . . . . . . . . . . .
5.3.3 Wind and solar time series . . . . . . . . . . . . . . . . . . .
5.3.4 Other simulation details . . . . . . . . . . . . . . . . . . . . .
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.1 Dispatch profiles . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.2 Demand function for transmission . . . . . . . . . . . . . . .
5.4.3 Further analysis . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
54
57
58
60
62
62
62
65
65
69
71
71
73
73
73
74
74
74
75
77
77
77
78
78
79
79
82
82
89
6 Renewable energy sources and responsive demand. Do we need
congestion management in the distribution grid?
91
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Problem analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.1 The need for congestion management due to the weakening
corellation between wholesale electricity prices and demand . 93
6.2.2 Minimum cost EV charging formulation . . . . . . . . . . . . 94
6.2.3 Simulation of the current situation (flat grid tariff) . . . . . . 96
6.3 Congestion management mechanism design . . . . . . . . . . . . . . 97
6.3.1 Dynamic network tariff . . . . . . . . . . . . . . . . . . . . . 98
6.3.2 Advance capacity allocation . . . . . . . . . . . . . . . . . . . 99
6.3.3 Distribution grid capacity market . . . . . . . . . . . . . . . . 99
6.3.4 Proxies for optimal tariff . . . . . . . . . . . . . . . . . . . . . 100
6.3.5 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Contents
6.4
.
.
.
.
.
.
.
103
103
103
105
107
107
108
7 A refined view on electric vehicle charging
7.1 Equivalence of centralized and decentralized demand scheduling . . .
7.1.1 Theoretical analysis of EV dispatch . . . . . . . . . . . . . .
7.1.2 Simulations comparing centralized and decentralized EV dispatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1 Inter-temporal generation constraints . . . . . . . . . . . . .
7.2.2 Influence of the forecast horizon . . . . . . . . . . . . . . . .
7.2.3 Influence of charging availability . . . . . . . . . . . . . . . .
7.3 System level networks impacts of minimum cost charging . . . . . .
7.4 Other settings and applications for demand response . . . . . . . . .
7.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
111
112
113
8 Conclusions and recommendations
8.1 Conclusions and answers to research questions . . . . .
8.2 Contours of a new paradigm for a clean and intelligent
8.3 Recommendations . . . . . . . . . . . . . . . . . . . .
8.3.1 Future work . . . . . . . . . . . . . . . . . . . .
8.3.2 Considerations for policy makers . . . . . . . .
133
134
137
139
139
141
6.5
Results and discussion . . . . . . . . . . . . .
6.4.1 Simulation setup . . . . . . . . . . . .
6.4.2 Simulation results . . . . . . . . . . .
6.4.3 Comparison of results to the literature
6.4.4 Uncertainty . . . . . . . . . . . . . . .
6.4.5 IT infrastructure requirements . . . .
Conclusions . . . . . . . . . . . . . . . . . . .
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . . . . . .
power system.
. . . . . . . .
. . . . . . . .
. . . . . . . .
118
122
122
124
126
127
128
130
Appendix A: The potential of EVs in an isolated power system
143
Appendix B: EV impacts in residential low voltage grids
149
Appendix C: Carbon emmissions due to EV charging
151
Appendix D: Synthetic driver profiles
159
Appendix E: Cold storage as another resource for demand response 163
Bibliography
169
Nomenclature
176
Summary
181
Samenvatting
187
List of publications
193
iv
Contents
Curriculum vitae
195
NGInfra PhD Thesis Series on Infrastructures
197
Acknowledgements
Like most scientific work, this thesis, too, could only have been realized with the
invaluable contributions of many others. First and foremost I want to thank my
supervisor Zofia Lukszo. She has brilliantly guided me through this research by
letting me be free yet always showing me the right direction in a remarkably sharp
and subtle way. It is an honor and great pleasure to work with her.
I also wish to thank Marija Ilić for being my promotor and for giving me the
opportunity to leave Delft and perform research both at Carnegie Mellon and at
MIT. The insights gained during those inspiring periods form the basis on which
this thesis is build.
I want to express my gratitude to Margot Weijnen for trusting me to be part of
the section E&I. I am also grateful for the freedom in which I was allowed to do my
research. It is greatly appreciated.
I would also like to acknowledge the important role that Laurens de Vries has
played in this research. By sharing his exceptional knowledge about the electricity
sector he contributed directly and indirectly to this thesis.
Working together with Carlo Brancucci Martı́nez-Anido is something I truly
enjoyed. I think we managed to combine our models to arrive at new insights in a
strikingly fast, efficient and most of all fun way. I hope we will continue the good
work.
I thank Han Slootweg from Enexis for giving me the opportunity to work in his
team to assess the impacts of electric vehicles on their distribution networks. The
insights, discussions and data from that period were very valuable for this research
and chapter 4 relies almost completely on it. Continuing this work together with
Else Veldman has been a great pleasure. I hope the fruitful cooperation between
Delft and Enexis will continue to exist.
For the largest part of this research I shared office with Amineh Ghorbani and
Chang Yu, who I thank for their inspiration, their wonderful company and the
pleasant atmosphere that always fills our office. Among many other colleagues that
have contributed directly and indirectly to this thesis, I would especially like to
thank Reinier van der Veen for the insightful discussions on balancing markets,
Chris Davis for all his Linux related help, Rob Stikkelman for his unconventional
yet always sharp advice and Michiel Houwing for being my initial office-mate which
kick-started my research enormously. The Power Rangers are highly appreciated
for their sharp and enthusiastic input in our weekly meetings. All my other E&I
colleagues, too, are greatly appreciated for their good advice, shared knowledge,
v
vi
Acknowledgements
friendliness, humor and wit. You are a wonderful team to work amongst and I am
glad that I have the opportunity to continue doing this.
The sometimes cumbersome process of doing a PhD research is made much lighter
in the times outside working hours. For this I thank all my friends in Rotterdam
and elsewhere, although the early working hours sometimes did not feel particularly
light because of you. I am also truly grateful to my dear family for their love and
endless support. And finally, I thank you, Elise, because being with you makes me
feel so happy. That keeps me going more than anything.
Remco Verzijlbergh,
Delft, September 2013.
Chapter 1
Introduction
1.1
1.1.1
The changing energy landscape
The growth of renewables and its drivers
Realizing a transformation to a sustainable energy based economy is one of the
great challenges of our time, because it addresses one of the biggest threats to life
on earth as we know it: anthropogenic global warming. The International Panel on
Climate Change (IPCC) reports are the remarkable materialization of many years of
climate science, and they leave little room for doubt: in order to maintain a livable
planet, carbon emissions should drastically be reduced [1]1 . This calls for fundamental changes in our society, and most notably in our energy system. A quote
from Nobel-prize winning chemist Sherwood Rowland related to the ozone debate
can be considered appropriate in the discussion on climate science, global warming
and renewable energy policy, too:
‘What is the use of having developed a science well enough to make predictions if,
in the end, all we are willing to do is stand around and wait for them to come true?’
However, despite the overwhelming scientific evidence for anthropogenic global
warming, a number of skeptical voices are still being heard. In the end, their argument often has an economic character: the costs of preventing an uncertain global
warming scenario to happen are simply too high. Nevertheless, next to the environmental arguments, economic and geopolitical considerations are equally important
drivers towards a departure from a fossil fuel based economy. In the longer run,
prices of finite natural resources will inevitably rise as the most easily accessible resources will become depleted. When zooming in on shorter timescales, one observes
developments that can temporarily alter long term trends, such as financial crises
or technological breakthroughs like the new shale gas production techniques in the
1 In September 2013 a draft version of the 5th IPCC report was published. The officially approved
version is expected in January 2014. The approved executive summary of the draft version shows
roughly the same conclusions as the 4th IPCC report.
1
2
1 Introduction
2
Natural Gas Price (−)
1.5
1
0.5
EU
USA
0
2000
2002
2004
2006
Year
2008
2010
2012
Figure 1.1 – Normalized (with respect to 2001 levels) natural gas price development for
large industrial consumers in the US and Europe. Data from [2] and [3].
US. As an illustration, Fig. 1.1 shows the recent trends in natural gas prices for
industrial consumers, relative to the levels of 2001. One observes how in Europe the
trend is clearly upwards, with a little dip that marks the 2010 post-crisis dip in oil
prices. On the other hand, after 2008, the massive deployment of shale-gas extraction
technology has seriously lowered US gas prices. This pictures also reveals a slight
fraction of the complex geopolitical issues that play a role around energy. While the
US have increasingly become an independent producer with a domestic production
that meets a large portion of demand, Europe has become more dependent on other
countries. Two opposite trends in gas prices are the result.
The largest economic driver towards renewable energy sources (RES) are, however, not so much the rising costs of fossil fuels, but the spectacularly decreasing
costs of RES themselves. In particular the cost of solar photo-voltaic (PV) energy
has dropped dramatically in recent years. Both wind and solar PV now have similar
levelized costs per MWh as most conventional generation technologies. This point
is illustrated clearly in Fig. 1.2, that displays the total levelized costs of different
generation technologies. One observes 1) the enormous reduction in solar PV costs
in only three years time and 2) the fact that wind energy has already the third lowest
cost, after the modern gas turbines, whose costs have mainly dropped because of
the shale-gas revolution. Looking at these figures, one can state that it is possible
or even likely that RES will soon simply become the cheapest way of generating
electricity.
Because of the environmental, economic and geopolitical concerns outlined above,
governments around the world are taking action and ambitious decarbonization targets have been formulated. For example, the EU has a long term goal of 80-95%
reduction of greenhouse gas emission in the power sector by 2050 [4]. The recently
presented energy plan from the Obama administration wants to double the share of
RES by 2020 [5]. In the Netherlands, a recent ‘national energy agreement’ outlines
1.1 The changing energy landscape
3
2013
2010
Solar Thermal
Wind−Offshore
Solar PV
Advanced Coal with CCS
Conventional Gas Combustion Turbine
Advanced Coal
Biomass
Advanced Nuclear
Advanced Gas Combustion Turbine
Conventional Coal
Advanced Gas CC with CCS
Hydro
Geothermal
Wind
Levelized Capital Cost
Fixed O&M
Variable O&M (inc. Fuel)
Transmission Investment
Conventional Gas Combined Cycle
Advanced Gas Combined Cycle
0
50
100
150
200
250
2008USD/MWh
300
350
400
Figure 1.2 – Comparison of generation cost estimates from 2010 (lower bars) and 2013
(upper bars). Numbers denote USA average levelized costs (2008 $/MWh) for plants
entering service in 2016 (for the 2010 numbers) and 2018 (2013). Generation types are
ranked according to the 2013 costs. Data from [3].
4
1 Introduction
300
120
World
EU27
China
USA
World
EU27
China
USA
100
Installed capacity (GW)
Installed capacity (GW)
250
200
150
100
50
80
60
40
20
0
2005
2006
2007
2008 2009
Year
2010
2011
(a) Cumulative installed wind power
2012
0
2005
2006
2007
2008 2009
Year
2010
2011
2012
(b) Cumulative installed PV power
Figure 1.3 – Cumulative installed capacities from 2005 to 2012 of wind power (a) and PV
power (b). Data from [9], [10] and [11].
strategies for a fully sustainable energy system in 2050.2
In some countries, RES policies have already led to a sharp increase in the
installed capacities of clean generation technologies. Fig. 1.3 shows the installed
capacities of wind and PV power worldwide and in some key regions. The installed
wind generation capacity in 2012 was more than 5 times the one in 2005; for solar the
installed capacity in 2012 was more than 10 times the capacity in 2005. While Europe
contributed to most of the observed growth in the earlier years of the 2005-2012
period, other countries are catching up rapidly. Most projections show continuing
strong growth of both wind and solar, see e.g. [7], [8] and [9].
The fast and inevitably growing shares of RES have a profound effect on the functioning of power systems. Traditionally, the stable and secure operation of power
systems relies on forecasting electricity demand and scheduling the necessary power
generation in the most economic way, taking into account appropriate reliability
margins and technical constraints. The typical characteristics of wind and solar
power introduce a number of complexities to this model. The chaotic and intermittent nature of atmospheric processes is the main source of these complexities:
not only is the output of wind and solar power plants very variable by nature, it is
also hard to predict. These two characteristics, variability and uncertainty, pose a
number of challenges to the planning and operation of power systems, see e.g. [12].
In this report, among many others, it is argued that flexibility 3 is key in dealing
with the variability and uncertainty of wind and solar generation. Four sources of
flexibility are identified: flexible generation, storage, interconnection and demand
response. For the latter to play a serious role, a large source of flexible electricity
demand is required, but today this source is virtually non-existent, since electricity
demand has proved to be almost completely inelastic. This premise may well change
2 The
short-term targets are, however, not nearly as ambitious: 16% renewable energy by 2023,
which is a lower target for a later moment than previously stated goals [6].
3 In [12] flexibility is defined to ”express the extent to which a power system can modify electricity
production or consumption in response to variability, expected or otherwise”
1.1 The changing energy landscape
5
in the coming years.
1.1.2
The advent of electric vehicles
Roughly the same concerns that push RES can also be considered to drive the
introduction of electric vehicles (EVs)4 : rising oil prices, a large dependency on a
small number of oil producing countries and greenhouse gas emissions caused by road
transport. Reductions of tail-pipe emissions form another important advantage of
EVs because they can significantly reduce local air pollution problems.
Governments worldwide are acknowledging the potential of EVs and are therefore formulating ambitious EV penetration targets [13]. Various measures to promote their introduction are proposed, some of which are already being implemented
in various countries. They include tax benefits, research programs, but also initial
investments in charging infrastructure. Two milestones that are mentioned are a
worldwide 50% market share in 2050 and at least 5 million EVs and PHEVs sold
per year as of 2020. In [13] it is also argued that the path to a large-scale EV
introduction is not without obstacles. Most notably, improved battery performance and reduced battery costs are necessary for EVs to successfully compete with
conventional vehicles and achieve the large market shares that are so ambitiously formulated. Estimates on the pace of the introduction of EVs and total market volumes
therefore vary markedly and are subject to many uncertainties regarding raw material reserves, oil prices, stimulating policy instruments, technological breakthroughs,
etc.
As an illustration, Fig. 1.4 shows the EV penetration scenario as envisioned by
the Dutch government in 2009 [14]. In this scenario, the market eventually saturates
at 75% of all passenger vehicles. Although the significance of such predictions in the
early stages is questionable, it is interesting to remark that in the first few years the
actual observed number of registered EVs is higher than this government forecast
from 2009. As of July 2013 there are already more than 10.000 electric vehicles on
the road in the Netherlands [15].
1.1.3
The potential synergy between electric vehicles and renewable energy sources
The two trends described above, the large scale adoption of RES and EVs, have some
interesting potential synergies. The key to this potential lies in the potential flexibility in the charging process of EVs, i.e. to vary charging power and/or postpone
charging. For example, for a typical EV and average driving behavior, an EV owner
needs to re-charge his car every four days. This flexibility can play a crucial role
in power systems with a large RES penetration, since it could replace the flexibility
that is normally provided by conventional generation units but that most RES are
only able to provide to a limited extent. Instead, flexible electricity demand can be
adjusted according to the availability of RES output. The contours of this paradigm
shift begin to appear: from a fixed demand that is met by controlling the generation
4 In this thesis we will not differentiate between plug-in hybrid electric vehicles (PHEVs) and
full electric vehicles and we will denote all with the acronym EV
6
1 Introduction
0.8
EV share of passenger cars (%)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2010
2015
2020
2025
Year
2030
2035
2040
Figure 1.4 – EV penetration scenario forecasted by Dutch government [14].
side, towards a controllable demand side that follows the fixed but time varying
generation. Although a complete reversal of the traditional paradigm is not likely,
i.e. flexible generators will always be needed to some extent, responsive demand is
expected to play a major role in high RES power systems [16].
1.1.4
Changing roles in future power systems
The more active role of electricity demand could well lead to a re-definition of traditional rules and roles in power systems. With the introduction of distributed
generation and demand response, the demand side of the sector will become more
actively involved. For instance, new types of services based on the flexibility in demand and/or distributed generation could emerge. Any of these services will need
to be accommodated by the electricity networks, and some services might even be
specifically aimed at the network. The new market models can only be created
in a meaningful way if the techno-economic characteristics of the new paradigm are
thoroughly understood. This thesis aims to contribute to this understanding by analyzing the potential of flexible EV demand in the multi-actor context of liberalized
power systems with high shares of RES.
1.2
1.2.1
This thesis
Problem description
In liberalized power systems, different tasks regarding the planning and operation of
the power system concern different actors. The flexibility of EV charging therefore
also contains a value for a variety of actors. Distribution system operators (DSOs),
for example, have an interest in controlling the EV charging process in such a way
that sharp peaks in network load are avoided, since they could require reinforcements
1.2 This thesis
7
of networks. On the other hand, retailers who buy electricity on wholesale markets
could benefit from lower off-peak wholesale prices if they can postpone EV charging
to low price periods. Yet another perspective is if EV charging power is being
adjusted with regard to the variable output of renewable energy sources.
1.2.2
Research objectives
The question thus arises how EV charging flexibility can add the most value, and,
consequently, how this flexibility can best be ‘shared’ among different actors. The
research objective of this thesis can be therefore be formulated as gaining a better
understanding of the potential value of controlled EV charging in liberalized multiactor power systems with high shares of renewable energy.
This objective motivates the following research question:
How can the flexibility of EV charging best be utilized in multi-actor power systems
with high shares of renewable energy sources?
In order to answer the main research question, a number of subquestions have been
formulated.
1. How can the controlled charging of EVs reduce their impacts on the distribution grid?
2. How can controlled EV charging reduce generation costs in power systems with
a high share of renewable energy sources?
3. How can the costs of EV charging be minimized within distribution grid constraints?
1.2.3
Thesis outline, structure, research methods and scope
This thesis addresses the questions formulated above in the following structure:
chapter 2 provides the necessary background knowledge of the system under consideration. It treats the relevant technical and economic aspects of power systems
and describes relevant characteristics of EV charging. In chapter 3 we present a
literature analysis to identify knowledge gaps and position the work described in
this thesis relative to the literature.
Chapter 4 treats the first sub-question listed above. It first assesses the impacts
on the distribution grid caused by EV charging and then analyzes the potential
cost savings of controlled EV charging due to lower network investments and energy
losses. The networks analyzed in this chapter cover a large part of the complete
Dutch distribution network, and the time-horizon extends to 2040.
Chapter 5, which deals with the second sub-question, looks at the flexibility of EV
charging from the perspective of electricity generation in system with a high share
of RES. By extending a unit-commitment model with EV charging as optimization
variable, the flexibility of EV demand with respect to variable RES output is analyzed in combination with cross-border transmission capacity. The system analyzed
in this model is based on projections for the German power system in 2025.
8
1 Introduction
Ch3
Ch1
Introduction
Theory, Ch2
Systems,
Data
Literature
Ch4
Modeling
chapters
Ch5
EVs &
networks
EVs &
generation
Ch6
EVs &
networks +
generation
Ch7
Refined view
Ch8
Conclusions
Figure 1.5 – Schematic representation of the thesis structure and chapter numbers.
The final sub-question is treated in chapter 6. Here, the effect of EV charging
based on wholesale electricity prices on the distribution network is analyzed for
an example distribution network. Moreover, possible mechanisms to prevent EV
charging from overloading the networks are discussed.
Chapter 7 treats a number of additional aspects of EV charging and aims to
connect the different viewpoints of the earlier chapters. Most notably, we investigate
the differences between a centralized and a decentralized approach to EV charging.
Furthermore, a sensitivity analysis on various assumptions made in earlier chapters,
as well as an analysis of the effect of cost-minimizing EVs on the same set of networks
that was used in chapter 4 are presented. The thesis ends with conclusions, reflection
and recommendations.
The thesis structure is summarized schematically in Fig. 1.5. Chapters 4, 5 and
6 have been published or submitted as journal papers and we have chosen to include
them integrally in this thesis. As a consequence, these chapters themselves start
with introductory texts which will inevitably contain some repetitions compared
with earlier chapters.
Different research methods and data have been used in the work presented in
this thesis and they will be explained in more detail in subsequent chapters. In
short, we have mainly used mathematical optimization models combined with EV
data that has been derived from current driving patterns from conventional vehicles
users. We assume rational, cost-minimizing entities and, throughout the thesis, we
model all optimization problems as deterministic, so we do not take various types
of uncertainties into account. Some further limitations of scope are the following:
• We do not consider EV charging optimization for the provision of balancing
services. Although this clearly is a promising venue for flexible EV demand,
we choose not to include it in this thesis for a number of reasons. One of the
important reasons are that there already exists a large body of scientific liter-
1.2 This thesis
9
ature on this topic, which is discussed in further detail in chapter 3. Secondly,
a meaningful analysis of the potential value of EVs with respect to balancing
services requires stochastic optimization methods combined with realistic data
of forecast uncertainties, which were not readily available.
• The penetration rate of EVs and/or RES is taken as given. We do not consider
strategies to promote the adoption of EVs, renewable energy policies, etc.
• The IT infrastructure needed to control EVs, communicate price signals, etc,
is largely out of scope of this thesis. In chapter 6 we briefly comment on the
IT requirements of different congestion management schemes for distribution
grids. Issues like robustness, safety and topology of EV related IT infrastructures are not treated.
• Consumer behavior with respect to EV charging is based solely on driving
patterns. We do not focus on what incentives are most effective for consumers
to provide demand response services.
• Vehicle-to-grid, or V2G, where EVs can feed electricity into the grid is not
considered in the main text of this thesis, except for appendix A.
• In the optimization formulations of chapters 5, 6 and 7 we look at minimizing
short term variable costs, so investment in new assets as decision variables,
either in generation or network capacity, are not considered.
The work described in this thesis was performed at the section Energy and Industry of the department of Technology, Policy and Management at Delft University
of Technology. Its signature can be recognized in this work, since multi-actor infrastructure systems lie at the very heart of this group. Parts of this research have
been performed in close cooperation with Dutch DSO Enexis.
10
1 Introduction
Chapter 2
Electric vehicles in future
power systems
In this chapter we aim to provide some elementary background that is considered
helpful for a better understanding of the remainder of this thesis. To this end, we
start with a brief review of the technical and non-technical aspects of today’s power
systems1 and we will especially emphasize ongoing trends and expected changes.
Then we focus on the role of EVs by discussing relevant actors concerned with EV
charging, presenting a model for EV charging based on a dataset of current driving
patterns, and showing how various EV charging strategies can be formulated as
mathematical optimization problems. This will constitute the modelling framework
used throughout this dissertation.
Regarding the first section of this chapter that gives some background on power
systems, we note that this is only a very concise overview with a limited scope fitting
the issues treated in this thesis. As a consequence, many important aspects are not
discussed here and we refer the reader to a number of textbooks on technical and
non-technical details of power systems, such as [16], [17], [18], [19], [20] and [21].
2.1
Power systems
2.1.1
Technical aspects
Traditional functioning The main technical functions of a power system are
generation, transmission, distribution and consumption. Current power systems are
characterized by a hierarchical top-down structure, as depicted schematically in Fig.
2.1.
Power is generated mostly by large power plants, where it is immediately transformed to higher voltages and fed into the transmission network. The main function
of the transmission network, usually a meshed network to ensure redundancy, is
1 When
we refer to power system we actually mean the electric power system.
11
12
2 Electric vehicles in future power systems
~
~
Interconnector
~
~400kV
~
Large consumer
~200kV
Transmission
~
Distribution
~10kV
~0.4kV
Figure 2.1 – Schematic view of the structure of current power systems.
to transport power at a high voltage from the generation sites towards load centers. Typically, voltages range between 150 kV and 400 kV, but in some countries
higher voltages are common as well. Interconnectors - transmission lines to neighboring countries and/or power systems - are found both in alternating current (AC)
and direct current (DC) form, the latter being used mainly for longer distances or
to connect non-synchronous regions. The boundary between the transmission grid
and the distribution grid lies at the high voltage (HV) substations, where voltage
is transformed down to lower levels. From the HV substations a number of MVtransmission (MV-T) cables (typically 10 kV or 20 kV) transport power further to
MV substations, where a number of MV distribution (MV-D) cables are fed. MV-D
cables are often laid out in a ring structure, with a net opening that is open under
normal operation. In case of a fault on the MV-D cable, the net opening closes automatically such that no interruption of supply is experienced by the loads connected
to the MV-D cable. Connected to the MV-D cables are MV/LV transformers, that
typically serve 50-100 households through a number of LV feeder cables. Household
electricity consumption hence takes place at the lowest voltage level, but medium
sized and large industrial customers can be connected to higher voltage levels, up to
the high voltage transmission grid for very large consumers. Furthermore, a limited
amount2 of distributed generation is connected to the distribution grid, either at medium voltage (e.g. medium sized wind turbines, combined heat and power (CHP)
installations) or at low voltage (PV panels, micro CHP). It should be noted that
many variations of the typical topography described above exists between countries,
2 This
is true for most countries. Germany with its 35 GW of solar capacity forms an exception.
2.1 Power systems
13
~
~
~
Interconnector
~400kV
~
Large consumer
~200kV
Transmission
~
~
Distribution
~10kV
Flexible
demand
~0.4kV
Figure 2.2 – Schematic view of the structure of future power systems with increased RES
penetration, interconnection capacity and flexible demand.
and even within countries and regions. For example, on the MV-level one finds many
combinations of ring, meshed and radial configurations.
Trends towards future power systems A number of changes that are taking
place in power systems across the globe can be identified, depicted schematically in
Fig. 2.2. Presumably the most profound one is the sharp increase in RES, as Fig.
1.3 clearly demonstrates. The two main renewable generation technologies are wind
power and, more recently, solar PV power. While the former is mostly embedded
at medium, and increasingly so, at high voltage levels, the latter is predominantly
connected to the grid at LV level. The variable, unpredictable and non-dispatchable
nature of RES makes them hard to integrate. Additional electro-technical complexities such as a lower system inertia provided by rotating mass and voltage stability
play a role as well.
Partly as a reaction on and anticipating growing RES penetration levels, a number of other changes are taking place. One observes an increased level of interconnection between countries, both AC and DC, and ENTSO-E scenarios [8] show
that this trend is likely to continue. Interconnectors, traditionally used mostly for
reliability reasons, now increasingly facilitate the coupling of electricity markets. In
the light of high RES levels this becomes particularly interesting due to the geographic smoothing effect: the RES output over a larger geographic area shows a
more constant production profile. Furthermore, pumped hydro resources in neighboring countries can act as buffers for RES production - the case of Denmark and
14
2 Electric vehicles in future power systems
20
22
Winter
Summer
18
18
16
16
14
Demand (GW)
Demand (GW)
20
14
12
10
8
12
10
8
6
6
4
4
2
2
0
0
1
2
3
4
Time (days)
5
6
7
0
0
(a) Load profiles
2000
4000
Hours
6000
8000
(b) Load duration curve
Figure 2.3 – Demand profiles in a summer and winter week (a) and load duration curve (b)
in the Netherlands in 2012.
Norway is exemplary. As RES levels are increasing and power systems are merging,
an important role is also foreseen for FACTS technology (Flexible AC Transmission System). FACTS provide a means to, to some extent, control flows in the
transmission grid, thereby enabling a more efficient operation of the grid.
On the demand side, the electrification of transport and domestic heating are
leading to the introduction of new loads that have a more flexible character than
traditional household loads: electric vehicles and heat pumps. In addition to this,
micro CHP creates another source of flexibility embedded at household level. Together with developments in IT that enable an infrastructure for communication and
control, the new flexible electricity demand can become an important part of the
electricity system, that partially takes over functions that are traditionally supplied
by the large conventional generation units.
2.1.2
Load and generation profiles
System load, renewable energy and residual demand Electricity demand
typically varies with time depending on the season, day of the week and hour of
the day, but also on prevailing weather conditions. In moderate climates the system
peak usually occurs on a winter evening, while in warmer climates the use of air conditioning causes demand peaks in warm summer afternoons. Fig. 2.3(a) shows the
system demand in the Netherlands in a typical winter week and a typical summer
week. Information on the yearly load profile is often expressed in a load duration
curve, that basically ranks the hourly demand in descending order and can be interpreted as a probability distribution. Fig. 2.3(b) shows the load duration curve for
the Netherlands in 2012. The minimum load lies around 8 GW whereas the maximum is a little under 18 GW. The load duration curve also shows at a glance how
many hours the system demand is larger than a certain value. This information is
important since it determines what mix of power plants can meet the time varying
2.1 Power systems
15
20
22
18
Demand
Wind Power
Residual Demand
16
Power (GW)
0 GW Wind
5 GW Wind
10 GW Wind
20 GW Wind
15
Residual demand (GW)
20
14
12
10
8
6
10
5
0
4
2
0
0
1
2
3
4
Time (days)
5
(a) Residual load profile
6
7
−5
0
2000
4000
Hours
6000
8000
(b) Residual load duration curve
Figure 2.4 – Modeled residual demand profile in a winter week (a) and modeled residual
load duration curves for different wind capacities (b) in the Netherlands in 2012. A negative
residual demand denotes a surplus of wind generation.
yearly demand in the most economic way. Coal plants, for example, have relatively
high investment cost and low variable costs and are cheaper than gas plants when
they can run for more than, say, 6000 hours per year.
A concept often used in relation with RES is the residual demand, which is
defined as the demand minus RES production. This is the demand curve that has
to be met by dispatchable generators. As the share of RES grows, it has a large
impact on the residual demand curve. Fig. 2.4(a) shows the same winter week as Fig.
2.3(a), but this time also wind production scaled to 10GW installed wind capacity
and the resulting residual demand have been plotted3 . It can be seen that since wind
power is variable, and in principle uncorrelated with demand, a residual load curve
emerges that still shows some daily pattern, but also the randomness induced by the
variable wind power. Naturally, the higher the installed capacity of wind power, the
stronger this effect is. For solar power a similar effect can be expected, although,
due to the daily cycle of the sun, solar power is more correlated with demand than
wind power.
The residual load duration curves shown in Fig. 2.4(b) demonstrate the effect of
more RES in another way. One notes that especially the amount of base-load hours
(in the right of the figure) decreases quickly when more wind power is installed.
The peak demand does hardly decrease however. This is due to the fact that, since
wind power and demand are uncorrelated, there will always be some hours with high
loads and low wind. By looking at Fig. 2.4(b) one is able to understand the potential
value of demand response and interconnection capacity. The former effectively shifts
demand from low residual demand periods to high residual demand periods, which
would result in a flattening of the residual load duration curve. Interconnection, on
3 Since aggregated wind power time series for the Netherlands are not available, the wind power
time series are based on data from the western Danish system that has similar wind characteristics
as the Netherlands.
With RES
Hours
(a) Residual load curves
with and without RES.
Residual Load
Without RES
Residual Load
2 Electric vehicles in future power systems
Residual Load
16
Hours
(b) Effect of demand response on residual load
duration curve.
Hours
(c) Effect of interconnection on residual load
duration curve.
Figure 2.5 – Residual load duration curves and the (exaggerated) effect of demand response
(b) and a smoother RES profile due to more interconnection (c). The dashed line in (b)
and (c) denotes the residual load duration curve with RES from (a)
the other hand, leads to a more constant aggregated RES production profile, so that
the residual load duration curves looks more like a shifted version of the original load
duration curve depicted in Fig. 2.3(b). Fig. 2.5 shows these effects in a schematic
and exaggerated way.
Load profiles in distribution networks and energy losses The total system
load profile that was shown in Fig. 2.3 represents the sum of all electricity demand
in the Netherlands. On lower levels of the network, a differently shaped profile is
observed, due to the fact that the large consumers with flatter load profiles are
connected at the higher voltage levels. The exact combination of loads connected
to a certain distribution assets determines the load profile (i.e. the power flow as a
function of time) on that asset. Typical load profiles for a winter week and a summer
week on a LV/MV transformer with approximately 250 households connected to it
are depicted in Fig. 2.6(a). The yearly load duration curve is shown in 2.6(b).
Compared with the national system load profile depicted in Fig. 2.3 the household
load profile shows much more variation between peak demand and low demand.
Due to the random nature of loads, the combined peak of many loads Pi (t) is
usually much smaller than the sum of the individual peaks. The ratio of these two
is defined as simultaneity factor g (sometimes its reciprocal diversity factor is used):
max
g = ∑N
∑N
i=1
i=1
Pi
max Pi
(2.1)
The value of g depends on the network level and is usually smallest at the lower
levels. Working formulas exist for relations between consumed yearly energy of loads
(this is typically what is measured by DSOs) and the expected combined peak of
those loads. Such formulas are being used by network planners if new networks
have to be constructed. If, for example, a new residential area is being built, one
estimates expected electrical energy use based on the type of housing and dimensions
the networks to be able to supply expected peak loads. Naturally, higher capacity
assets are more expensive.
2.1 Power systems
17
250
250
Winter
Summer
200
Load (kW)
Load (kW)
200
150
100
50
0
150
100
50
1
2
3
4
Time (days)
(a) Load profile
5
6
7
0
0
2000
4000
Hours
6000
8000
(b) Load duration curve
Figure 2.6 – Standard household load profile of 250 households in a summer and winter
week (a) and the load duration curve (b). Data from [22].
Next to the capital costs of assets, a second large cost associated with the distribution networks are energy losses. Energy losses in a conductor, say a line l are
given by Ohm’s law:
Ploss,l (t) = Il2 (t)Rl
(2.2)
where Il is the current in the conductor and Rl its resistance. If ones assumes a
constant voltage, the energy losses scale with Pl2 (t) where Pl (t) is the instantaneous
power in the line. The ratio between energy and peak load is called service time and
is a measure for the ‘flatness’ of the load profile. A flat load profile (higher service
time) hence leads to lower energy losses. Furthermore, the service time is a useful
measure to estimate yearly energy losses based on only a measurement of the yearly
peak load.
2.1.3
Non-technical aspects: organizational, economical, and
regulatory
Liberalization and unbundling: a multi-actor system Many power systems
around the world have undergone a transition from being centrally operated to
allowing for competition in the generation and retail of electricity. This process,
often referred to as restructuring or liberalization, has also led to the unbundling
of generation and transport of electricity in many countries. In addition, transport
of electricity is often divided in a transmission and a distribution network, where
transmission networks are operated by one or a few transmission system operators
(TSOs) and distribution grids by distribution system operators (DSOs). A number
of textbooks describe liberalized (also referred to as restructured) power systems in
much more detail, see e.g. [16, 18, 19, 20].
The resulting structure of the electricity sector can hence be described as a
multi-actor system with different actors operating in different technical areas of the
system. Naturally, different actors also have different objectives that are specific
18
2 Electric vehicles in future power systems
Generator
Markets
Supplier
Consumer
DSO
TSO/ISO
System operation &
Network Management
(a) Actor overview current situation
Producer
Generator
Markets
Supplier
Aggregator
Consumer
EV owner
DSO
TSO/ISO
System operation &
Network Management
(b) Possible actor overview future situation
Figure 2.7 – Schematic representation of actors in the current electricity sector (a) and a
possible representation of the future situation with small producers and flexible demand
(b). Arrows denote contractual relationships and flows of information. Figures based on
[23].
2.1 Power systems
19
System Price
Figure 2.8 – Schematic representation of electricity price setting by the marginal unit. M C
denote marginal costs of generation units and Q the demand of electricity.
for their tasks. Fig. 2.7(a) provides a schematic overview of the actors operating in
todays unbundled power systems.
With the changes described in the previous section - the advent of multi-national
markets, more RES and increased volumes of distributed generation and flexible demand - the traditional roles in the electricity sector are changing, too. As an example,
one could think of the market for balancing services. Currently, generators offer bids
for balancing power and are dispatched according to the system imbalance, and, to
some extent, large consumers that have a suitable consumption profile might offer
balancing power in the form of interruptible load contracts. In the future, small
consumers with flexible demand and/or distributed generation, possibly represented
by some aggregating entity, might begin offering those services as well. Also, functions related to the networks like congestion management on the transmission grid,
or energy balancing on the distribution network level (currently not taking place),
could be expected to be fulfilled by demand response and distributed generation in a
cost-effective way. In a nutshell, one might say that if the controllable conventional
generation will be replaced by less flexible RES and, simultaneously, more flexibility
on the demand side, it is a logical consequence that a part of the services provided
traditionally by the conventional power plants will now be transferred to the demand
side. Fig. 2.7(b) provides a possible schematic overview of the actors in such a future
power system.
Supply and demand, electricity price, economic dispatch One of the fundamental results of the theoretical underpinning of power systems restructuring is
that centralized dispatch of electricity generation leads to the same outcome as in
a perfectly operating electricity market, which is that all generating units in the
system will increase their output until the point their marginal cost is equal to the
system marginal cost which is sometimes called ‘system lambda’. The derivation of
this result can be found in standard textbooks, e.g. [17] and [20]. The implication
of this result is that the optimal electricity price will reflect the marginal cost of
the marginal unit in the system, i.e. the most expensive (in terms of marginal cost)
20
2 Electric vehicles in future power systems
Demand response
Wind Power
(a) Merit order wind power
Wind Power
(b) Merit order with wind power and demand
response
Figure 2.9 – Schematic view of how electricity price changes due to wind power and demand
response. In the left figure the dashed line represents the original merit order without wind,
in the right the second dashed line represents the original demand curve without demand
response.
power plant that is needed to meet the demand. In a market this price is the result
of supply bids by the generators - economic theory suggests that the optimal bids
are exactly at marginal cost. Fig. 2.8 shows schematically how the electricity price
emerges from demand and supply bids and reflects the marginal cost of the marginal
generator.
In reality, next to the marginal cost of generators, there is a number of other
factors that are of importance for how electricity prices emerge. Most notably the
so-called inter-temporal constraints play a role. These express, loosely speaking,
constraints on generators’ output between different time-steps. In other words, if a
generator is now producing at a certain operating point, this restricts the possible
operating points in future time-steps. Examples of such constraints are that units
have a start-up times (and/or costs) and ramping limits, i.e. they cannot adjust
their output infinitely fast. In a market environment, a large variety of other issues
plays a role.
With the ongoing changes in the electricity sector, mostly the much higher RES
penetration, electricity price formation will be influenced, too. In principle, RES
have a marginal cost of zero, or, when taking non-fuel related variable operational
and maintenance cost into account, very close to zero. If renewable energy is, like
conventionally generated electricity, traded on the spot market, the merit order will
hence start with a large amount of zero bids, see Fig. 2.9(a). In some countries, as
a means to promote their use, RES are paid directly through subsidies and are not
traded on the wholesale market. Referring to the merit order, the effect is the same,
since RES production can now be subtracted from demand and only the resulting
residual demand has to be met by the conventional generators.
RES output varies with time depending on weather conditions, but, if it exceeds
2.1 Power systems
21
demand, then in principle a market clearing price of zero will be the result. However, due to the inter-temporal constraints described above and a number of other
technical and non-technical complexities, prices can even become negative - this is
a phenomenon that is already observed in systems with high amounts of renewables
like Denmark and Germany. Box 1 describes the cases of negative wholesale prices
in Denmark and Germany in more detail.
A high amount of renewables will thus create moments with very low, or negative
electricity prices. Increased interconnection and flexible demand will, on the other
hand, have a damping effect on electricity prices. Depending on the exact nature
and physical characteristics of the flexible demand, it could, for example, be possible
to shift a portion of the demand by several days. This would allow to anticipate on
low prices caused by high RES output, and to schedule demand in those periods.
Referring to Fig. 2.9(b), this would mean that the demand bids would be shifted to
the right in case of a large amount of zero bids. The effect of the zero bids would
thus be partially offset by the price responsive demand. Market coupling enabled
by interconnectors has a similar, though slightly different damping effect on price.
Box 1 - Negative wholesale prices The figure below shows the demand, wind
energy production and the resulting wholesale electricity price in the Western
Danish System, for a period in spring 2012. One observes how prices briefly
become negative in one of the periods where wind power exceeds demand. It is
also interesting to note that in other periods with negative demand prices did not
go to or below zero, indicating once more that the emergence of electricity price is a
complex phenomenon and cannot only be explained by looking at the merit order
and system demand. Among other things, exchange with neighboring systems,
maintenance scheduling and the already mentioned inter-temporal constraints play
an important role, too.
Power (MW)
6000
Demand
Wind Power
Residual Demand
4000
2000
0
Price (EUR/MWh)
87
88
89
90
Time (days)
91
92
93
88
89
90
Time (days)
91
92
93
40
30
20
10
0
87
Demand, wind power and prices in the western Danish system in a period around April
1st 2012. Data from [24]
22
2 Electric vehicles in future power systems
Box 1 - Continued More recently, Germany coped with an even extremer case
of negative wholesale prices. On Sunday June 16th 2013, a very low system load
coincided with a fairly large output of wind and solar power. The results was that
the coal, lignite and nuclear generation units had to be ramped back to unusually
low levels of approximately 20 GW. This led to a situation with extremely low
wholesale prices of -200 e/MWh for a sustained period of time.
Generation (GW)
60
Conventional
Wind
Solar
50
40
30
20
10
0
2
4
6
8
10
12
14
16
18
20
22
24
2
4
6
8
10 12 14
time (hours)
16
18
20
22
24
price (EUR/MWh)
50
0
−50
−100
−150
−200
−250
Conventional, wind and solar power and spot prices for the German system on June 16th
2013. Data from [25]
Distribution networks regulation Unlike the production and retail of electricity, transport and distribution are not subject to open competition in unbundled
power systems. DSOs are responsible for the distribution networks and, since they
form a natural monopoly in their service area, their activities are regulated. Some
variations in the exact form of regulation are possible, but generally this means that
the competition authority determines the (change in) tariffs the DSO is allowed to
charge its customers for a certain period of time, referred to as the regulatory period.
For instance, in the Netherlands this is 4 years. In the regulatory framework in the
Netherlands and many other countries the change in tariffs is determined by the
following formula:
T It,i = (1 + cpit − xi + qi )T It−1,i
(2.3)
where the T It,i is the allowed total income of DSO i in regulatory period t, cpi is
the consumer price index (a measure of inflation), x is the efficiency factor and q
is the quality factor. The x-factor basically determines how much more efficient a
DSO need to be in the next regulatory period. It is determined in a complicated
way, accounting e.g. for differences between the service areas of the different DSOs,
2.2 Electric vehicles
23
but the guiding idea is a benchmark that is set by the DSO with the lowest cost
levels.
DSOs are in principle profit maximizing entities, but their income is determined
by the regulator, so a profit maximizing objective implies that they will generally aim
to minimize costs. Costs of a DSO are largely determined by CAPEX and OPEX on
the assets, of which the former is dominant. Energy losses form an important part of
OPEX related to assets. For instance, the 2012 annual report of Enexis, one of the
three largest DSOs in the Netherlands, covering roughly 33% of the Netherlands,
lists investments in assets as 247 MA
C and costs of energy losses at 90 MA
C [26].
Reducing investments in new or replacements of old assets and costs of energy losses
are thus a vital tasks for DSOs in a regulated environment.
2.2
Electric vehicles
The following sections describe the role of EVs in future power systems. We present
a model EV charging and the mobility data that is the input for these models. In addition, we describe how EV charging can be viewed as a mathematical optimization
problem and we show how the optimization objectives depend on the perspectives
of the actors involved in it. We start the chapter by an analysis that explores the
various actors involved in EV integration and different institutional arrangements.
2.2.1
Actor analysis
Fig. 2.7(b) shows a schematic overview of an unbundled electricity sector in which
a number of interacting actors can be distinguished. Below we briefly describe the
actors that will play a role around EV charging. After that we describe three possible
configurations in which EV charging could take place, each with different interactions
between the actors. This analysis is partially based on [27] that gives a thorough and
comprehensive description of different actors involved in EV charging and possible
institutional/contractual arrangements. For a more elaborate treatment of this topic
we thus refer to [27]. Next to the actors specifically involved in EV charging we
also summarize the role of the other power system actors discussed in the previous
sections.
• Generator (also referred to as electricity producer). Produces and sells electricity. Electricity sales can be either through various markets (day-ahead, intraday, balancing), or in bilateral contracts with suppliers or large consumers.
Producers are balance responsible parties (BRPs), which means they have the
obligation to comply with a production profile submitted usually 24 hours in
advance.
• Supplier (retailer). Intermediate party selling electricity to end-consumers.
Suppliers can own generation capacity, or buys electricity from generators
through spot market or bilateral contracts. Suppliers are BRPs, too.
• DSO. Owns and maintains distribution networks. Regulated entity and legally
unbundled from generation and retail of electricity. Consumers pay a grid tariff
24
2 Electric vehicles in future power systems
to DSOs included in their supplier tariff, who pays the DSO. Large consumers
sometimes pay directly to DSO.
• TSO/ISO. Transmission system operator or independent system operator.
Manages high-voltage transmission networks and is responsible for various
other system functions. Maintains system balance, manages congestion, organizes different markets (day-ahead, intraday, balancing).
• Consumer (final customer). User of electricity. Connected mostly to LV distribution network, but large customers (industrial, commercial) can be connected
at higher voltage levels. Has a contract with a supplier, and pays through supplier to DSO for grid access and taxes.
• EV owner. Owns (or possibly rents/uses) and charges an EV. Might have a
contract with supplier or possibly an EV aggregator. Might have possibility
to charge at home, or else at public or private charging stations.
• EV aggregator. Intermediate party between EV owners and other actors, most
notably suppliers and/or other market parties and DSOs. Manages charging
of a fleet of EVs and benefits from economies of scale through aggregation.
Does not necessarily own or operate physical charging infrastructure.
• Charge point manager. Owns and operates physical charging infrastructure.
Could have possibilities for smart charging strategies. Pays DSO and either
suppliers or directly to the market for network capacity and energy.
Possible EV charging arrangements
Below we discuss a few possible configurations of EV charging. We loosely follow [27]
who classify the different arrangements on three aspects: the location of the charging
point, the intermediate actor between EV owner and other actors/functions in the
system (e.g. DSOs, suppliers, ‘the market’), and the level op sophistication in the
management of the charging process (uncontrolled, controlled). It is emphasized
that more possible configurations are possible than discussed here. Furthermore,
the different arrangements are likely to co-exist and even a single EV owner might
charge in several different configurations.
Uncontrolled home charging Due to the lack of a large-scale EV charging infrastructure, many EVs are currently being charged at the EV owners home. In this
scheme, depicted schematically in Fig. 2.10, the EV can be seen as simply another
domestic electrical appliance without a special connection or meter and its energy
is paid for through the regular electricity bill. One could say that the intermediate
actor in this scheme is the same one as for the consumer: the electricity supplier.
Since in most countries there are no advanced time of use tariffs, it is likely that
most EV owners will start charging their car upon arrival at home. This scheme is
therefore labeled uncontrolled charging, as opposed to the more advanced schemes
where charging is controlled or postponed according to some objective.
2.2 Electric vehicles
25
Figure 2.10 – Schematic view of economical and physical layers related to the uncontrolled
charging scheme. Arrows denote exchanges of information and/or money.
Controlled home charging with aggregation A more sophisticated scheme
than uncontrolled home charging is a configuration where charging physically still
takes place at the EV owners home, but with an EV aggregator as intemediate
actor. This scheme is depicted schematically in Fig. 2.11. More advanced metering
infrastructure will likely be required in this configuration. Furthermore, EV owners
without private parking spaces like a garage or driveway could use a public space
charging point near their house. The aggregator benefits from economies of scale
and more predictable demand patterns through the process of aggregation. He will
purchase energy from the (various) market(s) and could optimize EV charging in
order to reduce energy costs. Possibly the aggregator still has energy contracts with
the conventional suppliers of electricity. In more advanced schemes the aggregator
can even offer regulating power to balancing markets, or other products based on
EV charging flexibility. Of course, any form of controlled charging would need to
be within boundaries indicated by the EV owner. Such boundaries are technical
specifications like the power and battery limits on EV charging, but more importantly the individual preferences in terms of driving needs. Differentiated tariffs with
respect to costs, priority and speed of charging, environmental aspects, etc, could
be possible. Aggregators could have direct contractual relationships with DSOs, or,
alternatively, this can be via the electricity suppliers.
Controlled charging in a charging station Another form of EV charging that
may emerge as the number of EVs grows is charging at a dedicated charging station, shown in Fig. 2.12. Various different settings are possible, ranging from e.g. a
single charging point in an office parking lot to complete charging stations with multiple charging points and possibilities for fast charging. The latter would resemble
current day gasoline stations to some extent. The actor that owns and manages
the charging station or single charging points was named the charge point manager
(CPM) in [27]. The CPM could also engage in controlled charging strategies, pos-
26
2 Electric vehicles in future power systems
Figure 2.11 – Schematic view of economical and physical layers related to the controlled
charging with aggregation scheme.
sibly enabled by on-site distributed and/or renewable generation and energy storage
technologies. The CPM has a contractual agreement with the DSO for network
capacity. Such contracts could be based on ToU tariffs, or more advanced dynamic
grid tariffs4 . Depending on the number of charging points in the charging station, an
MV connection together with a MV/LV transformer might be required. In addition
to network related contracts, the CPM also manages energy sales, either through a
supplier or by directly interacting with the market and/or generators. Time-varying
wholesale prices, the intermittent character of installed on-site RES generation and
dynamic grid tariffs could all be incentives for the CPM to charge time-varying charging rates. To the extent forecasting allows it, these could be announced ahead of
time. Drivers can then plan when to enter the charging station. In this way, the
combined demand of many charging stations might still exhibit the elasticicity that
is beneficial for the electricity system.
Simplified arrangements treated in this thesis
The different charging configurations described in the previous section are already a
gross simplification of the complex interplay between different actors that would be
observed in reality. Nevertheless, many of these inter-relations between actors are
considered to be outside the scope of this thesis, and simplified institutional arrangements are therefore assumed. Since different chapters have different perspectives on
EV charging, a number of simplified configurations are assumed, see Fig. 2.13. In
chapter 4, where the impacts of EV charging on distribution networks are assessed,
the DSO is assumed to have control over the charging process. In this sense, only
the interaction between EV owners and the DSO is relevant, as depicted in Fig.
2.13(a). In chapter 5, on the other hand, EV charging is controlled from the perspective of lowering generation costs by including EV charging as decision variable
4 See
chapter 6 for a more elaborate discussion on this topic
2.2 Electric vehicles
27
Figure 2.12 – Schematic view of economical and physical layers related to the controlled
charging in charging station scheme.
in a unit-commitment model. Here, all complexities regarding markets, intermediate
actors and distribution networks are ignored. In this perspective only the interaction between generation units and the EVs is of importance, as shown schematically
in Fig. 2.13(b). Yet another perspective is found in chapter 6. Here it is assumed
that EVs are charged based on wholesale market prices. Furthermore, information
exchange between EVs and the DSO is assumed. In addition, in this chapter the
role of the aggregator is treated more prominently. This simplified arrangement is
schematically sketched in Fig. 2.13(c). Chapter 7 investigates additional aspects of
EV charging, but mostly from the perspective of the arrangement depicted in Fig.
2.13(c).
While the institutional arrangements regarding smart EV charging are not the
main topic of this thesis, and therefore strongly simplifying assumptions have been
made to represent them, this approach does provide useful insights on how such
arrangements could be organized. In this sense, the analyses from this thesis can be
considered as input for more detailed studies on the institutional aspects of EV charging. Eventually, such studies should lead to a meaningful institutional arrangement
where the potential of EVs can fully contribute to the realization of future sustainable energy systems.
2.2.2
Driving data
The following section describes the driving data that has been used to model EV
demand profiles throughout this thesis. A more detailed description can be found
in [28]. The Mobility Research Netherlands gives a large dataset of individual trips
by various transport means. The data is collected by means of a survey of roughly
40.000 people in the Netherlands [29]. The dataset consists of over 130.000 individual
movements (one way trips), from which approximately 40.000 are car movements of
28
2 Electric vehicles in future power systems
(a) Chapter 4
(b) Chapter 5
(c) Chapters 6 and 7
Figure 2.13 – Simplified institutional arrangements assumed in different chapters of this
thesis
2.2 Electric vehicles
29
800
600
400
200
0
20
0
24
40
18
60
12
80
6
0
100 or more
Distance driven (km)
Arrival Time
Figure 2.14 – Joint distribution of home arrival times and daily driving distances. Values
on the z-axis denote the number of occurrences in the dataset of approximately 18.000
individual drivers.
roughly 18.000 individuals. Important variables for EV charging are (for each of
the 18.000 individual cars): daily driving distance, home arrival time and home
departure time.
To get some insights in the driving patterns, Fig. 2.14 shows the distribution of
car trips based on daily driving distance and the time at which the last arrival at
home takes place. From Fig. 2.14 it can be concluded that on average, the majority
of car drivers covers only modest distances. Furthermore, it is noteworthy that the
distribution of the shorter trips shows two distinct peaks, one around noon and one
around 1800h. Apparently, a significant fraction of the people tend to use their
car only during the morning, since we have considered only the last arrival time at
home. For the longer distances, the time of the last arrival at home is mostly in the
early evening or late afternoon. This can intuitively be understood by considering
the daily commuting cycle of driving to work in the morning and arriving back home
in the evening.
2.2.3
EV battery model
In the following section we present a simple model of an EV battery, with the
objective of relating technical battery parameters, the energy needed for driving
and the power demand of the EV as seen from the grid. More detailed models on
EV batteries can e.g. be found in [30], [31] and in [32].
The state-of-charge of a battery is a dimensionless number representing the
30
2 Electric vehicles in future power systems
charge content of a battery and is defined as:
SoC(t) =
Q(t)
Q0
(2.4)
where Q(t) (with units Ah) is the amount of charge at time t and Q0 is the nominal
capacity of the battery in Ah. Differences in SoC due to charging or discharging
within a period from t0 to tf are given by:
1
∆SoC =
Q0
∫
tf
I(t)dt
(2.5)
t0
For various numerical simulations on EV charging it is convenient to use a discretized
version equation 2.5 and to use energy content rather than SoC . If we assume a
constant battery voltage Vbatt and charging/discharging with a constant current,
we can write the following expression for the energy content EEV,k = Qk Vbatt at
time-step k of the EV battery:
EEV,ik+1 = EEV,ik + Pbatt,ik ∆t
(2.6)
where Pbatt,ik = Iik Vbatt,ik denotes the power flow into or out of the battery and
subscript i identifies different EVs. There are, however, various losses associated
with charging or discharging a battery.
Discharging Most notably, the battery capacity Q0 actually depends on the magnitude of the discharge current, and hence on the driving behavior. We are, however,
most interested in the EVs from the point of view of the electricity grid, and not
in the processes taking place while driving. Therefore we define a constant driving
efficiency ηd with units km/kWh. One way to estimate the value of this parameter
is to compare the nameplate battery capacity with the reported range of the vehicle.
In [33] a number of different values for the range of a Nissan Leaf - one of the
standard EV models on the market today - with a battery capacity of 24kWh is
reported. These are the results of different tests under different driving conditions.
We assume a rather conservative range of 120km, which yields a driving efficiency
of ηd,i = 5 km/kWh .
We can now readily relate the discharges dik (battery discharge due to driving)
to driving patterns Lik (number of kilometers driven at time-step k):
dik = ηd,i Lik
(2.7)
The driving patterns Lik are modeled based on the mobility data presented in the
previous subsection.
Charging Charging of EVs is also associated with inefficiencies such as inverter
losses and various losses inside the battery. Although some of these losses depend on
the magnitude of the charging power, we assume a constant charging efficiency ηc .
Often, however, only round trip efficiencies of a charge-discharge cycle√are reported.
In [34], a round trip efficiency of 85%, resulting in a charge efficiency of 0.85 ≈ 0.93,
2.2 Electric vehicles
31
is assumed, although it is also noted that laboratory experiments showed a DC-DC
round trip efficiency of over 95%. Now from the point of view of the grid, the power
drawn by the EV and the power actually flowing into the battery are related by
Pbatt,ik = ηc PEV,ik . The complete equation that relates battery energy content of
EV i to its charging power and driving patterns is thus given by:
EEV,ik+1 = EEV,ik + ηc PEV,ik ∆t − ηd,i Lik
2.2.4
(2.8)
Uncontrolled charging
The dataset of driving patterns described above provides the basis for controlled
charging strategies described later in this thesis, but also for an uncontrolled charging
scenario that has an important function as reference case to compare other scenarios
with. The uncontrolled charging scenario assumes that people will charge their EV
at home, every day after the last arrival of the day. The exact shape of the power
profile drawn by a single EV will differ per vehicle, depending on factors such as the
battery type and the battery management system. Common charging profiles start
with a phase of constant current and end with a phase of constant voltage, see e.g.
[30]. Some vehicles have fast charging capabilities that yield even different power
profiles.
A simple approximation to real observed charging profiles would be to assume
constant power charging throughout the complete charging process. If this constant
power charging is fixed at a certain value, an individual EV charging profile is
completely determined by the amount of energy that needed to be recharged. With
the assumptions stated above (charging at home after last arrival), the profile of
single EV is hence a block starting at home arrival time, with a height equal to the
fixed power and a width such that the area (which denotes energy) is corresponding
to the daily driven distance. Adding many of these blocks yields the profile of a
group of EVs as depicted schematically in Fig. 2.15.
The profiles constructed in the way described above can be used to represent the
demand of a group of EVs of any given size if they are scaled appropriately. This
means that the shape of the profile is assumed to be independent of the number of
EVs, but the magnitude is scaled according to the energy demand of the EVs. For
very low numbers of EVs this approach will not be accurate, since the profiles have
a more spiky shape. The question is for how many EVs one can assume that the
aggregate profile will be a good approximation. Fig. 2.16 shows the simultaneity
factor defined by Eq. 2.1 as a function of the number of EVs. We conclude that
this approach is more or less valid if the number of EVs considered is larger than
approximately 50.
Furthermore, from Fig. 2.16 we can also deduce how the combined peak load of a
group of EVs depends on the charging power of the individual EVs. The higher the
charging power, the shorter the time needed to recharge the battery and, hence, the
lower the probability that charging of different EVs overlaps. For a charging power
of 10 kW, the simultaneity factor approaches a value of 0.1, which implies that the
combined peak of, say, 1000 EVs charging with 10kW is only 1000kW. This number,
which is based on actual driving patterns, is an order of magnitude lower than what
2 Electric vehicles in future power systems
charge rate (kW)
32
N=1
6
12
18
charge rate (kW)
0
24
N=2
6
12
18
charge rate (kW)
0
24
N=18000
0
6
12
Time
18
24
Figure 2.15 – Schematic representation of the construction of the aggregated charge profiles
from individual car energy needs in the uncontrolled, 3kW charging scenario
a calculation that simply assumes all EVs will be charging at the same time would
yield.
2.2.5
Electric vehicle charging as optimization problem
Eq. 2.8 relates the battery state of energy to discharges due to driving. Taking
into account the battery limits on energy and power, one observes that the charging
process inhibits some flexibility - the time dependent charging power PEV,k can
in principle be chosen freely, as long as the energy content allows for making the
planned trips Lk . Due to this flexibility, EV charging can be approached as an
optimization problem, where the challenge is to find a charging schedule PEV,k to
minimize some objective. The constraints of the optimization problem are given by
the battery energy and power limits, in combination with Eq. 2.8 that relates the
optimization variables Pk with the battery state-of-energy. In the following section
we will discuss a number of possible charge strategies that result from different
objectives of different stakeholders in liberalized power systems. For the theoretical
foundations of mathematical optimization a large body of literature exists, see e.g.
[35], [36] and [37]. The optimization formulations discussed below are discussed in
more detail in [38].
Aggregators as price takers An aggregator representing a number of EV
owners has the task to charge the EVs while taking into account the driving patterns of the EV owners. We will consider here only the case where the objective is
2.2 Electric vehicles
33
1.4
1 kW
3 kW
10kW
1.2
Simultaneity Factor (−)
1
0.8
0.6
0.4
0.2
0
0
20
40
60
Number of cars
80
100
Figure 2.16 – Simultaneity factor as a function of the number of cars.
to minimize the charging costs of EVs, although, of course, other strategies could
be envisioned as well. First the aggregators are modeled as price takers, i.e. the
combined demand of the EVs does not influence market prices. The optimization
problem then takes the following form:
min
PEV,ik
Nk N
EV
∑
∑
λk PEV,ik
(2.9)
k=1 i=1
subject to EEVmin ,i ≤ EEV,ik ≤ EEVmax ,i
∀i, k
(2.10)
PEVmin ,i ≤ PEV,ik ≤ PEVmax ,i
∀i, k
(2.11)
with λk the (anticipated) real time electricity price. The battery parameters
EEVmax ,i , EEVmin ,i , PEVmin ,i and PEVmax ,i are given for each EV.
The battery state equation given by Eq. 2.8 is repeated here in slightly different
format:
EEV,ik+1 = EEV,ik + ηc PEV,ik ∆t − dik
(2.12)
This equation can be used to express the constraints 2.10 solely in terms of the
optimization variables PEV,ik . Note that in this formulation a driver only has to
specify his driving schedule (the values of dik ) and never finds his battery empty.
If desired, the minimum battery level EEVmin ,i could be raised by a certain safety
margin, for example to always allow an EV driver to reach a fast-charging station.
If necessary, an additional constraint can be added to require the battery being
full at the end of the considered period, or, alternatively, one could assign a cost to a
not fully charged battery at the final step. In the simulations discussed in the later
chapters we have added an inequality constraint to make sure the battery state of
charge at the end of the optimization horizon, denoted with time T , was at least as
full as at the beginning:
EEV,iT ≥ EEV,i0 ∀i
(2.13)
34
2 Electric vehicles in future power systems
Problem 2.9 with constraints 2.10, 2.11 and 2.13 form a linear programming
problem.
Aggregators influence market prices Now we consider the situation where
an aggregator charges a number of EVs, but the aggregate demand of the EVs is
such that it influences market prices. Market prices are modeled to depend linearly
on the total electricity demand. Demand of electricity PD,k is given by:
PD,k = PD0,k + PEV,k
(2.14)
∑
where PD0,k is the original demand and PEV,k = i PEV,ik is the extra demand
of the EVs. Note that from now on we replace PEV,ik by PEV,k in the optimization formulations, but the optimization is still done with the individual PEV,ik as
variables.
If the market price of electricity is assumed to be some function of demand
λk = λ(PD,k ) then a first order Taylor approximation of the electricity price for a
certain period is given by:
λ(PD ) ≈ λ(P¯D ) + λ′ (P¯D )(PD − P¯D )
(2.15)
where the overbar denotes averaging over the period under consideration. After
substitution of Eq. 2.14 in Eq. 2.15, the expression for λk can be written as:
λk = αk + βPEV,k
(2.16)
αk = λ(P¯D ) + λ′ (P¯D )(PD0,k − P¯D )
β = λ′ (P¯D )
(2.17)
with
(2.18)
The coefficient β represents the sensitivity of electricity price to demand. It can for
example be estimated using statistics of historical data, see e.g. [39]. Here we use
the exponential fit of the merit order that is shown in Fig. 2.17.
With these definitions, the optimization problem takes the following form:
min
PEV,k
Nk
∑
2
αk PEV,k + βPEV,k
(2.19)
k=1
with the same constraints as in 2.10, 2.11 and 2.13 and the battery state equation
2.12. This is a quadratic programming problem. If desired, methods to include nonlinear price dependencies can be implemented. Such methods will generally require
iterations where price parameters are updated after each iteration step, see e.g. [39].
Fig. 2.18 shows an example solution of the optimization problem 2.19 with constraints 2.10, 2.11 and 2.12 for two different EVs, one with a small daily driving
distance, one with a large daily driving distance. The aggregate load of a large fleet
of EVs, approximately half of the total passenger car fleet in the Netherlands, on the
Dutch electricity demand can be seen in Fig. 2.19. As expected, the EV demand is
scheduled mainly during the night hours with low prices. It can be seen that the EV
demand ’fills’ the demand profile in the night, so that a much flatter profile results.
2.2 Electric vehicles
35
120
Based on merit order
Exponential Fit
price(EUR/MWh)
100
80
60
40
20
0
0
2
4
6
8
10
demand(GW)
12
14
16
80
60
40
20
2
3
Battery SOC (kWh)
Charge Speed (kW)
Price (EUR/MWh)
Figure 2.17 – Supply curve in the Netherlands and exponential fit.
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
2.5
3
3.5
4
4.5
Days
5
5.5
6
6.5
7
2
1
0
2
30
20
10
0
2
Figure 2.18 – Example of the quadratic programming formulation of the minimal charging
costs problem where EVs are modeled influencing market prices. From top to bottom:
electricity price, battery power, battery state of charge. Two different driver types are
plotted: the grey line corresponds with a driver with a low daily driving distance, the black
with a large daily distance.
36
2 Electric vehicles in future power systems
1.6
Original Demand
With EVs
Demand(MW)
1.4
1.2
1
0.8
0.6
2
3
4
5
6
7
Time (days)
Figure 2.19 – Original demand and EV demand resulting from the quadratic programming
formulation of the minimal charging costs problem where EVs are modeled influencing
market prices. The EV demand corrseponds with 3.5 million EVs or approximately 50%
of the total passenger car fleet in the Netherlands.
DSO: Minimize peak load and network losses If a distribution network operator were to control charging of EVs, its objective will generally be to minimize
losses and reduce (or at least not increase) peak loads. It can be shown that, under
most circumstances, an as flat as possible load profile will fulfill these objectives [40].
The optimization problem can then be defined as follows:
min
PEV,kl
Nk ∑
Nl
∑
k
Rl (Pl,k + PEV,kl )2
(2.20)
l
where index l is used to denote different lines, PEV,kl is the EV load on line l and Rl
is the resistance of line l. In principle this formulation is also valid for transformers.
The same constraints dictated by individual driving behavior and battery limits as
defined in Eqs. 2.9 to 2.13 apply. With these constraints and objective function
Eq. 2.20, this, too, is a quadratic programming problem. If desired, additional
constraints on network limits can be incorporated.
Fig. 2.20 shows the aggregate network load on a distribution asset together with
the EV demand scheduled on the basis of optimization problem 2.20. Here EV
demand, much like the previous case, is scheduled exactly in the hours with the
lowest network load. However, because network load on a distribution cable has
a different profile than national electricity demand, the resulting EV demand also
differs. This difference in perspective and its possible consequences are the main
subject of chapter 6. Chapter 4 will mainly focus on the impacts of EV charging on
the distribution system and how controlled EV charging can relieve those impacts.
In principle, the EV demand resulting from the formulation given in 2.20 should
be used to minimize the distribution system impacts, but in chapter 4 a slightly
different approach was used, which does, however, lead to a similar EV demand
2.2 Electric vehicles
37
120
Original Network Demand
With EVs Minimizing Losses
100
Demand(kW)
80
60
40
20
0
0
1
2
3
4
5
Time (days)
Figure 2.20 – Network load and EV demand for the quadratic programming formulation of
the minimal energy losses problem.
profile. Chapter 5 deals with the problem of including EV demand scheduling in the
optimization of power plant outputs, the so-called unit commitment problem.
38
2 Electric vehicles in future power systems
Chapter 3
Literature review
This thesis is about the integration of electric vehicles in power systems with a
large amount of renewable energy sources. This topic does, however, cover such
a large range of scientific aspects, that it is hard to sketch a complete picture of
the scientific literature describing this area. We will nevertheless attempt to give
an adequate overview, and identify the knowledge gap that this thesis aims to fill.
Following the lines of this thesis, we will focus on EVs and the distribution grid on
the one hand, and EV impacts related to renewable energy on the other hand. How
this thesis should be viewed in perspective of the scientific literature is described in
more detail at the end of this chapter. Next to this more general literature analysis,
the core chapters 4, 5 and 6 provide in their introductory texts an analysis of the
literature of their specific topics.
3.1
Trends in literature on the role of EVs in smart
grids
The topic of this thesis can, somewhat arbitrarily, be defined as EVs in smart grids,
both from a point of view of the grid and the integration of RES. We are therefore
interested in literature in a field spanned by the following query:
EVs AND Smart Grids AND (RES OR Grids)
However, since, on the one hand, different authors use many different terms and
keywords for the same things, and, one the other hand, ‘smart grid’ is a very broadly
defined notion, we use, after some trial and error, the following query in Scopus1 to
arrive at a list of articles that, in our opinion, contains the most relevant literature:
TITLE-ABS-KEY(
("Electric vehicle" OR EV OR PHEV OR "Electric transportation") AND
("smart grid" OR "demand response" OR "load management"
OR "responsive demand" OR V2G OR "vehicle-to-grid") AND
(("Renewable energy" OR wind OR solar OR "stochastic generation")
1 Although other knowledge databases than Scopus are available, and Scopus does not cover
everything, we consider the query results sufficiently complete as a basis for this literature analysis.
39
40
3 Literature review
120
100
Articles
80
60
40
20
0
1985
1990
1995
2000
Year
2005
2010
Figure 3.1 – Trend in the number of articles concerning of intelligent EV management.
OR ((power OR electricity OR distribution) AND (grid OR system OR network))))
AND (LIMIT-TO(DOCTYPE, "ar"))
We discuss some meta-level aspects of this literature list first, and then we discuss
some of the most important papers individually.
The query result counts 293 articles.2 Fig. 3.1 shows the number of records of
the above query per year until 2012. Clearly, this is a very new topic: the number of
articles from before 2009 is very low. The articles from 1994 and earlier turned out
to be on battery chemistry and were thus not so relevant in the light of this thesis.
Interestingly, the two earliest after that (from 2005) were also the most cited articles
from the list: [41] and [42], see also Table 3.1. One could conclude from this that
these articles really were the pioneering work that paved the way for many other
researchers - including the author of this thesis.
We observe from the other studies mentioned in Table 3.1 that they cover quite
a broad range of topics: from distribution grid impacts to frequency regulation.
Looking more closely at the complete list of 293 papers that form the query result,
one could make an attempt to define sub-fields. In Fig. 3.2 one such attempt of a
division of the completely field in subfields is shown, with some of the key papers
indicated that we will discuss in more detail in the next section. They are encoded
by the the first three letters of the first author and the year of appearance. The
figures shows two axes: the y-axis that ranges from ‘technical detail’ to ‘system view’
and the x-axis that ranges from ‘RES integration’ to ‘EV integration’. There is a
degree of arbitrariness in this division, but in the light of this thesis we consider
it meaningful. Another possible categorization was made in the overview article of
[50]. There, the distinction is made between ‘EV grid models‘, ‘EV impacts’ and
‘EVs and renewables’, which were each further divided in lower categories.
2 When
performing this query in June 2013.
3.1 Trends in literature on the role of EVs in smart grids
41
Table 3.1 – Most cited journal articles concerning intelligent EV management.
First Author, Year, Citations Title
Reference
Kempton2005, [41]
434 Vehicle-to-grid power fundamentals: Calculating capacity and
net revenue
Kempton2005a,
403 Vehicle-to-grid power implementation: From stabilizing the
[42]
grid to supporting large-scale renewable energy
Ipakchi2007, [43]
297 Grid of the future
Clement-Nyns
289 The impact of charging plug-in hybrid electric vehicles on a
2010*, [44]
residential distribution grid
Tomić2007 [45]
219 Using fleets of electric-drive vehicles for grid support
Lund2008 [46]
159 Integration of renewable energy into the transport and electricity sectors through V2G
Guille2009 [47]
152 A conceptual framework for the vehicle-to-grid (V2G) implementation
Han2010 [48]
137 Development of an optimal vehicle-to-grid aggregator for frequency regulation
Peterson2010 [49]
97 Lithium-ion battery cell degradation resulting from realistic
vehicle and vehicle-to-grid utilization
Sortomme2011 [40]
92 Coordinated charging of plug-in hybrid electric vehicles to
minimize distribution system losses
* Did not appear in the Scopus query result, but was added to the list because it is one of the most
cited papers by all other papers in this field.
System view
Institutional
Frameworks
Sov2009
Gom2011
V2G
Valley filling
Gui2009
Gal2010
Kri2011
Lun2009
Pec2010
Pet2010
Had2009
Kem2005
Renewables integration
EV integration
Rot2009
Sab2010
Don2012
Den2006
Mar2012
Ili2010
Fer2012
Ban2013
Sor2011
Balancing
Unit committment
Distribution grid impacts
Sun2012
Gre2011
Cle2010
Pie2011
Pec2010
Tay2010
Technical detail
Figure 3.2 – Schematic overview of the scientific literature around the topic of EV and
renewable integration.
42
3.2
3 Literature review
Discussion of some important papers per subfield
In this section we will discuss some of the most relevant articles from the field
described above. We will do so according to the sub-fields of Fig. 3.2. Furthermore,
we note that not all articles we discuss were found in the list of 293 that resulted
from the Scopus query. In the following discussion we also indicate per paper the
three letter year combination that was used in Fig. 3.2.
Vehicle to grid The work described in [41] (Kem2005) and in the accompanying
article [42] could be considered as the first to investigate the potential of EVs and
RES integration. These articles describe the first systematic exploration of the
vehicle-to-grid concept, since then often referred to as V2G. It should be noted,
however, that the first mention of this idea was already in earlier work [51]. The
more recent articles [41] and [42] quantify the potential for V2G to participate in
ancillary service markets and high values up to 4000$ per vehicle per year are found
as potential revenues for offering such services.
In [46] (Lun2008), the benefits of V2G in a power system with a very high RES
penetration were assessed. The main advantages are a sharp decrease of excess wind
production and a strong reduction of CO2 emissions. The latter can mostly be understood from a replacement of gasoline by clean electricity as fuel for vehicles. [46]
considered a number of EV scenarios, which are, in increasing order of complexity: uncontrolled charging, intelligent charging without discharge and, finally, V2G,
where EVs were allowed to discharge to the grid. Interestingly, almost all benefits
are already realized in the intelligent charging without V2G, leading to the conclusion that the possibility to postpone demand to high wind periods is more important
than the possibility to store and discharge electricity from the EVs.
In [49] (Pet2010) the economic potential of V2G for energy arbitrage purposes
considering PJM wholesale prices is investigated. Noteworthy is the fact that they
include realistic battery degradation costs to their analyses, that were determined in
laboratory experiments described in [34]. Due to these relatively high degradation
costs in combination with a small price spread in the PJM market, they find very
modest, if not negligible V2G revenues.
Valley filling A number of studies have evaluated the potential to dispatch EV demand in the hours with low electricity demand3 , since this is generally considered to
have a positive system impact. Most studies do this through the signal of electricity
wholesale prices since these should in principle reflect the magnitude of electricity
demand, others use more direct load scheduling algorithms.
[39] (Kri2011) discusses optimal EV charging based on Danish wholesale prices.
Important contributions of this work are that the effect of the extra EV load on
wholesale prices is taken into account, which leads to a quadratic programming
3 This is often referred to as ‘valley filling’ to distinguish it clearly from ‘peak shaving’ which
would only be possible in a V2G setting, i.e. vehicles delivering energy to the grid.
3.2 Discussion of some important papers per sub-field
43
problem. In addition to this, a method to group EVs by driving patterns is introduced to make computations tractable.
One of the earlier studies to optimal EV charging from the point of view to
‘fill the valley’ of electricity demand was done by [52] (Den2006). Instead of the
more frequently used mathematical programming tools, they use a ’valley-filling
algorithm’ that assigns EV load to the moments with low demand. They show,
using load duration curves, that EVs can indeed be charged in such a way that only
base-load is increased and peaks do not change.
In [53] (Mar2012) the intelligent control of a parking lot with a fleet of EVs was
considered. Interesting in this setup is that the vehicles can also deliver energy to
each other (denoted with vehicle-to-vehicle, V2V), preventing to pay an extra grid
fee for delivering energy back to the grid as in the ‘standard’ V2G setting. Spanish
wholesale prices were used for the time varying energy tariff, and it was found that
in their particular case study setting, there was a cost reduction compared to dumb
charging, but V2V and V2G were found not to be profitable.
Integration frameworks A number of papers have treated the subject of how
EVs and its smart management should be embedded in the current technical and
organizational structures of power systems. Usually these type of studies formulate
it as presenting a ‘framework’ for EV (or V2G) integration. Such studies usually
have a rather abstract or general character, which is the reason why they have been
placed at the top of Fig. 3.2.
One of the earlier works to present such a framework is described in [47]
(Gui2009). They identify load leveling (‘valley filling’), regulation power and reserver power as the most important aspects in which EVs could provide value. Then
possible approaches for the information, communication and control infrastructures
are discussed. The aggregator, as the intermediate and coordinating entity between
individual EVs and other stakeholders, plays a central role in this framework.
A second paper on EV integration frameworks is [54] (Gal2010), which, one could
argue, goes somewhat further than [47]. Interestingly, when discussing the possible
role of EVs in power systems, a remark is made that for an EV aggregator, although
playing a central role, it is not quite clear yet on which layer of the power system
structure (generation, transmission or distribution) or in which markets and with
what actors they will interact mostly - a notion that is highly relevant for this thesis
as well. One of the main ideas put forward in [54] is the operational framework based
on a state description for the EVs, similar to states of a power system like ‘normal
operation’ or ‘emergency’. In this framework, possible EV states are for example
‘charging’, ‘driving’, but also ‘feeding power’, as the framework incorporates V2G.
The framework is demonstrated with an example case study where EVs are providing
regulation services.
A comprehensive analysis discussing a conceptual regulatory framework as well
as new business models for charging EVs is given in [27] (Rom2011). A variety
of different settings of infrastructures, actors (denoted ‘agents’ in the paper) and
commercial relationships are explored in-depth. The paper distinguishes two new
actors that will be involved in EV charging besides EV owners: EV aggregators,
denoted as EV supplier aggregator (EVSA), and charge point managers (CPMs).
44
3 Literature review
One of the difference between these two is that the contracts of the EVSA are
‘not location based or bound to a single outlet’. The aggregator benefits from
aggregation advantages and economies of scale, but the EV owners it has contracts
with are dispersed geographically. The CPM, on the other hand, owns and operates
physical charging infrastructure on private property. The latter is an important
distinction. Here it is the CPM who buys energy and/or network capacity and
thus has to deal with DSOs and suppliers (or the market). Besides the various
instutional arrangments also technical properties of different charging modes are
discussed. V2G services are considered only to be feasible for the longer term. For
the shorter term, three possible arrangements are considered as the most likely: 1)
uncontrolled charging at home without any EVSA or CPM (basically what many
of the early adopters are doing now) 2) controlled charging in public areas with an
EVSA as intermediate and 3) controlled charging in private areas with the CPM as
intermediate.
Another EV integration framework is presented in [55] (Pec2010). In the same
paper, the authors also perform some technically detailed studies regarding possible
voltage problems and the potential of EVs to enhance dynamic stability in isolated
power systems. This twofold perspective is reflected in Fig. 3.2 where the same paper
can be found both at the very bottom and the top of the figure. The most distinctive
feature of the framework proposed in this paper is to distinguish explicitly between
a technical and a market domain. The idea is that EVs can provide services to
market participants, or benefit from certain conditions on the markets, only as long
as they do not violate technical constraints. Following this philosophy, it is the DSO
(responsible for the technical side) who can override the aggregator signals focusing
on the market domain. One could thus argue that this framework ascribes more
power to DSOs than in other frameworks, or, more importantly, than is allowed
in some unbundled power sectors where a DSO is more or less obliged to facilitate
and cannot interfere with the market. New possible organizational models to align
technical distribution network limits with market signals are also discussed in chapter
6 of this thesis.
Finally we note that the papers [56] and [57] (Ili2011) that we discuss more
in-depth below in fact also provide frameworks of EV integration, although their
scope extends beyond EVs specifically. A common element with other frameworks
is the key role of the aggregator. Compared to the other frameworks, [56] and
[57] provide a valuable contribution by more explicitly focusing on the potential of
flexible demand in the light of RES integration. Furthermore, the level of technical
and mathematical detail can be considered much higher than especially the first two
papers discussed in this subsection.
Institutional and social aspects Since, during recent years, EVs have started to
become a promising and possibly high impact new technology, its adoption has been
subject of many social science oriented papers as well. Since this perspective is not
really the viewpoint of this thesis, we will only discuss one of the earlier and mostly
cited paper in this field, which is [58] (Sov2009). The paper discusses the benefits and
potential barriers of large EV adoption and a V2G transition. It rightfully points out
that next to technical barriers, important institutional and socio-technical obstacles
3.2 Discussion of some important papers per sub-field
45
must be overcome as well. Such obstacles are not only the result from consumer
resistance against new technologies, but, and more seriously so, also from current
stakeholders’ interests to keep in place the existing infrastructure. Especially the
automotive industry might show a strong resistance against the large scale adoption
of EVs.
Distribution grid impacts of EVs An extensive overview of studies of EV
impacts on distribution grid was given in [59] (Gre2011). We do not consider it
particularly useful to repeat a large part of the studies mentioned in [59], so below
we only discuss a few of the most important EV grid impact papers.
One of the most cited articles on distribution grid impacts of EVs is [44]
(Cle2010), which compares a quadratic and dynamic programming approach to study
EV impacts (primarily voltage deviations and power losses) on a distribution feeder.
Both deterministic and stochastic optimizations are proposed and used in an example case study to minimize the energy losses and voltage deviations. The PhD
thesis of the same author extends the analysis to cover a broader range of aspects
[60].
[55] presents a conceptual framework for EV integration into power systems (this
part was discussed earlier in this section) and additionally studies both a static
distribution grid analysis (main output: voltage deviations) and dynamic behavior
(output: system frequency response). Both case studies have an isolated grid as
object of study. The distribution grid analysis shows, as expected, that congestion
and/or unacceptable voltage deviation occurs in the uncontrolled (the authors refer
to it as ‘dumb charging’) charging scenario. Almost all congestion and voltage
issues can be managed by applying a smart form of charging, for which in this case
no mathematical programming approach was used, but an iterative approach that
temporarily halts EV charging in congested lines.
[61] (Tay2010) analyzes a large set of distribution assets in a stochastic way to
present probability distributions of asset impacts. An important conclusion in this
work is that due to the EV load diversity, i.e. the randomness in the timing of
charging EVs leading to a large smoothing of the EV load, especially those assets
that have a lower number of customers connected to them are most at risk of being
overloaded.
One of the very few studies expressing the impacts of EV charging on distribution networks in monetary terms related to additional investments and energy
losses was presented in [62] (Pie2011). In this paper, additional EV related load
was imposed on two reference distribution networks, one in rural and one in urban
area. Contrary to many studies, the networks are real distribution grids with tens
of thousands of consumers connected to them and covering relatively large areas.
A large-scale distribution grid planning model was used to asses the required investment in additional grid assets. The results show marked differences between
the rural and urban areas. According to the precise EV penetration scenario, the
estimated cost savings of moving EV charging to off-peak hours are estimated in the
range of 5-35% of the investment costs. Energy losses were found to increase up to
40% due to EV charging.
46
3 Literature review
Co-optimization with respect to balancing power and wholesale prices A
number of papers describe how EVs can simultaneously minimize charge costs based
on time varying electricity prices and provide minute to minute balancing services.
[32] (Rot2010) used a dynamic programming algorithm to find optimal charge schedules that are a balance between charging against low electricity prices and leaving
the opportunity to provide regulation services. Indeed it is found that this combination has a higher financial potential than charging solely based on electricity prices.
Furthermore, [32] uses a much more complex battery and vehicle model (in driving
mode) than many of the EV studies described above.
In [40], [63] and [64] (Sor2011) a variety of optimal EV charging strategies is
discussed, among which are combined bidding of ancillary services (two types) and
energy purchases, but also load (and energy losses) minimizing strategies. In most of
the above papers, the authors refer to their EV charging strategies as unidirectional
V2G, since no power flows from EV to the grid, but still both regulation up and down
can be provided. This should be understood by realizing that lowering the power
demand of a group of EVs from, say 200kW to 100kW, is equivalent to delivering
100kW of power when no demand was scheduled. Although the range of regulation
power is clearly smaller in this unidirectional approach, there are also important
benefits because battery degradation and/or consumer reluctance for this type of
service play a less important role.
A true stochastic approach to the combined energy cost minimization and the
provision of regulation services was discussed in [65] (Don2012). Here a dynamic
programming approach was used to appropriately weigh costs associated with all
possible charging trajectories when the EVs are charging and providing regulation
services at the same time. A distinct value of the stochastic approach compared to
a deterministic method was reported.
EVs in unit commitment One of the first studies to explicitly consider EVs in
the unit commitment (UC) problem was described in [66] (Sab2010). In this work,
EVs were assumed to have V2G capabilities and, as such, they were modeled as small
portable power plants. As far as we can judge based on the description of the paper,
however, no details about the driving behavior were taken into account, and only
the optimal number of EVs providing V2G services was calculated. We therefore
interpret the focus of this paper to be more on the computational method (based on
a particle swarm optimization), than on the higher level impacts and benefits from
flexible EV demand.
In [67] (Fer2012) EVs with V2G capabilities were included in a UC model of
the Spanish power system. A number of scenarios with varying RES and EV penetration were considered. Some of the positive EV impacts that were observed are
the reduction of spilled renewable energy, the reduction of the use of pumped hydro storage, fewer hours with high system marginal cost and lower power reserve
requirements. Almost of all of these benefits were much greater in the high RES
scenario, indicating once more the potential synergy of EVs and RES.
An extension of the work [67] is presented in [68] (Ban2013). Here, too, the effect
of including EVs in the UC of the Spanish power system was studied. The model
includes a stochastic element by considering wind forecast errors and unplanned
3.3 Relative positioning of this thesis regarding the literature
47
thermal unit outages. In addition to an uncontrolled charging profile, three slightly
different controlled charging strategies were evaluated: minimizing the total marginal
generation costs, maximizing the minimum demand (valley filling) and minimizing
spillage of wind generation. The general picture that emerged was similar to that
of [67]: controlled EV charging leads to lower generation costs, less hydro pumping
and better wind utilization than uncontrolled charging. Results show also that the
three different controlled charging strategies lead to very similar outcomes, which
could be considered as not surprising since their objectives actually do not differ
much. Another noteworthy result is that the EVs could only reduce wind spillage to
some extent and a saturation point was observed at some level of EV penetration.
Furthermore, although CO2 emissions per MWh increased in the controlled scenarios
due to the increase in cheap coal generation, this effect was offset by the higher hydro
pumping losses leading to higher CO2 emissions in the uncontrolled scenarios.
Another approach to integrating EVs is proposed in [56] and [57]. Here EVs are
not explicitly discussed in detail, but they are treated as one form of elastic demand.
The idea put forward in the framework proposed in these papers is, contrary to
e.g. [66], a distributed and decentralized approach where elastic demand creates
and exchanges demand functions by performing optimizations based on a range
of price scenarios. In this way, the user preferences and physical constraints of
the inelastic demand, are internalized in the demand functions. While this is a
promising method as it circumvents the intractable computational problem of a
centralized entity scheduling all demand, the inter-temporal constraints4 of EVs are
not treated in detail. These constraints can be a complicating factor, since they
create a dependence between the demand functions of different time-steps. This
method will also be discussed in chapter 6 of this thesis, where a similar method,
but here for the demand function for network capacity, is treated.
3.3
Relative positioning of this thesis regarding the
literature
As a means of sketching a coherent picture of the elements of this thesis, we will
discuss how, in our opinion, this thesis should be seen relative to the scientific
literature on this topic. In chapter 3 we provided an overview of the scientific
literature on a research area that can be described as ‘the cross-section of integration
of EVs and RES’. In Fig. 3.3 we show the same schematic overview with the chapters
of this thesis placed in the diagram. We emphasize that, since we want to elaborate
on the contributions of this thesis, we have exaggerated the size of the grey zones
denoting the chapters in comparison with the other literature, where a light grey zone
denotes a complete ‘sub-field’ and individual author contributions are represented
by a point. We do not consider our chapters covering entire ‘sub-fields’ or much
broader scopes than other authors.
4 Inter-temporal constraints have been discussed in chapter 2 regarding generation units. An
example in the context of EV charging is that when a battery needs to be fully charged at a certain
time-point, this restricts the allowed battery states and thus charging power in previous time-steps.
48
3 Literature review
System view
Institutional
Frameworks
Sov2009
Gom2011
V2G
Valley filling
Gui2009
Gal2010
Kri2011
Lun2009
Pec2010
Ch6
Pet2010
Had2009
Kem2005
Ch5
Renewables integration
Ch7Den2006
Rot2009
Sab2009
Don2012
EV integration
Ch4
Mar2012
Ili2010
Fer2012
Ban2013
Sor2011
Balancing
Unit committment
Distribution grid impacts
Sun2012
Gre2011
Cle2010
Pie2011
Pec2010
Tay2010
Technical detail
Figure 3.3 – Schematic representation of the elements of this thesis relative to the scientific
literature on EVs and renewables integration.
The essence that we aim to express in Fig. 3.3 is that the chapters aim to arrive at system level conclusion from a more technical starting point. The goal in
chapters 4, 5 and 6 is not to provide an in-depth technical grid analysis or a new
optimization algorithm, but rather to take existing modeling techniques and apply
them consequently to one part of the electricity system to arrive at high level system conclusions. This is why indicated the chapters stretching out vertically, hence
covering a large part of the y-axis (from technical viewpoint to system view). In
chapter 7 the goal was to connect the individual pieces provided by the previous
chapters, hence its horizontal shape.
The resulting image could be perceived to bear some resemblance with a sparsematrix structure. We do not pretend to cover the complete field of renewable and EV
integration from technical to system point of view, but rather we studied different
aspect stretching between the ends of this field. By this rather wide, though sparse,
scope, we hope to have reached conclusions that would have been less obvious when
focusing on smaller parts of the scientific field.
Other contributions of this thesis
As explained above, we consider the integrated view that we take, i.e. from distribution grid impacts to EVs supporting renewable energy and the interrelations
between these two, one of the core contributions of this thesis since most literature
only focuses on single aspects. Secondly, the core chapters themselves have more
specific contributions with respect to the state-of-the-art. Below we list the novel
research elements that are described in this thesis, in the conclusions sections we list
results and new insights that contribute to the state-of-the-art.
• Chapter 4: An EV distribution grid analysis based on realistic driving patterns
and a large number of distribution grids using realistic and measured grid loads.
This allows to quantify system benefits, rather than benefits on an individual
(and hypothetical) distribution grid level.
3.3 Relative positioning of this thesis regarding the literature
49
• Chapter 4: A financial analysis of EV load induced distribution asset replacements and energy losses.
• Chapter 5: Analysis of the interrelations between (often seen as alternative
technologies of) smart EV management and cross-border electricity transmission expansion.
• Chapter 6: Mathematical formulations and analysis of different congestion
management schemes for aligning demand response with distribution grid capacity in a high RES power system.
• Chapter 7: Analysis of potential market power issues by large volumes of
flexible demand.
• Chapter 7: An exploration of various sensitivities related to demand response.
50
3 Literature review
Chapter 4
Network impacts and cost
savings of controlled EV
charging
This chapter has been published as [69] and has its own nomenclature that slightly deviates from
the rest of the thesis.
4.1
Introduction
Electric vehicles could be an important contribution to the reduction of greenhouse
gases in the transport sector, but concerns have been raised about the impacts of a
large fleet of EVs on the electricity distribution grid. A recent overview of studies
on different aspects of the impacts was given in [59]. Usually these analyses consider
aspects such as energy losses, voltage profiles, reduced lifetime of network components, thermal loadings of cables and transformers, etc. Three fundamental things,
however, seem to have received less attention: 1) detailed and realistic charging
profiles of EVs based on real life driving data, 2) analyses of the EV impacts on
large numbers of operational distribution grids and 3) economic implications of the
distribution grid impacts. The work presented in this chapter aims to fill this gap.
Some exceptions regarding the three points above are the work described in [61]
and in [70]. In the former study, analyses based on statistics of driving patterns
and actual grid assets show that even at low penetrations of EVs, a significant
portion of grid assets will be overloaded. Furthermore, it is concluded that controlled
charging of EVs can potentially reduce network impacts. The latter study considers
an operational distribution grid and shows that controlled charging of EVs can lead
to reduced investments in network reinforcements, but the analysis was done for
only one grid.
The study presented in this chapter introduces a method to investigate the impacts and economic consequences of EV charging on distribution grids by means of
extensive analyses of large datasets of distribution grids and transportation data.
51
52
4 Network impacts and cost savings of controlled EV charging
The main emphasis is on comparing controlled and uncontrolled charging scenarios.
By considering a large number of distribution grids and different grid assets, we
acknowledge the diversity between networks and we aim to point out where impacts
are most severe. Furthermore, this approach allows us to quantify the economic impacts on a system level, rather than on the individual network level. Especially for
distribution system operators (DSOs) that often operate in a regulated environment,
the economic viability of smart grids is a crucial aspect [71].
Our study uses Dutch grid and transportation data, but the methods and most
results are more generally applicable. This study partly builds on, but also extends
the work presented in [72], which is summarized in appendix B. Whereas in the
latter, the emphasis was on the low voltage (LV) part of distribution networks,
here we extend the analysis to the medium voltage (MV) parts of the distribution
networks and we add economic figures to it. Moreover, our analysis contributes
to the broader research topic on how smart grids can lead to more efficient use of
electricity grids. Although the flexibility of EV charging might be used for a whole
range of applications, the focus of this approach is on reducing the peak load of
distribution network assets by throttling the charge rate of the EVs.
4.2
Research method
The approach of this study is to consider a large number of networks, rather than
investigating a few sample networks in detail. Because of the large number of networks, the construction of the load profiles in all network nodes and the computation
of the power flows are handled in an automated procedure. Inherently, some approximations are made that could produce less accurate results for individual networks.
The findings of this study should therefore be understood to be mainly applicable
to the system level, rather than for individual network components.
The fact that we use an aggregated approach also has some consequences for the
data that we use, which should be recent, readily available and covering the largest
possible set of networks. We work mainly with the most recent yearly peak load
measurements – a method that is in line with current network planning practice.
4.2.1
Distribution networks
The typical structure of electricity distribution grids often has a historical path
dependency and as a result differs from country to country and often even from
region to region or town to town. Usually there are also differences between rural
and urban regions. We investigate in this chapter a large number of distribution grids
in the Netherlands operated by Enexis, one of the largest DSOs in the Netherlands.
The typical structure of electricity distribution grids in the Netherlands is shown
in Fig. 4.1. The distribution grids extend from the HV/MV transformer station to
the individual household connection. We consider in this chapter four levels of the
grid: the HV/MV transformers, the MV transmission cables, the MV distribution
cables and the MV/LV transformers. Some essential features of the Dutch distribution networks should be emphasized, because they can differ strongly from distri-
4.2 Research method
53
Table 4.1 – Overview of the network data that has been used
Data Type
Number of Re- Relevant Known/Measured
cords
properties
Trans- 12.000
Measured yearly maximum
power S, nominal capacity,
zip code, transformer type
properties
Comments
MV/LV
formers
Number of households is estimated on the basis of average electricity use per zip code.
MV Cables
HV/MV
stations
13.000 km
Measured yearly maximum The imbalance between the sum
power S, cable type proper- of all maximum transformer
ties, connected transformers loads and the maximum of the
cable load is corrected for by a
coincidence factor and/or fictitious loads.
Sub- 55 (150 Trans- Measured yearly maximum Coincidence factor and/or fictiformers)
power S, transformer type tious load accounts for imbalproperties, connected cables ances.
Figure 4.1 – Typical structure of distribution networks in the Netherlands. Picture from
[73].
bution networks in other countries. Almost all MV and LV cables are underground
cables, which is different from many countries in the world.1 The typical number
of households connected to one MV/LV transformer is 80 and the typical amount
of MV/LV transformers connected to one MV distribution cable is 10. Finally, we
make a distinction between MV distribution (MV-D) and MV transmission (MV-T)
cables. MV-T cables form the connection between the HV/MV substation and the
MV distribution station, without any loads (or MV/LV transformers) connected to
them. Sometimes the distinction between MV-D and MV-T cables is somewhat arbitrary, when for example the first part of an MV-D cable functions as the MV-T
cable to cover a reasonable distance, but at some point of the cable there are loads
or transformers connected to it.
1 This
matters especially for the cost calculations since overhead lines are much cheaper than
underground cables. Data reported in [62] show differences in the order of a factor 3 to 5. In [62],
the values for underground cables are, on the other hand, much higher than the value of 60 e/m
that has been used in this analysis.
54
4 Network impacts and cost savings of controlled EV charging
Figure 4.2 – Locations of analyzed networks in the Netherlands. Each dot represents a
MV/LV transformer.
The networks under consideration in this study are described in Table 4.1. If one
defines a distribution grid as everything behind a HV/MV substation (which Fig.
4.1 implicitly does), we consider in total 55 networks that cover a substantial part
of the Netherlands, see Fig. 4.2.
4.2.2
New load profiles
The following quantities are used in the procedure to calculate the future load profiles. Bold faced quantities denote time profiles, i.e. a vector with values for each 15
minute interval of a day.
Smax (i, t0 )
S(i, t)
SEV
Shouse
SEV s (i, t)
Shouses (i, t)
Nhouses (i)
NEvs (i, t)
f (t)
Measured peak demand (apparent power)at transformer i in year t0
Demand (apparent power) profile at transformer i in year t
Aggregated EV demand profile normalized to a single EV
Aggregated household demand profile normalized to a single household
EV demand profile at transformer i, year t
Household demand profile at transformer i, year t
Number of houses connected to transformer i
Number of EVs connected to transformer i in year t
Fraction of EVs of the total passenger car fleet in year t
4.2 Research method
55
The idea is to change the loads of the smallest building blocks (the MV/LV
transformers) based upon an estimate of the growth of the household load plus the
new EV part. The starting point of the future load profiles is the most recent
measured peak demand per transformer (2010 in our case) Smax (i, t0 ). To obtain
a household demand profile Shouses (i, t) it is assumed that the transformer load
follows a standard household load profile Shouse that is used extensively for network
planning purposes in the Netherlands [22].
In principle, Shouse is defined for each quarter of an hour of the year, but since we
are interested in the worst case scenario, we only consider the day with the highest
network load. We assume furthermore that the transformer load Shouses (i, t) will
grow exponentially with a rate of a = 1% per year and the shape of the profile
does not change, so the time evolution of the demand profile without the EVs for
transformer i will be as follows:
Shouses (i, t) = Smax (i, t0 ) ·
Shouse
· (1 + a)t−t0
max{Shouse }
(4.1)
It is emphasized that, strictly speaking, Shouses (i, t) does not necessarily contain
only household electricity demand, but also load from public lighting, small stores
etc. The name is chosen to distinguish it from the EV load.
To calculate the future EV demand profiles SEV s (i, t), we use additional information on the amount of expected EVs per transformer and information about driving
patterns. For the introduction speed of EVs, an S-curve scenario f (t) as envisioned
by the Dutch government is assumed, which saturates at approximately 75% EVs
(out of all passenger cars) in 2040 [14]. The electricity demand profile of the EVs
will depend on how the EVs are being charged. In this study we consider the different scenarios as presented in [28], where different EV charge profiles SEV have been
derived on the basis of a large dataset of driving patterns in the Netherlands [29].
The fundamental assumptions made to derive the different EV charging profiles are
that driving patterns (trip lengths, arrival times, etc) will not change with respect
to today and charging is done only at home, after the last arrival of the day. Fig. 4.3
shows the different SEV profiles added to the standard household profile Shouse . As
Fig. 4.3 illustrates, the controlled charging scenario was derived in such a way that
the bulk of the EV load is shifted to the night hours, when network load is low. In
this control strategy two objectives are met: 1) the combined peak load of household
demand and EV demand is minimal 2) the energy losses are minimal. One could
imagine many different control strategies in a smart grid setting, but here the point
of view of the distribution network operator is chosen. In a regulated environment,
DSOs will generally aim at minimizing costs associated with reinforcing existing
grid assets and energy losses. In the remainder of this chapter we will furthermore
assume that any IT technology needed for controlled charging of EVs is in place and
we will not include its costs in our analysis.
The profiles SEV represent the aggregate demand of a large number of EVs, but
they are normalized to a level of a single EV, so they should be multiplied with the
appropriate number of EVs to obtain the correct demand profile SEV in kVA. To
predict the evolution of number of EVs per transformer, the number of houses connected to the transformer is needed. This number, denoted by Nhouses (i), is either
4 Network impacts and cost savings of controlled EV charging
Power per Household (kW)
56
household only
uncontrolled, 3kW
uncontrolled, 10kW
controlled charging
2
1.5
1
0.5
0
0
4
8
12
time (h)
16
20
24
Figure 4.3 – Aggregated demand profiles of household with EV (normalized to one household)
known or else it is estimated by dividing the measured peak at the transformer by
the peak value of the standard profile of one household. The estimate will generally be larger than the actual number, due to the fact that we have ignored other
loads connected to the transformer such as public lighting, elevators in apartment
buildings, etc. To correct for this, the estimate is multiplied by a correction factor
so that the average number of estimated households per transformer is equal to the
known average actual number of households per transformer. This implies that our
estimate of Nhouses (i) could be less accurate for a single transformer, or even MV
cable with a number of transformers connected to it, but the overall picture will be
accurate.
The number of EVs connected to transformer i in year t is the product of the
penetration rate f (t) and Nhouses (i), where the fact that the average number of cars
per household is one has also been used. So the evolution of the EV demand profile
takes the following form:
SEV s (i, t) = NEV s (i, t) · SEV
(4.2)
Although not explicitly indicated in Eq. 4.2, we take into account that people in rural
areas generally drive larger distances per day, so the magnitude of SEV depends on
the location of the transformer. The zip code that is known for each transformer
allows us to connect it with other databases from which we retrieve information in
housing density, a measure how rural an area is.
The combined demand profile at transformer i in year t is thus given by:
S(i, t) = Shouses (i, t) + SEV s (i, t)
Shouse
= Smax (i, t0 ) ·
· (1 + a)t−t0 + NEV s (i, t) · SEV
max{Shouse }
(4.3)
where the addition of Shouses (i, t) and SEV s (i, t) is the addition of the P and Q
components of S. As an illustration of how fast the peak load grows, Fig. 4.4 shows
4.2 Research method
Maximum Transformer Load (kVA)
180
160
140
57
1% growth
1% + EVs uncontrolled 3kW
1% + EVs uncontrolled 10kW
1% + smart EVs
EVs uncontrolled 3kW
EVs uncontrolled 10kW
EVs Smart
120
100
80
60
40
20
0
2010
2015
2020
2025
Years
2030
2035
2040
Figure 4.4 – Maximum transformer load in different scenarios of a transformer with 100
houses connected to it
the evolution of the demand peak of a transformer with 100 houses connected to
it. In the uncontrolled EV charging scenarios, the EV load is a substantial part of
the total load. The profiles of the cable and HV/MV substation loadings are not
specified, but follow from the power flow analysis. Fig. 4.5 summarizes the procedure
to calculate the future load profiles that have been described in this section.
4.2.3
Power flow
The power flows in the networks have been determined with the commercial package
Vision [74], which uses the Newton-Rhapson method to solve the power flow equations. This package is widely used for network planning in the Netherlands, and
an important advantage is that the networks we consider are already represented in
full detail in the proper format. This means that the most recent measurements of
transformer and cable loadings are processed in the Vision network files and the networks are ‘calibrated’ to account for the coincidence factors of loads, non-measured
loads and losses. To simulate the new loads, the measured transformer loads are
replaced by the new values dictated by Eq. 4.3. The new EV loads are modeled as
constant power loads with a power factor of one. These assumptions are in line with
the models this DSO uses for network planning purposes and with how the loads are
already characterized in the network files.
The output consists of node voltages, line currents, transformer loadings and peak
losses in both lines and transformers, which have been calculated using the new load
profiles as in Fig. 4.3. The network status is always assumed to be as in normal
operation mode, i.e. with network openings (see Fig. 4.1) in open mode. Although
node voltages are readily available from the output of the power flow analysis, we
will not present them in this chapter. One of the reasons is that unacceptable
voltage excursions do not necessarily lead to extra costs, because transformer settings
can possibly be changed. Other network impacts such as unsymmetrical loads and
58
4 Network impacts and cost savings of controlled EV charging
Figure 4.5 – Block diagram illustrating the procedure to obtain the future load profiles
S(i, t), Scable (j, t) and Sstation (k, t). Quantities in grey blocks are input data, white blocks
denoted calculated values.
harmonics have also been discarded.
We emphasize that only the values of the MV/LV transformers have been changed
and the new peak loads in all network parts higher than the MV/LV transformers
follow directly from the power flow simulations. In general, the relative increase
in peak load due to the EVs will be lower for the higher network levels, since the
relative share of the residential load is lower due to the presence of commercial and
industrial consumers.
4.2.4
Energy loss estimation
The following quantities will be used in the next two sections to calculate the energy
losses and various economic figures. Note that index i is now used to denote any
component including cables and HV/MV transformers.
S(i, t)
tol (i)
TL,peak (i, t)
TL,0 (i)
PL,peak (i)
PL,0 (i)
Snom (i)
ACrpl (i, t)
Demand (apparent power) profile at component i in year t
Year component i will be overloaded
Service time of the peak loss at component i in year t
Service time of component i
Peak loss of component i
No load loss of component i
Nominal thermal capacity of component i
Annuity charges of component i in year t
4.2 Research method
Casset (i)
Closs (i, t)
Ctotal (t)
NPV
59
Asset costs of component i
Loss costs of component i in year t
Total costs in year t
Net present value
For network planning purposes, energy losses in different network assets are usually
estimated based on information that is readily available for the DSO: yearly peak
loading of the network assets [75]. The yearly losses can then be estimated with the
help of the asset properties (including the distribution of loads along a cable) and
an assumed yearly load profile. The commercial power flow solver used in this study
uses the asset properties and peak load values to calculate the peak loss Ploss,peak
and the no-load loss Ploss,0 . These numbers can be converted to the yearly energy
losses by invoking the service time of the peak loss TL , which contain information
of the shape of the load profile and the standard service time of transformers T0 (see
e.g. [75] and [73]). For a transformer the yearly energy losses are then given by:
Eloss (i, t) = α2 (i, t)PL,peak (i)TL (i, t) + PL,0 (i)T0
(4.4)
Eloss (i, t) = α2 (i, t)PL,peak (i)TL (i, t)
(4.5)
and for a cable
where α(i, t) = max{S(i,t)}
is the utilization factor of the asset and T0 = 8760h. Load
Snom (i)
independent losses, also referred to as iron-losses are mainly present in transformers
and are associated with phenomena such as eddy currents and hysteresis effects,
which depend on frequency and/or voltage. Here, they are assumed to be constant
for one type of asset.
The evolution over time of the losses is then determined by the product of the
evolution of the service time of peak loss and the utilization factor. These are known
exactly for the MV/LV transformers (since they are dictated by the load growth and
EV profile), and for the higher network levels they follow from the relative sizes of
the residential and industrial/commercial loads. The time dependence of Ploss,peak
and Ploss,0 has not been stated explicit in Eqs. 4.4 and 4.5, because it is assumed
constant for one type of asset. However, when in our cost analysis overloaded assets
are replaced, the values of Ploss,peak and Ploss,0 change accordingly. Usually the load
dependent losses will be lower for a higher capacity asset (at the same transported
power), because Ohmic resistance is reduced. Alternatively, the load independent
iron losses of transformers could increase slightly after replacement, this depends on
the exact type of transformer. As an example, Fig. 4.6 shows the evolution of the
yearly energy losses of one specific transformer whose threshold value is exceeded
after some years in the case of the uncontrolled charging scenarios. The 20-25 %
reduction in energy losses due to the replacement are clearly visible. It will depend
per asset if the overall losses over the 30 year horizon end up lower or higher in the
controlled or uncontrolled scenario; in the example of Fig. 4.6 the integral over the
losses is similar for the controlled and uncontrolled cases.
It should also be emphasized that we consider no loss minimization by controlling
node voltages and/or reactive power injections. Although this could clearly be an
60
4 Network impacts and cost savings of controlled EV charging
interesting application in future smart grids, we consider this outside the scope of
this study.
4.2.5
Costs
We calculate expected differences in costs between the EV charging scenarios by
considering replacement costs if an overloaded asset has to be replaced and the
energy loss costs. In our model, an asset is replaced by a type with a higher capacity
in the year the threshold value defined in Table 4.3 is exceeded, so the investment
(or cash flow CF) due to the replacement of component i in year t is given by:
{
Casset (i) if t = tol (i)
(4.6)
CFrpl (i, t) =
0
else
For the replacement costs, we take the annuity charges associated with the investment in new assets. The annuity charges can be considered to represent the
yearly amount that, over the life of the asset, has a net present value exactly equal
to the asset’s initial cost:

C
(i) ·


 asset
ACrpl (i, t) =



0
r
−T
(i)
1 − (1 + r) lif e
if tol (i) ≤ t < tol (i) + Tlif e (i)
else
(4.7)
where r = 3% is the interest rate based on the current value of Dutch treasury bonds
and the expected lifetimes are 50 years for MV/LV transformers and cables and 40
years for HV/MV transformers.
The asset costs Casset (i) incorporate all possible costs associated with a replacement: material, labor, taxes, etc. In general, reinforcement of a certain network
component can be done in various ways, but we assume a straightforward replacement by a heavier type of the same component. Cables are replaced by a higher
capacity type over the entire cable segment that was overloaded. It is stressed that
we focus only on the reinforcement of existing networks, thereby neglecting costs
associated with the construction of completely new networks. The values of the
asset costs are assumed to be constant over time and they are based on figures that
are currently being used for network planning purposes at the DSO controlling the
networks that this study considers and are given in Table 4.2.
The costs of energy losses for component i in year t are given by
Closs (i, t) = Eloss (i, t)Cel
(4.8)
where Cel denotes the electricity price, which is based on average Dutch electricity
prices and we assume it to be constant.
Total yearly costs are found by summing up replacement annuity charges and
energy loss costs of all network components:
Ctotal (t) =
N
∑
(ACrpl (i, t) + Closs (i, t))
i=1
(4.9)
4.2 Research method
61
Table 4.2 – Asset costs and energy loss costs. The asset costs differ from asset to asset;
values in this table are typical values.
Cost Type
MV/LV Transformer
MV Cable
HV/MV Substation
Energy Loss
1.8
1.7
EVs uncontrolled 3kW
EVs uncontrolled 10kW
EVs controlled
No EVs
Price
10.000 e
60 e/m
1.200.000 e
0.06 e/kWh
Replacement
Relative Losses (−)
1.6
1.5
1.4
1.3
1.2
1.1
1
0.9
2010
2015
2020
2025
Years
2030
2035
2040
Figure 4.6 – Time evolution of the losses in one of the MV/LV transformers for the different
scenarios. Losses are reduced when the transformer is replaced by a type with higher
capacity.
Eq. 4.9 can be converted to net present value according to:
tf
∑
Ctotal (t)
NPV =
(1 + r)t−t0
t=t
(4.10)
0
where tf = 2040 in our case. The fact that our time-horizon lies at the year 2040
means that a significant portion of the replacement costs as defined by 4.7 fall outside
the scope of this analysis. Also, unrecoverable sunk costs resulting from assets being
replaced before the end of their economic lifetime are not taken into account. These
notions should be taken into consideration when interpreting the results.
Fig. 4.7 schematically summarizes the procedure to calculate expected costs that
has been described in this section.
62
4 Network impacts and cost savings of controlled EV charging
Figure 4.7 – Block diagram illustrating the procedure to calculate the total costs.
4.3
4.3.1
Results
MV/LV Transformers
Fig. 4.8 shows histograms of expected MV/LV transformer loadings in the three
EV charging scenarios, together with the scenario of 1% growth only. The figures
also denote the percentage of overloaded transformers, for an overload criterion of
1.16. This value reflects the fact that we are considering an instantaneous peak
value, which can be higher than the nameplate capacity for some time. The number
of overloaded transformers can be interpreted as the fraction of transformers that
would have to be replaced. The data represents a subset of only those transformers
connected to residential customers, i.e. transformers connected to large industrial
or commercial customers have been omitted. The effect of the control of the EV
charging is pronounced: the amount of overload transformers that would have had
to be replaced decreases substantially. Compared to the situation with no EVs (the
thin black line in the histograms of Fig. 4.8), there are hardly any extra replacements
needed.
4.3.2
MV cables
Fig. 4.9(a) shows the distribution of the loadings of MV distribution (MV-D) cables
in the uncontrolled 3kW scenario and the controlled scenario. The histograms of the
other scenarios are not shown anymore, because their shape will resemble the ones
presented here. The overload criterion has been set to 60% of the nominal value of
the current in the cable. This reflects the fact that typically the MV-D cables are
4.3 Results
63
% of transformers
20
Uncontrolled 3kW
Overloaded : 40 %
15
10
5
0
0
0.5
1
1.5
2
2.5
3
% of transformers
20
Uncontrolled 10kW
Overloaded : 49 %
15
10
5
0
0
0.5
1
1.5
2
2.5
3
% of transformers
20
Controlled
Overloaded : 20 %
15
10
5
0
0
0.5
1
1.5
2
2.5
Transformer Utilization Factor (%)
3
Figure 4.8 – Distribution of MV/LV transformer loadings. The thin black line behind the
histogram bars denotes the situation with 1 % growth without EVs. The color change in
the histogram bars denote the values where the threshold value has been exceeded.
64
4 Network impacts and cost savings of controlled EV charging
30
Uncontrolled 3kW
Overloaded : 13 %
20
% of Cables
% of Cables
30
10
0
0
30
60
90
120
40
80
120
160
200
30
Controlled
Overloaded : 7 %
20
% of Cables
% of Cables
30
10
0
0
10
0
0
150
Uncontrolled 3kW
Overloaded : 42 %
20
30
60
90
Cable loading (%)
120
(a) MV distribution cables
150
Controlled
Overloaded : 25 %
20
10
0
0
40
80
120
Cable loading (%)
160
200
(b) MV transmission cables
Figure 4.9 – Distribution of MV cable loadings.
laid out in a ring structure (see Fig. 4.1) and in case of a fault somewhere in the
ring, the net opening will be closed and the cable on one side of the ring has to serve
the entire ring to comply with the N-1 criterion.2 In such mode of operation, cable
loadings up to 120% are tolerable for some time.
It can be seen that the number of overloaded MV-D cables is much lower than
number of overloaded MV/LV transformers. This leads to the conclusion that there
is much more ‘room’for extra load on the MV distribution cables than on the transformers. A possible explanation for this observation is the following, which is a result
of our method of bookkeeping: in a typical MV distribution ring as depicted in Fig.
4.1, the cable segments between the MV/LV transformers are counted as separate
cables. It is then likely that the last segments of the cables have a much lower load
because they are effectively serving only one or even no transformer at all in normal
operation. One could speculate, however, that the differences are also the result of
the fact that due to much higher labour costs of installing a new MV cable (since
they generally are underground cables), DSOs have typically installed cables with
much higher capacity to avoid costly reinforcement, which involves digging along the
entire length of the old cable.
Fig. 4.9(b) shows the expected loadings of the MV transmission (MV-T) cables.
Here, the overload criterion is 68% of nominal value rather than 60% in the case
of MV-D cables. This reflects the fact that usually there are two or more parallel
MV-T cables, that serve as backup to comply with N-1 standards, see [73] for more
details. Most notably, the amount of overloaded cables is much higher in all scenarios
than in the case with MV-D cables. Apparently, the MV-T cables have much less
spare capacity, which is expected since they transfer the power for an entire MV
distribution ring. It should be noted that if one considers the distributions of all
MV cables, they strongly resemble the distributions of the MV-D cables, since there
are far less MV-T cables.
2 The N-1 reliability criterion states, loosely speaking, that all loads should still be served after
failure of one asset.
4.3 Results
65
Table 4.3 – Overview of the fraction of transformers/cables whose threshold value has been
exceeded for the different charging scenarios. The values denote the situation in 2040 with
roughly 75% of all households owning an EV.
Property
MV/LV Transformers
MV Distribution Cables
MV Transmission Cables
HV/MV substations
Threshold No EVs Unc.3kW Unc.10kW Cont.
1.16
18%
40%
49% 20%
0.6
6%
13%
15%
7%
0.68
22%
42%
46% 25%
varies
35%
61%
66% 42%
In both the MV-D and the MV-T cables, the effect of applying the EV charge
control is imminent: much fewer overloaded cables in the controlled scenario. It will
be clear that this could lead to a dramatic reduction in investment costs that would
be needed to accommodate new EV loads and/or growth of the household electricity
demand.
4.3.3
HV/MV substations
Fig. 4.10 shows the distribution of the loading of the HV/MV substations in the
different EV charging scenarios. The overloading criterion depends on the configuration of the specific substation by taking into account the N-1 criterion and has
been determined for each substation individually. Usually there are a number of
actual transformers in one substation, so if for example the substation has three
40MVA transformers, the overload criterion is 80MVA, because that is the highest
load than can be supplied by two of the three transformers. In the Fig. 4.10 a
utilization factor of one hence denotes the safe N-1 capacity of the substation. The
amount of overloadings is clearly highest for the HV/MV substations compared to
any other types of grid asset we considered. A possible explanation for this could
the fact that, since a HV/MV transformer requires by far the highest cost to install
(or replace) and its costs are dominated by material rather than labor costs, DSOs
usually choose not to build much extra capacity. A new transformer is only installed
in a substation when the safe N-1 capacity threatens to be exceeded. This also implies that a substation that is counted as overloaded does not have to be replaced
altogether, but its capacity can be enhanced by installing another transformer with
accompanying switchgear. The most important results in terms of asset overloadings
are summarized in table 4.3.
4.3.4
Economic figures
Table 4.4 shows the NPV for all the scenarios compared to the baseline scenario
without EVs but with a yearly 1% load growth. The potential cost savings for controlled charging of EVs are around 20 % compared to the uncontrolled scenarios. It
is emphasized that the absolute numbers are somewhat arbitrary because they only
represent a part of the networks in a part of the Netherlands. It can be seen that
the energy loss costs dominate the total figures. It should be realized, however, that
66
4 Network impacts and cost savings of controlled EV charging
% of Stations
40
EVs Uncontrolled 3kW
Overloaded : 61 %
30
20
10
0
0
0.5
1
1.5
2
2.5
3
% of Stations
40
EVs Controlled
Overloaded : 42 %
30
20
10
0
0
0.5
1
1.5
2
2.5
Station Utilization Factor (%)
3
Figure 4.10 – Distribution of HV/MV transformer station loadings
this can partly be explained by the fact that the replacement costs are spread over
a large period which lies largely outside the time-horizon. Although in the NPV
figures the replacement costs are dwarfed by the energy loss costs, the differences
between the scenarios are distributed more evenly across the various cost components. Since we are interested in the potential cost savings due to controlling the EV
charging process, we discuss this point further by comparing the controlled and the
uncontrolled 3kW scenario. From Table 4.4 it can be deduced that the cost savings
in the controlled scenario are approximately 60% due to lower replacement costs
and 40% due to lower energy losses. Fig. 4.12 shows specifically what components
are responsible for the difference in costs between the uncontrolled 3kW and the
controlled scenario. The cable replacements and cable losses are the main cause for
the cost difference. Transformer losses, replacements and station replacements are
similar in size. Station losses actually have a slightly negative contribution: they are
higher in the uncontrolled scenario due to the replacements which lower the energy
losses.
To get some insight in the pattern of required investments, Fig. 4.13 shows the
annual cash flows defined by Eqs. 4.6 and 4.8 for the different cost components for
the uncontrolled 3kW scenario. Whereas the energy loss costs increase steadily, the
investments in replacements show a distinct peak around the year 2025.
Fig. 4.14 shows the total annual cash flows for all the scenarios. The investment peak present in the uncontrolled charging scenarios is absent in the controlled
charging scenario. Hence, we can conclude that the replacement cost differences as
reflected in Table 4.4 are almost solely caused in the period between 2015 and 2035.
It is again emphasized that the annual payments as a result of these replacements
(Eq. 4.7) will last much longer.
4.3 Results
67
1.6
1.4
Net Present Value
1.2
MV/LV Trans. Losses
MV Cables Losses
HV/MV Stations Losses
MV/LV Trans. Replacements
MV Cables Replacements
HV/MV Stations Replacements
1
0.8
0.6
0.4
0.2
0
No EVs
EVs Controlled
EVs 3kW
EVs 10kW
Figure 4.11 – Break-down into different components of the NPV for all scenarios.
Fraction of Difference
0.4
0.3
0.2
0.1
0
CR
CL
TR
SR
TL
SL
Figure 4.12 – Break-down of the difference in NPV between in the uncontrolled 3kW and the
controlled scenario. Colors match with Fig. 4.11. C, T, and S stand for cables, transformers
and stations, respectively. L and R stand for losses and replacements.
Table 4.4 – Overview of the net present value (NPV) of the different scenarios compared
to the base case scenario without electric vehicles.
Scenario
Total Replacements Energy Loss
No EVs (absolute)
265M e
25M e
240M e
No EVs
100 %
100 %
100 %
EVs controlled
105 %
113 %
104 %
EVs uncontr. 3 kW
124 %
240 %
112 %
EVs uncontr. 10 kW 129 %
257 %
116 %
68
4 Network impacts and cost savings of controlled EV charging
3.5
3
MV/LV Transformers Losses
MV Cables Losses
HV/MV Stations Losses
MV/LV Transformers Replacements
MV Cables Replacements
HV/MV Stations Replacements
Cash Flow
2.5
2
1.5
1
0.5
0
2015
2020
2025
Years
2030
2035
2040
Figure 4.13 – Annual cash flow for different components for the uncontrolled 3 kW scenario.
The cash flows are relative to the yearly cash flow in 2010.
3.5
3
EVs uncontrolled 3kW
EVs uncontrolled 10kW
EVs controlled
No EVs
Cash Flow
2.5
2
1.5
1
0.5
0
2015
2020
2025
Years
2030
2035
2040
Figure 4.14 – Total (replacements + losses) yearly cash flow for all scenarios.
4.4 Conclusions
4.4
69
Conclusions
A method to assess the network impacts of EV charging and its economic consequences has been presented in this chapter. This approach, that combines recent
network load measurements and modeled future EV profiles, has been used to analyze a large number of operational distribution networks and calculate the resulting
costs for energy losses and reinforcement of network components. Because our analysis covers a large set of different distribution networks consisting of thousands of
different grid assets, our findings should be interpreted to reflect system level figures
rather than predictions for individual networks or assets. Within the validity of our
assumptions, however, a number of important conclusions can be drawn.
Most importantly, it has been shown that controlled charging of EVs leads to
a significant reduction of overloaded network components compared to the uncontrolled charging scenarios. There are, however, marked differences between the impacts on various parts of the networks. The most severe impact, in terms of percentage of overloaded components, is to be expected on the HV/MV transformer station
level, followed by MV/LV transformers and MV cables. Regarding the MV cables,
the effects on MV transmission cables are more severe than on the MV distribution
cables.
The economic analysis shows that the expected NPV of all network reinforcements and energy losses differs approximately 20% between the uncontrolled and
controlled charging scenarios. The cost savings due to controlled charging are most
pronounced on the MV cable level. Furthermore, the energy loss costs were found
to be the most important component of the NPV and they were found to differ
only moderately between the controlled and uncontrolled charging scenarios. The
differences in costs between the controlled and uncontrolled scenarios, however, are
approximately 60% due to reduced investment costs and 40% due to lower energy
losses.
The implications of these results for DSOs are not straightforward and depend
largely on the institutional setting the DSO is operating in. One could argue that the
results presented in this study give a strong indication that there is a societal benefit
in allowing DSOs to apply some form of charge control – something that is not allowed in many regulated electricity sectors. In a regulated environment where DSOs
compete through yardstick price regulation, this study furthermore suggests that
if an EV penetration scenario like the one considered in this chapter materializes,
DSOs should anticipate a period with distinctly higher investment levels.
A number of venues for future research can be identified from this chapter. A
logical step to take would be to assess other types of network impacts such as voltage
excursions, reduced lifetimes of components, harmonics, unsymmetrical loads, etc.
In principle, the same method to derive the future load profiles can be applied for
these and other types of network impacts, although additional information will be
required. It would also be interesting to asses the network impacts of charge profiles
that result from different optimization objectives. For example, a situation where
EVs are used for frequency regulation might lead to completely different profiles
with possibly much higher peaks in the network. In addition to this, a stochastic
approach could provide more insights into probabilities of certain network impacts.
70
4 Network impacts and cost savings of controlled EV charging
The economic analyses have shown that although the reduction of the amount of
assets replacements due to controlling the charge process is significant, the resulting
difference in NPV is quite moderate. This demonstrates that it is not easy to capture
the costs of EV charging in a single number. One could argue that this is partly
due to the fact that such costs are not only the result of events happening in the
period under consideration, but also from the current state of the network - a result
of decisions in the past - and future developments that lie far beyond the end of the
considered period. It would be interesting to see if an economic analyses that uses
more detailed financial data and accounting methods on the individual asset level,
would produce different insights. One could speculate that such insights could lead
to a shift in optimization objectives that a DSO aims to fulfill with the control of
EV charging or with regard to grid expansion strategies.
The policy implications of this study are not straightforward and offer another
possibility for future research. Nevertheless, despite the fact that we have only focused on a single aspect of the broad concept of smart grids, our study strengthens
the confidence that the deployment of an information and communication infrastructure to enable more flexible and intelligent electricity grids is worthwhile.
Chapter 5
Impacts of controlled EV
charging on cross-border
electricity flows
This chapter is based on [76]
5.1
Introduction
The continuing growth of the share of renewable energy sources (RES) increasingly
poses a challenge to the traditional operation of the power system. In particular
the intermittent character of wind and solar PV generation, combined with wind
and irradiation forecast uncertainties, poses a tough problem that could hamper
the integration of RES. A few, possibly complementary, solution approaches are
often proposed in this context, see e.g. [77]: a large enough interconnected grid
to smoothen out fluctuations in renewable energy production, large-scale electricity storage, (fossil fuel fired) back-up plants, and responsive demand of electricity.
Essentially these options fall in two main solution approaches: inter-temporal vs.
inter-locational arbitrage.
Large-scale hydro storage options are not economically available everywhere
while thermal back-up plants are costly and polluting. Regarding the option of
demand response, electric vehicles (EVs) are a good candidate to fulfill an important role as flexible loads, because typical EV battery sizes and daily driving distance
usually allow the charging process to be controlled and/or postponed for a few hours
up to several days. Because of this potentially large source of flexible electricity demand, controlled charging of EVs has received much attention lately in a variety of
different contexts [50]. The natural combination of RES and EVs was first described
extensively in [42] and was further explored in e.g. [46]. Other work has focused
on relieving distribution network impacts (see e.g. [59]), using EVs for balancing
purposes (or other ancillary services) [78], EVs reacting on anticipated day-ahead
electricity prices (e.g. [39]) but also in unit commitment models [66].
71
72
5 Impacts of controlled EV charging on cross-border electricity flows
The other category of approaches to ease the integration of RES focuses on
increased interconnection capacity between different power systems. A number of
studies have reported on how large interconnected power systems help to smoothen
out the variable production profiles of wind and PV. In the European context an
important contribution was given by [79]. In this report, it was shown how very
high RES penetration together with heavily increased interconnection capacity has
the potential to almost complete curb carbon emissions of the electricity sector.
Although the focus was on cross-border transmission, it was also acknowledged that
demand response could play a key role in dealing with variability of renewables.
Other studies were carried out more from a perspective of meteorological timeseries, and it was shown in e.g. [80] and [81] that even with an infinite-capacity
transmission network, a very large storage capacity or back up generation to cover
periods with low wind and sun are still needed, mainly in winter periods with blocking high pressure systems over Europe. In [82], a European grid model and future
(renewable) generation capacities scenarios have been used to show how cross-border
electricity flows will change. The general picture that emerges from these studies
is that, indeed, interconnection of power systems can play an important role in a
sustainable Europe, but the cross-border transmission capacity that is needed is
very large, and additional measures may well prove to be a cost-efficient addition to
transmission capacity.
Despite the large body of literature related to RES and EV integration issues, a
number of things seemed to have received less attention. For example, many of the
studies that investigated the relation between EVs and RES have focused on regional
or national electricity systems and ignored the dependence of neighboring power
systems. The large-scale oriented studies have, on the contrary, often ignored EVs
as potential responsive demand. Next to this, many RES integration studies have
primarily focused on wind, since the share of solar energy only grew to substantial
numbers in the past few years.
The goal of this work is to investigate to what extent controlled charging of
EVs and extra transmission capacity are related to each other in systems with a
high share of renewable energy production. Intuitively, since they are both forms of
arbitrage, the two options are often seen as substitute technologies: increasing the
one would lead to a lower value of the other. If they were complements, they would
actually strengthen each other. Hence, this study aims to shed more light into the
question to what extent EV control impacts the needs for cross-border transmission.
We do so by using a unit commitment model where the EV charging is explicitly
taken into account as decision variable. We then apply our model to a case study of
a conceptual two-node system in order to get qualitative insights in dispatch profiles,
generation costs and transmission needs. We aim to identify relevant conditions that
influence the extent of substitution between the two technologies and we explore
sensitivities of the results to various parameters. A second case study using the
same model is described in more detail in [76] and [83]. This case study focuses
on the full European electricity system using realistic transmission and generation
portfolio scenarios.
Results of this chapter show that the demand for arbitrage is key in understanding the value of transmission and EV control. Under modest RES and transmission
5.2 Model formulations
73
scenarios, EVs and transmission can be seen as partially substitute technologies, although the value of both technologies is mostly independent of the other. However,
when RES penetration increases to very high levels, both control and transmission
become necessary and the two technologies actually become complementary, i.e. the
value of both technologies simultaneously is higher than the sum of their values
individually.
5.2
Model formulations
In the following section we will describe our least-cost unit commitment model. The
EVs are an explicit part of the models, so we will start this section by describing
the EV charging model.
5.2.1
EV charging model
To include EVs in the unit commitment model, we need information on the charging
needs of the EVs. These are dictated by their driving needs in combination with
technical vehicle parameters such as their battery capacity. Currently, the number of
EVs on the road is very modest, so we model EV driving patterns based on current
driving data which represent gasoline powered vehicles. Hence, the fundamental
assumption that we make by using conventional vehicle driving data is that EV
driving patterns will be similar to those of conventional gasoline vehicles.
EV charging can be described with a linear state equation that relates the battery
state-of-charge (SoC) to the charging power and discharges due to driving. The SoC
(expressed in terms of energy) of vehicle i at time-step k is given by
EEV,ik+1 = EEV,ik + ηc PEV,ik − dik
∀i, k
(5.1)
Here ηc represents the charging efficiency and −dik represents the discharges
due to driving. The technical limits of EEV,ik and PEV,ik are denoted by EEVmin ,i ,
EEVmax ,i , PEVmin ,i and PEVmax ,i . In the remainder of this chapter we assume that
vehicles cannot deliver energy back to the grid, hence PEVmin ,i = 0. The upper limit
PEVmax ,i is either dictated by the grid connection, the inverter limits or even charge
acceptance by the battery material. We are, however, not interested in the nature
of the limiting factor and assume a constant value for each vehicle.
5.2.2
EV data
The driving data originates from a large mobility survey performed in the Netherlands [29]. This data is also used and described more extensively in [28]. The data
provide trip lengths, departure and arrival times, trip durations, trip destinations
etc. Furthermore, we assume a constant kWh usage per km driven (0.2 kWh/km), so
the discharges due to driving −dik follow in a straightforward way from the driving
data.
We assume the technical vehicle parameters to be PEVmax ,i = 3kW, PEVmin ,i = 0,
EEVmax ,i = 24kWh, and we let these vary randomly within 10% of these average
74
5 Impacts of controlled EV charging on cross-border electricity flows
values to introduce some differences between the vehicles. Furthermore, we assume
lossless charging, hence ηc = 1 in Eq. 5.1.
5.2.3
Charging scenarios
We will consider two charging scenarios: 1) optimal charging, where PEV,ik is an optimization variable to minimize total generation costs, and 2) uncontrolled charging,
where vehicles only charge at home with a constant PEV,ik after the last arrival at
home. The uncontrolled charging scenario is described in more detail in [28] and
chapter 2 of this thesis. Briefly described, in the uncontrolled charging scenario the
charging power is given by PEV,ik = PEVmax,i between the moment of the last arrival
at home until the battery is fully charged.
For the optimal charging scenario, the time-dependent EV charging power PEV,ik
will be an optimization variable in a unit commitment model. Hence, this model
finds the right moments to charge the EVs in order to minimize the total electricity
generation costs, while still respecting the driving needs of the EVs. The driving
data thus enter the unit commitment problem through the dik term in Eq. 5.1. The
smaller this term in comparison with the battery size (EEVmax ,i ), the less flexibility
there is in postponing the moment of charging.
5.2.4
Typical EV fleet
In principle, every single EV can be taken into account in the optimization, but the
number of variables would become enormous, so some sort of aggregation to lump
many individual EVs in a single one is needed. We divide the EV fleet in 25 typical
EVs (since this was computationally tractable), by means of a K-means clustering
algorithm that is described in [39] for the same purpose. A small deviation from [39]
is that we choose to work with equal cluster sizes, which allows us to simply multiply
all the single EV equations with a single scaling factor that represents the ratio of
total EVs in the system divided by the amount of EVs in the model. To obtain
equal cluster sizes, we use a pragmatic way to redistribute entries in overpopulated
clusters to an underpopulated cluster with the closest Euclidean distance from it.
This procedure is describe in more detail in appendix D.
Scaling the aggregated vehicle demand to represent national levels is done by
invoking the total amount of passenger car vehicles per country given in [84] and we
assume an EV penetration of 25%. The largest differences between the vehicles are of
course the driving patters (the dik in 5.1), and as explained above, they follow from
the dataset described in [28]. We thus use Dutch data and hereby we assume that
the driving patterns (except for the distances) do not differ much between countries.
5.2.5
One node unit commitment model
The basic unit commitment model is based on [85], chapter 2.7. We have adapted the
model slightly by ignoring reliability constraints and adding pumped hydro equations
and renewable energy generators.
5.2 Model formulations
minimize
Nk ∑
NG
∑
PG,nk ,unk ,ynk ,znk ,PHjk
75
PG,nk M Cn + ynk SU Cn + znk SDCn
(5.2)
k=1 n=1
subject to
unk PGmin ,n ≤ PG,nk ≤ unk PGmax ,n
∀n, k
(5.3)
PG,nk−1 − PG,nk ≤ RDn
∀n, k
(5.4)
PG,nk − PG,nk−1 ≤ RUn
∀n, k
(5.5)
ynk − znk = unk − unk−1
∀n, k
(5.6)
PHmin ,j ≤ PHjk ≤ PHmax ,j
∀j, k
(5.7)
Hmin,j ≤ Hjk ≤ Hmax,j
{
Hjk−1 + Hin,jk + ηH PHjk
Hjk =
Hjk−1 + Hin,jk + η1H PHjk
∀j, k
(5.8)
∀j, k
(5.9)
∀k
(5.10)
NG
∑
n=1
PG,nk −
NH
∑
if PHjk ≥ 0
if PHjk < 0
PHjk = PD,k
j=1
We give a brief description of the meaning of the equations: the objective function
5.2 represents the costs of all plants NG and all time-steps Nk and is the sum of
marginal costs M C (linear with output Pnk ) and the start-up SU C and shut-down
costs SDC. The binary variables u, y and z represent if a unit n is on-line, in start-up
mode and in shut-down mode, respectively. Eq. 5.3 denotes the power limits Pmin,n
and Pmax,n of the generators. Eqs. 5.4 and 5.5 denote the ramping limits RD (down)
and RU (up) of the generators. The start-up and shut-down logic is expressed in
Eq. 5.6. The pumped hydro power PH for all units j (positive for pumping, negative
for producing) has limits PHmin and PHmax ; these are expressed in Eq. 5.7. The
hydro reservoir level H has to be within limits Hmin and Hmax given by constraint
5.8. The relation between hydro power and reservoir level in Eq. 5.9, where one also
observes the inflow term Hin . Finally, the load balance equation is given in Eq.
5.10, which says that the sum of generation equals demand PD . The optimization
problem given by Eqs. 5.2 to 5.10 is mixed-integer linear programming problem. It
has been programmed in Matlab and solved with the IBM ILOG CPLEX solver [86].
5.2.6
Multi node unit commitment model with flexible EV
load
The full model for a multi node systems with controllable EVs reads
76
5 Impacts of controlled EV charging on cross-border electricity flows
minimize
Nk ∑
NG
∑
PG ,u,y,z,PH ,PEV
PG,nk M Cn + ynk SU Cn + znk SDCn
(5.11)
k=1 n=1
subject to
unk PGmin ,n ≤ PG,nk ≤ unk PGmax ,n
∀n, k
(5.12)
PG,nk−1 − PG,nk ≤ RDn
∀n, k
(5.13)
PG,nk − PG,nk−1 ≤ RUn
∀n, k
(5.14)
ynk − znk = unk − unk−1
∀n, k
(5.15)
PHmin ,j ≤ PHjk ≤ PHmax ,j
∀j, k
(5.16)
Hmin,j ≤ Hjk ≤ Hmax,j
{
Hjk−1 + Hin,jk + ηH PHjk
Hjk =
Hjk−1 + Hin,jk + η1H PHjk
∀j, k
(5.17)
∀j, k
(5.18)
PEVmin ,i ≤ PEV,ik ≤ PEVmax ,i
∀i, k
(5.19)
EEVmin ,i ≤ EEV,ik ≤ EEVmax ,i
∀i, k
(5.20)
∀i, k
(5.21)
∀k
(5.22)
∀k
(5.23)
if PHjk ≥ 0
if PHjk < 0
EEV,ik+1 = EEV,ik + ηc PEV,ik − dik
NG
∑
n=1
PG,nk −
NH
∑
j=1
PHjk = PD,k +
N
EV
∑
PEV,ik
i=1
|FLk (PG,nk , PH,jk , PEV,ik , PD,k )| ≤ KL
The EV equations thus enter the unit commitment problem as an extra set of
constraints. They express that EV charging power needs to be within limits PEVmin
and PEVmax and the battery state-of-charge (SoC) needs to be within limits EEVmin
and EEVmax . The charging power and SOC are related through Eq. 5.1 and this is
where the driving data dik enter the problem.
Furthermore, in the multi-node model there is an additional set of constraints
related to the line capacities. These extra constraints (Eq. 5.23) express in a general
form that line flows FL need to be within limits KL . The exact form of this constraint
depends on how the power flows are modeled. In the lossless two node example that
we consider it is simply the surplus of generation minus demand at the one node
that flows to the other node. In a DC load flow approximation one usually uses
a linear relation of the form FL = HP where H (not to confuse with the hydro
levels in Eqs. 5.17 and 5.18) represents a matrix with so-called distribution factors
that relate nodal injections to line flows, calculated on the basis of electrical line
properties. In a true AC load flow formulation Eq. 5.23 will be non-linear and the
optimization problem becomes very difficult to solve. In an approximation method
to take losses into account, one could add an extra term to the objective function
that accounts for the losses.
By comparing constraints 5.16 to 5.18 with 5.19 to 5.21, one observes an interesting analogy. The equations for the pumped hydro plant are the same as for the EVs,
except for the discharge due to driving term −dik instead of the inflow term Hin,jk .
In this analogy, the EVs can be considered as a series of leaky (negative inflow)
5.3 Simulation setup
77
Table 5.1 – Overview of generation mix for the two different nodes. Installed capacity per
fuel type in GW.
Type
North South
Lignite
8
8
Coal
13
13
Gas
25
25
Wind Onshore
52
0
Wind Offshore
18
0
Solar
0
60
Hydro RoR
1
1
Hydro Pumped
20
20
hydro reservoirs that are (in our formulation) not capable of producing power, but
whose reservoir levels need to be maintained within limits. How fast the reservoir is
leaking thus depends on the driving patterns.
5.3
5.3.1
Simulation setup
Two node conceptual system
As a case study we perform simulations on a hypothetical two-node system that
reflects the integration challenges of German RES. The installed capacities for the
different generation technologies are based on German data, see Table 5.1. The idea
behind this scenario is to take the 2025 scenario for Germany from the European
Network of Transmission System Operators for Electricity (ENTSO-E), which gives
the expected generation mix and the demand time series for that year. We then
divide Germany in two nodes and assign all wind production to the northern node
and all solar production to the south. Clearly, in reality wind and solar are more
evenly distributed, but the point here is to get more or less realistic profiles of wind
and solar production, fossil fuel generator capacities and retain the daily and seasonal
time-correlation between wind and solar. Furthermore, we scale the EV demand data
according to the German figures on passenger car usage. We emphasize that the aim
of this case study is not to obtain quantitative conclusions for the German system,
but to gain qualitative insights in the relations between transmission capacity and
EV control.
5.3.2
Generator parameters
The capacities of the various generation technologies are listed in Table 5.1. Since
data on the individual plants was not readily available, we divided the coal, lignite and gas plants in 1 GW units to run the unit commitment model. The plant
characteristics are modeled based on some general assumptions, e.g. that gas plants
have larger flexibility than coal and gas plants. The start-up costs for the fossil fuel
generators have been determined by assuming a fixed start-up time per generation
78
5 Impacts of controlled EV charging on cross-border electricity flows
Table 5.2 – Overview of generator parameters. Individual fossil fuel plant marginal costs
vary within 10 % of the average values.
generator type PGmin PGmax
RU
RD
SU C
SDC
MC
(MW) (MW) (MW/h) (MW/h)
(e)
(e) (e/MWh)
Coal
300
1000
300
300 54000 ± 3600
0
30 ± 2
Lignite
300
1000
300
300 45000 ± 3600
0
25 ± 2
Gas
200
1000
1000
1000 12000 ± 1000
0
60 ± 5
Wind
2
Solar
0
Hydro
3
VOLL
1000
type and setting the start-up costs equal to
SU Cn = start-up time · M Cn · PGmin ,n
(5.24)
For coal and lignite, the fixed start-up times have been set to 6 hours, for gas to 1
hour. All generator data is shown in Table 5.2. We do not consider shut-down costs
of plants, although the description in the unit commitment model allows us to do
so. Renewable generators can ramp infinitely fast but their maximum output varies
with time according to the wind and solar time-series that we use. It is important
to emphasize that we do not model the renewable generators as negative load, so
curtailment of them is possible. This can sometimes be attractive in order to avoid
having to shut down and restart a fossil fuel plant. A virtual plant with a marginal
cost equal to the value of loss of load (VOLL) has been included to ensure a feasible
solution of the unit commitment. A small amount (2 GW) of run-of-river hydro
(RoR) that was assumed to be non-dispatchable has been included as negative load.
The overall efficiency (sequence
√ of pumping and generating) of the pumped hydro is
assumed to be 0.75, so ηH = 0.75.
5.3.3
Wind and solar time series
The meteorological solar data come from [87] and the wind data originate from
[88]. The method to convert the meteorological time-series to power output has
been described in [83]. Obviously the output of wind and solar depend on the
weather conditions and large variations can be expected between the seasons. Fig.
5.1 shows the output (curtailment not accounted for) of wind and solar throughout
the year. We run the model for all weeks of the year to get some insights in the
seasonal differences. Furthermore, Fig. 5.1 shows that the maximum wind output
is roughly 65GW (compared to the 70GW installed), whereas the maximum solar
output amounts to little more than 45GW (60GW installed).
5.3.4
Other simulation details
Each simulation spans a whole year, and we run the model in periods of one week.
The initial settings of the generator outputs for the next week are passed on from
5.4 Results
79
70
Wind
Solar
Wind Smoothed
Solar Smoothed
Wind + Solar Smoothed
60
Generation(GW)
50
40
30
20
10
0
0
2
4
6
Time (months)
8
10
12
Figure 5.1 – Renewable energy generation time series
Table 5.3 – Overview of relevant simulation results for 6 scenarios.
Quantity
Total Generation Costs (EUR)
Lignite Costs (EUR)
Coal Costs (EUR)
Gas Costs (EUR)
Hydro Generation (MWh)
Wind Curtailment (MWh)
Solar Curtailment (MWh)
0GW,
Contr.
1.46e10
3.27e9
5.71e9
5.65e9
4.92e6
4.60e6
1.99e5
10GW,
Contr,
1.36e10
3.41e9
6.15e9
4.29e3
4.54e6
1.17e6
2.74e3
Scenario
0GW,
10GW,
Unc.
Unc.
1.52e10 1.42e10
3.19e9
3.33e9
5.56e9
5.95e9
6.34e9
4.97e9
7.14e6
7.42e6
6.22e6
2.02e6
1.33e6
2.17e5
0GW,
No EVs
1.36e10
3.15e9
5.37e9
4.90e9
6.99e6
1.07e8
1.53e6
10GW,
No EVs
1.26e10
3.31e9
5.73e9
3.51e9
7.39e6
2.55e6
2.53e5
the final setting of the previous week. For the EVs, an additional constraint that the
SOC at the final time-step of the week has to be equal or higher than their initial
SOC. This is to prevent that the batteries are always depleted by the end of the
week. A similar constraint was formulated for the hydro reservoirs. The time-step
of the simulation is 1 hour.
5.4
5.4.1
Results
Dispatch profiles
Table 5.3 shows simulation results for 6 scenarios, a combination of the three EV
scenarios (controlled, uncontrolled and no EVs) and two scenarios for transmission
capacity between the nodes (0 GW and 10 GW). Total generation costs here denote
the yearly dispatch costs of all generators in the system, so there are no fixed costs
included in this number. Dispatch profiles for a typical week in spring for the same
80
5 Impacts of controlled EV charging on cross-border electricity flows
RoR
Lignite
Coal
Gas
Hydro Out
Wind
Solar
60
North
Demand
" + Pump
" + EVs
" + Export
South
Generation (MW)
Generation (MW)
Import
60
Controlled EVs
40
20
0
1
2
3
4
5
6
40
20
0
7
60
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
Time (days)
5
6
7
60
Generation (MW)
Generation (MW)
Uncontrolled EVs
40
20
0
1
2
3
4
5
6
20
0
7
60
60
No EVs
Generation (MW)
Generation (MW)
40
40
20
0
1
2
3
4
Time (days)
5
6
40
20
0
7
Figure 5.2 – Dispatch profiles for different vehicles scenarios without transmission between
the nodes.
RoR
Lignite
Coal
Gas
Hydro Out
Wind
Solar
Import
Demand
" + Pump
" + EVs
" + Export
60
North
South
Generation (MW)
Generation (MW)
Controlled EVs
40
20
0
1
2
3
4
5
6
60
40
20
0
7
60
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
Time (days)
5
6
7
60
Generation (MW)
Generation (MW)
Uncontrolled EVs
40
20
0
1
2
3
4
5
6
60
No EVs
Generation (MW)
Generation (MW)
20
0
7
60
40
20
0
40
1
2
3
4
Time (days)
5
6
7
40
20
0
Figure 5.3 – Dispatch profiles for different vehicles scenarios with 10 GW transmission
capacity between the nodes.
5.4 Results
81
6 scenarios are shown in Fig. 5.2 and 5.3. We discuss the results in increasing order
of complexity, so we start with the situation without EVs and without transmission
- the bottom graphs of Fig. 5.2.
The most prominent feature of the dispatch profiles is probably the daily cycle
of solar production and the much flatter wind profile. Of course, the latter is not
always the case, but in this particular week in spring it is. We observe in the dispatch
profile how during the periods with high renewables the thermal generators need to
be shut down or ramped down. During the moments when demand is high and there
is no solar or wind production, gas plants and pumped hydro plants also produce.
One can also observe that hydro reservoirs are filled at the renewable production
peaks.
In the case without transmission and uncontrolled EVs, we notice a distinctly
higher evening peak caused by EV charging. It can be seen that this extra demand
is mainly met by dispatching extra gas plants - the darker profiles of coal and lignite
are almost identical between the case without EVs and with uncontrolled EVs.
The most notable effect of the EV control (top graphs of Fig. 5.2) is to shift the
EV load to the moments when renewable outputs are highest; during windy conditions in the northern node and during sunny conditions in the southern node. One
also observes how EV control effectively shifts generation from the more expensive
gas plants to the cheaper coal plants. Although this makes sense from a cost point
of view (and total costs are what was minimized in the optimization), this leads to
worse outcomes in terms of CO2 emissions. In this chapter we do not further pursue
the issue of CO2 emissions, but appendix C explores the CO2 emissions due to EV
charging in more detail.
Furthermore, since EVs are mostly charged when renewable output is highest,
one should note that in summer this can be at the peak demand of the system.
A possible consequence of this fact, of which the further analysis is outside the
scope of this chapter, is that when the renewable generation is not embedded at the
distribution level, this might pose a challenge to the distribution grid.
Fig. 5.3 shows the dispatch profiles when a 10 GW transmission capacity between
the nodes is in place. We see similar patterns as in the case without EVs, but the
main difference is that the interconnection capacity equalizes the use of the fossil
fuel plants in the nodes. As expected, the flows between the nodes are mainly
present when renewable outputs are high. Interestingly, at first sight the flows do
not seem to differ much between the EV scenarios. Summarizing, one could say that
the inter-locational arbitrage equalizes the thermal generation profiles in the nodes
(the envelope of all blue colors in Fig. 5.3) whereas the EV control (inter-temporal
arbitrage) further shifts load from gas to coal (attempts to flatten the blue envelope).
Table 5.3 shows furthermore that essentially all cost differences between the
scenarios are the results of shifting gas to coal generation plus the fact that for the
same volume of gas generation, more efficient plants are being used. More detailed
analysis showed that lower start-up costs contributed only marginally.
5 Impacts of controlled EV charging on cross-border electricity flows
1.55
Controlled EVs
Uncontrolled EVs
No EVs
Generation Cost (EUR)
1.5
1.45
1.4
1.35
1.3
1.25
0
2
4
6
8
Transmission Capacity (GW)
(a) Generation costs
10
Marginal Change Generation Costs (EUR/GW)
82
0
−0.5
−1
EVs Controlled
EVs Uncontrolled
No EVs
−1.5
−2
0
2
4
6
8
Transmission Capacity (GW)
10
(b) Demand function for transmission
Figure 5.4 – Total generation costs as a function of transmission capacity and marginal
change in generation costs (negative demand function) for transmission capacity for different EV scenarios
5.4.2
Demand function for transmission
Fig. 5.4(a) shows the total generation costs as a function of transmission capacity
both for the controlled and the uncontrolled EV cases. As expected, the total
generation costs decrease as a function of transmission capacity until some point
where the effect saturates. Interestingly, however, the difference between the EV
scenarios does not depend much on transmission capacity. In other words: whether
or not EV management is present does not affect the value of transmission capacity.
Fig. 5.4(a) also shows that, dependent on how much transmission capacity is in place,
EV control leads to the same amount of savings on generation costs as a substantial
amount of transmission capacity.
Based on Fig. 5.4(a) we can also deduce some insights in the socially optimal
amount of transmission capacity. In theory, at the optimum, the marginal value
of the transmission capacity should equal its costs. The derivative of the curves in
Fig. 5.4(a) can be interpreted as the (negative) demand function for transmission
capacity and they are plotted in Fig. 5.4(b). If one unit of transmission capacity
would cost, say 1.4 · 108 EUR/GW, then the intersection of a horizontal line at
−1.4 · 108 EUR/GW with the demand functions would denote the optimal transmission capacity between the nodes. We observe that it differs approximately 1GW
between the uncontrolled and controlled EV cases. So EV control always lowers
generation costs, more or less independent of transmission capacity. Because of this
independence, the optimal amount of transmission capacity only modestly depends
on whether or not EV control is in place.
5.4.3
Further analysis
In the preceding analysis we considered results regarding the interrelations between
transmission capacity and EV control. Now we further explore conditions to which
those interrelations of transmission and EV control are particularly sensitive.
5.4 Results
83
10
Generation cost (EUR)
x 10
EVs Uncontrolled
EVs Controlled
No EVs
2
1.5
1
0.5
0
Hydro, No Trans. No Hydro, No Trans.
Without Transmission
Hydro, Trans.
No Hydro, Trans.
With 10GW Transmission
Cost Difference (%)
20
EVs Uncontrolled
EVs Controlled
15
10
5
0
Hydro, No Trans. No Hydro, No Trans.
Without Transmission
Hydro, Trans.
No Hydro, Trans.
With 10GW Transmission
Figure 5.5 – Percentual difference in total generation cost for EV scenarios in comparison
with no EVs, for different pumped hydro and transmission capacities.
Sensitivity to pumped hydro storage Fig. 5.5 shows how total generation costs
depend on transmission and whether or not pumped hydro storage is in place. In the
bottom graph, only the percentual cost difference with the no EV scenario are shown.
We interpret these graphs by first recalling that there are basically three mechanisms available to deal with the variable renewables: transmission (inter-locational
arbitrage), EV management and pumped hydro storage (both inter-temporal arbitrage). It can be seen that the value of EV control (the difference between the grey
and black bars in the bottom graph) is only markedly lower in the case that there is
both transmission and hydro-capacity in place, although even in this case the cost
difference is substantial. From this we conclude that especially if one of the two
alternative arbitrage mechanisms of transmission and pumped hydro is lacking, the
value of EV management becomes more prominent.
Sensitivity to RES penetration levels Fig. 5.6 shows the total generation costs
for increasing RES penetration. The extra 50% and 100% RES are with respect to
the values quoted in Table 5.1. Shown in the lower part of the same figure are
the percentual differences between the EV cases and the case without EVs. They
can be interpreted as the additional generation costs to accommodate the extra EV
demand.
The figures show that total generation costs decrease only modestly for the higher
RES levels. The percentual extra costs of accommodating the EVs actually increase
84
5 Impacts of controlled EV charging on cross-border electricity flows
10
Generation cost (EUR)
2
x 10
EVs Uncontrolled
EVs Controlled
No EVs
1.5
1
0.5
0
base
+50%
+100%
Without Transmission
base
+50%
+100%
With 10GW Transmission
Cost Difference (%)
25
EVs Uncontrolled
EVs Controlled
20
15
10
5
0
base
+50%
Without Transmission
+100%
base
+50%
+100%
With 10GW Transmission
Figure 5.6 – Total generation costs and percentual difference of the cases with EVs for the
base case and case where RES has been increased by 50% and 100%.
slightly in the uncontrolled scenarios, but they decrease quite significantly for the
controlled case. This can, however, not be regarded as surprising, since the increased
RES penetration makes the demand for arbitrage larger.
The most important point that Fig. 5.6 reveals is, however, the following. We
observe that in the high RES the value of control (the difference between the two
bars in the bottom graph) actually increases with higher transmission capacity. This
is a counter-intuitive result, since the common understanding, and the picture that
emerged from this analysis so far, is that transmission and control are substitute
technologies. One can explain this result as follows: at some point RES production
is so high that we need the inter-temporal arbitrage potential in both nodes to absorb
the excess wind or solar in one node. In this case, one needs the transmission capacity
to use the EV control in the other node. At these very high levels of RES penetration,
the two technologies thus in fact need each other and are complementary.
The value of EV control under different conditions The value of EV control
and its relation to transmission capacity can be analyses by considering the difference
in total generation costs between the uncontrolled and the controlled EV scenarios.
Figs. 5.7(a) and 5.7(b) shows these in both absolute and relative terms as a function
of transmission capacity for two cases: the base case and the high (200%) RES case.
One observes how, in absolute terms, the value of control decreases with transmission capacity, although the decrease in the high RES case is very small. This
5.4 Results
85
9
10
8
Cost reduction by control (%)
Value of control (EUR)
9
Base scenario
High Renewables
7
6
5
4
3
2
0
2
4
6
8
Transmission Capacity (GW)
(a) Absolute value of control
10
8
Base scenario
High Renewables
7
6
5
4
3
0
2
4
6
8
Transmission Capacity (GW)
10
(b) Relative value of control
Figure 5.7 – Absolute value of control (difference in total generation costs) and relative
value of control (percentual reduction in total generation costs) as a function of transmission
capacity
decreasing trend can partially be explained by the fact that the absolute value of
generation costs itself decreases with increasing transmission capacity. By contrast,
Fig. 5.7(b) shows how much EV control reduces generation costs relatively compared
to the uncontrolled EV scenario. Here we can confirm the notion that we found in
the RES sensitivity analysis: for large RES penetration levels, the value of EV control becomes higher with increased transmission capacity. Under such conditions,
EV control and cross-border transmission thus become complementary technologies.
Seasonal dependencies Fig. 5.8 shows how generation costs and the differences
between various EV and transmission scenarios depend on the time of year. These
figures provide additional insights on the value of EV control and transmission and
their seasonal dependencies. When interpreting these results, it is instructive to
recall Fig. 5.1 that shows the wind and solar output throughout the year. In the top
graph of Fig. 5.8 the total weekly generation costs are plotted for the different EV
and transmission scenarios. It can be seen that generation costs are highest in winter,
when demand is highest and the combined generation of wind and solar is not so
high. Loosely speaking, the generation costs reflect how much fossil fuel generation
is needed to meet the demand, so it is not surprising that this is determined by the
residual load (demand minus renewable generation).
The difference in generation costs between the scenarios with and without EVs
(displayed in the second graph of Fig. 5.8) show only some small seasonal dependence for the controlled cases, and the effect of transmission on this dependence is
negligible. For the controlled cases, the extra generation costs are lowest in spring
and summer, due to the fact that EV control is especially effective to handle the
more periodic solar generation patterns.
The value of 10GW transmission capacity can be determined by considering the
difference in generation costs with and without the 10GW transmission capacity.
86
5 Impacts of controlled EV charging on cross-border electricity flows
Costs (M EUR)
500
No EVs, 0GW Trans.
No EVs, 10GW Trans
Contr. EVs, 0GW Trans.
Contr. EVs, 10GW Trans.
Unc. EVs, 0GW Trans.
Unc. EVs, 10GW Trans.
Generation costs
400
300
200
1
2
3
4
5
6
7
8
9
10
11
12
8
9
10
11
12
8
9
10
11
12
8
9
10
11
12
Cost diff. (M EUR)
60
Contr. EVs, 0GW Trans.
Contr. EVs, 10GW Trans.
Unc. EVs, 0GW Trans.
Unc. EVs, 10GW Trans.
Costs of EV charging
40
20
1
2
3
4
5
6
7
Cost diff. (M EUR)
40
20
0
1
Cost diff. (M EUR)
Contr. EVs
Unc. EVs
No EVs
Value of transmission
20
2
3
4
5
6
7
0GW Trans.
10GW Trans.
Value of EV control
10
0
1
2
3
4
5
6
7
Time (months)
Figure 5.8 – Weekly generation costs, costs of EV charging, value of transmission and value
of EV control for all weeks of the year. All graphs are smoothed by using a moving average
filter with a window of 5 weeks.
5.4 Results
87
The third graph in Fig. 5.8 shows this quantity for the whole year. For most of
the year, the value of transmission is lower when there are controlled EVs - this
is in line with the conclusions from Fig. 5.4. Also, there is a distinct peak in the
value of transmission around month 10. This is exactly when the combined wind
and solar generation was at its highest (see Fig. 5.1). Interestingly, at this peak the
value of transmission capacity did not depend on EV control anymore. Actually,
the unsmoothened graphs (not shown here) revealed that the value of transmission
was even slightly higher in the controlled EV case. More detailed analysis of these
situations showed that in this case wind generation in the northern node was considerably higher than demand in that node and EVs in the other node were able
to increase their demand, thereby replacing fossil fuel. So in excess wind and/or
solar generation the combination of transmission and EV control is beneficial and
the ability to control EVs in fact adds to the value of transmission. This shows once
more the complementary nature of the two technologies for those situations. In the
base case RES scenario these periods are quite rare, but from the RES sensitivity
analysis we draw a similar conclusion.
The value of EV control, given by the difference in EV charging costs between
the controlled and uncontrolled scenario, is depicted in the bottom graph of Fig.
5.8. This number is highest in spring and summer, when there is a reasonably balanced mixed of wind and solar generation. In these situations the EVs are optimally
benefiting from renewable production. The reason that the value of EV control fluctuates has to do with the price differences between gas, coal and lignite. Sometimes,
dependent on the combination of inelastic electricity demand and renewable generation, EV control leads to a shift from gas to coal, which results in a large price
difference. On other moments, the shift is from coal to lignite, which has a much
smaller difference and hence the shifting of load has a lower value associated with
it.
A conceptual interpretation of the results We consider it instructive to illustrate the results on the complementarity of EV control and transmission capacity
using a schematic diagram of a hypothetical two node system. Fig. 5.9 shows the
dispatch profiles for this system. There are four time-steps in which the demand
varies between one and two. There are two dispatchable generation technologies
(say, coal and gas) and one renewable technology (say wind) with zero marginal
cost. Furthermore, there is an opportunity for demand response to move one block
of demand per node. For four different wind situations, the diagrams indicate the
value of transmission (when the two nodes become one), the value of control (when
one block of demand is shifted) and the value of both transmission and control. In
case 5.9(a), when there is only little wind, there is no value in any of the technologies.
In case 5.9(b) there is more wind and in the base scenario wind needs to be curtailed
(indicated by the white blocks below the x-axis). In this case both transmission and
control have a value: they effectively displace the use of gas by coal and prevent curtailment. There is, however, no extra value in the combination of the two. In case
5.9(c), the wind in the two nodes produces at the same moments and one observes
that there is no value in transmission, only in control. Again, the combination of
transmission and control adds no extra value. In the final case 5.9(d) where there is
88
5 Impacts of controlled EV charging on cross-border electricity flows
(a) Medium renewable
output, spread in time.
(b) High renewable output, spread in time.
(c) High renewable output, coinciding.
(d) Very high renewable
output, only in node 1.
Figure 5.9 – Schematic representation of dispatch results in a hypothetical two node system
for 4 time-steps. VT , VC and VT +C denote the value of transmission, value of control and
value of transmission and control, respectively. Control means that one unit of demand
can be shifted in time. Per unit costs of the black, grey and white technologies are 10, 20
and 0 respectively. Curtailment occurs if a white block is below the x-axis.
5.5 Conclusions
89
only a very high wind production in the one node, we observe that the combination
of transmission and control has a higher value than the sum of the individual values.
In this case, one can consider the two technologies complementary.
Based on these diagrams one can understand the observed results in Figs. 5.6,
5.7 and 5.8 better. It explains why only in the high RES case or in those weeks of
the year with very high RES output there exists a complementarity effect between
control and transmission. It also explains, to some extent, the importance of the
exact timing of the RES output in both nodes. If wind and solar production happen
to coincide, there is less value in transmission. Indeed, in Fig. 5.8 we observe that
the value of transmission is lowest in spring, when wind and solar are more or less
in balance.
5.5
Conclusions
In this chapter we aimed to gain insight in the interrelations between demand response in the form of EV control and transmission capacity between neighboring
nodes. We formulated a least cost unit commitment model that includes EV charging as optimization variables and studied a conceptual two node power system
with a high penetration of renewable energy sources. Below we summarize the most
important findings of this work:
• In our case study, controlled EV charging always reduces generation costs and
hardly depends on how much transmission capacity is in place.
• In the base case scenario, cost savings due to EV control come almost entirely
from a shift from gas to coal. For the remaining gas generation, the more
efficient plants are being used. Fewer start-ups contribute only marginally
to the cost reduction. When RES penetration is very high, the avoidance of
curtailment leads to extra cost reductions.
• Effectively, in our modeled system there are three options for arbitrage:
pumped hydro, EVs and cross-border transmission. The value of EV control
becomes more pronounced if one of the other two are not present.
• When the penetration of renewables is higher than the base case scenario,
transmission capacity and EV management become complementary in the
sense that their combined effect is larger than the sum of their individual effects. Transmission is needed to transport power to where the EVs can absorb
it.
• The value of transmission and control can be largely understood in terms of
the demand for arbitrage. We showed, however, that one cannot solely explain this demand for arbitrage in terms of the RES penetration level. Details
like the timing of renewable generation in neighboring nodes and the relative
magnitudes of different renewable technologies do play an important role.
• One could argue that the extreme cases that we considered are unrealistic, but
in the light of European RES targets for the year 2050 this is not the case.
90
5 Impacts of controlled EV charging on cross-border electricity flows
Moreover, this study stresses the importance of research of fully renewable
power systems, since much conventional knowledge and intuition on traditional
power systems no longer hold.
A number of remarks should be made in reflection of our results and findings.
Clearly the conceptual two node system, though inspired by German figures, is a
model of reality and it is hard to draw quantitative conclusions from it. Therefore, the second case study presented in [76] focuses on a more realistic setting by
considering real European generation and transmission scenarios for the year 2025.
Furthermore, an important assumption that we have made is the assumption of
perfect forecasts of renewable energy production. Loosening this deterministic assumption is a crucial aspect for future research in this area since it will target the
second (next to variability) difficult characteristic of RES: uncertainty. Related to
this is the assumption of perfectly known driving patterns and the willingness of EV
drivers to control their EV charging. More detailed analyses that investigate this
aspect are therefore also recommended. In such analyses that take uncertainty into
account, more advanced optimization and control algorithms need to be applied.
Finally, we consider it worthwhile to include investment costs of new generation in
the analysis. In this study we focused only on variable production costs of a given
portfolio, but from a systems point of view one wants to minimize total generation
costs including fixed costs. Moreover, in such an analysis it will be more natural to
include the vehicle-to-grid concept, since it can reduce the need for back-up generation capacity. Within the scope of the present study, however, the results strengthen
the confidence that an interconnected and intelligent electricity infrastructure is a
promising way to realize a fully sustainable energy system.
Chapter 6
Renewable energy sources
and responsive demand. Do
we need congestion
management in the
distribution grid?
This chapter is based on [89].
6.1
Introduction
Two major technological trends that may play an important role in a transition towards a low-carbon energy system are the introduction of renewable energy sources
(RES) and the advent of electric vehicles (EVs). As the penetration of RES
progresses, electricity will increasingly be supplied by fluctuating, weather-driven
sources like solar and wind energy. This has consequences for the functioning of
the electricity system. One paradigm for dealing with variable RES is the concept
of a smart grid, which may facilitate demand responsiveness. Currently, the priceelasticity of electricity demand is low and the ability to shift load is, especially for
small consumers, limited. EVs, on the contrary, could form a flexible load of significant magnitude. They are therefore capable of reacting to and influencing wholesale
electricity prices. By doing so they have the potential to support renewable energy
sources [46].
However, when fleets of EVs are reacting to wholesale prices, they will do so in
a correlated way, since they react to the same electricity price. When EVs postpone
charging until the electricity price is at its lowest, they will create a peak demand at
that moment. Of course there is a feedback mechanism: when the EVs are expected
to add to demand significantly at that moment, the resulting price increase will be
91
92
Distribution grid congestion management
taken into consideration ex ante and EV demand will spread to some extent, see
e.g. [39]. However, as was for example shown in [38], when the share of variable
renewables increases, electricity demand and the wholesale price become increasingly
decorrelated because the wholesale price depends on the instantaneous load and
production of electricity. In other words, when wind and/or solar generation are
high, there could be relatively low wholesale prices even if electricity demand is
high.
Owing to the potentially large impacts of a high share of EVs in power systems,
a large body of literature on this topic has been developed in recent years. A
broad overview of EV studies was given in [50]. The problem of an aggregator who
minimizes the charging costs of EVs by reacting on time varying prices (potentially
driven largely by renewable energy production) was studied in e.g. [39], and in
combination with providing frequency regulation in e.g. [78]. EVs also received
much attention because of the potential role in supporting renewable energy sources,
e.g. in [42] and [46]. Furthermore, there has been a strong focus on the grid impacts
of EVs and how charging can be controlled in order to alleviate potentially severe
impacts, such as [55], [44], [69] and [59], which gives an overview of such studies.
A common finding of many of these studies is that controlled charging of EVs
can have significant value for various actors in the power system. Only a few studies,
however, have analyzed the problem of satisfying different objectives such as minimizing the cost of vehicle charging, frequency regulation and providing balancing
services, in combination with grid constraints.
In [90], the problem of EV charging under grid constraints was investigated. The
essence of the method that was applied in this paper was first to minimize the cost
of charging, then to evaluate if all grid constraints are satisfied, and if not, then
to communicate available grid capacity to the EV aggregator who, finally, performs
the charging cost optimization again, now including the EV power constraints that
follow from the grid conditions. It was shown that this method is successful in not
exceeding grid limitations and can even be applied to quite complex grid topologies.
[91] investigates how dynamic day-ahead grid tariffs can be used to alleviate
distribution grid congestion caused by electric vehicle charging. In this case, the
exact congestion tariff is set by the distribution system operator (DSO) using forecasts of wholesale prices and EV trips. In this formulation of the problem, the DSO
first mimics the EV aggregator and computes the minimum-cost EV charging profile. Next, he computes the congestion tariff based on the locational marginal price
formulation using the shadow price of grid congestion in the grid. However, this
approach does not take into account the inter-temporal constraints of EV charging.
A somewhat heuristic method for determining the appropriate ‘congestion’ level was
applied to circumvent the problem that arises when inter-temporal constraints are
ignored: if one only charges a congestion fee on the expected time of congestion,
then the EVs will ajust their behavior and congestion will likely occur just before
and after the time with congestion fee.
[92] provides a comprehensive overview of different methods for relieving distribution grid congestion caused by EV charging. Three methods are discussed: 1) a
distribution grid capacity market, 2) advance capacity allocation and 3) a dynamic
grid tariff. Of the two formerly discussed papers, [90] falls in category 1 and [91]
6.2 Problem analysis
93
in category 3. [92] concludes that there is always a trade-off between complexities,
values and risks for the stakeholders. No single strategy has a clear advantage over
the others.
The goal of this chapter is to investigate possible mechanisms for aligning EV
responsive demand with constraints resulting from limited distribution grid capacity
more thoroughly. To this end, we first present a mathematical description of optimal
EV charging in the current situation and analyze the problem that arises when the
share of variable renewable energy increases. We then describe and formalize a
number of congestion management mechanisms that could alleviate the problem. We
analyze the different mechanisms in a case study in which we make use of empirical
data on renewable energy production and driving behavior. We will also comment
briefly on the challenges caused by uncertainty and on the IT requirements of the
different congestion management mechanisms.
This chapter adds a number of novel aspects to the scientific literature. The first
is that, as the chapter title suggests, we focus explicitly on power systems with a high
share renewable energy sources, because especially in these systems price-responsive
demand is expected to play an important role. A central theme of this chapter
is that the influence of variable RES on the electricity price makes the need for
congestion management in the distribution grid more urgent. Secondly, we present
a mathematical formalization of different distribution grid congestion management
methods in relation with EV charging, which may become the single largest source of
flexible electricity demand at household level. In addition, we perform a numerical
case study in order to compare the performance of the different congestion management mechanisms, producing new insights into the advantages and limitations of
the various methods. Because the case study is based on empirical data of driving
patterns, electricity prices, renewable energy production and network load, we are
able to evaluate the congestion management methods in realistic situations.
6.2
6.2.1
Problem analysis
The need for congestion management due to the weakening corellation between wholesale electricity prices
and demand
In an electricity system without much variable renewable energy, electricity prices
rise when demand is high. This provides an incentive for EVs to be charged during
off-peak hours, so the charging of EVs should not add significantly to the peak flows
through the electricity network. However, in the presence of a significant volume of
wind energy, which will likely be the case by the time that these large numbers of
EVs have been introduced, this dynamic can be expected to change.
A consequence of a large volume of wind energy will be that the current correlation between the wholesale electricity price and network flow will become weaker.
This means that if the charging of electric vehicles is driven by wholesale electricity prices, this may lead to network flows in excess of the capacity of distribution
94
Distribution grid congestion management
grids. A typical case is when wind output is maximal during peak consumption
hours. This may result in low wholesale prices despite the fact that consumption is
high, and responsive demand reacting to the price could therefore cause even higher
network loads. For a typical time period, Fig. 6.1 shows how system load, wind
generation, the electricity price and network load are correlated. The data are taken
from the Dutch electricity system [93], but we expanded the current electricity generation portfolio with two different installed wind capacities. For the network load
we use the standard household load profile that is used for net planning purposes
in the Netherlands [22]. During the observed period, there is a distinct peak in
wind power production that lasts roughly one day. It is important to observe how
wind generation reduces the electricity price significantly in the case with 15 GW of
wind generating capacity and, as a result, the electricity price profile shows quite a
different shape than the network load profile: at some moments with high network
load the electricity price is low. In the case with only 2 GW of wind, the electricity
price follows the demand curve more closely. One could argue that this phenomenon
of decoupling, due to the specific timing of the wind power production and demand,
is a relatively rare event, but since the networks need to accomodate the maximum
demand, even a limited hours per year can be problematic.
In the case that price-responsive EV demand will react on a low wholesale price
while network load is high, a congestion management mechanism will be necessary
in order to prevent network congestion. This mechanism will need to influence EV
charging behavior in such a way that network constraints are not violated. An
additional objective is that it should minimize the additional cost to EV owners as
compared to charging when wholesale prices are lowest. A final consideration is the
feasibility in terms of information, computation and communication requirements.
6.2.2
Minimum cost EV charging formulation
We will first summarize our EV charging model. A more elaborate description of
the optimization formulation of an aggregator that minimizes charge costs of a fleet
of EVs is provided in [38] and chapter 2 of this thesis.
We assume that real-time pricing of electricity will have been implemented and
that EV owners, represented by an aggregator who manages a fleet of EVs, therefore
have their vehicles programmed to minimize the electricity cost of charging. The aggregate demand of the EVs is assumed to be large enough to influence the wholesale
electricity price, see also [39]. We differentiate between electricity consumed by EVs
and electricity consumed for other purposes. We will refer to the latter as ‘baseline
electricity consumption’, which we assume to be perfectly price-inelastic.
Demand of electricity PD,k is given by:
PD,k = PD0,k + PEV,k
(6.1)
where PD0,k is the baseline demand and PEV,k is the extra demand of the EVs at
time-step k.
We include an EV-load dependent part in the electricity price:
λk = αk + βPEV,k
(6.2)
System demand (GW)
6.2 Problem analysis
95
20
15
10
5
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1.5
2
2.5
3
3.5
4
4.5
5
1.5
2
2.5
3
3.5
4
4.5
5
Wind (GW)
20
15GW wind
2GW wind
10
Net Demand (GW)
0
0
Price (EUR/MWh)
1
10
0
0
15GW wind
2GW wind
0.5
1
100
15GW wind
2GW wind
50
0
0
Network Load (kW)
0.5
20
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0.5
1
1.5
2
2.5
Days
3
3.5
4
4.5
5
1000
500
0
0
Figure 6.1 – From top to bottom: total system demand, wind generation, net demand,
electricity price and distribution network load. All graphs are for the same 5 day period.
The second, third and fourth figures show the differences between the installed wind power
capacity of 2GW and 15GW.
96
Distribution grid congestion management
where αk is the baseline electricity price at time k and βPEV,k the EV dependent
part. As outlined in chapter 2, the idea is to linearize the electricity price around
a certain average price and use the merit order of power plants to estimate the
sensitivity parameter β. This approximation allows for a quadratic programming
formulation of the EV charging problem, while it still includes the feedback of EV
charging on the electricity prices. If this feedback would not be taken into account,
all EV demand would be programmed in a short time interval with the lowest prices
in the optimization period, causing unrealistically high peaks in demand.
Next to the electricity price λk we will also consider the time-dependent network
tariff µk . The minimum cost charging problem then takes the following form:
min
PEV,ik
Nk
∑
2
(µk + αk )PEV,k + βPEV,k
(6.3)
k=1
N
EV
∑
∀k
(6.4)
PEVmin ,i ≤ PEV,ik ≤ PEVmax ,i
∀i, k
(6.5)
EEVmin ,i ≤ EEV,ik ≤ EEVmax ,i
∀i, k
(6.6)
EEV,ik+1 = EEV,ik + ηc PEV,ik − dik
∀i, k
(6.7)
s.t. PEV,k =
PEV,ik
i=1
where optimization variable PEV,ik denotes the charging rate of vehicle i out of
a total of NEV at time k and state variable EEV,ik denotes the battery state-ofcharge (actually state-of-energy). Technical vehicle parameters are the following:
ηc denotes the charging efficiency per vehicle, EEVmin ,i and EEVmax ,i the minimum
and maximum state-of-charge and the charging power limits are given by PEVmin ,i
and PEVmin ,i . The state equation Eq. 6.7 relates battery state to charging rate and
it implicitly contains the assumption that vehicles cannot deliver energy back to
the grid: PEVmin ,i = 0. The term di,k represents the discharges due to driving and
thus depends on the driving patterns of the EV owners. In a way, this term can be
considered to reflect individual consumer preferences, since it determines the level of
flexibility of each EV. Drivers covering large daily distances can not afford to wait
several days to recharge.
6.2.3
Simulation of the current situation (flat grid tariff )
To demonstrate why there could be a need for congestion management in distribution grids, we run a simulation where a fleet of EVs charges according to the
formulation above. Part of this fleet is connected to a certain distribution grid with
a certain capacity and a perfectly inelastic electricity load that representes roughly
1000 households. We assume that 50% of the households owns an EV and we represent the resulting fleet of approximately 500 EVs by 25 typical driver profiles, see
also [39] and appendix D of this thesis. Furthermore, we assume that the cable has
a peak load (occurring during only one hour of the simulated year) that is exactly
equal to the cable’s safe capacity. The simulation spans a period of a year in a
so-called rolling horizon optimization scheme. This means we solve the optimization
problem described by Eqs. 6.3-6.7 for a period of 5 days (using a 1 hour time-step,
6.3 Congestion management mechanism design
97
2000
1800
1600
Baseline demand
With 50% EVs, Min Cost, 15GW wind
With 50% EVs, Min Cost, 2GW wind
With 50% EVs, Uncontrolled
Power (MW)
1400
1200
1000
800
600
400
200
0
0
1
2
3
Time (days)
Figure 6.2 – Comparison between load profiles resulting from minimum cost EV charging
for two wind capacities and uncontrolled EV charging.
so 120 time-steps), then implement the control actions for the first day, then move
the horizon by one day, etc. The rolling horizon optimization has two important advantages: there are no end-point effect towards the end of each optimization period
and secondly, this formulation allows for an easier adaptation to include uncertain
forecasts that are updated each day. We emhasize that this uncertainty is not taken
into account in this chapter, which implicitly means we assume perfect RES and
load forecasts.
We run a simulation of a year. The EVs are charged as described above and
the network tariff µk is assumed not to depend on time. For comparison, we also
simulate an uncontrolled charging profile that occurs if EVs simply start charging
upon arrival at home. (See [28] for more information on the construction of the
uncontrolled charging profile.) Fig. 6.2 shows the network load that results from the
minimum cost charging formulation described above for the same period that was
shown in Fig. 6.1. One observes a strong peak in network demand in the evening of
day 2. This is the result of the fact that the EVs postpone their charging until the
period with the lowest price. In the case with only 2GW of wind installed, this peak
is absent. One also observes how the minimum-cost charging strategy can even lead
to a higher network peak than the uncontrolled charging scenario.
6.3
Congestion management mechanism design
We will analyse the effectiveness and implementation issues of the main three congestion management mechanisms that are discussed in the literature: a dynamic
grid tariff, advance capacity allocation and a distribution grid capacity market. In
this section we will discuss how we modeled these mechanisms.
A congestion management mechansim would limit the combined load of the EVs
plus the inelastic baseline load to the capacity Kl of a certain distribution asset, say
98
Distribution grid congestion management
a cable l. The mathematical formulation of this constraint reads:
∑
PEV,ik + Pl,k ≤ Kl
∀k, l
(6.8)
i∈EVl
where EVl denotes the subset of EVs connected to a particular cable l and Pl,k
the inelastic baseline load. The minimization of equation 6.3 in combination with
constraints 6.4 to 6.7 and this extra constraint leads to the theoretical minimum
energy costs of a group of EVs on a certain cable while respecting the cable limits. An
optimal congestion management mechanism, either price-based or capacity-based, is
thus expected to yield the same EV charging profile and costs.
6.3.1
Dynamic network tariff
The first mechanism that we discuss is a dynamic network tariff that introduces
a time-varying network price per kWh for using the network. The theoretically
optimal network tariff would be the lowest tariff that would cause the EV load plus
baseline demand to be just lower than network capacity. This can be formulated as
a bi-level programming problem of the following form.
min
µk
s.t.
Nk
∑
µk
(6.9)
k=1
∑
PEV,ik + Pl,k ≤ Kl
(6.10)
i∈EVl
min
Nk
∑
PEV,k
2
(µk + αk )PEV,k + βPEV,k
(6.11)
k=1
s.t. EEVmin ,i ≤ EEV,ik ≤ EEVmax ,i
(6.12)
PEVmin ,i ≤ PEV,ik ≤ PEVmax ,i
(6.13)
EEV,ik+1 = EEV,ik + ηc PEV,ik − dik
(6.14)
PEV,k =
N
EV
∑
PEV,ik
(6.15)
i=1
This problem formulation states that the task of the DSO (the leader) is to
find the lowest time-varying tariff such that when the EV aggregator (the follower)
minimizes his charging costs, the sum of inelastic demand and EV demand does not
exceed line capacity Kl . Note that this formulation assumes a welfare-maximizing
DSO, which is an approximation of a perfectly regulated or an ideal publicly owned
DSO. Relaxation of this assumption is outside the scope of this chapter, as network
regulation is a different and well-developed line of research.
In e.g. [94] it is shown that even the linear version of a bi-level programming
problem is NP-hard and generally requires sophisticated algorithms to solve. In this
case, however, because the follower problem is quadratic in nature, the complexity increases further, even in the case of perfect information on the leader’s side.
6.3 Congestion management mechanism design
99
The presence of inevitable uncertainties makes this a stochastic problem that could
render some solution approaches infeasible. We therefore consider two alternative
approaches that have certain practical advantages: one capacity-based approach and
one distributed and iterative price-based approach.
6.3.2
Advance capacity allocation
The idea behind the advance capacity allocation method is that the DSO announces
ahead of time what the free capacity of the network is, i.e. the capacity that is not
needed to serve inelastic demand. A practical implementation of this formulation
would involve the network operator communicating the forecasted time-dependent
available capacity Kl − Pl,k to the EV aggregator ex-ante. Another possibility would
be for the network operator to take on the role of the EV aggregator himself, but
this would constitute a departure from the principle of unbundling network-related
and commercial activities. In case of a single aggregator, the aggregator can simply
include the announced free capacity in the optimization formulation. Essentially he
thus includes constraint 6.8 in the optimization problem described by 6.3 to 6.7.
In a case with multiple EV aggregators, matters become more complicated. A
natural approach would be for the DSO to collect the demand bids (for each timestep) for network capacity by the aggregators and to auction the free network capacity. Once the market is cleared, the DSO communicates the allocated capacity per
time step to each aggregator. An issue that complicates the allocation by auction
is that the demand bid for capacity in a certain time-step depends on what whas
allocated in previous and future time-steps. It would thus require iterations to circumvent this problem, which seriously increases the complexity of this mechanism.
In the next subsection we discuss a mechanism that is more suitable for the case of
multiple aggregators.
Other allocation methods could also be possible, for instance on the basis of the
maximum willingness to pay for network capacity. If the aggregators would submit
their demand for network capacity ex ante (computed periodically, e.g. once per
year) to the DSO, in case of congestion he would allocate the available network capacity based on these bids. The aggregators could be asked to submit bid functions,
but perhaps a single maximum willingness to pay could also work. They would not
need to update their demand frequently, as in principle their maximum willingness
to pay should remain constant over time. This method has simplicity and transparency as advantages, but does not guarantee that all EVs always have enough energy
in their batteries that all intended vehicle trips can be made. As we take the driving
behavior as a constraint in our analysis, we will not pursue this option further.
6.3.3
Distribution grid capacity market
The problem of finding the optimal dynamic grid tariff described by Eqs. 6.9 to
6.15 can also be approached iteratively and in a distributed way. This method is
more suitable in case of multiple aggregators. Essentially, this approach consists
of the following steps: first the aggregators perform the optimization without a
network tariff and communicate their charging schedule to the DSO. The DSO then
100
Distribution grid congestion management
evaluates whether the network constraints are satisfied; if not, he raises the network
tariff during the moments when network capacity is exceeded. The aggregators recalculate their charging schedules based on the new grid tariff, communicate their
demand to the DSO, etc. This procedure is repeated until it converges, which
results in a certain grid tariff and a binding charging schedule. This scheme was
labeled a distribution grid capacity market [92] and was also investigated by [95],
who shows that this works in a situation with multiple aggregators by demonstrating
that a distribution grid capacity market can be solved in a distributed way and
yields the optimal load profiles without the need for information sharing between
the aggregators. We use the same algorithm as [95], which is briefly described
below. For notational convenience, instead of using subscript k, we now use bold
fonts to denote vectors with a length of the number of timesteps in the optimization.
Furthermore, we drop superscript N to denote the network tariff, so CtN ≡ C. The
index j denotes the iteration number. The algorithm consists of the following steps:
µ0 = 0
Solve Eq. 6.3 s.t. Eqs. 6.4 to 6.7 to obtain Fj = P∗EV + Pl
µj+1 = µj + γ max(0, Fj − Kl )
Stop if |µj+1 − µj | < ε
In [95] it is shown that with a convex objective function and affine constraints
the optimal EV charging profile can be found by this method. We will therefore
refer to the tariff that results from this algorithm as the ‘optimal tariff’.
6.3.4
Proxies for optimal tariff
The three congestion management mechanisms that we discussed so far should, in
theory, all lead to the same optimal EV profile. They have, however, certain practical
difficulties associated with their implementation, so we will investigate a number of
proxies for the optimal tariff that are simpler and easier to implement. We consider
three possible proxies and discuss them in the order of increasing complexity. We add
a constant part to all tariffs, including the tariffs described above, in order to make
the average tariff equal to 0.04EUR/kWh, which is equal to the current networkrelated part in the Dutch electricity tariff. This is, however, for the optimization
irrelevant since the constant part of the tariff drops out of the optimization objective.
A day-night tariff The simplest time-dependent network tariff is a day-night
tariff. In the Netherlands, the night tariff is in effect from 11 PM to 7 AM, so we
assume the same period. A second degree of freedom is the price difference between
the periods. Here we assume a value of 0.02 EUR/kWh.
This tariff structure has a number of desirable features. Day-night electricity
tariffs are already in place in many countries and consumers are therefore familiar
with it. In addition, it is transparent for consumers since it is always the same.
Furthermore, it does not require network load to be metered or forecast. At most,
one might adjust the starting hours of the different tariff periods and the price
difference periodically.
6.3 Congestion management mechanism design
101
A time of use tariff based on historical network load A more differentiated
time of use (ToU) tariff would take into account the shape of the network load profile.
This varies from day to day, depending on many factors such as the weather, the
composition of the loads connected to the network (residential, commercial, etc),
type of houses, etc. Ignoring these variations, one could simply take the average
load profile over a period in the past and base the tariff on that profile.
The reason for taking the load profile into account is the following: the tariff
should be high when network load is high and vice versa. We take a simple approach and scale the load profile (measured in kW) linearly with a conversion factor
(unit: EUR/h) to obtain a time-differentiated tariff in EUR/kWh. The value of the
conversion factor is chosen such that the average tariff again equals 0.04 EUR/kWh.
This tariff, although a bit more sophisticated, has most of the positive features
of the day-night tariff: it is transparent for consumers since it is the same every day
and easy to implement (as it does not require an IT infrastructure). In addition, it
only requires past network load measurements.
A time of use tariff based on real network load A next step in the complexity
of a proxy tariff is to base the tariff on expected real network load. It is the same as
the previous tariff, except that now we use the real network load profile, instead of a
historic average. This should give a more precise economic signal, but requires more
enabling IT infrastructure. The most serious requirement is that now it becomes
necessary to forecasted network load, because the tariffs need to be announced ahead
of time. How far ahead exactly is a point that deserves further study that we consider
outside the scope of this chapter. We emphasize that we assume perfect network
load forecasts in our simulations (which we present in the next section). In reality,
one needs to find a balance between forecasting as far ahead as possible, which allows
consumers to adjust their demand in a timely fashion and accepting larger forecast
errors.
This tariff also requires a more sophisticated IT infrastructure, because forecasts
need to be announced to the consumers frequently. One can think of a variety of
possibilities to do this, ranging from announcements on web pages to dedicated IT
systems. We consider discussion on this theme important, but outside our scope.
For consumers, this tariff is the least transparent of the proxy tariffs, since it changes
every day.
6.3.5
Comparison
Fig. 6.3 schematically shows the four (counting the proxy tariffs as one) different
congestion management schemes that have been discussed. The drawings indicate
the information flows between the DSO and the aggregator(s). The variables that
the stakeholders are required to forecast in the different schemes are indicated as
grey boxes. We distinguish between network load (indicated with load), electricity wholesale prices (price) and all EV related information, i.e. planned trips and
technical EV parameters, which determine the demand for EV charging.
Fig. 6.4 shows the optimal tariff as well as the three proxy tariffs that we consider
in this chapter. One observes how the three proxies give quite a strong signal in
102
Distribution grid congestion management
Load Price
EVs
Load
DSO
DSO
EV aggregator
Price
EV aggregator
EVs
Price
(a) Advance capacity
allocation.
EVs
(b) Dynamic grid tariff.
Load
Load
DSO
DSO
EV aggregator
Price
EV aggregator
EVs
EVs
Price
(c) Iterative distribution grid
capacity market
(d) Proxies
tariff
for
grid
Figure 6.3 – Schematic representation of the different congestion management schemes. µk
denotes a time varying network tariff , µ̂k a proxy network tariff, Kkf ree available network
capacity, Dk (K) a demand function for network capacity at time k and PEV,ik denotes
EV load. The gray blocks denote forecasting of information that has to be done by the
stakeholders. Arrows denote information flows. Dashed lines mean to indicate that this
arrow/block may not be necessary.
120
Day−Night
ToU Real Load
ToU Average Load
Optimal
Price(EUR/MWh)
100
80
60
40
20
0
0
0.5
1
1.5
2
2.5
Days
3
3.5
4
4.5
Figure 6.4 – Comparison of the different network tariffs.
5
6.4 Results and discussion
103
comparison with the optimal tariff, which is essentially always constant except when
congestion occurs. This observation suggests that the proxies may perform well in
terms of reducing the peak load, but will distort the economic signal of the energy
prices and therefore result in higher energy costs.
6.4
Results and discussion
We will now discuss the results of the simulations with which we evaluated the
various congestion management methods. In the presentation of the results we focus
on two output parameters: 1) the peak load, which is a measure of the extent to
which the congestion management mechanism was able to prevent congestion and
2) the total costs of charging the EVs, which is measure of the economic efficiency
of the congestion management mechanism. The unilaterally determined dynamic
grid tariff is not simulated explicitly because of the difficulties of its application
in practice. However, the optimal tariff that was found using the iterative grid
capacity market can be considered to approximate it closely. In the presentation of
the simulation results, the method labeled ‘optimal tariff’ can hence be considered
to represent both the dynamic network tariff from section 6.3.1 and the distribution
grid capacity market from section 6.3.3.
6.4.1
Simulation setup
We evaluated the various congestion management mechanisms discussed above with
our simulation model. The setup is similar to the one described in section 6.2.3. The
network tariffs, both the optimal tariff and those of the proxy methods, are contained
in the µk term in Eq. 6.3. We emphasize that in the simulations with the network
tariffs, we did not include the network constraint (Eq. 6.8) in the optimization.
Furthermore, in the simulation of the advance capacity allocation method it was
assumed that there was only one aggregator on the network under consideration.
Hence, the allocation of the capacity could be done in a straightforward manner by
directly including the network constraint 6.8 in the optimization problem. In the
algorithm for the iterative grid capacity market, the convergence criterion ε was set
to 0.01.
6.4.2
Simulation results
The lower panel of Fig. 6.5 shows the resulting load profiles for the simulated distribution cable for a typical period. One observes several interesting features. Most
notably, we see that the dashed lines that correspond to the proxy tariffs all show
network peaks similar to the flat tariff. These tariffs therefore perform poorly with
respect to network peak reduction. Secondly, the load profile resulting from the optimal tariff and the load profile representing the advance capacity allocation method
are identical. This shows that both methods indeed converge to the same EV charging profile.
Load profiles for an entire year are commonly summarized in a load-duration
curve. The upper panels of Fig. 6.5 show the load duration curves for all mechanisms.
104
Distribution grid congestion management
Network Load (kW)
1500
Baseline demand
w. EVs, Flat Tariff
w. EVs, Capacity Allocation
w. EVs, Optimal Tariff
w. EVs, Day−Night Tariff
w. EVs, ToU Tariff Real Load
w. EVs, ToU Tariff Average Load
w. EVs, Uncontrolled
1000
500
0
0
200
400
4000
6000
8000
2000
Hours
Network Load (kW)
1500
1000
500
0
1
2
3
Time (days)
Figure 6.5 – Comparison of the different load profiles. The upper panels shows the yearly
load duration curve (a detail of the first part in the left panel) and the lower panel shows
a typcial period of three days.
The left side of the figure is an enlargement of the first 400 hours of the full loadduration curve, which is shown on the right. The figure confirms that the loadduration curve that results from an optimal tariff, as obtained through an iterative
capacity market, is practically identical to the advance capacity allocation case. We
also note how the three proxy tariffs all lead to a higher load peak. The reason is
that the proxy network tariffs do not make use of feedback between the network
tariff and EV load (like the quadratic term in Eq. 6.3) so, as a result, EV load
concentrates in moments with low tariffs. This result is similar to what one observes
when the dependence between EV load and the electricity price is not included in a
minimum electricity cost formulation, as was e.g. shown in [38].
The second important performance criterion of the different tariffs, in addition
to the peak load on the cable, is economic efficiency. This can be evaluated from
the total yearly electricity costs for EV charging. We emphasize that these do not
contain the network costs. Table 6.1 shows the total yearly energy costs for the
EV fleet. We note that the energy costs for a fleet of approximately 500 EVs are
58,600 EUR in case of uncontrolled charging (when EVs do not respond to wholesale
prices nor any network tariff signal), whereas the flat tariff with EVs minimizing
their costs in response to the wholesale price signal, whithout consideration for grid
constraints, yields the lowest cost of 41,900 EUR. Interestingly, the application of an
efficient congestion management method causes the charging costs to be only slightly
higher (42,100 EUR) than the unconstrained flat tariff case. We draw the important
conclusion that the network constraint is a ‘cheap constraint’, much cheaper than
investing in the additional network capacity that would be needed to avoid the
6.4 Results and discussion
105
Table 6.1 – Comparison of simulation results
Case
No EVs
Uncontrolled
Flat tariff
Advance capacity allocation
Optimal tariff
Day-night tariff
Historic ToU tariff
Real ToU tariff
Peak demand Charging Costs
(MW)
(kEUR)
1.00
1.35
58.6
1.47
41.9
1.00
42.1
1.00
42.1
1.69
49.0
1.58
49.7
1.55
49.3
congestion. This may not always be the case, but given the fact that we started
with a minimally-dimensioned network (just large enough to match current peak
demand) and high volumes of RES and EVs, it is likely to be true in many other
cases.
We observe that the proxies perform considerably worse with respect to charging
costs, approximately 20% higher than in the optimal tariff case. This is due to the
fact that the proxy network tariff always distort the electricity price signal, while
this is not needed most of the time. Combining the fact that the proxies lead to a
higher network peak and higher energy costs, we can only conclude that they do not
work. The business-as-usual option (a flat tariff) is even preferable to proxy network
tariffs.
6.4.3
Comparison of results to the literature
An assumption that underlies the ‘dynamic grid tariff’ that was proposed in [91]
and also discussed in [92] is that the DSO can accurately forecast the electricity
price, network load and EV preferences, and correctly compute the optimal grid
tariffs from them. The stochastic, non-linear bi-level optimization problem that is
the mathematical formalization of this scheme may actually be very difficult, if not
impossible, to solve with acceptable accuracy and computing time. Future research
should shed more light on the practical implementation of this. Furthermore, one
may question if a DSO should even engage in the forecasting and optimization tasks
that are involved with this scheme because they are fundamentally different from
his core tasks of network operation and maintenance.
The iterative distribution grid capacity market was described in [90], [92] and [95].
As already pointed out in [92], this approach also creates a large computational and
communicative burden for both the DSO and EV aggregator(s). The convergence of
this procedure is an issue that is worthwile discussing, as was also done in [95], who
illustrated it with a numerical example of a small conceptual system. The system
studied in this chapter can be considered somewhat richer (with 25 EVs and a 120
timestep optimization horizon), so it is interesting to evaluate the convergence of the
algorithm in this particular setting. In 166 out of the 360 simulations we performed
there was grid congestion. Fig. 6.6 shows the distribution of the number of iterations
106
Distribution grid congestion management
45
40
occurences (−)
35
30
25
20
15
10
5
0
0
5
10
15
20
25
iterations until convergence (−)
30
35
Figure 6.6 – Number of iterations until convergence for the iterative capacity market algorithm.
needed to converge (according to the criterion outlined in section 6.3.3) to a network
tariff that solved the congestion. It can be seen that in the majority of the cases
the algorithm converges within 10 iterations and the highest number of iterations
required to converge was 33. This is in line with [95] who also find convergence after
approximately 10 iterations.
The advance capacity allocation method described in this chapter is straightforward if one assumes the presence of a single EV aggregator on each distribution
network branch. It seems, however, unlikely but also undesirable from a competition point of view that this is the case. In the case of multiple aggregators, the
DSO allocates the free capacity according to some distribution ratio. How this ratio
should be chosen in a fair and economically efficient way is an important question.
Alternatively, free capacity could be auctioned among the various aggregators who
will need to communicate their demand function for network capacity to the DSO
at every time step. It is important to mention here again that the free network
capacity is not a single quantity but a function of time (the free capacity is given
by Kl − Pl,k ). Combined with the inter-temporal constraints associated with the
demand response optimization by the EVs, this makes the advance allocation by
ex-ante communication of demand complicated. A secondary market in which capacity rights can be traded could prove to be a helpful mechanism in this respect.
An allocation based on an off-line computed maximum willingness to pay is another
possibility.
The congestion management schemes investigated in this chapter are also interesting in the light of proposed methods for integrating price-responsive energy
demand with electricity generation such as the ones discussed by [56]. The main
difference in scope is that [56] discusses methods for matching supply and demand
of energy whereas we concern ourselves mainly with supply and demand of network
capacity. [56] proposes to replace point by point iterations between suppliers and
6.4 Results and discussion
107
demand of electricity with the exchange of demand functions, which are constructed
by varying energy prices uniformly and calculating corresponding values for demand.
This approach could also be of interest for the grid capacity market discussed here,
but the inter-temporal constraints remain an issue to be investigated further.
6.4.4
Uncertainty
The optimizations that were formulated in the previous sections are all deterministic:
all actors had perfect knowledge of future electricity prices, driving behavior and
network load. In reality this is not the case and all of the above optimization tasks
are questions of decision making under uncertainty. This justifies the question which
of the proposed congestion management schemes performs best under the assumption
of imperfect information.
A dynamic grid tariff set by the DSO requires forecasts of EV driving schedules,
technical EV parameters, electricity prices and inelastic network load, as one can see
from the optimization formulation of Eqs. 6.9 to 6.15 and schematically in Fig. 6.3.
Therefore many uncertainties enter the problem. Except for network load, it seems
difficult for a DSO to forecast these variables since they do not belong to his core
business. The distribution grid capacity market appears more promising because it
requires actors only forecast variables in their own ‘domain’, i.e. network load for the
DSO and EV parameters and electricity prices for the EV aggregator. The advance
capacity allocation mechanism appears to create the smallest forecasting burden: the
DSO only needs to forecast network load. Of course, if network capacity is auctioned,
the EV aggregator needs to forecast price and EV parameters, but in this case wrong
forecasts only lead to economic inefficiency, not to network overloads. Given the low
frequency of congestion these costs will probably be limited.
6.4.5
IT infrastructure requirements
Another implementation issue is the requirement for IT infrastructure that is associated with the different schemes. The number of distribution network assets that
a DSO handles is typically large. Depending on the method of bookkeeping, or on
what network level the congestion management is to be applied, this can be in the
orders of tens of thousands of networks. The costs related to the IT infrastructure
may therefore be substantial and should be an important criterion in the assessment
of different congestion management schemes.
For the ex-ante determined dynamic grid tariff, the IT requirements appear to
be rather modest. The main burden in this scheme is for the DSO, who can perform
the optimizations ‘off line’ and only needs to communicate the resulting grid tariff.
There does not need to be any communication from aggregator to DSO.
In the case of advance capacity allocation, the DSO communicates the volume
of free network capacity to the aggregator(s). In case of only one aggregator, this
is everything that is needed, but in case of multiple aggregators they need to send
their demand functions to the DSO, who then clears the market. Here, too, the IT
requirements do not appear to be excessive, unless, as mentioned in the description
of this mechanism, iterations would be required to circumvent the problem of the
108
Distribution grid congestion management
inter-temporal constraints that make the demand functions in different time-steps
dependent on each other.
By far the heaviest requirements on IT infrastructure are associated with the iterative network capacity market. In this scheme, information has to flow iteratively
between DSO and aggregator. Also, on the aggregators’ side, each iteration requires
a new optimization. With large amounts of EVs and especially when stochastic
optimization methods are applied, this is a large computational effort in itself. The
entire scheme could thus take considerable time, during which information is constantly flowing between the actors.
6.5
Conclusions
This study presents an analysis of methods for managing the congestion of distribution networks that may arise when a large quantity of responsive EV demand
reacts to wholesale electricity prices which are influenced by a large share of variable
renewable energy sources. We present a mathematical formulation of the EV optimization problem and analyze a number of possible congestion management methods.
The most important conclusions from this study can be summarized as follows:
• Variable renewable energy generation weakens the correlation between wholesale electricity prices and electricity demand. As a consequence, large network
flows caused by cost-minimizing EV demand may cause distribution network
overload.
• The constraint that limited network capacity puts on EV charging has a low
cost associated with it. Shifting demand peaks through an optimal congestion management mechanism increases the energy costs of EV charging only
marginally.
• Tariffs that are fixed ex ante, based on historic network load profiles, do not
solve congestion efficiently and may not be effective at all. They distort the
economic signal of the wholesale electricity price, leading to unnecessarily high
electricity costs for EV charging. The reason is that network capacity is only a
constraint during a limited number of hours per year, while these tariffs force
continuous changes in EV charging. Therefore we do not recommend the use
of grid tariffs that are fixed ex-ante for managing network congestion that is
caused by EVs.
• The unilateral determination of an optimal dynamic grid tariff can be formulated as a non-linear bi-level programming problem where the leader’s task is
to find a tariff that minimizes the economic distortion of wholesale prices but
keeps network loads within bounds. Even in the case of perfect information
this is a difficult problem to solve. The additional complexity of uncertainty
may render this mechanism infeasible.
• An algorithm for an iterative grid capacity market may solve congestion in an
economically efficient way, but its implementation requires frequent exchange
6.5 Conclusions
109
of information between DSO and aggregator(s) and therefore poses a heavy
computational burden to both the DSO and the EV aggregator.
• Advance capacity allocation is a straightforward method for dealing with network constraints if there is only a single EV aggregator. If more aggregators
are active in the same distribution network, a capacity auction could be a
possible mechanism, but the existence of inter-temporal constraints for the
EV optimization complicates such an auction. Possibly, a secondary market
in which capacity rights can be traded among aggregators could contribute to
this solution.
We summarize the answer to the question raised in the title of this chapter as
follows: due to the limited additional costs of the network constraint on EV charging
and the poor performance of grid tariffs that are determined ex ante, an efficient
congestion management mechanism will likely be needed for the distribution grid.
The introduction of a such a mechanism will postpone or even avoid the need for
more costly network capacity upgrades. The design of this mechanism depends on
many factors that need to be addressed carefully in future research, for which we
recommend a number of possible directions. Most importantly, uncertainties in electricity prices, network load and EV driving schedules need to be taken into account.
Furthermore, adding a large share of solar energy would be a valuable addition to
the wind-based analysis performed here. Two factors make this particularly interesting: solar energy has a more regular and markedly different timing than wind,
and, secondly, it is largely embedded at the distribution level itself. Other important
issues to be investigated concern the relations between congestion management on
the one hand and investment in new distribution assets and capital cost recovery on
the other hand. Another topic that deserves more attention is the question how to
embed the distribution grid congestion management mechanisms in the new energy
market designs as discussed in [56]. Finally, we recommend more detailed studies of
the exact design of the different congestion management mechanisms. The specifics
of capacity allocation by auctioning, demand functions for network capacity, bids
representening willingness to pay, the iterative grid capacity market, etc, should be
investigated in such studies.
110
Distribution grid congestion management
Chapter 7
A refined view on electric
vehicle charging
In this chapter we aim to connect the individual elements provided by chapters 4, 5
and 6. Those chapters not only dealt with several aspects of the role of EVs in power
systems, but there were also different assumptions and/or viewpoints regarding the
EV charging process. Table 7.1 lists the most important assumptions concerning
the EV charging formulation. The main differences in modeling assumptions are
the availability of the EVs for charging, inter-temporal constraints in the power
plant scheduling and differences in the optimization objectives and the optimization
horizon.
Chapter 6 already deals with the combination of the optimization objectives related to distribution grid capacity and charging costs based on wholesale prices with
a control horizon of 5 days. The main difference between chapter 5 and chapter 6 is
that in the former, EVs were taken into account in the generation scheduling problem from a centralized viewpoint, where in the latter, EVs react to wholesale prices
that reflect marginal generation costs. To what extent, and under what conditions
the decentralized and centralized formulations lead to the same outcomes in terms of
EV dispatch is therefore investigated in this chapter. Furthermore, the sensitivity to
Table 7.1 – Overview of most important modeling assumptions in different chapters
Chapter EV charging availability
4
Charging is assumed to be
done only at home, after the
last arrival
5
EVs are assumed to be always available for charging
6
Optimization objective
Optimization horizon
Minimize network load. One day
Centralized formulation
Minimize generation costs. Five days
Centralized
formulation.
Full unit commitment.
EVs are assumed to be al- Minimize EV charging costs Five days
ways available for charging within network limits. Decentralized formulation
111
112
7 A refined view on electric vehicle charging
Ch4
Networks
Ch5
RES
Sensitivity analysis
on control horizon and
charging availability
Analysis of the equivalence between
centralized and decentralized
optimal charging formulation
Large scale impacts
of minimum cost
EV charging
Inter-temporal
generation constraints
Ch6
RES+Networks
Figure 7.1 – Schematic representation of the elements discusses in this chapter and their
relation with previous chapters.
the inter-temporal constraints that are taken into account in the unit commitment
problem of chapter 5 is analyzed.
The main difference between chapter 4 and 6 is that in the latter, EVs are
assumed to be always available for charging and they optimize over a horizon of five
days, whereas in the first EVs charge only at home and require a full battery every
day. A sensitivity analysis on these two aspects is performed to shed more light
on them. Fig. 7.1 shows schematically the relations between the chapters and the
main research elements investigated in this chapter that provide the link between
them. We begin this chapter by analyzing the differences between charging from a
centralized viewpoint (a social planner) and a decentralized viewpoint of aggregators
reacting to electricity wholesale prices.
7.1
Equivalence of centralized and decentralized
demand scheduling
In chapter 5 the scheduling of EV demand was done from a centralized point of
view: one entity scheduled the EV demand to minimize the total generation costs
of the entire system. In chapter 6, on the other hand, a decentralized approach in
which EVs were only reacting to wholesale electricity prices was taken. Chapter 4
had yet a somewhat different approach because there was EV charging was formally
not considered as optimization problem, but the philosophy of a DSO controlling
EV charging can also considered to be a centralized approach. The question arises
to what extent the centralized and decentralized approaches are equivalent. We
approach this question first by considering a simple conceptual system. After that
we present simulation results of a case study based on the Netherlands.
7.1 Equivalence of centralized and decentralized demand scheduling
7.1.1
113
Theoretical analysis of EV dispatch
In this section we use a number of basic concepts from mathematical optimization
theory that can be found in standard textbooks like [35] and [37].
Centralized EV dispatch Consider a power system with one generator with
a strictly convex cost curve C(PG ). This is a conceptualization of the situation
where a finite number of generating units with different marginal cost curves serve
electricity demand. There are two time-steps with different electricity demand PD1
and PD2 . The basic economic dispatch problem is to meet the demand for the lowest
generation costs. The mathematical formulation reads:
min C(PG1 ) + C(PG2 )
(7.1)
s.t. PG1 = PD1
(7.2)
PG2 = PD2
(7.3)
PG1 ,PG2
This problem can be solved by forming the Lagrangian and differentiating with
respect to PG1 and PG2 and the Lagrange multipliers λ1 and λ2 that correspond to
constraints 7.2 and 7.3. The latter two denote the marginal system cost (and hence
electricity price) at the two time-steps. The solution to this problem is trivial since
it is immediately dictated by the constraints.
Now consider that there is an extra portion of, say, EV related electricity demand
B, but it can be scheduled in a flexible way between time-step one and two. This
problem reads:
min
PG1 ,PG2 ,PEV 1 ,PEV 2
C(PG1 ) + C(PG2 )
(7.4)
s.t. PG1 = PD1 + PEV 1
:λ1
(7.5)
PG2 = PD2 + PEV 2
:λ2
(7.6)
PEV 1 + PEV 2 = B
:κ
(7.7)
The Lagrangian associated with this problem reads:
L(PG1 , PG2 , PEV 1 , PEV 2 , λ1 , λ2 , κ) = C(PG1 ) + C(PG2 )
− λ1 (PG1 − PD1 − PEV 1 ) − λ2 (PG2 − PD2 − PEV 2 )
− κ(PEV 1 + PEV 2 − B)
(7.8)
larger and the first-order necessary optimality conditions are given by:
∂C
− λ1 = 0
∂PG1
∂C
− λ2 = 0
∂PG2
λ1 − κ = 0
λ2 − κ = 0
(7.9)
(7.10)
(7.11)
(7.12)
114
7 A refined view on electric vehicle charging
∂C
together with the constraints 7.5, 7.6 and 7.7. From this we find readily that ∂P
=
G1
∂C
∂PG2 . The assumption of strict convexity implies that the derivative evaluated at
two points can only be equal if those points are equal. This leads to the following
solution:
λ1 = λ2 = κ
(7.13)
1
1
B + (PD2 + PD1 )
(7.14)
2
2
1
1
PEV 1 = B + (PD2 − PD1 )
(7.15)
2
2
1
1
(7.16)
PEV 2 = B − (PD2 − PD1 )
2
2
The flexible demand is thus scheduled such that the total demand is exactly
equal in both time-steps. This makes sense in the light of the convex nature of the
cost function. Fig. 7.2 shows a conceptual representation of this problem and the
solution. When a non-negativity constraint is added for the EV demand, we find by
writing the Karush-Kuhn-Tucker (KKT) conditions1 the following solutions. In the
case that B > PD2 − PD1 we have
1
1
PG1 = PG2 = B + (PD2 + PD1 )
(7.17)
2
2
1
1
PEV 1 = B + (PD2 − PD1 )
(7.18)
2
2
1
1
PEV 2 = B − (PD2 − PD1 )
(7.19)
2
2
and when B ≤ PD2 − PD1
PG1 = PG2 =
PG1 = B + PD1
(7.20)
PG2 = PD2
(7.21)
PEV 1 = B
(7.22)
PEV 2 = 0
(7.23)
In the first case the solution is the same as in the case without inequality constraints,
in the second case all flexible demand is scheduled in the time-step with the lowest
inelastic demand.
Decentralized EV dispatch In the decentralized demand dispatch, flexible demand reacts to an electricity price. The relation between electricity price and demand is essentially given by the marginal cost curve of electricity generation, which,
in our conceptual representation, would be simply the derivative of the curve C(PG )
at a given demand PD . We thus define the EV load dependent price as:
dC C ′ (PEV ) =
(7.24)
dPG PG =PD +PEV
1 The
KKT conditions are necessary conditions for the optimal solution of non-linear optimization problems with inequality constraints. Although in most cases a solution can not directly be
derived from them, they do often provide useful insights of the solution.
7.1 Equivalence of centralized and decentralized demand scheduling
115
Figure 7.2 – Schematic representation of the economic dispatch with flexible demand. The
flexible demand is shifted such that the demand at both time-steps is equal. Note that
∗
PEV
2 is negative in this example.
where the load balance constraint PG = PD + PEV has been used to express the
electricity price as a function of demand. In this case the optimization problem is
to minimize the charge costs which, for each time-step, are given by the product of
the volume and price. It reads:
min
PEV 1 ,PEV 2
s.t.
C ′ (PEV 1 )PEV 1 + C ′ (PEV 2 )PEV 2
PEV 1 + PEV 2 = B
(7.25)
:κ
(7.26)
The Lagrangian of this problem is given by:
L(PEV 1 , PEV 2 , κ) = C ′ (PEV 1 )PEV 1 + C ′ (PEV 2 )PEV 2
− κ(PEV 1 + PEV 2 − B)
(7.27)
The first-order optimality conditions are found by differentiating Eq. 7.27 with
respect to the optimization variables. Using the chain rule for derivatives we find:
∂C ′ PEV 1 + C ′ (PEV 1 ) − κ = 0
∂PEV 1 PD1 +PEV 1
∂C ′ PEV 2 + C ′ (PEV 2 ) − κ = 0
(7.28)
∂PEV 2 PD2 +PEV 2
Without an exact expression for C(PG ) we cannot derive a meaningful solution
from the above formulation. However, if we compare Eqs. 7.9 to 7.12 with Eq. 7.28
we notice the extra term with second derivative of the cost curve C(PG ). Only
if this term is zero, the two conditions are identical, but in this case there is no
unique solution to the problem since the price is constant and equal for both timesteps. Graphically we interpret this problem as depicted in Fig. 7.3. The sum of
grey areas denote the objective functions to be minimized. If we compare this with
116
7 A refined view on electric vehicle charging
Figure 7.3 – Schematic representation of the decentralized dispatch of flexible demand.
The total grey area is represents the costs to be minimized. The dark grey area denote the
difference with the centralized formulation.
the centralized formulation and recall that the price at a certain electricity demand
C ′ (PD ) was the derivative of the cost function C(PG ), we note that the dark grey
areas are exactly the difference with the objective function in that formulation, which
are represented by the integral of C ′ (PD ) (the light grey area). This also explains
why the difference in the two formulations becomes smaller as the second derivative
of C(PG ) approaches zero, because then the dark grey areas become smaller as well.
An example with a quadratic cost function Before moving on to the case
with multiple EV aggregators, we consider it to be instructive to first evaluate the
previously treated general cases with an exact expression for the cost curve C(PG ).
To this end we define a quadratic function:
C(PG ) =
1 2
aP + bPG + c
2 G
(7.29)
Furthermore we have again the inelastic demand in the two time-steps PD1 and PD2
and the energy constraint on the flexible demand PEV 1 + PEV 2 = B. Working out
the solutions in the centralized case yields the solution already found in the general
case which we repeat here for convenience
1
B+
2
1
= B−
2
PEV 1 =
PEV 2
1
(PD2 − PD1 )
2
1
(PD2 − PD1 )
2
(7.30)
(7.31)
In the decentralized case we now have
C ′ (PEV ) = a(PD + PEV ) + b
(7.32)
so the optimization problem reads:
min
PEV 1 ,PEV 2
(a(PD1 + PEV 1 ) + b)PEV 1 + (a(PD2 + PEV 2 ) + b)PEV 2
s.t. PEV 1 + PEV 2 = B
(7.33)
(7.34)
7.1 Equivalence of centralized and decentralized demand scheduling
117
We can use the constraint 7.34 to eliminate PEV and then differentiate the resulting expression with respect to PEV 1 to find a condition for the optimum:
4aPEV 1 + a(PD1 − PD2 ) − 2aB = 0
(7.35)
which yields the following solution:
1
B+
2
1
= B−
2
PEV 1 =
PEV 2
1
(PD2 − PD1 )
4
1
(PD2 − PD1 )
4
(7.36)
(7.37)
We note that this solution indeed differs from the one found in the centralized case.
Decentralized EV dispatch with multiple aggregators In the preceding sections we compared a centralized demand scheduling entity with an aggregator who
schedules demand on the base of expected prices. In the latter case, however, the
aggregator scheduled all demand and was clearly not a price taker, because the price
dependency was explicitly taken into account. Now we consider the case where there
is not a single aggregator anymore, but a second one with an equal amount of flexible demand. Both aggregators will aim to minimize the charge costs, anticipating
on the behavior of the other. The resulting situation can be considered a Cournot
model of a duopoly, see e.g. [18]. We split the flexible demand exactly in half for
each aggregator, so PEV 1 + PEV 2 = B/2 and P̂EV 1 + P̂EV 2 = B/2, where we labeled
the second aggregators demand with P̂ .
We use the same quadratic expression Eq. 7.29 for the cost function. Furthermore, for notational convenience we define c1 = aPD1 + b and c2 = aPD2 + b so that
we can write for the first aggregators objective function in the duopoly:
min
PEV 1 ,PEV 2
(c1 + aPEV 1 + aP̂EV 1 )PEV 1 + (c2 + aPEV 2 + aP̂EV 2 )PEV 2
s.t. PEV 1 + PEV 2 = B/2
(7.38)
(7.39)
By using the same procedure as before, i.e. eliminating PEV 2 and P̂EV 2 using
the constraints and setting the derivative with respect to PEV 1 to zero, we find the
following condition for the maximum:
PEV 1 =
c2 − c1
3B
1
+
− P̂EV 1 = f1 (P̂EV 1 )
4a
42
2
(7.40)
We find a similar expression for the second aggregator:
P̂EV 1 =
c2 − c1
3B
1
+
− PEV 1 = f2 (PEV 1 )
4a
42
2
(7.41)
For the Nash equilibrium2 it holds that
2 In the Nash equilibrium none of the two competing aggregators would benefit from changing
its decision unilaterally. This is where the Cournot duopoly should eventually converge to.
118
7 A refined view on electric vehicle charging
∗
∗
PEV
1 = f1 (P̂EV 1 )
(7.42)
∗
P̂EV
1
(7.43)
=
∗
f2 (PEV
1)
To solve this system, we write in more compact form
1 ∗
∗
PEV
1 = C − P̂EV 1
2
1 ∗
∗
P̂EV
1 = C − PEV 1
2
(7.44)
(7.45)
which leads, after some rewriting to the solutions:
(
)
1B
2 c2 − c1
1
2
1B
∗
∗
C
=
+
+ (PD2 − PD1 )
PEV
=
P̂
=
=
1
EV 1
3
22
3
4a
22
6
(7.46)
In a straightforward manner we can extend the analysis for N aggregators and find
for the demand of aggregator i:
(i)∗
PEV 1 =
1B
1 PD2 − PD1
+
2N
2 N +1
(7.47)
To compare the total EV demand with the case where there was only a single
aggregator, we now sum the EV demand in the first time-step of the all aggregators
and find for the aggregated EV demand:
1
B+
2
1
= B+
2
1
= B+
2
PEV 1 =
PEV 1
PEV 1
1
(PD2 − PD1 )
4
1
(PD2 − PD1 )
3
1 N
(PD2 − PD1 )
2N +1
for one aggregator
(7.48)
for two aggregators
(7.49)
for N aggregators
(7.50)
The expression for the demand in the second time-step follow from the constraints
that we used to eliminate the PEV 2 variable.
The interpretation of this result is that for a large number of aggregators the last
expression approaches the solution of the centralized case and the flexible demand is
dispatched in an manner that maximizes social welfare. This is a result that could
be expected on the basis of the economic intuition that if the number of market
players is very large, they are not able to benefit from influencing prices.
7.1.2
Simulations comparing centralized and decentralized
EV dispatch
The previous section presented a theoretical analysis on centralized vs. decentralized
demand dispatch in a simple conceptual system. We now repeat the analysis by
comparing simulations in a model representing the Netherlands. The decentralized
7.1 Equivalence of centralized and decentralized demand scheduling
119
model is similar to the one used in chapters 2 and 6 and described in [38] with
the main difference that we now allow for multiple aggregators. Each aggregator
performs the following optimization:
min
PEV,ik
Nk
∑
2
αk PEV,k + βPEV,k
k=1
∑
(7.51)
∀k
(7.52)
PEVmin ,i ≤ PEV,ik ≤ PEVmax ,i
∀i, k
(7.53)
EEVmin ,i ≤ EEV,ik ≤ EEVmax ,i
∀i, k
(7.54)
EEV,ik+1 = EEV,ik + ηc PEV,ik − dik
∀i, k
(7.55)
s.t. PEV,k =
PEV,ik
i∈EVj
Here EVj denotes the subset of vehicles belonging to aggregator j. Before performing
this optimization, each aggregator updates the electricity price parameters αk and
β by including the demand of the other aggregators of the previous iteration.
αk = λ(PD0 + P̂EV ) + λ′ (PD0 + P̂EV )(PD0,k + P̂EV,k − (PD0 + P̂EV ))
β = λ′ (PD0 + P̂EV )
(7.56)
(7.57)
where P̂EV,k denotes the anticipated EV demand of all other aggregators, PD0,k
denotes the residual electricity demand (demand minus wind power), and the overbar denotes an average over the period of optimization. Within one iteration all
aggregators perform the above optimization. In the first iteration the demand of
all aggregators is set to zero, so all aggregators react on the same electricity price.
With this procedure we expect to reach a Nash equilibrium after several iterations.
For reasons of comparison it is instructive to treat the centralized version first. In
this formulation it is the task of a single entity to schedule all EV demand such that
total generation costs are minimized, whereas in the decentralized version charging
costs are minimized. In the above QP formulation, we have linearized the price
around the average demand, using the supply curve that was shown in Fig. 2.17.
Since price reflects the supply curve (the marginal generation costs), the total or
cumulative generation costs are given by the integral of the supply curve, see also
subsection 7.1.1 on the conceptual system. For a linearized supply curve this results
in a quadratic formulation of the objective function:
min
PEV,k
Nk
∑
k=1
1
2
αk PEV,k + βPEV,k
2
(7.58)
The difference with the objective function in the decentralized version (Eq. 7.51) is
the factor ½ in the quadratic term3 .
3 This formulation is based only on the marginal costs and effectively ignores inter-temporal
constraints of all generation units. In the next section it is shown that this leads to only slightly
different EV demand profiles compared to the case with inter-temporal constraints.
Demand (MW)
7 A refined view on electric vehicle charging
15000
10000
0
Inelastic demand
EV demand
20
40
60
Centralized
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Decentralized, 50 aggregators
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Decentralized, 1 aggregator
80
100
Time (hours)
120
(a) Demand and EV demand
140
Demand (MW)
5000
Demand (MW)
Demand (MW)
Demand (MW)
Demand (MW)
120
160
15000
Inelastic demand
EV demand
Centralized
10000
5000
0
15000
20
40
60
Inelastic demand
EV demand
80
20
40
60
Inelastic demand
EV demand
80
100
120
140
160
Decentr. 50 aggregators
10000
5000
0
15000
100
120
140
160
Decentr. 1 aggregator
10000
5000
0
20
40
60
80
100
Time (hours)
120
140
160
(b) Residual demand and EV demand
Figure 7.4 – Comparison between the dispatched EV load in the centralized case and the
decentralized cases with 50 and 1 aggregator. Simulations with 15GW wind and 50% EV
penetration.
Fig. 7.4 shows a comparison between the centralized formulation (Eq. 7.58) and
the decentralized formulation (Eq. 7.51) with 1 aggregator and with 50 aggregators.
One clearly observes that the EV demand in the centralized version is practically
identical to the case with many aggregators. This confirms the intuition that when
EV aggregators are too small to influence market prices, the EV demand will converge to the social optimum. In the case with a single aggregator, however, we observe a somewhat different EV demand profile. One distinguishes the same feature
as in the conceptual example of the previous section: a single aggregator benefits
from scheduling a part of the demand in periods with higher prices. This rather
counter-intuitive notion can be understood by realizing that an extra price is paid
for a small portion of the demand, but as a consequence, a lower price is paid for
a much larger portion of the demand. Fig. 7.3 illustrates this effect schematically.
Fig. 7.4(b) furthermore shows how in the socially optimal cases (the top two figures)
the EV demand is scheduled in such a way that the residual demand, and hence the
system marginal cost, tends to become equal between different time-steps - an effect
usually referred to as ‘valley filling’. This was also shown in the conceptual analysis
from subsection 7.1.1.
To get some more insight in the influence of the amount of aggregators, Fig.
7.5(a) shows the difference between the centralized (socially optimal) EV dispatch
and the decentralized EV dispatch as a function of the number of aggregators. The
difference indicates the summed RMS difference between the profiles for a given
week. As expected, we see the difference approaches zero for a large number of
aggregators. The initial decline is quite fast, indicating that even a small number
of aggregators leads to a significant improvement of the EV demand profile. The
extra generation costs due to the non-optimal dispatch of the EVs are shown in Fig.
7.5(b). It can be seen that even in the case with a single aggregator scheduling all EV
demand, the extra generation costs are quite modest. The cost difference falls very
7.1 Equivalence of centralized and decentralized demand scheduling
0.7
5
Generation Cost Difference (%)
EV Profile Difference (MWh)
6
4
3
2
1
0
0
121
10
20
30
Number of aggregators
40
(a) Energy profile difference
50
0.6
0.5
0.4
0.3
0.2
0.1
0
0
10
20
30
Number of aggregators
40
50
(b) Generation cost difference
Figure 7.5 – Difference between the centralized EV profile and the combined EV profile of
all aggregators as a function of the number of aggregators. The energy profile difference
is the summed absolute difference between the profiles for a week. The generation costs
difference are the extra generation costs for this week compared to the centralized case.
rapidly towards zero, once again indicating that even a small number of aggregators
is sufficient for an efficiently working demand response market. It should of course
be noted that these numbers are based on the residual demand and the supply curve
shown in Fig. 2.17. One could speculate that in real systems there are more volatile
electricity prices due to the inter-temporal effects. As noted earlier in chapter 2, in
Denmark, for example, negative wholesale prices have been observed and this is a
phenomenon that our model of the electricity price is unable to capture.
Finally, we look at the number of iterations needed to converge to a Nash equilibrium. Fig. 7.6 shows the difference in energy profiles as a function of the number
of iterations. We recall that in the first iteration, the other aggregators demand was
assumed to be zero, so all aggregators reacted to the same electricity price that did
not include any EV load. In the subsequent iterations, each aggregator added the
combined EV demand of all the aggregators to the electricity demand and updated
the electricity prices accordingly. We observe that the Nash equilibrium is reached
already after a few iterations. When a low amount of aggregators is present, the
equilibrium is approached more slowly.
As a final remark, we note that it is interesting to conclude that dispatchable
demand can actually offer a possibility for market parties represting only demand
and no generation to exert market power. Usually, demand response is seen as a
remedy against market power exerted by generators, see e.g. [96]. This is still true,
of course, since the flexible demand adjusts itself to avoid price peaks. But when the
number of aggregators is very small, the resulting demand profile does not exactly
match the social optimum.
122
7 A refined view on electric vehicle charging
5
5 aggregators
10 aggregators
20 aggregators
30 aggregators
50 aggregators
Difference (MWh)
4
3
2
1
0
1
2
3
Number of iterations
4
5
Figure 7.6 – Difference between the centralized EV profile and the combined EV profile of
all aggregators as a function of the iteration.
7.2
7.2.1
Sensitivity analysis
Inter-temporal generation constraints
In the previous section we compared centralized and decentralized scheduling of EV
demand. Minimizing generation costs in the centralized situation was, however,
solely based on the combined marginal cost curve of all power plants and ignored
inter-temporal constraints such as start-up costs, ramping rates and minimum power
output levels. Therefore we compare in this section the cases of a unit commitment
model that takes all this into account with the simplified version ignoring the intertemporal constraints. The unit commitment (UC) model is described in chapter 5,
but this time we use Dutch power plant data that is based on [97]. We recall that
the UC model is a mixed-integer linear programming model, which means that we
assume constant efficiencies for the generation units, whereas a more sophisticated
version could include more detailed efficiency curves. The model without intertemporal constraints is given by Eqs. 7.58 with the usual EV constraints 7.52 to
7.55. In this section we also compare situations with a different amount of installed
wind power and a different EV penetration degree.
Fig. 7.7 shows the dispatch profiles for the UC model and the model without
inter-temporal constraints in the case with 5 GW of installed wind capacity and
25% EV penetration. It can be seen that the differences are small. Part of the
difference can be explained by the fact that in the UC model the generation units
have constant marginal costs, so in this formulation it does not matter if a unit is
generating at, say 40% or 100% of its output. Because of this, we observe a somewhat
less smooth EV profile. Furthermore, we note that EV demand is scheduled almost
exclusively in the night hours when electricity demand is low.
Fig. 7.8 shows the same plots, but now for a situation with 15 GW of installed
Residual demand (MW)
15000
10000
5000
0
Demand (MW)
123
Inelastic demand
EV demand
20
40
60
Unit Commitment
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Without intertemporal contstraints
80
100
Time (hours)
120
140
Residual demand (MW)
Demand (MW)
7.2 Sensitivity analysis
160
(a) Demand and EV demand
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Unit Commitment
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Without intertemporal contstraints
80
100
Time (hours)
120
140
160
(b) Residual demand and EV demand
10000
5000
0
Demand (MW)
Residual demand (MW)
15000
Inelastic demand
EV demand
20
40
60
Unit Commitment
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Without intertemporal contstraints
80
100
Time (hours)
120
(a) Demand and EV demand
140
160
Residual demand (MW)
Demand (MW)
Figure 7.7 – Comparison between the dispatched EV load in the centralized case and the
decentralized case. Simulations with 5GW wind and 25% EV penetration.
15000
Unit Commitment
Inelastic demand
EV demand
10000
5000
0
15000
20
40
60
Inelastic demand
EV demand
80
100
120
140
160
Without intertemporal contstraints
10000
5000
0
20
40
60
80
100
Time (hours)
120
140
160
(b) Residual demand and EV demand
Figure 7.8 – Comparison between the dispatched EV load in the centralized case and the
decentralized case. Simulations with 15GW wind and 50% EV penetration.
124
7 A refined view on electric vehicle charging
wind capacity and 50% EV penetration. Once again, the differences between the UC
model and the simplified merit order dispatch are small. Moreover, a noteworthy
observation is that in this case EV demand is actually scheduled at the system peak,
which leads to a significant increase of the peak demand, as can be seen in Fig. 7.8(a).
One can readily understand this by looking at Fig. 7.8(b) which shows the residual
demand, given by the electricity demand minus wind generation. When wind power
is sufficiently high, the actual minimum of residual demand can occur at the peak
load hours. This effect, and its consequences for the distribution networks have been
discussed as well in chapter 6. These figures show it in much more instructive way
though, and they demonstrate that the effect also occurs when considering intertemporal constraints of electricity generation. It can therefore not be attributed
to the simplified electricity price model based on the merit order that we used in
chapter 6.
7.2.2
Influence of the forecast horizon
The benefits of a flexible electricity demand are for a large part due to the ability
to await favorable conditions (e.g. low prices or low network load) to consume
electricity. In the cases that have been treated in this thesis so far, in particular the
minimization of charging costs based on wholesale prices, the optimization horizon
was set to 5 days. This effectively allowed the EVs to postpone charging up to 5
days to anticipate the low price periods. This, as we saw, lead to large financial
benefits in terms of lower energy prices, but also high peaks in network demand
because, effectively, the demand of 5 days worth of EV charging is squeezed in a
much smaller period. The question thus arises to what extent the energy costs
and network peaks are dependent on the horizon over which the optimization is
performed. In particular, one could argue that a smaller optimization period would
lead to much smaller peaks in the networks, especially since we impose the constraint
that all EV batteries have to be full at the end of the optimization period so all energy
consumed by EVs on a day has to be recharged the same day. The extreme case of
this question has already been considered in the form of the uncontrolled charging
scenario, since these can be considered to represent the case with the lowest possible
flexibility in charging.
Fig. 7.9(a) shows how the peak network demand (EV demand plus inelastic
demand) depends on the control horizon. The same setting was used as described
in chapter 6, i.e. a network with approximately 1000 households and 500 EVs,
reacting to the electricity wholesale prices based on 15GW installed wind capacity
in the Netherlands. Note that we performed simulations spanning a whole year, so
the peak demand refers to the highest peak of the year. We observe that, indeed,
for an optimization horizon of one day, the network peak is markedly lower. For
an optimization period of 2 days the peak increases sharply and there is only little
difference for an even longer optimization horizon. It is also interesting to see that
the highest peak actually occurs for a horizon of 4 days, and not for 7 days, the
longest period we considered. We hypothesize that this is due to the fact that for
a very long optimization horizon, another period with low prices might actually
enter the horizon and the EV demand can be divided over the two low price periods
7.2 Sensitivity analysis
125
6
1.6
1.55
5
Energy Costs (EUR)
Peak Load (MW)
1.5
1.45
1.4
1.35
4
3
2
1.3
1
1.25
1
2
3
4
5
Optimization Horizon (days)
(a) Network peak load
6
7
0
0
1
2
3
4
5
Optimization Horizon (days)
6
7
(b) Energy costs
Figure 7.9 – Sensitivity of peak load of combined controllable EV demand and inelastic
demand (a) and charging energy costs (b) as a function of the control horizon of the EVs.
In the right figure, a control horizon of 0 days denotes the uncontrolled charging profile.
instead of ‘squeezing’ all the demand in the single low price period.
Given the much lower network peak in the case with a one day optimization
horizon, one could argue that the existence of, say, a day ahead gate closure time
of markets for (flexible) electricity demand would enforce EVs to plan within this
horizon. As a result, the distribution grid congestion issue that would result would
be much less severe. We deem this incorrect, however. There is (and should be)
no rule or law that forbids market parties to look ahead and plan further than
cleared day-ahead prices. Especially in systems with a high RES penetration, we
feel that it will actually become increasingly important to look ahead multiple days,
in anticipation of high wind or solar periods.
Fig. 7.9(b) shows how energy costs for EV charging depend on the control horizon. Again, energy costs refer to the annual energy costs, since we simulated a
whole year. Here, a horizon of 0 days denotes the uncontrolled scenario (charging
immediately after arrival at home). We observe that, as expected, the energy costs
decrease monotonically as a function of the control horizon. After approximately
3-4 days the effect more or less saturates. This 3 to 4 day period is interesting in
the light of the typical battery size of 24 kWh (that was used in our model) in combination with the average daily driving distance of 35 km/day (equivalent with 7
kWh/day), because those numbers imply a timescale of approximately 3.5 days associated with the flexibility in the charging process. This time reflects how long the
average driver can postpone charging. This timescale seems to be well reflected in
Fig. 7.9(b). One could speculate that for larger EV battery sizes or for stand-alone
grid storage systems there would be more value associated with looking even further
ahead.
Fig. 7.9(b) is also interesting in the light of predictability of wind driven electricity
prices. Reasonably accurate forecasts up to a few days ahead are feasible, so it seems
that wind forecast errors would not severely undermine the potential of demand
response of EVs.
15000
10000
5000
0
Demand (MW)
Residual demand (MW)
7 A refined view on electric vehicle charging
Inelastic demand
EV demand
20
40
60
Always charge
80
100
120
140
160
15000
10000
5000
0
Inelastic demand
EV demand
20
40
60
Only charge at home
80
100
Time (hours)
120
140
160
(a) Demand and EV demand
Residual demand (MW)
Demand (MW)
126
15000
Inelastic demand
EV demand
Always charge
10000
5000
0
15000
20
40
60
Inelastic demand
EV demand
80
100
120
140
160
Only charge at home
10000
5000
0
20
40
60
80
100
Time (hours)
120
140
160
(b) Residual demand and EV demand
Figure 7.10 – Comparison of EV demand profiles between the case where EVs are always
able to charge and the case where only charging at home is possible.
7.2.3
Influence of charging availability
In all EV charge optimization problems we have considered so far, we assumed that
EVs were always available for charging. In the distribution network analysis from
chapter 4 though, we assumed that EVs were only charged at home, after the last
arrival of the day. We will therefore consider what the effect of the more limited
availability for charging is on the cost minimizing charge strategies.
We expect two counter-acting mechanisms to influence the results in terms of the
height of the network peak. On the one hand, one could reason that a more limited
charging availability would lead to higher peaks, since the same amount of energy for
all EVs has to be charged in a smaller time frame. On the other hand, network peaks
could turn out to be lower, since some EVs might already be done charging when
others arrive. This ‘randomness’ is also the reason that the uncontrolled charging
profiles have a lower combined peak than the cost minimizing charge strategy.
To model the limited charging availability we simply add the following constraint
to the usual EV constraints (Eqs. 7.52 to 7.55):
PEV,ik = 0
∀i, k ∈
/ Ti
(7.59)
where Ti denotes the set of all time-steps for which vehicle i is plugged in. In this
case Ti is determined fully when EVs are at home, but one could consider other
configurations as well.
Fig. 7.10 compares the demand profiles in the case with and without charging
availability constraints. The differences are rather modest, although one notices, as
expected, that during the day time the EV load is somewhat smaller in the case with
the availability constraints. This is due to the fact that during the day the number
of EVs parked at home is smaller than during the night.
To analyze the effect on the network peak, we show the load duration curve of
network load in Fig. 7.11. From this figure, too, we conclude that the differences
in load profile are quite small. We note that the number of hours with a high load
7.3 System level networks impacts of minimum cost charging
127
4
2
x 10
Charge always
Charge only at home
1.8
Network Load (kW)
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
200
400 0
2000
Hours
4000
6000
8000
Figure 7.11 – Comparison of load duration curves between the case where EVs are always
able to charge and the case where only charging at home is possible. The network load
denotes combined household and EV demand.
has decreased slightly (visible in the left panel of the figure), so apparently the first
effect discussed above is more prominent. The highest peak is equal in both cases
though, because it occurs at a moment where all cars are actually plugged in. We
thus conclude that the charging availability has little influence on the EV induced
peak load. One reflective remark is considered appropriate here: in case EVs are
also allowed to charge at work or parked near shopping centers, etc, the network
load at the residential grid will obviously be lower. An analysis that takes such
spatial effects into account is considered outside the scope of this thesis and forms
an interesting venue for future research.
7.3
System level networks impacts of minimum
cost charging
Chapter 4 evaluated the network impacts and financial consequences of EV charging
by considering uncontrolled charging profiles as well as a controlled charging profile
aimed to minimize the network load. We showed in chapter 6 and this chapter how
minimizing the charge costs based on wholesale electricity prices can actually lead
to even higher network peaks than the uncontrolled case. The question thus rises
what the distribution network impacts and its financial consequences of these charge
profiles are. Precisely this question was analyzed in [98]. Here we summarize a few
key findings of this study; for more details we refer to the paper. The method used
to convert network profiles to network impacts and cost figures was roughly the same
128
7 A refined view on electric vehicle charging
as the one described in chapter 4.
In [98] three different EV scenarios have been analyzed: uncontrolled (identical to
the one used in this thesis), minimizing network peak and energy losses (essentially
the formulation in chapter 2, Eq. 2.20) and minimizing energy costs (7.58). A similar
EV adoption curve as shown in Fig. 1.4 has been assumed, leading to roughly 50%
EV penetration in 2030 which was the time horizon of this study.
Fig. 7.12 shows the NPV of all capital expenditures related to asset replacements
and energy losses up to 2030. We observe that the higher peaks in the cost minimizing case indeed leads to significantly higher costs, in terms of higher energy losses
and but most prominently in higher replacement costs. Interestingly, one also observes how energy losses are even higher in the minimum loss scenario that in the
uncontrolled scenario. We explain this by recalling that the replacement of assets
generally leads to lower losses. So the uncontrolled scenario may have lower losses,
but at the expense of more costs for new assets.
These figures are also interesting to interpret in the light of the results shown in
Table 6.1. Here it was shown that for applying a congestion management mechanism
to limit EV load to free asset capacity leads to negligible extra energy costs. Since not
limiting the load indeed leads to much higher network costs (as shown clearly in Fig.
7.12), we conclude that it makes sense to consider a form of congestion management
in the distribution grid. Although we already arrived at this conclusion in chapter
6, the figures presented here significantly strengthen this conclusion.
7.4
Other settings and applications for demand response
This thesis describes a range of asects of controlled EV charging. Other settings in
which EV charging might offer great benefits are, however, also possible. Furthermore, although EVs have a number of characteristics that make them particularly
suitable for demand reponse purposes, other forms of flexible electricity demand
could start playing an important role, too.
One type of power systems that may particularly benefit from (EV) demand reponse are small isolated power systems on islands or remote communities [99]. By
definition, they lack interconnection with neighbouring systems, so those benefits
in terms of reliability or a smoother RES availability profile are not available. Furthermore, they often rely on expensive and polluting diesel or fuel oil generators for
electricity supply. Marginal generation costs of such units can be in the order of
200-300 $/MWh, which is a considerably larger than total levelized costs of wind
and solar energy, see Fig. 1.2. However, installing large capacities of wind and solar
power will quickly lead to curtailment4 , because there are no ways to export surpluses. Being able to adjust a portion of the demand to the wind power avaialibility
thus has a great economic potential. In addition, it leads to much lower emissions
4 Based
on the exact total levelized costs of wind and diesel generation, say these are 100 and
200 $/MWh, it would still be economically viable to install a capacity of wind power that heavily
exceeds the average load, because even if 50% has to be curtailed, wind power still has a lower
overall cost than diesel.
7.4 Other settings and applications for demand response
180
160
Net Present Value (%)
140
129
MV/LV Trans. Losses
MVD Cables Losses
MVT Cables Losses
HV/MV Stations Losses
MV/LV Trans. Replacements
MVD Cables Replacements
MVT Cables Replacements
HV/MV Stations Replacements
120
100
80
60
40
20
0
No EVs
EVs MinLoss EVs MinCosts EVs NoControl
(a) Total costs components
45
MV/LV Trans. Losses
MVD Cables Losses
MVT Cables Losses
HV/MV Stations Losses
MV/LV Trans. Repl.
MVD Cables Repl.
MVT Cables Repl.
HV/MV Stations Repl.
40
NPV difference (%)
35
30
25
20
15
10
5
0
EVs MinLoss
EVs MinCosts
EVs NoControl
(b) Cost differences components
Figure 7.12 – (a) NPV components of capital expenditures and energy losses for the HV/LV
transformers, the MV distribution and transmission cables and the MV/LV transformers in
the various cases. The values are normalized with respect to the NPV in the case without
EVs. (b) Difference in NPV between EV scenarios and the case without EVs.
130
7 A refined view on electric vehicle charging
of CO2 and other pollutants. Appendix A, based on the work described in [100],
explores the potential of controlled EV charging in such a system in more detail.
Important findings are that controlled EV charging indeed can replace a lot of diesel
generation by wind and solar power, leading to lower generation costs and lower
CO2 emissions. Compared to diesel powered vehicle use and electricity generation,
CO2 reductions up to 85% are possible.
The demand response potential of EVs is largely due to the flexibility in the
charging process, or, in other words, to shift its electricity demand in time. Many
other types of electric loads have this ability, too. One that has received a lot of
attention in the context of demand response and/or direct load control - long before
EVs were even in the picture, see [101] - is the control of climate systems for heating and cooling of buildings. For such systems, too, optimization problems may be
formulated based on economic objectives under the constraints of certain temperature bounds and the dynamics of the system under consideration. Whereas EVs
have electrochemical storage in their batteries, thermal storage is their equivalent in
climate control of buildings. The level of flexibility (how much and for how long one
is able to shift the electricity demand) depends on physical parameters such as the
heat capacity of the object and the heat transfer with the environment. The level of
flexibility of, say, one single refrigerator is modest. One cannot pre-cool a refrigator
to very low temperatures and then postpone the next cooling cycle by a few days.
For larger systems with more thermal mass, such as cold storage warehouses or office buildings, the level of flexibility may be larger. Appendix E, based on the work
described in [102], explores such a setting in more detail: a cold storage warehouse
in combination with solar PV generation. Some of the main findings of this work
are that the optimal cooling trajectories, the economic value of intelligent cooling
and the maximum network flows all depend strongly on the electricity tariff structure. This, once again, demonstrates how the physical demand response potential is
heaviliy interlinked with the economic environment it is embedded in.
7.5
Conclusions
This chapter explored a number of refinements to the analyses on EV charging from
previous chapters. We considered differences in managing a fleet of EVs from a social
planners point of view and decentralized price-based charging by cost-minimizing
EV aggregators. It was found that, as economic intuition suggests, if the number of
aggregators is small they can benefit from influencing market prices at the expense
of higher generation costs. The differences with the socially optimum profile are
modest, however, and quickly vanish when the number of aggregators grows. In
a sensitivity analysis on inter-temporal generation constraints we compared a unit
commitment formulation including EV scheduling with an approach that ignores all
individual generator characteristics and minimized EV charging based on a linearized
marginal cost curve. The results were found to be rather similar between both cases,
which strengthens the confidence that EV aggregators minimizing their charging
costs facilitate the integration of variable RES sources. The optimization horizon
did not prove to be very critical, which gives some confidence that e.g. forecasting
7.5 Conclusions
131
errors of wind production will not significantly hamper the potential of EV flexibility.
Availability of EVs for charging had an equally limited effect on the EV profiles.
Furthermore, a large-scale grid analysis similar to chapter 4, but this time focussing
on the network peaks induced by cost-minimizing EVs reacting to wind power based
wholesale prices, showed that these peaks indeed lead to substantial grid costs.
Together with the finding of chapter 6 that limiting EV load to grid capacity leads
to negligible extra energy costs, this essentially completes the argument for applying
a form of congestion management in distribution grids.
132
7 A refined view on electric vehicle charging
Chapter 8
Conclusions and
recommendations
This final chapter concludes the work described in this thesis by summing up its
main results and insights. In section 8.1 we formulate a coherent answer to our
main research question posed in chapter 1 and treat its underlying sub-questions
associated with chapters 4, 5 and 6 in more detail. Furthermore, in section 8.2
we sketch the contours of a possible new paradigm for clean and intelligent power
systems. We end this chapter with recommendations for future research and some
considerations relevant for policy makers.
We start by listing the contributions of this thesis to the state-of-the-art
in the scientific fields around the role of EVs in smart grids.
• A large-scale analysis of distribution grid impacts of EVs and its
financial consequences for DSOs has been presented in [69] and chapter
4 of this thesis. Results show that when controlling the charging process
the replacement costs are reduced most markedly, costs for energy losses are
much closer between controlled and uncontrolled scenarios and an overall cost
reduction in the order of 20% can be realized, the largest part of which to be
found at the level of medium voltage cables.
• An analysis of the combined potential of controlled EV charging and
cross-border transmission capacity for integration of variable RES
has been presented in [76] and chapter 5 of this thesis. The main finding is
that these technologies, that are often seen as substitutes, can complement
each other in high RES scenarios.
• Possible congestion management mechanisms to efficiently allign the potential of EVs for RES integration and the distribution networks have
been proposed and analyzed in [89] and chapter 6 of this thesis. Results indicate that applying an optimal congestion management scheme is economically
efficient, but in the design a trade-off exists between simplicity and efficiency
that needs to be considered more closely.
133
134
8 Conclusions and recommendations
• Various sensitivity analyses adding to the robustness of the conclusions were presented in chapter 7. Particular findings are that decentralized
price based charging only converges to socially optimal EV profiles for a large
number of aggregators and the optimal EV charge profiles depend only slightly
on 1) inter-temporal generator constraints, 2) an optimization horizon beyond
two days and 3) the availability for charging during day-time. Furthermore,
the large-scale distribution grid impacts of EVs reacting in a correlated way to
wholesale electricity prices are costly even compared to uncontrolled charging,
which strenghtens the case for congestion management.
8.1
Conclusions and answers to research questions
The central question of this thesis reads:
How can the flexibility of EV charging best be utilized in multi-actor power systems
with high shares of renewable energy sources?
The two important perspectives from which controlled EV charging can add significant value are its ability to be shifted in time according to fluctuating RES output on
the one hand, and to avoid peaks in network demand to defer or postpone network
investments on the other hand. With flat network tariffs and wholesale prices that
will be influenced strongly by fluctuating RES output, price responsive EV demand
can lead to even higher demand peaks than uncontrolled EV charging. The required
network reinforcements are costly and unnecessary because limiting the load to free
network capacity through an efficient congestion management mechanism has negligible additional energy costs. There are various congestion mechanisms possible
to allign the cost minimizing EVs with network constraints, either based on shadow
prices associated with the network constraints or an ex-ante allocation of free network capacity, but in both approaches there exists a trade-off between simplicity
and economic efficiency. All schemes, however, seem to have in common that the
function of the DSO is extended beyond its current role.
Description of the role of electric vehicles in multi-actor power systems
The large-scale introduction of EVs and the continuing growth of variable RES poses
a number of threats and opportunities to power systems. In liberalized power systems, different actors have different objectives that translate into different strategies
with respect to these technologies. Distribution network operators generally have
incentives, through some form of regulation, to minimize costs related to asset replacements and energy losses. Retailers and/or aggregators representing consumers
have the incentive to minimize charging costs based on electricity prices. In the
absence of market power, such objectives also lead to a minimization of variable
electricity generation costs.
The flexibility of EV charging can be understood mostly by comparing the typical daily battery discharge of approximately 6 kWh with a typical battery size of
24 kWh. This gives EV owners the option to postpone the charging process by a
8.1 Conclusions and answers to research questions
135
few days. With the help of a linear equation that relates the battery energy content to charging power and discharges due to driving, the problem of finding an EV
charging schedule that maximizes some objective can be described as a mathematical optimization problem. The objective function of this problem depends on the
perspective of the relevant actor. An EV aggregator minimizes charge costs based
on wholesale electricity prices and takes into account that the extra EV demand
influences wholesale prices as well. A DSO minimizes peaks in network load, which
results in minimizing the square of network load plus EV demand. EV charging can
thus be scheduled intelligently to maximize the value of its flexibility.
Sub-question 1: How can the controlled charging of EVs reduce their
impacts on the distribution grid? Large scale adoption of EVs leads to a
substantial increase in the demand of electricity, which, if it coincides with non EV
related demand, causes high peaks in network demand. If EVs charge solely at home,
starting immediately after arrival at home, the resulting uncontrolled charging profile
leads to a marked increase of the evening peak. In a controlled charging scenario,
EV demand is transferred largely to the night hours resulting in a much flatter
load profile and almost completely avoiding an increase in the evening peak. From
the point of view of a DSO, the financial benefits of controlled EV charging are
due to postponing and/or deferring grid asset replacements and lowering costs for
energy losses. The difference in the net present value of investments and energy
losses between the uncontrolled and the controlled charging scenarios is in the order
of 20%. The biggest savings are to be expected on the level of MV-cables. We
conclude that there is a strong financial incentive for controlled EV charging with
the objective to defer network investments and lower energy losses.
Sub-question 2: How can controlled EV charging reduce generation costs
in power systems with a high share of renewable energy sources? The
variable nature of renewable energy sources causes large fluctuations in the residual
load that is served by conventional, dispatchable generators. Flexible EV demand
can be explicitly included in the optimal scheduling of generation units, leading to
a flatter output profile of the dispatchable units. Compared to an uncontrolled EV
demand profile, controlled EV charging leads to a reduction in variable generation
costs by shifting load from expensive peak units to more efficient base-load units
and by reducing the number of start-ups.
Another way to deal with the intermittency of renewables is the exchange of
power between interconnected nodes, because averaging the RES output over a
larger area leads to a flatter RES production profile. In a setup with two interconnected power systems with different variable RES profiles, EV demand response
and interconnection between the nodes proved to be substitute technologies that
independently lead to a reduction in generation costs for moderate RES penetration scenarios. For high RES penetration, however, it was found that EV demand
response and cross-border transmission capacity actually strengthen each other, because transmission capacity is needed to transport power to locations where there
is sufficiently much flexible demand to absorb it. It is thus concluded that in future
136
8 Conclusions and recommendations
high RES scenarios there are financial benefits for both EV responsive demand and
increased interconnection capacity.
Sub-question 3: How can the costs of EV charging be minimized within
distribution grid constraints? Because RES lead to a reduced correlation
between electricity price and network load, low electricity prices can occur simultaneously with high network demand. As a result, flexible EV demand reacting to
wholesale electricity prices can create large peaks in network demand. Because these
peaks only occur during a few hours per year, limiting the flexible EV load to the
available network capacity increases energy costs only marginally. Hence, without an
additional mechanism that gives an incentive to take network load into account, the
EV peaks may cause unnecessary and costly reinforcements of distribution network
components.
A number of congestion management mechanisms that aim to avoid distribution
grid overloading have been analyzed by modeling an EV fleet reacting to electricity
wholesale prices in a system with a high RES penetration. The first mechanism is a
unilaterally determined dynamic grid tariff based on a shadow price associated with
the limited line capacity. Although an optimal tariff indeed limits EV load to the free
line capacity, it is, however, difficult for a DSO to determine because it requires full
knowledge of all EV preferences and wholesale price forecasts. Furthermore, the bilevel programming problem that should be solved to determine this optimal dynamic
grid tariff is hard even in the case of perfect knowledge and could be infeasible to
determine in practice in the presence of uncertainties.
A distributed approach where DSO and EV aggregators iteratively exchange
updated grid tariffs and EV demand profiles also converges to the optimal dynamic
grid tariff. The main advantage of this scheme is that it requires a DSO only
to forecast (non EV related) network load, whereas EV aggregators predict EV
preferences and wholesale prices. Because these types of information lie much closer
to the core tasks of the two respective actors, it is a more natural setup than the
unilaterally determined grid tariff. The main disadvantage is that this scheme is of
higher complexity and requires more IT infrastructure since information has to flow
iteratively between DSO and EV aggregator.
A third possible mechanism, advance capacity allocation, is not price based but
capacity based. Here the main challenge lies in the allocation of the capacity in the
case where multiple EV aggregators represent EVs on the same distribution line.
An auction of capacity, where EV aggregators announce time-dependent demand
curves for distribution grid capacity, is complicated by the inter-temporal constraints
related with EV charging. These constraints make the demand curves in future timesteps dependent on the amount of allocated capacity in previous time-steps.
Simple proxies for the optimal dynamic grid tariff have also been analyzed and
were found neither to solve the congestion issues, nor to be economically efficient
because they unnecessarily distort the economic signal of wholesale electricity prices.
Their use is therefore not recommended.
We conclude that there exists a trade-off between the simplicity and the efficiency
of possible congestion management mechanisms for responsive demand in the distribution grid. Issues related to uncertainty and requirements on IT infrastructure
8.2 Contours of a new paradigm for a clean and intelligent power system.
137
should therefore be taken into account when further investigating these mechanisms.
A refined view on EV charging EV demand can either be scheduled centrally
by a social planner aiming to minimize total generation costs, or in a decentralized
approach where EV aggregators aim to minimize the charging costs. In the extreme
case of one single aggregator who schedules all EV demand and thereby minimizes
his charge costs, an EV demand profile that is different from the social planner’s
optimum results. This can be ascribed to the fact that a single aggregator can
schedule a small portion of demand against higher prices and then benefits from
lower prices for the much larger remaining part of the demand. Multiple aggregators
that take each other’s actions into account, do lead to an aggregated EV demand
profile that converges to the socially optimal profile. A well functioning market that
includes price-elastic electricity demand therefore requires a number of competing
aggregators.
From a sensitivity analysis that explores the effects of inter-temporal generation
unit constraints, the length of the forecast horizon and the availability of EVs for
charging, it was concluded that these factors have in general a rather modest effect
on the EV charging profile. These results thus strengthen the confidence in the
conclusions described above and show that also under very different assumptions
there can still be large benefits in controlled EV charging with respect to grids
and/or generation costs. Furthermore, we showed that the impacts of EV demand
responding to wholesale prices indeed incurs substantial costs related to distribution
grid upgrades.
8.2
Contours of a new paradigm for a clean and
intelligent power system.
In this thesis we explored how flexible EV demand can provide value with respect to
different functions and for different actors in a power system. In principle, most of
our analysis of the potential value of flexibility was done from the perspective of the
current institutional design. One could question whether these current settings allow
this flexibility to be exploited optimally. Whereas the reforms towards restructured
power systems in the 80s and 90s were largely driven by ‘the need for growth in
productivity and efficiency’ [17] , the main objective of today must be to develop a
set of rules in which the use of clean electric energy, with its typical and irregular
characteristics, can be used optimally.
We have shown that the main value of flexible demand lies in its ability to adjust according to the fluctuating and uncertain RES output. When recalling the
definition of flexibility as the extent to which a power system can modify electricity
production or consumption in response to variability, expected or otherwise [12], we
note that there are already many possibilities in today’s power system for flexibility: bilateral contracts, day ahead markets, intraday trading, balancing markets,
interruptible load contracts, etc. But as flexibility will become more and more crucial, the notion of some type of market for ‘flexibility products or services’, however
those may exactly be defined, could become necessary, like also briefly discussed in
138
8 Conclusions and recommendations
[16], chapter 14. Here it was also noted that the development of a market for such
products will attract the necessary investments in flexibility enabling technologies.
One natural suggestion in this line could be something like a ‘flexible demand
quotum’, defined as a contract for a certain amount of energy in a certain time frame,
regardless of the exact timing. For example, if a consumer parks his electric vehicle
at 6PM and leaves the next day at 8AM and he needs 6kWh to recharge, he could
buy such a flexible demand quotum: 6kWh (6PM-8AM). Quota spanning a longer
time frame could be offered at a lower price. Similarly, for storage one could think
of a ‘flexible supply quotum’: a storage facility has an amount of energy to offer, but
it does not care when exactly it will produce. When being able to offer such types
of products, the burden of forecasting and planning is transferred from the parties
offering flexibility to the ones needing it. The existence of a market where flexibility
services are rewarded appropriately could also accelerate the emergence of aggregators - the intermediate parties intelligently operating flexible demand resources of
consumers.
However, one of the main points of this thesis was that demand flexibility has a
value both with respect to variable RES generation and the distribution networks.
The question thus remains what the future role of the DSOs should look like. Should
they, for example, also be allowed to purchase flexibility products like the ones
described above? Or alternatively, are other congestion management mechanisms
like the ones discussed in chapter 6 of this thesis adequate? We showed in chapters
6 and 7 how price responsive EV demand under the current rules and tariffs can
lead to unnecessary and costly investments in distribution grid upgrades. All of the
possible congestion management mechanisms to remedy this issue have in common
that there is role for DSOs that extends beyond their current mandate. The DSO as
a ‘passive’ copper building entity seems not to fit the image of an efficiently operating
power system where all functions in the value chain of electricity supply are aligned.
With the right governance structure, unnecessary investments can be avoided while
still facilitating the optimal use of clean generation technologies.
When DSOs will indeed will be more actively managing the distribution networks, their role becomes more similar to that of today’s TSOs, although there are
fundamental differences too. Congestion management on the distribution networks,
for example, involves not a few large companies like the generators/retailers for the
transmission system. Instead, many small players, possibly represented by a much
smaller number of aggregators, on thousands of different network would engage in
economic activities for electrical energy and/or network capacity. Transaction costs
will thus matter. Furthermore, small consumers can not be assumed to be always
pursuing minimization of costs, which could result in larger fundamental uncertainties regarding their load profiles and response to economic signals. Another complicating factor, as discussed in chapter 6 are the inter-temporal constraints associated
with demand response and/or storage, which makes it much harder to quantify energy and network capacity needs for a certain period ahead of time. Uncertainties
on local production and demand complicate this even further. In summary, if DSOs
will become a sort of mini TSOs, they will do so in a much more complex playing
field.
One could even speculate that the complexity of facilitating demand response
8.3 Recommendations
139
with competing aggregators raises the question if the benefits outweigh the transactions costs that are inevitably part of such a complex system. If, for example, a DSO
simply assigns one single EV aggregator (or acts as one itself) for a part of its service
area, the complexity and hence transaction costs would be reduced strongly. The
disadvantage is that this aggregator thus has a natural monopoly position, and will
need to be regulated. This will be a departure from the very unbundling philosophy,
since now commercial activities will fall in the regulated network domain. In a way,
these reflections touch upon the ‘limits of the unbundling’, because the fact that
centralized coordination can be so much easier and more efficient should be weighed
against the invisible hand of the market.
A more pragmatic approach would be to allow the EVs (or other flexible demand)
to induce network loads that are higher than the safe N-1 capacity. In case of the rare
simultaneous event of a line fault together with a high EV load, the EVs can simply
be curtailed. This does require an infrastructure to physically curtail the EV load,
but this will likely be much less costly than an IT infrastructure enabling ‘dynamic’
congestion management plus its transactions costs. Generalizing this notion, one
could question whether the N-1 criterion makes sense in the clean and intelligent
power systems we should be moving towards.
8.3
8.3.1
Recommendations
Future work
A number of recommendations for future research can be extracted from the work
described in this thesis. The two fundamental characteristics of RES making them
more difficult to integrate in the power system are variability and unpredictability.
For both these problems, flexible electricity demand might offer solutions to some
extent. While the former has been treated extensively in this thesis, the issue of
uncertainty and the potential of EVs to deal with it have not been treated. This
issue has received quite some attention in the literature (see also chapter 3), but
many open questions remain. The deterministic optimization approach that was
used for the work described in this thesis should then be abandoned and stochastic
techniques should be used instead. Essential for these models are realistic forecasts - of RES production and demand - including probability distributions of the
forecast errors. Ensemble weather forecasts could provide the basis for these probabilistic RES output predictions. The resulting problems of decision making under
uncertainty require more advanced optimization and planning techniques and will
computationally be more demanding too.
Closely related to the above is the issue of EV participation in balancing markets. This topic, too, has already received quite some attention in the literature,
but important questions remain. For example, balancing and/or intraday markets
are often organized differently in different countries or power systems. An exploration of the potential of flexible demand in different contexts is considered a valuable
addition to existing literature. Such analyses can also contribute to the broader research question how (balancing) markets with extremely large penetrations of RES
140
8 Conclusions and recommendations
can best be organized in the first place. An issue that follows directly from the
above considerations is how distribution grid impacts of EVs participating in balancing markets would turn out. One could speculate that without proper network
related financial incentives, the grid impacts of responsive (EV) demand reacting
on balancing signals could prove to be even worse compared to the case described
in this thesis because balancing markets have more volatile prices than day-ahead
markets.
Another topic that is worth investigating in more detail is how investment in generation and/or transmission capacity is influenced by the large potential of flexible
EV demand. In chapter 5 we only considered the effect of transmission capacity and
EV management on variable generation costs. The question how optimal portfolios of RES, conventional generation and transmission capacity look in relation with
demand response is therefore considered worth investigating in more detail. Following this issue are related questions such as what policy instruments give proper
incentives to arrive at the right technology mix when a significant volume of demand
response is present.
Also related to the topic of investment is an analysis of efficient planning and
cost recovery in distribution grids. We analyzed in chapter 6 how congestion management in a given distribution grid could be designed, but we left untouched the
question of investment and cost recovery in new grids. These topics are also tightly
related through the collection of congestion rents by the DSO. Care should be taken
that DSO do not have incentives to underbuild capacity. It is not unthinkable that
the introduction of large amounts of price responsive demand changes the landscape
in electricity distribution so profoundly that new regulation schemes should be investigated. One could, for example, think of incentives for price responsive demand
to reveal their long-term demand curves for network capacity. We noted in chapter
6 that it is more logical that the responsibility for short-term forecasting of EV demand and electricity prices should lie with market parties instead of DSO’s, but the
same could be true for long-term demand for network capacity. More research in
this area is therefore recommended.
Throughout this thesis, we explored the potential of EV demand control under
the assumption of ‘rational’ EV owners. The word rational is enclosed by quotes since
equating ‘cost minimizing’ and ‘rational’ would not do justice to many factors that
drive people’s behavior. For instance, one could hardly call it irrational to always
want to have a full EV battery in case an unexpected event requires the immediate
use of the vehicle. Instead of ‘cost minimizing’, one could thus better use the term
‘utility maximizing’, but then the questions arises: ‘what is the utility function?’.
How much, to continue the example, are people willing to pay to always have a full
battery? Determination of a ‘utility function’ is difficult, and the answer should not
be looked for in the realm of exact sciences, but rather in social, behavioral and psychological fields. In this light it is also interesting to note that there are indications,
in the context of an energy management system experiment, that people are equally
or sometimes even more motivated by environmental than financial incentives [103].
Alternatively, instead of focusing on ‘utility’, another promising approach could be
to consider ‘regret’. Empirical evidence in a travel-mode choice experiment shows
that regret-minimization is sometimes a better way of representing people’s choices
8.3 Recommendations
141
than utility-maximization [104]. Similar experiments in relation to demand response
could shed more light on the question how to unlock the demand response potential,
the benefits of which have been explored in this thesis.
8.3.2
Considerations for policy makers
In this thesis EVs were considered not simply as electric loads, but rather, by tapping
into their inherent flexibility, as building blocks of intelligent energy systems. One
advice to policy makers as therefore to do the same and formulate new EV policies
from the broader perspective of intelligent power systems. Especially in the light
of the transition towards a renewable energy based power sector it makes sense to
coordinate RES and EV policies - not only because of natural combination of flexible
demand and variable supply, but also since EVs powered by RES have the potential
to substantially reduce emissions in the transport sector. With regard to the networks, we showed that congestion management of price-responsive EV charging has
a very low extra energy cost but saves considerably in terms of additional network
investments. When the EV load reaches considerable magnitude, the application of
some form of congestion management and its required changes in the institutional
and regulatory structure are therefore recommended.
Although we have only briefly touched upon the issue of CO2 emissions and their
relation with EV charging in this thesis, some concluding remarks and recommendations are justified. In chapter 5 and appendix D we showed that as long as coal
(and/or lignite) generation have the lowest marginal cost of generating electricity,
any form of price based control of EV charging will lead to an increase in CO2 emissions, which will strongly affect the CO2 reduction potential of a large-scale switch
to electric mobility. Furthermore, it will weaken the business case for cleaner and
more flexible gas generators and, instead, increase the number of hours coal plants
are running. We conclude that the current CO2 policy (or more specifically: ETS
and its current price levels) not only fails in attracting enough investment in clean
energy technologies, it also gives incentives to operate a given portfolio in the most
polluting way. We therefore recommend a re-thinking of current CO2 policies in the
light of the intelligent power systems of the future.
A recommendation that follows from the analysis in chapter 7 of this thesis, is
that governments should warrant that no market power on the demand side of electricity is exercised, as we showed that this leads to economic inefficiencies. Current
regulation focuses more on preventing market power on the supply side of the electricity market, but as large volumes of price responsive EV demand may one day
become reality, there might emerge a need to guarantee a well functioning market
for the demand side as well. We demonstrated how in systems with only a small
number of EV aggregators the combined EV demand quickly approaches the socially
optimal demand profile, so keeping this market reasonably liquid seems desirable.
Perhaps the most important advice, and also the most firmly funded by the work
in this thesis, is to reconsider the rules and roles of today’s power sector in the light
of clean an intelligent power system. We sketched above some of the contours of
such a system and we advice policy makers to start looking in the right direction,
because the long lifetime of infrastructure assets requires todays policies to be aimed
142
8 Conclusions and recommendations
at tomorrows objectives.
Final thoughts
Scientific progress, by increasing labor productivity and technological efficiency, has
led to an enormous growth in prosperity for a large part of the world. Among the
darker sides of this picture are a dangerously close approach to ecological limits and
a deeply unequal distribution of wealth. New scientific developments and technological breakthroughs, not in the least those in the energy domain, are now creating
renewed opportunities for a better world. Focusing on the ecological dimension, it
seems that technology is no longer the limiting factor to a more sustainable world.
Renewable energy sources are becoming ever cheaper, their growth in many countries
is spectacular and new ways of operating highly renewable systems are being studied
by many enthusiastic scholars and professionals. The crucial economical and institutional aspects, too, are subject of study by researchers and policy advisors worldwide
and contours of new governance structures are beginning to appear. Although much
work still needs to done, this thesis has hopefully contributed a minuscule piece to
coming one step closer to a sustainable energy system.
Appendix A: The potential of
EVs in an isolated power
system
This appendix is based on the work described in [100].
Introduction
In this appendix we briefly discuss the role of EVs on an island with a high RES
penetration. The system is modeled using data from the Azores island Flores, which
has approximately 4000 inhabitants, a peak load of roughly 2 MW and electricity
generation by wind, hydro and, mostly, diesel power.
The analysis presented here are treated in more detail in [99], chapter 11. The
main new elements compared with the rest of this thesis are:
• EVs were assumed to have V2G capabilities, i.e. they can discharge power to
the grid.
• A deterministic dynamic programming algorithm instead of a linear or quadratic programming was used to calculate the optimal EV dispatch.
• A small isolated power system with a very high penetration of renewables was
studied. The characteristics of this power system are such that there is a high
curtailment of RES, and, secondly, expensive diesel generators are used for
periods with too little RES output.
• Reduction of CO2 emissions was analyzed.
• The effect of controlled EVs on the optimal generation mix was investigated.
Residual demand curves
The residual demand curves depicted in Figs. 1(a) and 1(b) show that there are
many hours with a surplus of renewable generation. The EVs, when charging is
143
144
Appendix A: The potential of EVs in an isolated power system
2
3
2
Residual Demand (MW)
Residual Demand (MW)
1.5
1
0.5
0
−0.5
−1
1
0
−1
−2
−1.5
−2
100
No EVs
EVs Uncontrolled
EVs Controlled
without EVS
with EVs controlled
with EVs uncontrolled
101
102
103
Time (Days)
104
105
−3
0
2000
(a)
4000
Hours
6000
8000
(b)
Figure A.1 – Residual demand for the moderate wind and solar scenario and 1000 EVs in
a five day spring period (a) and the load duration curves (b).
Table A.1 – Total (electricity generation + vehicle emissions) yearly CO2 emissions in kton
for different scenarios.
Vehicle Scenario
Electricity Scenario
All Diesel 50%EVs
ICE
Uncont.
50%EVs
Cont.
100%EVs
Uncont.
100%EVs
Cont.
Current generation mix
Moderate Wind and Solar
Aggressive Wind and Solar
8.38
6.18
5.52
8.06
4.65
3.29
7.80
4.26
3.13
7.76
3.05
1.29
8.08
5.37
4.42
controlled, manage to absorb a significant portion of this surplus. Nonetheless,
a significant share of renewable generation has to be spilled. Fig. A.2 shows the
different generation technologies for the same five day spring period. These graphs
readily show how the controlled EV charging leads to an increase in demand in times
of a surplus wind and solar generation. Also, since V2G capabilities were assumed
here, the amount of diesel generation is reduced, even compared to the case without
EVs.
Effects on emissions
From the obtained time series of dispatched diesel generation, the CO2 emissions
can be calculated in a straightforward manner, assuming that the diesel generator
emits 0.7 ton/MWh [105]. We compare the total emissions of the island under
different electricity generation and vehicle scenarios. Currently the fleet of roughly
2000 passenger cars is mostly powered by an internal combustion engine (ICE) fueled
with diesel with a typical emission of 150 g/km [106]. Table A.1 gives an overview
of the total emissions in the various scenarios.
The most important conclusion from this table is the large emission reduction
potential that EVs offer in combination with renewable generation. In the most
aggressive scenario, the total reduction of CO2 emission is more than 85% compared
to the current situation. In this scenario, the value of controlled charging is also the
145
Wind
Solar
Hydro
Diesel
Spilled
Demand
Generation (MW)
4
EVs Controlled
3
2
1
0
100.5
101
101.5
102
102.5
103
103.5
104
104.5
105
101
101.5
102
102.5
103
103.5
104
104.5
105
101
101.5
102
102.5
Days
103
103.5
104
104.5
105
Generation (MW)
4
EVs Uncontrolled
3
2
1
0
100.5
Generation (MW)
4
No EVs
3
2
1
0
100.5
Figure A.2 – Use of different generation types for a period in spring with 1000 EVs in
different scenarios for the case with maximum wind and solar distribution.
146
Appendix A: The potential of EVs in an isolated power system
Fit
CO2emissions vs. wind, solar
Equal Investment
Optimal Distribution
8000
Fit
CO2emissions vs. wind, solar
Equal Investment
Optimal Distribution
7500
7000
7000
6500
6000
CO2emissions
CO2emissions
6000
5000
4000
5500
5000
4500
4000
3000
3500
2000
3000
0
1000
0
1
0.5
1
1.5
2
2
2.5
3
3
0
2500
0
1
0.5
1
1.5
2
2
2.5
3
3
solar
wind
(a) 1000 EVs Controlled
solar
wind
(b) 1000 EVs Uncontrolled
Figure A.3 – CO2 emissions as a function of installed wind and solar capacity in both the
controlled (a) and uncontrolled scenario (b). Also shown are the line of the optimal mix
and lines of equal investment. The units of wind and solar capacity are dimensionless and
are such that the maximum of 3 correspond to 4.1 MW for wind and 4.0 MW for solar.
most prominent: a reduction of more than 60% (from 3.13 kton to 1.29 kton), only
due to shifting the energy needs of the vehicles. This confirms the intuition that the
value of ‘smarts’ increases when more RES in installed. The table shows also that
with the current generation mix that is dominated by diesel, replacing ICE diesel
cars with EVs leads to only modest emission reductions.
Optimal wind and solar mix
If one is interested in reducing CO2 emissions, it is instructive to compare the cost effectiveness of investments in wind and solar generation with respect to the emissions
of CO2 . By varying the amount of installed wind and solar and running the model
for the whole year for each combination, the emissions as a function of installed wind
and solar have been determined. To take into account that wind has lower total levelized costs than solar [3], the extra capacity of solar has been scaled according to
the ratio of levelized costs, so that one unit of capacity of wind has the same costs as
one unit of capacity in solar. In terms of installed capacity, the maximum value of
3 corresponds to roughly 4 MW for both technologies. Figs. 3(a) and 3(b) show the
resulting CO2 emission as a function of installed wind and solar capacity for the case
with 1000 EVs (controlled and uncontrolled). Since the capacity scale was choses
such that one unit of wind has the same costs as one unit of solar, the lines given by
wind + solar = Const. denote the lines of a certain investment. The values of wind
and solar where these lines are minimal then correspond to the optimal distribution
of a given amount of investment in new capacity. So in both the uncontrolled and
the controlled EV charging scenario, it is better to invest in more wind capacity first.
At some point, however, building more wind leads to much more spilled generation
and it is better to diversify the generation mix by adding some solar, despite the
fact that this is roughly two times more expensive. An interesting observation is
that the optimal mix depends on whether or not there is controlled charging of EVs
147
Table A.2 – Percentage of spilled renewable generation (wind + solar) for different scenarios.
Recall that the amounts of installed renewables are larger in the case with 100% EVs by
approximately 20%.
Vehicle Scenario
Electricity Scenario
No EVs
50%EVs
Uncont.
50%EVs
Cont.
100%EVs
Uncont.
100%EVs
Cont.
Current generation mix
Moderate Wind and Solar
Aggressive Wind and Solar
0
28 %
49 %
0
21%
42%
0
10%
30%
0
23%
45%
0
8%
29%
in place. In the controlled EV scenario, the EVs are able to avoid spilling energy
much longer, so here it is beneficial to build more of the cheaper wind capacity. But
even in this case, at some point it becomes more beneficial to invest in extra solar
instead of building more wind. To understand the results on the cost effectiveness
of new wind and solar generation, it is useful to invoke the amount of curtailed
renewable energy (hydro has not been counted as such). Table A.2 lists the amount
of spilled renewables in the different considered scenarios. It is important to notice
that in the current generation mix there is never any spilling of wind or solar, but
this is partly a result of the fact that we did not include start-up or ramping costs of
diesel generators. In practice, some diesel generators will probably be kept running
while wind is spilled, mainly for reliability reasons. The table also shows that in the
moderate wind and solar scenario, the EVs are very effective to avoid curtailment
of wind or solar. At some point though, there is simply to much extra energy to
absorb and the amount of spilled energy starts to increase dramatically. Considering
cost effectiveness, it is good to compare Table A.2 with the levelized cost of wind,
solar and diesel. As stated before, with current diesel prices, the marginal costs of
diesel generators are in the order of 250-300 $/MWh. In [3], levelized cost of wind
and solar are roughly 100 $/MWh and 200 $/MWh, respectively. This means that
if 60% of all wind is spilled, it is still cheaper than diesel generation. For solar this
is the case if 20% is spilled. These numbers strongly suggest that it does not only
make sense to invest in wind and solar from an environmental point of view, but
also from an economical.
148
Appendix A: The potential of EVs in an isolated power system
Appendix B: EV impacts in
residential low voltage grids
This appendix is based on the work described in [72]
In chapter 4 the effect of EV charging on distribution grid assets ranging from
the MV/LV transformers to HV/MV substations was analyzed. The lowest level
of distribution grid asset, the LV feeder cables, were not included in this analysis.
This appendix presents some results of an analysis of EV impacts on LV feeder
cables. The dataset used was substantially smaller (145 LV cables) than the one
used in chapter 4. This is explained by the fact that the load on LV feeders is not
measured in normal operation, so only data that has been collected during specific
measurements was available.
Fig. 1(a) shows the histograms of LV feeder cable loadings resulting from a 75%
EV penetration in 2040. Compared to the loadings of the other assets, only a modest
number of overloadings is observed. This can partly be explained by the fact that
LV feeder cables are generally over-dimensioned in comparison with the MV/LV
transformers.
Another issue that is relevant for LV feeder cables is the voltage drop along the
cable. Fig. 1(b) shows the voltage drops between the MV/LV transformer and the
end of the cable for different EV charging modes. Because of the small dataset, a
Weibull distribution has been fit to the data in order to estimate the probability of
a voltage drop larger than 20 V. We note that these probabilities are very small,
although, as expected, the extra EV load in the uncontrolled charging scenarios
increases the probabilities markedly.
In conclusion, we note that the LV cable overloadings or voltage problems are
much less pronounced than the impacts on the higher network levels that were
analyzed in chapter 4. One important assumption should be repeated here, though.
When calculating the impacts on the LV cables, we also used the aggregate EV
demand profiles shown in chapters 2 and chapter 4. In chapter 2 we also showed,
however, that the use of the aggregate profiles was only accurate for a number of EVs
exceeding approximately 50. A typical LV cable serves around 20-25 households, so
actually the use of the aggregate profiles was not quite accurate for this network
level. One could suggest to correct the peak load by a certain factor, or, even better,
to perform a stochastic analysis that assesses the probability of overloadings. This
is left as a recommendation for future research.
149
150
Appendix B: EV impacts in residential low voltage grids
40
# Cables
20
10
0.2
Probability Density
Uncontrolled 3kW
N > 1.2 = 13
= 9.0%
30
0.1
0.05
0
0.5
1
1.5
2
40
Uncontrolled 10kW
N > 1.2 = 19
= 13.1%
10
15
20
25
30
35
40
Uncontrolled 3kW
P > 20V = 4.37 %
0.1
0.05
0
5
10
15
20
25
30
35
40
0
0
0.5
1
1.5
2
40
Controlled
N > 1.2 = 7
= 4.8%
30
20
10
0
0
0.5
1
1.5
Cable Utilization Factor
2
(a) Cable loading distribution
Probability Density
0.2
10
# Cables
5
0.15
0
20
Uncontrolled 10kW
P > 20V = 5.42 %
0.15
0.1
0.05
0
0
5
10
15
20
25
30
35
40
0.2
Probability Density
# Cables
30
0
0.2
Probability Density
0
0
1 % growth only
P > 20V = 2.07 %
0.15
0.15
Controlled
P > 20V = 2.54 %
0.1
0.05
0
0
5
10
15
20
25
Voltage Drop (V)
30
35
40
(b) Voltage deviation distribution
Figure B.1 – Change in cable loadings (a) and voltage deviation (b) if 75% of all households
owns an EV in different charging scenarios. The line graphs in (a) denote the 1% growth
scenario without EVs. Figure (b) also shows estimated (line graphs) probability density
functions based on a Weibull distribution
Appendix C: Carbon
emmissions due to EV
charging
This appendix is based on the work described in [107].
Introduction
CO2 reduction is one of the important drivers for electric mobility. The exact CO2
reduction of EVs is, however, determined completely by the electricity generation
technologies employed to charge the EVs. When effectively powered by a coal plant,
an EV has a similar emission per kilometer driven than a gasoline powered conventional vehicle. For example, assuming an EV efficiency 0.15-0.2 kWh/km [13]
and a typical coal emission of 1000 g/kWh [108], one finds a CO2 emission range of
150-200g/km. By contrast, the average CO2 emissions of newly sold conventional
gasoline vehicles in the EU in 2013 is around 130 g/km, whereas the average CO2
emissions of the current passenger car fleet are around 180 g/km [109]. An EV
powered by a coal plant hence has a considerably higher emission than a modern
efficient gasoline vehicle.
The use of gas plants for EV charging reduces the emissions considerably by
roughly 50%. When powered by RES, an EV drives virtually CO2 free, except
for emissions related to car manufacturing, road construction and other life-cycle
effects. How charging EVs contributes to CO2 emission at the system level depends
on factors like the generation mix (how much coal, gas, wind energy etc is installed in
a country) and the timing of EV charging. Furthermore, the order in which various
power plants are dispatched (i.e. the merit order) depends on price of fossil fuels
and non-fuel related variable costs like CO2 pricing. In this appendix we analyse
how EV emissions depend on these factors. In particular we study how emissions
of EV charging depend on 1) the generation mix 2) the timing of charging 3) wind
energy and 4) CO2 price.
151
1.6
120
80
40
0
0
2
4
6
8
10
12
14
16
0.8
0.4
0
2
4
6
8
10
12
Cumulative Generation (GW)
14
(a) Dutch generation portfolio
16
120
2
80
40
0
18
1.2
0
Marginal Cost (EUR/MWh)
160
CO2 Emission (kg/MWh)
Marginal Cost (EUR/MWh)
Appendix C: Carbon emmissions due to EV charging
CO2 Emission (kg/MWh)
152
18
0
2
4
0
2
4
6
8
10
12
14
16
18
6
8
10
12
14
16
18
1.6
1.2
0.8
0.4
0
Cumulative Generation (GW)
(b) Alternative generation portfolio.
Figure C.1 – Merit order and marginal emission function of power plants for the Dutch
generation portfolio (a) and the alternative generation portfolio based on the German fuel
mix (b).
Research method
The analysis is done based on a simple model of the electricity system where it is
assumed that power plants are dispachted according to increasing marginal costs of
electricity production (i.e. the merit order). By combining the merit order of power
plants with the system demand time series, the dispatch of power plants is found
for each time-step. This model thus ignores start-up costs, minimum power output
levels, ramping rates etc. Two different generation portfolios have been analysed.
The first is the Dutch portfolio, with power plant characteristics originating from
[23, 93, 97]. The second portfolio is inspired by German fuel mix statistics [110],
which are characterized by a large share of coal and lignite (50%), nuclear (20%),
gas (20%) and hydroelectricity and wind (10%). Taking into account typical plant
sizes and efficiencies, we thus create a portfolio that is representative for the German
portfolio; we will, however, refer to it as alternative portfolio. The total installed
capacity still has the value of the Dutch total installed capacity used for the original
Dutch portfolio, for reasons of a fair comparison between the two cases.
Caloric values of the different fuel types (e.g. as given in [111]), together with
plant efficiencies allow us also to compute the instantaneous CO2 emissions of electricity generation. We do not consider upstream emissions. Emissions of nuclear,
hydro, waste incineration, biomass and wind have all been set to zero. It should be
noted here that because nuclear plants have a must-run character, we have artificially put their marginal costs at zero, so they are first in the merit order. Fig. C.1
shows the supply function of the both portfolios, together with the CO2 emissions
of the plants ranked according to marginal costs.
To assess the impacts of EV charging, EV demand profiles are added to the system demand of electricity. We consider two stylized EV load profiles: one where EV
charging is mostly done in the evening peak hours (referred to as peak charging) and
one where the EV demand is shifted to the night hours (night charging). Furthermore, we assume a peak EV load of 1 GW, an EV driving efficiency of 5km/kWh
153
2000
EV Night Charging
EV Peak Charging
Demand (MW)
1500
1000
500
0
12:00
16:00
20:00
00:00
time
04:00
08:00
12:00
Figure C.2 – Extra system load caused by EV charging.
Table C.1 – Simulation results for different portfolios and different times of charging.
Dutch portfolio
Alternative portfolio
Additional by EVs
Average Additional by EVs
Average
(kg/kWh) (kg/kWh)
(kg/kWh) (kg/kWh)
No EVs
0.60
0.60
EV Night
0.45
0.59
1.25
0.61
EV Peak
0.53
0.59
1.05
0.61
and a total of one million EVs. The two EV charging profiles have been plotted in
Fig. C.2. The yearly emissions caused by EV charging are found by the difference
in total emissions in the case with EVs and the case without EVs.
Results
Emissions of EV charging for different portfolios
The emissions caused by EV charging in both portfolios are listed in Table C.1. We
discuss the case with the Dutch portfolio first. For the average CO2 emission of
electricity production, we found a value of 0.60 ton/MWh, which is in close agreement with literature values of 0.55-0.60 ton/MWh [112]. We notice the difference
in CO2 emissions between night charging and peak charging due to the system load
(including EV charging) at night being mostly around 9 GW, whereas peak loads
range between 12 GW and 17 GW (summer and winter). Comparing this with the
marginal emission curve from Fig. C.1, we conclude that for night charging the most
efficient gas plants are the marginal plants and for peak charging the less efficient
gas plants are dispatched. Worth noting is the fact that both at night and at system
peak, EV charging emissions are still lower than the average CO2 intensity of all
electricity generation. This is due to the fact that the most polluting coal plants are
included in the average CO2 intensity, but do not contribute to the extra emissions
caused by EV charging.
We observe remarkable differences in the alternative portfolio in comparison with
the Dutch portfolio results. Most notable are the much higher emissions for both
peak and night charging. This can not be readily understood if one takes into
account that the average CO2 emission of electricity production is about equal to
CO2 emissions (kg/kWh)
Appendix C: Carbon emmissions due to EV charging
0.6
CO2 emissions (kg/kWh)
154
1.4
Dutch, Night
Dutch, Peak
0.5
0.4
0
2
4
6
8
10
12
Alternative, Night
Alternative, Peak
1.2
1
0.8
0
2
4
6
8
10
12
Installed Wind Generation (GW)
Figure C.3 – CO2 -emissions of EV charging as a function of installed wind power generation
the Dutch case. Invoking the marginal emission curve plotted in Fig. C.1 explains
this result: the most polluting generation is found towards the end of the merit
order. So, the extra emissions caused by EV charging are always higher than the
average emission. The marginal emission curve also explains the fact that peak
charging is in this case less polluting that night charging; a result that is opposite
to the Dutch situation. Night charging is mostly done at a system load where either
lignite or coal plants are the marginal plants, whereas peak charging takes place in
the regime where gas plants are marginal. The fact that we find equal average CO2
intensities of electricity generation for the Dutch and the alternative portfolio and
completely different emissions caused by EV charging, unmistakenly points out that
using average CO2 intensities for calculating the environmental impacts of EVs is
inaccurate.
The effect adding wind energy to the portfolio
When wind energy is added to the portfolio, emissions of EV charging will change
due to the fact that wind has zero marginal cost and hence shifts all other generation to the right in the merit order. In the merit order dispatch model that is
used in this analysis this is equivalent to subtracting wind power from demand. Fig.
C.3 shows the effect of installing extra wind generation on the EV charging emissions on the Dutch and the alternative portfolio respectively. The results are again
markedly different between the two different portfolios. In the Dutch case, initially
wind power replaces the least efficient gas plants at the end of the merit order. At
some point there is so much wind power that occasionally coal plants will be the
marginal plants and the average EV charging emissions will start to rise again. In
the alternative portfolio, initially the most polluting plants are pushed out of the
merit order by wind power and we observe rising emissions. At some point nuclear
plants will become the marginal plants and the EV charging emissions will decline
CO2 Emissions EV Charging (kg/kWh)
155
1.6
Dutch, Night
Dutch, Peak
Alternative, Night
Alternative, Peak
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
20
40
60
CO2 Price (EURO/ton)
80
100
Figure C.4 – CO2 emissions of EV charging as a function of CO2 price.
again. Another noteworthy feature is that in both cases the differences in emissions
between night charging and peak charging vanishes when more than 10GW of wind
is installed. This can be understood by realizing that the probability distribution
of the residual system load (electricity demand minus wind generation) of these two
cases will approach each other with increasing stochastic generation. The modest
change in system load due to EV charging is completely offset by the much larger
changes caused by wind generation.
Effect of CO2 price
Fig. C.4 shows the relation between CO2 price and EV charging emissions for the
different charging patterns and portfolios. For the Dutch portfolio, we observe that
there is only a noticeable effect for CO2 prices higher than 20-25 e/ton. An explanation for this can be found in the difference between coal and gas prices. In Fig.
C.1 we observe that coal is about 20 e/MWh cheaper than gas, and emits about 1
ton/MWh. So, only a CO2 price higher than 20 e/ton will start shifting the coal
plants in the merit order. For higher CO2 prices the effect is modest. For the alternative portfolio, there is immediately a large effect noticeable. The reason for this is
that already with low CO2 prices, nuclear and lignite plants will shift in the merit
order and these are exactly the two types of plants that have the largest difference
in CO2 emission.
These results also suggest that with low CO2 prices, demand response strategies,
i.e. shifting load to times with low system demand, might actually lead to higher
emissions. As long as the most polluting generation technologies are first in the
merit order, any strategy with the aim of shifting load to off peak periods will
hence increase the use of the most polluting plants and lead to an increase in CO2
emissions.
Appendix C: Carbon emmissions due to EV charging
Peak Charging, Dutch Portfolio
Installed Wind Capacity (GW)
Installed Wind Capacity (GW)
Night Charging, Dutch Portfolio
20
15
10
5
0
0
20
40
60
80
20
0.9
15
0.85
10
0.8
5
0.75
0
100
0
20
40
60
80
CO2 Price (Euro/ton)
CO2 Price (Euro/ton)
Night Charging, Alternative Portfolio
Peak Charging, Alternative Portfolio
100
0.7
20
Installed Wind Capacity (GW)
Installed Wind Capacity (GW)
0.65
15
10
5
0
0
20
40
60
80
CO2 Price (Euro/ton)
100
20
0.6
15
0.55
10
CO2 Emissions EV Charging (kg/kWh)
156
0.5
5
0.45
0
0
20
40
60
80
CO2 Price (Euro/ton)
100
0.4
Figure C.5 – CO2 emissions as a function of CO2 price and installed wind capacity for both
peak and night charging, and both portfolios.
Combined effects
In the previous sections we have shown how different aspects influence CO2 emissions
of EV charging. It is instructive to also see the combined effect of all these aspects.
Fig. C.5 essentially contains all information we have presented so far. An important
conclusion based on this figure would be the strong sensitivity of the emissions for
every single aspect we have considered. It is, however, especially the combination
of aspects that determines the exact numbers. This would make any prediction one
does about the future emissions of EV charging very difficult. The range in which
the CO2 emissions of EV charging can be found is approximately from 0.4 to 1.4
kg/kWh, or a factor of 3.5.
Conclusions
The most important overall finding from the work presented in this appendix is
that CO2 emissions due to electric vehicle charging are determined in a sometimes
counterintuitive way by a combination of the merit order and the timing of charging,
because these determine which plants are dispatched for charging the EVs. Using
the average CO2 intensity of electricity generation to estimate the emissions caused
by charging electric vehicles can thus lead to imprecise outcomes. The emissions
of charging electric vehicles depend strongly on the time of day of charging, the
157
generation portfolio, the amount of installed wind power and the price of CO2 .
With the most-polluting plants being first in the merit order, demand response
causing a shift in load to off-peak periods will lead to higher CO2 emissions.
Furthermore, it was found that emissions of EV charging may range between 0.4
and 1.4 kg/kWh. With an EV driving efficiency range of 0.15-0.2 km/kWh, this
would translate to 60-280 g/km. Compared with the typical emissions of modern
conventional gasoline vehicles, that are around 130 g/km, switching to electric mobility does by no means guarantee a large reduction in greenhouse gas emissions. A
much large share of RES combined with clean dispatchable fossil fuel generation will
be needed to achieve this.
158
Appendix C: Carbon emmissions due to EV charging
Appendix D: Synthetic driver
profiles
This appendix describes the method that was used to convert the driver data that
originates from [29] to a set of aggregated driver profiles. The reason to use these
synthetic profiles is that the various optimization problems that have been used
throughout this thesis are computationally only tractable with a limited amount
of drivers. The number of drivers still leading to reasonable computation times
was found to be in the order 10-50 drivers, depending mainly on the time-horizon
of the optimization. Given this relatively low number, it was found that random
sampling of the dataset resulted in too large deviations from the average numbers,
so a method of constructing aggregate driver profiles that nonetheless reflect the
actual distributions was chosen. The method that has been used is a slightly adapted
version of the K-means clustering algorithm, which we describe briefly below. More
details on this algorithm can e.g. be found in [113].
K-means clustering
In words, the central idea behind K-means clustering is to assign datapoints to a
cluster (a group of datapoints) based on the Euclidean distance of the point to
the mean of the cluster. After the point has been assigned, new cluster means are
computed and the algorithm proceeds to the next datapoint.
An N-dimensional point is denoted as x with xi denoting its N components. The
‘distance’ between two points x and y can be defined in various ways, but here (and
in most versions of the standard K-means algorithm) we use the squared Euclidean
distance defined as
1∑
d(x, y) =
(xi − yi )2
(1)
2 i
The mean of cluster k is denoted as m(k) . As initialization, the cluster means are
chosen to be equal to N randomly selected samples. The algorithm consists of an
assigment and an update step. In the assignemnt step, a datapoint x(n) is assigned
to a cluster k̂ (n) according to:
k̂ (n) = arg min{d(x(n) , m(k) )}
k
159
(2)
160
Appendix D: Synthetic driver profiles
i.e. find the cluster whose mean has the lowest squared Euclidean distance to the
datapoint.
In the update step, the new cluster means are computed by
m̂(k) =
1
Nk
∑
x(n)
(3)
x(n) ∈K(k)
where Nk denotes the number of datapoints in cluster k and K(k) denotes the set of
all observations in cluster k. The algorithm, which can be shown to always converge
(see [113]), stops if no more changes occur in the assignment step.
Adapted method for equal cluster sizes
For the purpose to use the synthetic driver profiles in the optimization formulations
used in this thesis it is convenient to work with equal cluster sizes (the number of
datapoints in each cluster) N . This allows the profiles to be scaled easily to represent
an aggregated number of vehicles. We have adapted the method in a pragmatic way
to obtain equal cluster sizes. The idea is simply to redistribute the overpopulated
clusters (i.e. with more than the predefined cluster size) to the underpopulated
clusters, again based on minimizing the squared Euclidean distance to it. This step
is performed after each loop through all the datapoints and after the redistribution
step the new cluster means are calculated accordingly.
The redistribution steps can be described as follows:
Find set of overpopulated clusters k + with Nk > N . The underpopulated clusters
are k − .
For all k ∈ k + find the Nk − N datapoints with the largest Euclidean distance to
the cluster mean. Denote this set of datapoints RD
−
For all x(n) ∈ RD assign to new cluster k according to k̂ (n) = arg min
{d(x(n) , m(k ) )}
−
and update k − after each new assigment
k
Repeat untill k − = ∅
Because this procedure has no guaranteed convergence, a heuristic stopping criterion based on the number of changes in the assignment step of the K-means algorithm (not the assignment step during the redistribution steps) was used.
Application to driver data
The original driving data that has been used consists of approximately 18.000 individual drivers and is described more extensively in chapter 2 of this thesis. Three
parameters were considered relevant for our purposes: first departure time from
home, last arrival time at home and distance driven that day. The datapoints are
thus three-dimensional and, furthermore, we have chosen a number of 25 clusters.
The algorithm was found to converge to an alternating state where approximately
50 datapoints (i.e. less than 0.3% of the total) were alternatively assigned to different
161
probability
1
Synthetic
Real data
0.5
0
0
5
10
15
departure time
20
5
10
arrival time
20
probability
0.4
0.2
0
0
15
probability
0.4
0.2
0
0
50
100
150
distance (km)
200
250
Figure D.1 – Comparison of probability distributions of home departure time, home arrival
time and daily driven distance between the original dataset and the synthesized dataset
clusters. After reaching this state the algorithm was stopped manually and the
remaining datapoints were discarded.
The resulting set of driver profiles is given in Table D.1. To compare the synthesized driver set with the original dataset, we plot the probability distributions of
the three different parameters in Fig. D.1. We observe that the distributions have
roughly a similar shape. However, due to the small number of drivers in the synthesized set, there are unevitably some differences between the distributions. Most
notably, the departure and the arrival times seem to be more centered towards the
average departure and arrival times. Still, the main characteristics of the driving
data have been preserved to a reasonable extent and we consider the synthetic profiles a better alternative than randomly sampling 25 drivers from the original data.
162
Appendix D: Synthetic driver profiles
Table D.1 – Synthetic driver profiles
Driver Departure time Arrival time Distance (km)
1
10.11
13.12
3.1
2
9.26
18.34
35.2
3
9.31
15.21
14.5
4
8.41
14.26
7.8
5
11.21
18.54
16.8
6
9.13
15.27
10.9
7
9.16
16.50
24.8
8
10.27
15.20
5.7
9
9.06
18.33
50.8
10
11.07
18.58
21.7
11
11.31
19.11
28.2
12
9.26
19.25
109.2
13
13.21
18.26
11.8
14
9.16
18.45
58.8
15
9.07
17.17
31.0
16
15.04
19.22
3.2
17
9.13
18.25
40.1
18
9.11
18.59
77.0
19
10.11
18.46
45.2
20
9.40
19.22
65.9
21
7.38
17.58
4.3
22
8.38
22.17
4.5
23
9.08
16.01
19.5
24
9.09
19.24
91.8
25
15.09
19.48
7.9
Appendix E: Cold storage as
another resource for demand
response
This appendix is based on the work described in [102]
This thesis, as its subtitle suggests, explores the potential of flexible electricity
demand. So far, though, we have only considered EV demand as the main source
of flexibility. In principle, other types of electricity demand could offer demand
response services as well. Similar types of optimization problems can be formulated
in terms of similar objectives described earlier in this thesis, either related to price
or to reduce network loads or losses. In this appendix we briefly describe another
type of responsive demand: a cold storage warehouse. Here we only summarize the
work that has been described in more detail in [102], and we briefly touch upon the
similarities and difference with the EV charging optimization.
A cold storage warehouse is basically a huge refrigerator where goods are stored
at very low temperatures. In a simple approximation of temperature dynamics, an
equation for the cold store temperature Tc in terms of the ambient temperature Ta
and cooling power Pc can be derived. It reads, in discrete time-step k:
Tc [k + 1] = (1 − a)Tc [k] + aTa [k] − bPc [k]
(4)
with
U A∆t
Cp
η∆t
b=
Cp
a=
(5)
(6)
where the different symbols, and their values used in an example simulation, are
listed in Table E.1. This equation is analogous to Eq. 2.8 that describe the energy
content of an EV battery. Indeed, instead of electro-chemical energy stored in a
battery we now have heat (or rather: cold) stored in the cold storage warehouse
- they are of course the state variables (a term often used in control theory) that
physically describe the system. The ‘sink’ term is now given by a(Tc [k] − Ta [k]),
163
164
Appendix E: Cold storage as another resource for demand response
Table E.1 – Physical and simulation parameters
Cp
UA
∆t
η
Pc,min
Pc,max
P Vmax
Tmax
Ta
Value
Unit Description
2500 kWh/K Cold store heat capacity
20 kW/K Heat transfer coefficient between
cold store and ambient
0.25
h Simulation time-step
3
- Cooling power efficiency
0
kW Minimum cooling power
2000
kW Maximum cooling power
2000
kW Installed PV capacity
◦
-18
C Temperature limit of cold store
◦
15
C Ambient temperature
whereas in the case of the EVs energy was ‘leaking’ through the term that described
battery discharge due to driving ηd Lk . The ‘source’ is now the cooling power Pc
whereas in the EV case it was the charging power PEV .
An evident optimization strategy would be to minimize the total costs for cooling, while maintaining the cold store temperature within limits. The optimization
formulation reads:
Nk
∑
min
Pc [k]
λ[k]Pc [k]
(7)
k=1
s.t. Tc,min ≤ Tc [k] ≤ Tc,max
(8)
Pc,min ≤ Pc [k] ≤ Pc,max
(9)
Tc [k + 1] = (1 − a)Tc [k] + aTa [k] − bPc [k]
(10)
This formulation is almost completely identical to the EV charging optimization
formulation given in Eqs. 2.9 to Eq. 2.12, except for the dynamic equation that is
slightly different.
Case study simulation
As an illustration of how the flexibility of a cold storage warehouse can be exploited,
we consider the setting depicted in Fig. E.1. The cold store has PV production on
the same site, and it can both withdraw power from the grid at a (time-varying)
cost Cin [k] and feed back against a certain tariff Cout [k]. In many countries, and
presumably even more so in the future, tariffs for consuming are higher than for
delivering energy to the grid. The challenge is thus to find the optimal cooling
schedule that maximizes the profits of energy delivery minus the cost of purchase.
The objective function is hence given by:
min
Pin [k],Pout [k]
Nk
∑
k=1
Cin [k]Pin [k] − Cout [k]Pout [k]
(11)
165
Figure E.1 – Schematic representation of the cold store with installed PV capacity. The
cold can withdraw power from the grid Pin at a cost Cin and feed power Pout into the grid
for a price Cout
If no electricity can be stored on site, we have an extra equation to link the PV
power and the in- and outflows of electricity:
Pc [k] = PP V [k] + Pin [k] − Pout [k]
(12)
Together with the constraints described above, these equations are the complete
optimization formulation of the problem. The main goal of the work described in
[102] was to assess the effect of different electricity tariff structures on cold storage
electricity demand. Five different tariff structures were studied, see Fig. E.2.
The price scenarios range from a flat tariff to a real-time price, modeled on the
basis of wholesale prices. In scenario E, a real-time price that would be observed in
a system with a large amount of solar power is modeled.
Fig. E.3 shows the optimization results for what could be considered the most
extreme cases: the flat tariff without a feed in penalty (3(a)) and the real time tariff
(with feed-in penalty) based on wholesale prices in a high solar system. The PV
power is modeled so as to represent two sunny and two cloudy days, see the upper
panels of Fig. E.3; the ambient temperature is assumed to be constant at 15 ◦ C.
One observes how in the flat tariff case, the cooling power is constant and keeps
the cold storage temperature at exactly the upper temperature limit, which makes
sense because the ‘heat loss’ to the environment is proportional to the temperature
difference Tc − Ta .
In the case of the real time wholesale based tariff, a markedly different cooling
trajectory is optimal. Here it pays off to use all PV power for cooling during the
sunny days and ‘ride through’ the cloudy days without having to buy electricity
from the grid. This is at the expense of higher thermal losses, of course, but the
price differences make it worthwhile.
The results for the other tariff cases are listed in Table E.2. One observes how, depending on the tariff, markedly different outcomes are found in terms of the amounts
of energy ands maximum power. From a network point of view, for instance, one
could argue that smaller maximum power withdrawal or feed-in are beneficial something that is not properly incentivised in many tariffs.
The economic value of the look-ahead energy management strategies can be
evaluated by considering the difference with the uncontrolled cooling schedule (this
166
Appendix E: Cold storage as another resource for demand response
Price (EURct/kWh)
15
Feed In
10
5
A: flat tariff
0
0
0.5
1
1.5
2
2.5
0
0.5
1
1.5
2
2.5
0
0.5
1
1.5
2
2.5
3
3.5
4
10
5
B: feed in penalty
0
3
3.5
4
3.5
4
3.5
4
15
10
5
C: day−night
Price (EURct/kWh)
0
15
Price (EURct/kWh)
Price (EURct/kWh)
15
Price (EURct/kWh)
Consume
15
3
10
5
D: APX based
0
0
0.5
1
1.5
2
2.5
3
10
5
E: APX, high solar
0
0
0.5
1
1.5
2
Days
2.5
3
3.5
4
Figure E.2 – Different price scenarios.
2000
PV Output (kW)
PV Output (kW)
2000
1500
1000
500
0
0
0.5
1
1.5
2
2.5
3
3.5
500
0
200
100
1
1.5
2
2.5
3
3.5
4
0
0.5
1
1.5
2
2.5
3
3.5
0
0.5
1
1.5
2
2.5
3
3.5
4
0
0.5
1
1.5
2
Days
2.5
3
3.5
4
1000
500
0
4
−15
Temperature (o C)
Temperature (o C)
0.5
1500
−17
−18
−19
−20
0
2000
Cooling Power (kW)
Cooling Power (kW)
1000
4
300
0
1500
0
0.5
1
1.5
2
Days
2.5
(a) Flat tariff
3
3.5
4
−20
−25
−30
−35
(b) Wholesale price based - high solar
Figure E.3 – PV production, optimal cooling trajectory and cold storage temperature for
two different tariffs.
167
Table E.2 – Overview of some relevant simulation results.
Price case
Profit
A: flat tariff
B: flat tariff, feed-in penalty
C: day-night tariff
D: APX based real time
E: APX based high solar
(EUR)
8500
5310
6220
7960
4960
Cooling
Energy
Used
(MWh)
19.4
20.3
19.7
20.7
22.5
Energy
Withdrawn
(MWh)
9.9
0
5.1
18.9
0
Energy
Fed In
(MWh)
31.2
20.4
26.2
38.9
18.2
Maximum
Power
Withdrawn
(kW)
220
0
220
2000
0
Maximum
Power
Fed In
(kW)
1780
1780
1780
2000
1504
9000
Controlled
Uncontrolled
8000
7000
Profit (EUR)
6000
5000
4000
3000
2000
1000
0
A
B
C
Price Case
D
E
Figure E.4 – Comparison of the profits between the different price cases and the effect of
the optimization.
would be the same as in scenario A, since this minimizes the total cooling energy
needed). Fig. E.4 shows these differences. Scenario E shows the highest value for
the look-ahead optimization, which can be understood by the specific combination
of PV production and time varying prices. If these happen to coincide, there is most
incentive in using all locally produced energy.
The example treated in this appendix shows the similarity between EV load
management and another form of demand response. In the end, it is the physical
properties of the system under consideration that determine the economic potential.
In this perspective, it is interesting to note that the physics of the cold storage
are essentially determined by two parameters: its thermal mass, represented by Cp
and the thermal leakage contained in U A. The ratio of these two defines a certain
fundamental ‘time constant’ associated with the system. In this case we find for this
time constant a value of 125 hours, but for example a household refrigeration system
a much smaller time constant is found, roughly in the order of several hours. This
time constant determines, to a large extent, how flexible the system is, i.e. how long
168
Appendix E: Cold storage as another resource for demand response
the consumption of energy can be postponed and how much value can be created
by looking ahead and awaiting favorable conditions.
In the case of EVs, on the other hand, a similar time constant can be determined
by comparing the battery capacity with the typical daily driving energy needed. In
the simulations considered in this thesis that amounted to approximately 4 days, so
a similar order of magnitude. This, to a large extent, explains the suitability of EVs
for demand side management compared to other forms like heating and cooling of
residential buildings.
Bibliography
[1] International Panel on Climate Change (IPCC), “IPCC Fourth Assessment
Report: Climate Change 2007 (AR4),” 2007.
[2] Eurostat, “Electricity and natural gas price statistics.”
epp.eurostat.ec.europa.eu, 2013. Last visited July 23rd 2013.
http://
[3] EIA, “Energy information administration.” www.eia.doe.gov.
[4] European Commission (EC), “A roadmap for moving to a competitive low
carbon economy in 2050,” March 2011. EU Commission (DG Climate),
COM(2011) 112 final.
[5] US Government, “President Obama’s Plan to Fight Climate Change.”
www.whitehouse.gov/share/climate-action-plan, 2013. Last visited July 23rd
2013.
[6] Social and Economic Council of the Netherlands (SER), “Energieakkoord voor duurzame groei (Energy agreement for sustainable growth).”
www.energieakkoordser.nl/, 2013. In Dutch. Last visited July 23rd 2013.
[7] Global Wind Energy Council (GWEC), “Wind energy outlook 2010,” 2010.
[8] European Network of Transmission System Operators for Electricity (ENTSOE), “Scenario Outlook and Adequacy Forecast 2012 2030,” 2012.
[9] European Photovoltaic Industry Association, “Global market outlook for
photovoltaics until 2016,” May 2012.
[10] Global Wind Energy Council (GWEC), “Global Wind Statistics 2012,” 2013.
[11] World Wind Energy Association, “Wind Energy International 2011/2012.”
http://wwindea.org, 2012.
[12] International Energy Agency (IEA), “Harnessing variable renewables.,” tech.
rep., 2011.
[13] International Energy Agency (IEA), “Technology roadmap electric and plug-in
hybrid electric vehicles,” June 2011.
169
170
Bibliography
[14] Ministry of Transport, Public Works and Water Management, “Plan van Aanpak Elektrisch Rijden (in Dutch).” last visited October 2010, 2009.
[15] Agentschap NL, “Cijfers elektrisch vervoer (in dutch).”
www.agentschapnl.nl/onderwerp/cijfers-elektrisch-vervoer, 2013.
visited July 24th 2013.
http://
Last
[16] I. Pérez-Arriaga, ed., Regulation of the Power Sector. Springer, 2013.
[17] F. Schweppe, M. Caramanis, R. Tabors, and R. Bohn, Spot Pricing of Electricity. Kluwer Academic Publishers, 1988.
[18] D. Kirchen and G. Strbac, Fundamentals of Power System Economics. WileyInterscience, 2004.
[19] S. Stoft, Power System Economics. Wiley-Interscience, 2002.
[20] M. Ilic, F. Galiana, and L. Fink, eds., Power Systems Restructuring. Kluwer
Academic Publishers, 1998.
[21] A. Gómez-Expósito, A. Conejo, and C. Cañizares, Electric Energy Systems:
Analysis and Operation. CRC Press, 2009.
[22] Energie Data Services Nederland, “Dutch consumption
www.edsn.nl/verbruiksprofielen. Last accessed September 2013.
profiles.”
[23] L. De Vries, Securing the public interest in electricity generation markets. PhD
thesis, Delft University of Technology, The Netherlands, 2004.
[24] Nord Pool Spot, “Elspot Prices.” http://www.nordpoolspot.com, 2013. Last
accessed june 2013.
[25] EEX, “European energy exchange.” www.eex.com.
[26] Enexis, “Jaarverslag 2012 (in dutch).” www.enexis.nl/jaarverslag2012.
[27] T. Gómez San Román, I. Momber, M. Rivier Abbad, and A. Sánchez Miralles, “Regulatory framework and business models for charging plug-in electric
vehicles: Infrastructure, agents, and commercial relationships,” Energy Policy,
vol. 39, no. 10, pp. 6360 – 6375, 2011.
[28] R. Verzijlbergh, Z. Lukszo, E. Veldman, J. Slootweg, and M. Ilic, “Deriving
electric vehicle charge profiles from driving statistics,” in Power and Energy
Society General Meeting, 2011 IEEE, pp. 1–6, July 2011.
[29] Ministry of Transport, Public Works and Water Management, “Mobiliteitsonderzoek Nederland (in Dutch),” April 2009. Last visited October 2010.
[30] K. Young, C. Wang, L. Wang, and K. Strunz, “Electric vehicle battery
technologies,” in Electric Vehicle Integration into Modern Power Networks
(R. Garcia-Valle and J. A. Peas Lopes, eds.), Power Electronics and Power
Systems, pp. 15–56, Springer New York, 2013.
Bibliography
171
[31] J. Dogger, B. Roossien, and F. Nieuwenhout, “Characterization of li-ion batteries for intelligent management of distributed grid-connected storage,” Energy Conversion, IEEE Transactions on, vol. 26, no. 1, pp. 256–263, 2011.
[32] N. Rotering and M. Ilic, “Optimal charge control of plug-in hybrid electric
vehicles in deregulated electricity markets,” Power Systems, IEEE Transactions on, vol. 26, pp. 1021 –1029, aug. 2011.
[33] J. Larminie and J. Lowry, Electric Vehicle Technology Explained. John Wiley
& Sons, Ltd., 2012.
[34] S. Peterson, J. Whitacre, and J. Apt, “The economics of using plug-in hybrid
electric vehicle battery packs for grid storage,” Journal of Power Sources,
vol. 195, no. 8, pp. 2377–2384, 2010.
[35] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge university
press, 2004.
[36] S. Rao, Engineering optimization: theory and practice. Wiley, 2009.
[37] T. Edgar, D. Himmelblau, and L. Lasdon, Optimization of chemical processes.
Second edition. McGraw-Hill, 2001.
[38] R. Verzijlbergh, Z. Lukszo, and M. Ilic, “Comparing different ev charging
strategies in liberalized power systems,” in European Energy Market (EEM),
2012 9th International Conference on the, pp. 1 –8, may 2012.
[39] T. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging of electric
drive vehicles in a market environment,” Applied Energy, vol. 88, no. 5,
pp. 1940–1948, 2011.
[40] E. Sortomme, M. Hindi, S. MacPherson, and S. Venkata, “Coordinated charging of plug-in hybrid electric vehicles to minimize distribution system losses,”
Smart Grid, IEEE Transactions on, vol. 2, pp. 198 –205, march 2011.
[41] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Calculating
capacity and net revenue,” Journal of Power Sources, vol. 144, no. 1, pp. 268–
279, 2005.
[42] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: From
stabilizing the grid to supporting large-scale renewable energy,” Journal of
Power Sources, vol. 144, no. 1, pp. 280–294, 2005.
[43] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power and Energy
Magazine, vol. 7, no. 2, pp. 52–62, 2009.
[44] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging plug-in
hybrid electric vehicles on a residential distribution grid,” IEEE Transactions
on Power Systems, vol. 25, no. 1, pp. 371–380, 2010.
[45] J. Tomic and W. Kempton, “Using fleets of electric-drive vehicles for grid
support,” Journal of Power Sources, vol. 168, no. 2, pp. 459–468, 2007.
172
Bibliography
[46] H. Lund and W. Kempton, “Integration of renewable energy into the transport
and electricity sectors through v2g,” Energy Policy, vol. 36, no. 9, pp. 3578–
3587, 2008.
[47] C. Guille and G. Gross, “A conceptual framework for the vehicle-to-grid (v2g)
implementation,” Energy Policy, vol. 37, no. 11, pp. 4379–4390, 2009.
[48] S. Han, S. Han, and K. Sezaki, “Development of an optimal vehicle-to-grid aggregator for frequency regulation,” IEEE Transactions on Smart Grid, vol. 1,
no. 1, pp. 65–72, 2010.
[49] S. Peterson, J. Apt, and J. Whitacre, “Lithium-ion battery cell degradation resulting from realistic vehicle and vehicle-to-grid utilization,” Journal
of Power Sources, vol. 195, no. 8, pp. 2385 – 2392, 2010.
[50] D. Richardson, “Electric vehicles and the electric grid: A review of modeling approaches, impacts, and renewable energy integration,” Renewable and
Sustainable Energy Reviews, vol. 19, no. 0, pp. 247 – 254, 2013.
[51] W. Kempton and S. Letendre, “Electric vehicles as a new power source for electric utilities,” Transportation Research Part D: Transport and Environment,
vol. 2, no. 3, pp. 157–175, 1997.
[52] P. Denholm and W. Short, “An evaluation of utility system impacts and benefits of optimally dispatched plug-in hybrid electric vehicles,” Tech. Rep.
NREL/TP-620-40293, NREL, 2006.
[53] P. Sanchez-Martin, G. Sanchez, and G. Morales-Espana, “Direct load control decision model for aggregated ev charging points,” Power Systems, IEEE
Transactions on, vol. 27, no. 3, pp. 1577–1584, 2012.
[54] M. Galus, M. Zima, and G. Andersson, “On integration of plug-in hybrid
electric vehicles into existing power system structures,” Energy Policy, vol. 38,
no. 11, pp. 6736 – 6745, 2010.
[55] J. Peças Lopes, F. Soares, and P. Almeida, “Integration of electric vehicles in
the electric power system,” Proceedings of the IEEE, vol. 99, no. 1, pp. 168–
183, 2010.
[56] M. Ilic, L. Xie, and J.-Y. Joo, “Efficient coordination of wind power and priceresponsive demand;part i: Theoretical foundations,” Power Systems, IEEE
Transactions on, vol. 26, no. 4, pp. 1875–1884, 2011.
[57] M. Ilic, J.-Y. Joo, L. Xie, M. Prica, and N. Rotering, “A decision-making
framework and simulator for sustainable electric energy systems,” IEEE Transactions on Sustainable Energy, vol. 2, pp. 37–49, 2011.
[58] B. Sovacool and R. Hirsh, “Beyond batteries: An examination of the benefits
and barriers to plug-in hybrid electric vehicles (phevs) and a vehicle-to-grid
(v2g) transition,” Energy Policy, vol. 37, pp. 1095–1103, Mar. 2009.
Bibliography
173
[59] R. Green II, L. Wang, and M. Alam, “The impact of plug-in hybrid electric vehicles on distribution networks: A review and outlook,” Renewable &
Sustainable Energy Reviews, vol. 15, pp. 544–553, 2011.
[60] K. Clement-Nyns, Impact of Plug-in Hybrid Electric Vehicles on the Electricity
System. PhD thesis, Katholieke Universiteit (KU) Leuven, 2010.
[61] J. Taylor, A. Maitra, M. Alexander, D. Brooks, and M. Duvall, “Evaluations
of plug-in electric vehicle distribution system impacts,” in IEEE Power and
Energy Society General Meeting, pp. 1–6, 2010.
[62] L. Pieltain Fernandez, T. Gómez San Román, R. Cossent, C. Domingo, and
P. Frias, “Assessment of the impact of plug-in electric vehicles on distribution
networks,” Power Systems, IEEE Transactions on, vol. 26, no. 1, pp. 206–213,
2011.
[63] E. Sortomme and M. El-Sharkawi, “Optimal scheduling of vehicle-to-grid energy and ancillary services,” Smart Grid, IEEE Transactions on, vol. 3, no. 99,
pp. 351 – 359, 2012.
[64] E. Sortomme and M. El-Sharkawi, “Optimal combined bidding of vehicle-togrid ancillary services,” Smart Grid, IEEE Transactions on, vol. 3, pp. 70 –79,
march 2012.
[65] J. Donadee and M. Ilic, “Stochastic co-optimization of charging and frequency regulation by electric vehicles,” in North American Power Symposium
(NAPS), 2012, pp. 1–6, Sept.
[66] A. Saber and G. Venayagamoorthy, “Intelligent unit commitment with vehicleto-grid -a cost-emission optimization,” Journal of Power Sources, vol. 195,
no. 3, pp. 898–911, 2010.
[67] C. Fernandes, P. Fras, and J. Latorre, “Impact of vehicle-to-grid on power
system operation costs: The spanish case study,” Applied Energy, vol. 96,
pp. 194 – 202, 2012.
[68] F. Banez-Chicharro, J. Latorre, and A. Ramos, “Smart charging profiles for
electric vehicles,” Computational Management Science, pp. 1–24, 2013.
[69] R. Verzijlbergh, M. Grond, Z. Lukszo, J. Slootweg, and M. Ilic, “Network
impacts and cost savings of controlled ev charging,” IEEE Transactions on
Smart Grid, vol. 3, pp. 1203 –1212, sept. 2012.
[70] J. Lassila, J. Haakana, J. Partanen, J. Koivuranta, and S. Peltonen, “Network effects of electric vehicles - case from nordic country,” in CIRED 21st
International Conference on Electricity Distribution, 2011.
[71] E. Veldman, D. Geldtmeijer, J. Knigge, and J. Slootweg, “Smart grids put into
practice: Technological and regulatory aspects,” Competition and Regulation
in Network Industries, vol. 11, no. 3, pp. 287–307, 2010.
174
Bibliography
[72] R. Verzijlbergh, Z. Lukszo, J. Slootweg, and M. Ilic, “The impact of controlled electric vehicle charging on residential low voltage networks,” in Networking, Sensing and Control (ICNSC), 2011 IEEE International Conference
on, pp. 14–19, april 2011.
[73] M. Grond, “Impact of future residential loads on medium voltage networks,”
Master’s thesis, Delft University of Technology, 2011.
[74] Vision Network Analysis. http://www.phasetophase.nl/nl products/vision
network analysis.html.
[75] Energiened, Elektriciteitsdistributienetten. Kluwer Techniek, 1996. in Dutch.
[76] R. Verzijlbergh, C. Brancucci Martı́nez-Anido, L. De Vries, and Z. Lukszo,
“Does controlled electric vehicle charging substitute cross-border transmission
capacity?,” Applied Energy, 2013. Accepted for publication.
[77] A. Purvins, A. Zubaryeva, M. Llorente, E. Tzimas, and A. Mercier, “Challenges and options for a large wind power uptake by the european electricity
system,” Applied Energy, vol. 88, no. 5, pp. 1461 – 1469, 2011.
[78] E. Sortomme and M. A. El-Sharkawi, “Optimal combined bidding of vehicleto-grid ancillary services,” Smart Grid, IEEE Transactions on, vol. PP, no. 99,
pp. 1 –10, 2011.
[79] European Climate Foundation, “Roadmap 2050 - a practical guide to a prosperous low-carbon Europe,” 2010.
[80] F. Steinke, P. Wolfrum, and C. Hoffmann, “Grid vs. storage in a 100% renewable europe,” Renewable Energy, vol. 50, no. 0, pp. 826 – 832, 2013.
[81] D. Heide, L. von Bremen, M. Greiner, C. Hoffmann, M. Speckmann, and
S. Bofinger, “Seasonal optimal mix of wind and solar power in a future, highly
renewable europe,” Renewable Energy, vol. 35, no. 11, pp. 2483 – 2489, 2010.
[82] C. Brancucci Martı́nez-Anido, M. Vandenbergh, L. De Vries, C. Alecu,
S. Purvins, G. Fulli, and T. Huld, “Medium-term demand for european crossborder electricity transmission capacity,” Energy Policy, vol. 61, no. 0, pp. 207–
222, 2013.
[83] C. Brancucci Martı́nez-Anido, Electricity without borders; The need for crossborder transmission investment in Europe. PhD thesis, Delft University of
Technology, 2013.
[84] European Union Road Federation, “ERF 2011 European Road Statistics,”
tech. rep., 2011.
[85] E. Castillo, A. Conejo, P. Pedregal, R. Garciá, and N. Alguacil, Building and
Solving Mathematical Programming Models in Engineering and Science. John
Wiley & Sons, Inc., 2002.
Bibliography
175
[86] “IBM ILOG CPLEX (Version 12.5).”
[87] T. Huld, R. Müller, and A. Gambardella, “A new solar radiation database for
estimating pv performance in europe and africa,” Solar Energy, vol. 86, no. 6,
pp. 1803 – 1815, 2012.
[88] E. Kalnay, M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, and L. Gandin et
al, “The NCEP/NCAR 40-year reanalysis project,” Bull. Amer. Meteor. Soc.,
vol. 77, pp. 437–471, Mar. 1996.
[89] R. Verzijlbergh, L. De Vries, and Z. Lukszo, “Renewable energy and responsive
demand: do we need congestion management in the distribution grid?,” IEEE
Transactions on Power Systems, 2013. Under review.
[90] O. Sundstrom and C. Binding, “Flexible charging optimization for electric
vehicles considering distribution grid constraints,” Smart Grid, IEEE Transactions on, vol. 3, no. 1, pp. 26–37, March.
[91] N. O’Connell, Q. Wu, J. Østergaard, A. Nielsen, S. Cha, and Y. Ding, “Dayahead tariffs for the alleviation of distribution grid congestion from electric
vehicles,” Electric Power Systems Research, vol. 92, pp. 106–114, 2012.
[92] P. Bach Andersen, J. Hu, and K. Heussen, “Coordination strategies for distribution grid congestion management in a multi-actor, multi-objective setting,”
in Innovative Smart Grid Technologies (ISGT Europe), 2012 3rd IEEE PES
International Conference and Exhibition on, pp. 1–8, Oct.
[93] TenneT, “Dutch transmission system operator.” www.tennet.org.
[94] J. Bard, Practical Bilevel Optimization. Kluwer Academic Publishers, 1998.
[95] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, Congestion management in a smart grid via shadow prices, pp. 518–523. IFAC Workshop Series,
Elsevier Science, 2012.
[96] K. Spees and L. B. Lave, “Demand response and electricity market efficiency,”
The Electricity Journal, vol. 20, no. 3, pp. 69 – 85, 2007.
[97] Enipedia, “Data on energy and industry statistics.” http://enipedia.tudelft.nl/
wiki/Netherlands/Powerplants.
[98] E. Veldman and R. Verzijlbergh, “Smart charging of electric vehicles: Opportunity or threat for distribution grids?.” Submitted to IEEE Transactions on
Smart Grid, 2013.
[99] M. Ilić, L. Xie, and Q. Liu, eds., Engineering IT-Enabled Sustainable Electricity
Services. The Tale of Two Low-Cost Green Azores Islands. Springer, 2013.
[100] R. Verzijlbergh, M. Ilic, and Z. Lukszo, “The role of electric vehicles in making
azores islands green,” in Engineering IT-Enabled Electricity Services. The Case
of Low-Cost Green Azores Islands (M. Ilic, L. Xie, and Q. Liu, eds.), ch. 11,
pp. 226–240, Springer, 2013.
176
Bibliography
[101] F. Schweppe, R. Tabors, J. Kirtley, H. Outhred, F. Pickel, and A. Cox,
“Homeostatic utility control,” Power Apparatus and Systems, IEEE Transactions on, vol. PAS-99, pp. 1151 –1163, may 1980.
[102] R. Verzijlbergh and Z. Lukszo, “Conceptual model of a cold storage warehouse
with pv generation in a smart grid setting,” in IEEE International Conference
on Networking, Sensing and Control (ICNSC), 2013.
[103] C. Kobus, R. Mugge, and J. Schoormans, “Washing when the sun is shining!
how users interact with a household energy management system,” Ergonomics,
vol. 56, no. 3, pp. 451–462, 2013.
[104] C. Chorus, T. Arentze, and H. Timmermans, “Spatial choice: a matter of
utility or regret?,” Environment and Planning B: Planning and Design, vol. 36,
no. 3, pp. 538–551, 2009.
[105] L. Gagnon, C. Belanger, and Y. Uchiyama, “Life-cycle assessment of electricity generation options: The status of research in year 2001,” Energy Policy,
vol. 30, pp. 1267–1278, November 2002.
[106] J. Sullivan, R. Baker, B. Boyer, R. Hammerle, T. Kenney, L. Muniz, and
T. Wallington, “Co2 emission benefit of diesel (versus gasoline) powered
vehicles,” Environmental Science & Technology, vol. 38, no. 12, pp. 3217–3223,
2004.
[107] R. Verzijlbergh and Z. Lukszo, “System impacts of electric vehicle charging
in an evolving market environment,” in Networking, Sensing and Control
(ICNSC), 2011 IEEE International Conference on, pp. 20 –25, april 2011.
[108] International Energy Agency (IEA), “Technology Roadmap: High-Efficiency,
Low-Emissions Coal-Fired Power Generation,” tech. rep., 2012.
[109] N. E. Ligterink and R. T. Smokers, “Praktijkverbruik van zakelijke personenautos en plug-in voertuigen (in dutch),” tech. rep., TNO, 2013.
[110] IEA, “International energy agency,
www.iea.org/stats/, 2007.
statistics and balances.” http://
[111] D. Weisser, “A guide to life-cycle greenhouse gas (ghg) emissions from electric
supply technologies,” Energy, vol. 32, pp. 1543–1559, Sept. 2007.
[112] CBS, “Statistics netherlands,” 2013.
[113] D. MacKay, Information Theory, Inference and Learning Algorithms. Cambridge University Press, 2003.
Nomenclature
Symbols
αk
Non-EV dependent part of electricity price at time-step k
β
Sensitivity of electricity price to EV demand
∆t
Simulation time-step
ηc
Charging efficiency
ηd
Driving efficiency
λk
Electricity price at time-step k
µk
Network tariff at time k
dik
Battery discharge due to driving of EV i at time-step k
EEV,ik Energy content of (the battery of) EV i at time-step k
EEVmax ,i Maximum energy content of EV i
EEVmin ,i Minimum energy content of EV i
FLk
Line flow in transmission line L at time-step k
g
Simultaneity factor
Hin,jk Water inflow into pumped hydro unit j at time-step k
Hjk
Water level of pumped hydro unit j at time-step k
Hmax,j Maximum water level of pumped hydro unit j
Hmin,j Minimum water level of pumped hydro unit j
Il
Current flowing through in line l
KL
Capacity of transmission line L
Lik
Distance driven by EV i at time-step k
NG
Total number of generators
177
178
Bibliography
NH
Total number of hydro generators
Nk
Number of time-steps
NEV
Total number of EVs
Pl
Real power flowing through line l
Pbatt,ik Power flow into EV battery i at time-step k
PD,k
Total electricity demand at time-step k
PD0,k Electricity demand (without EV demand) at time-step k
PEV,ik Power flow into EV i at time-step k
PEV,k EV demand at time-step k
PEVmax ,i Maximum charging power of EV i
PEVmin ,i Minimum charging power of EV i
PG,nk Power output of generating unit n at time-step k
PGmax ,n Maximum power output of generating unit n
PGmin ,n Minimum power output of generating unit n
PHmax ,j Maximum power output of pumped hydro unit j
PHmin ,j Minimum power output of pumped hydro unit j
PHjk
Power output of pumped hydro unit j at time-step k
Ploss,l Power loss in line l
Q
Amount of charge inside a battery
Q0
Nominal battery capacity
Rl
Resistance of line l
SoC
State of charge (of a battery)
unk
Binary variable expressing if unit n is on-line at time-step k
VC
Value of EV charging control
VT
Value of transmission capacity
VT +C Value of both transmission capacity and EV charging control
ynk
Binary variable expressing if unit n is in start-up mode at time-step k
znk
Binary variable expressing if unit n is in shut-down mode at time-step k
Bibliography
Acronyms
CHP
Combined heat and power
CPM
Charge point manager
DSO
Distribution system operator
EV
Electric vehicle
HV
High voltage
LV
Low voltage
MV
Medium voltage
MV-D Medium voltage distribution
MV-T Medium voltage transmission
NPV
Net present value
PHEV Plug-in hybrid electric vehicle
PV
Photo-voltaic
RES
Renewable energy sources
RoR
Run-of-river
ToU
Time-of-use
TSO
Transmission system operator
V2G
Vehicle-to-grid
VOLL Value of lost load
179
180
Bibliography
Summary
Introduction and problem statement
Environmental, economic and geo-political concerns continue to drive the decarbonisation of many economies worldwide. In the power sector this is leading to a
sharp increase in renewable energy sources (RES). The most important RES, wind
power and solar (PV) power, are, however, fundamentally different types of electricity production compared to conventional fossil fuel based generation: their output
is variable, less dispatchable and prone to forecast errors. One possible paradigm
to ease the integration of RES is to involve the potential flexibility on the demand
side of the electricity system. Currently, however, the amount of flexible electricity
demand is limited. Electric vehicles, by contrast, provide a source of flexibility of
significant magnitude. Moreover, when powered by green electricity, they pave the
way for strong reductions in greenhouse gas emissions in the transport sector. EVs
and RES are therefore natural candidates to co-exist in a powerful synergy.
Nonetheless, the large increase in electricity use associated with a large scale
adoption of EVs could also require significant infrastructure investments. Here, too,
the flexibility of EVs holds an economic potential because it could reduce the required
network investments. In liberalized power systems, however, generation and retail
on the one hand, and transport and distribution of electricity on the other hand,
are separated activities concerning different actors. The flexibility of EV charging
can provide a value with respect to all these activities, but the associated actor
objectives may be conflicting. This thesis thus aims to answer the following research
question:
How can the flexibility of EV charging best be utilized in multi-actor power systems
with high shares of renewable energy sources?
To answer this question we will first explore controlled EV charging from the point
of view of the distribution networks and from the perspective of integrating variable
RES. Hereafter we combine both perspectives and investigate mechanism to allign
the network and generation related objectives.
181
182
Summary
Research methods, results and insights
Different perspectives on EV charging
In this thesis we study the role of EVs both from a network point of view and in the
light of RES integration in unbundled and liberalized power systems. We therefore
analyse the most important technological, economic and institutional aspects of these
power systems and relevant EV characteristics to obtain a modelling framework that
is used in this thesis.
In the unbundled and liberalized power systems of many western countries, electricity prices emerge as the result of supply and demand bids, and according to
basic economic principles, wholesale prices tend to be correlated with demand: high
prices in periods of peak demand. RES alter this dynamic to some extent, because
they lower prices in periods with high RES output. For an aggregator that aims
to minimize the energy costs of charging a fleet of EVs, it is beneficial to shift the
EV demand to the periods with low prices. Price-responsive EV demand therefore
has the positive effect of creating a flatter residual load profile and dampening price
fluctuations. From the perspective of a distribution system operator (DSO), EV
charging would be scheduled differently, since the DSOs are regulated such that
they have incentives to reduce costs. This would translate in an objective to lower
energy losses and avoid unnecessary investments in distribution assets, which means
that peaks on local network load need to be avoided.
Comparing typical EV battery characteristics such as energy content and power
limits with the typical driving patterns people exhibit, one finds that there is a significant opportunity to postpone the charging process. Using a simple linear model
of EV battery dynamics and e.g. electricity prices or network load we can formulate EV charging as an optimization problem whose objective function depends on
what aspect of the power system and its associated actor are targeted. This thesis
explores the potential of this flexibility first from the point of view of the distribution networks and then from the point of view to lower generation costs, either
directly, or through the signal provided by the wholesale electricity price. Moreover,
we also combine these perspectives and analyze how and when the network and generation related objectives can be conflicting and investigate possible mechanisms to
align them. We assume simplified institutional arrangements where either EVs are
modeled to directly interact with the system or they are represented by an aggregator
as an intermediate party.
Distribution networks impacts
In chapter 4 we study the potential of controlled EV charging with respect to the
electricity distribution system. From this perspective, an EV charging strategy that
shifts the bulk of the energy transfer to the night hours when network load is low,
greatly reduces the extra peak load caused by EV charging that would result from an
uncontrolled charging strategy. Different EV charging profiles, controlled and uncontrolled, are superimposed on existing network load profiles to assess the impacts on
different types of distribution assets. It was found that in the uncontrolled charging
scenarios the extra EV load would lead to a significant amount of overloaded assets,
183
and the effect is most prominent on the MV/LV transformer level. In the controlled
EV charging scenario almost all of the extra replacements compared to the reference
case without any EVs can be avoided. The financial consequences of the different
scenarios have been evaluated by considering the net present value of the annuitized
investment costs and the costs for energy losses. The controlled charging scenario
proved to have an approximately 20% lower NPV than the uncontrolled scenario.
Whereas the energy losses dominate the total cost figures, the differences between
the scenarios are largely caused by the replacement costs. Furthermore, the costs
related to MV-cables were found to be most prominent compared to that of the
MV/LV transformers and HV-substations.
Generation costs and interrelations between controlled EV demand and
cross-border transmission
The objective of chapter 5 is to investigate the potential of EV responsive demand
with respect to lowering marginal generation costs in systems with a high penetration of RES. Furthermore, we look at interdependencies between controlled electric
vehicle charging and cross-border transmission capacity - two paradigms that are
usually seen as important in facilitating RES integration. We extend a unit commitment model with electric vehicle charging power as decision variable and we study
a conceptual two node system based on German data with wind power in one node
and solar power in the other node. The results show that EVs lead to significant cost
reduction because they shift demand to periods of high wind and/or solar power and
limit the use of expensive gas plants. For an expected renewable energy penetration
scenario for 2025, controlled charging and extra transmission capacity can be considered substitute technologies - they both lead to certain cost reduction, more or
less independent of each other. We conclude, however, that with a higher renewable
energy scenario, the demand for energy arbitrage increases and the two technologies
become complementary in the sense that there combined potential is higher than
the sum of their individual effects. The main reason for this is that cross-border
transmission capacity is needed to transport power to where the electric vehicles
can absorb it. These insights are relevant in the light of European renewable energy
targets towards the year 2050.
Aligning distribution network objectives and price-responsive demand
When EVs act as responsive loads that increase demand when electricity prices are
low, they could cause high peaks in network load. The effect becomes more prominent in high RES scenarios, because a high penetration of stochastic generation leads
to a weakened correlation between wholesale electricity prices and network demand.
Chapter 6 therefore investigates possible congestion management mechanisms for
price-responsive electric vehicle demand in electricity distribution networks. It was
shown that application of some form of congestion management is justified, because
limiting the EV load to available network capacity leads to a negligible increase in
energy costs. Simple grid tariffs based on network load were found to make the
problem worse compared to the base case scenario of flat grid tariffs. An optimal
dynamic grid tariff that is unilaterally determined by the DSO leads to desirable
184
Summary
outcomes but is difficult to determine under real life conditions where uncertainty
plays a role. A distribution grid capacity market - an iterative approach where DSO
and aggregator sequentially exchange dynamic tariffs and resulting electricity demand - converges to a final price and charging schedule, but requires a complex IT
infrastructure and has a heavy computational burden. Advance capacity allocation
is more straightforward to implement in the case of a single aggregator, but there
remain important issues related to the allocation of the capacity between multiple
aggregators and the inter-temporal constraints of the EVs. There thus exists a tradeoff between the complexity and efficiency of a congestion management mechanism
which should be subject of further investigation that takes uncertainty into account.
A refined view on EV charging
Chapter 7 aims to connect the different elements discussed in the preceding chapters.
Most notably, we analyze differences between a centralized approach in which generation costs are minimized and a decentralized approach where EV charging costs are
minimized based on wholesale prices. It was found that aggregators who schedule
a very large part of EV demand can exert market power by influencing wholesale
prices. If there is enough competition between aggregators, the aggregated EV demand profile converges to the socially optimal profile.
Furthermore, we show how the value of controlling EV demand depends on various aspects such as the control horizon of the optimization, inter-temporal constraints of generation units and the availability of EVs for charging. These analyses
suggest that also under milder assumptions regarding the predictability of electricity
prices and EV behavior, the value of controlling EV charging is still high, and there
is still a need for congestion management to avoid unnecessary peaks in network
demand. Analyzing the effects of cost minimizing EVs on a large number of distribution networks confirmed that the peaks caused by price-responsive EV load leads
to a significant amount of extra costs related to network investments.
Conclusions
The thesis addresses how the flexibility of EV charging can best be utilized in multiactor power systems with high shares of renewable energy sources. The answer this
question can be summarized as follows: the two important perspectives from which
controlled EV charging can add most value are its ability to be shifted in time
according to fluctuating RES output on the one hand, and to avoid peaks in network
demand to defer or postpone network investments on the other hand. With flat
network tariffs and wholesale prices that will be influenced strongly by fluctuating
RES output, price responsive EV demand can, ironically, lead to even higher demand
peaks than uncontrolled EV charging. The required network reinforcements are
costly and unnecessary because limiting the load to free network capacity through an
efficient congestion management mechanism has negligible additional energy costs.
There are various congestion mechanisms possible to allign the cost minimizing EVs
with network constraints, either based on shadow prices associated with the network
constraints or an ex-ante allocation of free network capacity, but in both approaches
185
there exists a trade-off between simplicity and economic efficiency. All schemes,
however, seem to have in common that the function of the DSO is extended beyond
its current role. A clean and intelligent power system might therefore require a rethinking of the rules and roles in today’s unbundled power systems. When objectives
related to different functions of the electricity system are aligned in such intelligent
IT enabled systems, demand response and EVs in particular can play a key role in
the transition to a cleaner energy system.
186
Summary
Samenvatting
Introductie en probleemstelling
Milieu-gerelateerde, economische en geopolitieke zorgen blijven de drijvende kracht
achter decarbonisering van economiën wereldwijd. In de elektriciteitssector leidt dit
tot een scherpe toename van duurzame energiebronnen. De meeste belangrijke duurzame energiebronnen, wind- en zonne-energie, zijn echter fundamenteel andere vormen van elektriciteitsproductie dan de conventionele centrales op basis van fossiele
brandstoffen: hun productie is variabel, minder goed regelbaar en onderhevig aan
voorspellingsfouten. Eén van de mogelijke paradigma’s om de integratie van duurzame energie te vergemakkelijken is om de potentiële flexibiliteit aan de vraagkant
van het elektriciteitssysteem aan te spreken. Momenteel is de hoeveelheid flexibele
elektriciteitsvraag echter gering. Elektrische auto’s (Electric vehicles EVs) bieden
daarentegen een bron van flexibiliteit van aanzienlijke grootte. Als ze door duurzame bronnen van stroom worden voorzien, banen ze bovendien de weg voor forse
reducties van broeikasgassen in de transportsector. Elektrische auto’s en duurzame
energie zijn daarom logische kandidaten om in een sterke synergie te co-existeren.
Desalniettemin kan de grote toename in elektriciteitsgebruik door de grootschalige introductie van EVs ook aanzienlijke investeringen in infrastructuur vereisen.
Ook hier bevat de flexibiliteit van EVs een economisch potentieel omdat het de benodigde netwerkinvesteringen kan verlagen. In geliberaliseerde en gesplitste1 elektriciteitssystemen zijn productie en handel van elektriciteit aan de ene kant en transport
en distributie aan de andere kant gescheiden activiteiten die verschillende actoren
toebehoren. De flexibiliteit van het laden van EVs kan een waarde voor al deze
actoren vertegenwoordigen, maar de bijbehorende doelstellingen kunnen onderling
strijdig zijn. Dit proefschrift probeert daarom de volgende onderzoeksvraag te beantwoorden:
Hoe kan de flexibiliteit van EVs het best worden benut in multi-actor
elektriciteitssystemen met een hoog aandeel duurzame energie?
Om deze vraag te beantwoorden zullen we eerst het gestuurd laden van EVs vanuit
het oogpunt van de distributienetten bestuderen, en dan vanuit het perspectief van
de integratie van duurzame energie. Hierna combineren we beide perspectieven en
onderzoeken mechanismen om de doelstellingen vanuit de oogpunten van netwerk en
duurzame energieproductie op een lijn te brengen.
1 Gesplitst
verwijst naar de splitsing van commerciële activiteiten en netwerkbeheer.
187
188
Samenvatting
Onderzoeksmethode, resultaten en inzichten
Verschillende perspectieven op het laden van elektrische auto’s
In dit proefschrift bestuderen we de rol van EVs vanuit een netwerkoogpunt en in het
licht van integratie van duurzame energie in geliberaliseerde elektriciteitssystemen.
Daarom analyseren we eerst de belangrijkste technische, economische en institutionele aspecten van deze systemen en de relevante karakteristieken van EVs om een
modelleerraamwerk dat in dit proefschrift gebruikt wordt te verkrijgen.
In de gesplitste en geliberaliseerde elektriciteitssystemen van veel westerse landen, ontstaan elektriciteitsprijzen als het gevolg van biedingen van vraag en aanbod.
Volgens economische basisprincipes zijn de prijzen gecorreleerd met de vraag: hoge
prijzen ten tijde van piekvraag. Duurzame energiebronnen veranderen deze dynamiek in zeker mate, omdat ze de prijzen drukken op momenten van veel duurzame
productie. Voor een aggregator die als doel heeft om de energiekosten van het laden
van een groep EVs te minimaliseren is het dus gunstig om de EV elektriciteitsvraag
te verschuiven naar periodes met lage prijzen. Prijs-responsieve vraag van EVs heeft
daarom een positief effect doordat het zorgt voor een vlakker netto belastingsprofiel
en een dempende werking op fluctuerende prijzen. Vanuit het perspectief van een
distributienetbeheerder (DNB) zou het laden van EVs anders gepland worden, omdat DNBs op zodanige wijze gereguleerd zijn dat ze een prikkel hebben om kosten
te reduceren. Dit zou zich vertalen naar een doelstelling om de de energieverliezen
te verlagen en onnodige investeringen in de netten te vermijden, wat erop neerkomt
dat pieken in de lokale netbelasting voorkomen dienen te worden.
Impacts op de distributienetwerken
In hoofdstuk 4 bestuderen we het potentieel van het gestuurd laden van EVs met het
oog op het distributienetwerk. Vanuit dit perspectief zorgt een oplaadstrategie die
de bulk van de energie-overdracht naar de nacht verplaatst als de netbelasting laag
is voor een sterke reductie van de extra piekbelasting die het ongecontroleerd opladen van EVs tot gevolg zou hebben. Verschillende EV laadprofielen, gecontroleerd
en ongecontroleerd, worden bij bestaande netwerkprofielen opgeteld om de impact
op verschillende typen netwerkcomponenten te evalueren. Er blijkt dat in de ongecontroleerde scenario’s de extra EV belasting tot een significant aantal overbelaste
componenten zou leiden. Dit effect is het meest prominent op het niveau van de
MS/LS-transformatoren (middenspanning/laagspanning). In het controleerde scenario kunnen bijna alle extra vervangeningen t.o.v. het referentie-scenario zonder
EVs voorkomen worden. De financiële consequenties van de verschillende scenario’s zijn geëvalueerd op basis van de netto contante waarde van de annuı̈teiten
van de vervangingskosten en de kosten voor energieverliezen. Het gecontroleerde
laadscenario bleek een ongeveer 20 % lagere netto contante waarde te hebben de
ongecontroleerde scenario’s. Hoewel de kosten van energieverliezen de totale kostenplaatjes domineren, worden de verschillen tussen de scenarios vooral veroorzaakt
door de vervangingskosten. Ook bleek dat de kosten gerelateerd aan de middenspanningskabels dominant waren t.o.v. de kosten voor laagspanningstransformatoren en
transformatorstations van hoogspanning naar middenspanning.
189
Productiekosten van elektriciteit en de onderlinge relatie tussen en het
gestuurd laden van elektrische auto’s en grensoverschrijdende transmissie
Het doel van hoofstuk 5 is om te onderzoeken in hoeverre de prijs-responsieve vraag
van EVs de marginale productiekosten van elektriciteit kan verminderen in een systeem met een hoog aandeel duurzame energie. Verder kijken we naar de onderlinge afhankelijkheden tussen het gestuurd laden van EVs en grensoverschrijdende
transmissie-capaciteit - twee paradigma’s die als belangrijk worden gezien in het faciliteren van de integratie van duurzame energie. We breiden een unit commitment 2
model uit met het opladen van EVs als beslissingsvariabelen en we bestuderen een
conceptueel systeem van twee knooppunten. Dit systeem is gebaseerd op data van
het Duitse systeem en heeft windenergie in het ene knooppunt en zonne-energie in
het andere. De resultaten laten zien dat gestuurde EVs tot een significante kostenbesparing leiden doordat de elektriciteitsvraag naar de periodes met veel winden zonne-energie wordt verschoven en zo de inzet van dure gascentrales wordt beperkt. In een scenario voor de penetratie van duurzame energie voor 2025 kunnen
het gestuurd laden van EVs en grensoverschrijdende transmissie vooral als substituten worden gezien. Ze leiden beide tot een bepaalde kostenreductie, onafhankelijk
van elkaar. Voor een scenario met meer duurzame energie concluderen we echter dat
de twee technologiën complementair worden. Hun gezamenlijke potentieel is groter
dan de twee afzonderlijk bij elkaar genomen. De belangrijkste reden hiervoor is dat
grensoverschrijdende transmissiecapaciteit nodig is om energie te transporteren naar
de locatie waar EVs het kunnen absorberen. Deze inzichten zijn relevant in het licht
van de Europese doelstellingen voor duurzame energie voor het jaar 2050.
Het op een lijn brengen van doelstellingen t.a.v. distributienetten en
prijs-responsieve vraag
Als EVs zich als responsieve belasting gedragen die hun vraag doen toenemen als
elektriciteitsprijzen laag zijn, kunnen ze hoge pieken in netwerkbelasting veroorzaken. Dit effect wordt nog prominenter in scenario’s met veel duurzame energie,
omdat een hoge penetratie van fluctuerende productie leidt tot een verminderde
correlatie tussen marktprijzen en netwerkbelasting van elektriciteit. Hoofdstuk 6
onderzoekt daarom mogelijke mechanismen voor congestie-management voor prijsresponsieve vraag van EVs in distributienetten. Er is aangetoond dat het toepassen
van een vorm van congestie-management gerechtvaardigd is omdat het limiteren
van de vraag van EVs tot de beschikbare netwerkcapaciteit verwaarloosbare extra
energiekosten met zich meebrengt. Simpele netwerktarieven bleken het probleem erger te maken in vergelijking met het referentiescenario van vlakke netwerktarieven.
Een optimaal dynamisch netwerktarief dat unilateraal door de netbeheerder wordt
bepaald leidt tot wenselijke uitkomsten maar is moeilijk te bepalen in de praktijk
waarin onzekerheden een rol spelen. Een capaciteitsmarkt voor een distributienet een iteratieve benadering waarin netbeheerder en EV aggregator sequentieel dynamische tarieven en de resulterende elektriciteitsvraag uitwisselen - convergeert naar
een uiteindelijke prijs en een laadschema, maar vereist een complexe ICT infrastruc2 Dit
type model wordt gebruikt om de inzet van de elektriciteitscentrales te plannen
190
Samenvatting
tuur en veeleisende computerberekeningen. Het vooraf alloceren van capaciteit is
meer rechttoe rechtaan om te implementeren in het geval van één enkele EV aggregator, maar er blijven belangrijke bezwaren bestaan gerelateerd aan het verdelen van
capaciteit tussen verschillende aggregatoren en de intertemporele afhankelijkheden
van het laden van EVs. Er bestaat dus een trade-off tussen de complexiteit en de
effectiviteit van een congestie-management mechanisme die aan nader onderzoek dat
onzekerheden meeneemt onderworpen zou moeten worden.
Een verfijnde kijk op het laden van elektrische auto’s
Hoofdstuk 7 beoogt de verschillende elementen uit voorgaande hoofdstukken aan
elkaar te verbinden. In het bijzonder analyseren we verschillen tussen een gecentraliseerde benadering waarin productiekosten worden geminimaliseerd en een decentrale benadering waarin de laadkosten van EVs geminimaliseerd worden op basis van
marktprijzen van elektriciteit. Er blijkt dat aggregatoren die een zeer groot deel van
de totale elektriciteitsvraag van EVs plannen marktmacht kunnen uitoefenen door
het beı̈nvloeden van marktprijzen. Als er voldoende concurrentie tussen aggregatoren is, zal het geaggregeerde vraagprofiel van de EVs het sociaal optimale profiel
benaderen.
Verder laten we zien hoe de waarde van het gestuurd laden van elektrische auto’s
van diverse aspecten afhangt zoals de optimalisatiehorizon, intertemporele afhankelijkheden in elektriciteitscentrales en de beschikbaarheid van EVs om te laden.
Deze analyses suggereren dat ook onder mildere aannames over de voorspelbaarheid van elektriciteitsprijzen en EV gedrag de waarde van het gestuurd laden nog
steeds hoog is, en dat nog steeds de noodzaak tot congestie-management bestaat
om onnodige pieken in netbelasting te voorkomen. Het analyseren van de effecten
die prijs-responsieve vraag van EVs op een groot aantal distributienetten bevestigde
dat de door kosten-minimaliserende EVs veroorzaakte vraagpieken leiden tot een
aanzienlijke hoeveelheid extra kosten gerelateerd aan netwerkinvesteringen.
Conclusies
Deze thesis behandelt de vraag hoe de flexibiliteit van het laden van elektrische
auto’s het best benut kan worden in multi-actor elektriciteitssystemen met een hoog
aandeel duurzame energie. Het antwoord op deze vraag kan als volgt samengevat
worden: de twee belangrijke perspectieven van waaruit het gestuurd laden van EVs
de meeste waarde kan toevoegen zijn het vermogen om het laden in tijd te verschuiven al naar gelang de productie van duurzame energie enerzijds, en om pieken
in netwerkbelasting te voorkomen om netwerkinvesteringen uit of af te stellen anderzijds. Met de huidige vlakke netwerktarieven en marktprijzen van elektriciteit
die sterk door duurzame energie beı̈nvloed worden, kan op prijs reagerende vraag
van EVs ironisch genoeg tot hogere vraagpieken leiden dan ongecontroleerd laden.
De benodigde netwerkinvesteringen zijn echter onnodig omdat het beperken van de
EV vraag tot de vrije netwerkcapaciteit middels een efficiënt congestie-management
mechanisme tot verwaarloosbare hogere energiekosten leidt. Er zijn verschillende
mechanismen voor congestie-management mogelijk om kosten-minimaliserende EVs
191
op een lijn te brengen met netwerkbeperkingen. Die zijn ofwel gebaseerd op schaduwprijzen geässocieerd met de netwerkbeperkingen of op het vooraf alloceren van
vrije capaciteit, maar in beide benaderingen bestaat er een trade-off tussen eenvoud
en economische efficiëntie. Al deze benaderingen lijken echter gemeen te hebben
dat de functie van de netbeheerder verder gaat dan zijn huidige rol. Een schoon
en intelligent elektriciteitssysteem zou daarom wel eens een heroverweging van de
rollen en regels van de huidige gesplitste elektriciteitssector kunnen vergen. Als de
doelstellingen met betrekking tot de verschillende functies van het elektriciteitssysteem op één lijn gebracht zijn in dergelijke intelligente, door ICT mogelijk gemaakte
systemen, kunnen vraag-respons en EVs in het bijzonder een sleutelrol vervullen in
de transitie naar een duurzamer energiesysteem.
192
Samenvatting
List of publications
Reviewed journal papers
1. R. A. Verzijlbergh, M. Grond, Z. Lukszo, J. Slootweg, and M. D. Ilić, “Network
impacts and cost savings of controlled ev charging,” IEEE Transactions on
Smart Grid, vol. 3, no. 3, pp. 1203 –1212, sept. 2012.
2. R. A. Verzijlbergh, C. Brancucci Martı́nez-Anido, L. De Vries, and Z. Lukszo,
“Does controlled electric vehicle charging substitute cross-border transmission
capacity?” Applied Energy, Accepted for publication, 2013.
3. E. Veldman and R. A. Verzijlbergh, “Smart charging of electric vehicles: Opportunity or threat for distribution grids?” IEEE Transactions on Smart Grid,
Under review, 2013.
4. R. A. Verzijlbergh, L. De Vries, and Z. Lukszo, “Renewable energy and responsive demand: do we need congestion management in the distribution grid?”
IEEE Transactions on Power Systems, Under review, 2013.
Book chapters
1. R. A. Verzijlbergh, M. D. Ilić, and Z. Lukszo, “The role of electric vehicles in
making azores islands green,” in Engineering IT-Enabled Electricity Services.
The Case of Low-Cost Green Azores Islands, M. D. Ilić, L. Xie, and Q. Liu,
Eds. Springer, 2012, ch. 11, pp. 226–240.
2. J. Donadee, J.-Y. Joo, R. A. Verzijlbergh, and M. D. Ilić, “Generation and
demand characteristics of the islands of flores and sao miguel,” in Engineering
IT-Enabled Electricity Services. The Case of Low-Cost Green Azores Islands,
M. D. Ilić, L. Xie, and Q. Liu, Eds. Springer, 2013, ch. 4, pp. 119–147.
Conference papers
1. R. A. Verzijlbergh, Z. Lukszo, J. Slootweg, and M. D. Ilić, “The impact of
controlled electric vehicle charging on residential low voltage networks,” in
193
194
List of publications
Networking, Sensing and Control (ICNSC), 2011 IEEE International Conference on, april 2011, pp. 14–19.
2. R. A. Verzijlbergh and Z. Lukszo, “System impacts of electric vehicle charging in an evolving market environment,” in Networking, Sensing and Control
(ICNSC), 2011 IEEE International Conference on, april 2011, pp. 20 –25.
3. R. A. Verzijlbergh, Z. Lukszo, E. Veldman, J. G. Slootweg, and M. D. Ilić,
“Deriving electric vehicle charge profiles from driving statistics,” in Power
and Energy Society General Meeting, 2011 IEEE, July 2011, pp. 1–6.
4. R. A. Verzijlbergh, M. D. Ilić, and Z. Lukszo, “The role of electric vehicles
on a green island,” in North American Power Symposium (NAPS), 2011, aug.
2011, pp. 1 –7.
5. R. A. Verzijlbergh, Z. Lukszo, and M. D. Ilić, “Various power system impacts
of the large scale adoption of electric vehicles (poster presentation),” in 7th
Annual Carnegie Mellon Conference on the Electricity Industry, 2011.
6. R. A. Verzijlbergh, Z. Lukszo, and M. D. Ilić, “Comparing different ev charging
strategies in liberalized power systems,” in European Energy Market (EEM),
2012 9th International Conference on the, may 2012, pp. 1 –8.
7. R. A. Verzijlbergh, M.O.W.Grond, Z.Lukszo, J.G.Slootweg, and M.D.Ilić, “Potential cost savings of controlled electric vehicle charging (poster presentation),” in 8th Annual Carnegie Mellon Conference on the Electricity Industry,
2012.
8. R. A. Verzijlbergh and Z. Lukszo, “Conceptual model of a cold storage warehouse with pv generation in a smart grid setting,” in IEEE International
Conference on Networking, Sensing and Control (ICNSC), 2013.
Curriculum vitae
Remco Alexander Verzijlbergh was born on June 10th 1981 in Hellevoetsluis, the Netherlands. He finished his pre-university education at the Erasmiaans Gymnasium
in Rotterdam in 1999 and commenced his study Applied Physics at Delft University of Technology in 2000. Towards the end of the bachelor program he became
interested in atmospheric physics. During his MSc graduation project he studied
and published a journal paper on the role of clouds in the dispersion of pollutants
through the atmosphere, under supervision of prof.dr. H.J.J. Jonker and dr. T. Heus.
Before receiving the MSc in Applied Physics Remco conducted an internship at
Dutch energy company Nuon in 2008, where he worked in field experiments on an
experimental micro-grid equipped with solar PV and battery storage. During this
period, the company, enforced by the Dutch unbundling law, split into a commercial
branch (Nuon) and a regulated network branch (Alliander). After his graduation
he continued working for what was now distribution system operator Alliander for
several months, before returning to Delft.
In 2009 he started his PhD research on the role of electric vehicles in future
power systems in the Energy & Industry group at the faculty of Technology, Policy
and Management, under supervision of dr.ir. Z. Lukszo and prof.dr. M.D. Ilić. In
2010 Remco worked closely together with Dutch DSO Enexis to assess the impact of
electric vehicles charging on their distribution networks. Later he worked as a visiting
PhD student at the Massachusetts Institute of Technology in 2011 and at Carnegie
Mellon University in 2012, where he mainly focused on the role of EVs in small-scale
renewable energy systems. During his PhD research, Remco enjoyed being involved
in education by supervising master students and giving several guest lectures on
EVs and smart grids in BSc, MSc and post-graduate courses. He initiated the
Power Rangers, a group of researchers on electricity related topics coming together in
informal weekly meetings. Remco’s ambition is to continue working in the scientific
fields around the economics and operation of power systems, with a special focus on
renewable energy sources.
195
196
Summary
NGInfra PhD Thesis Series
on Infrastructures
1. Strategic behavior and regulatory styles in the Netherlands energy industry
Martijn Kuit, 2002, Delft University of Technology, the Netherlands.
2. Securing the public interest in electricity generation markets: the myths of the
invisible hand and the copper plate
Laurens de Vries, 2004, Delft University of Technology, the Netherlands.
3. Quality of service routing in the internet: theory, complexity and algorithms
Fernando Kuipers, 2004, Delft University of Technology, the Netherlands.
4. The role of power exchanges for the creation of a single European electricity
market: market design and market regulation
François Boisseleau, 2004, Delft University of Technology, the Netherlands,
and University of Paris IX Dauphine, France.
5. The ecology of metals
Ewoud Verhoef, 2004, Delft University of Technology, the Netherlands.
6. MEDUSA, Survivable information security in critical infrastructures
Semir Daskapan, 2005, Delft University of Technology, the Netherlands.
7. Transport infrastructure slot allocation
Kaspar Koolstra, 2005, Delft University of Technology, the Netherlands.
8. Understanding open source communities: an organizational perspective
Ruben van Wendel de Joode, 2005, Delft University of Technology, the Netherlands.
9. Regulating beyond price: integrated price-quality regulation for electricity distribution networks
Viren Ajodhia, 2006, Delft University of Technology, the Netherlands.
10. Networked reliability: institutional fragmentation and the reliability of service
provision in critical onfrastructures
Mark de Bruijne, 2006, Delft University of Technology, the Netherlands.
197
198
NGInfra PhD Thesis Series on Infrastructures
11. Regional regulation as a new form of telecom sector governance: the interactions with technological socio-economic systems and market performance
Andrew Barendse, 2006, Delft University of Technology, the Netherlands.
12. The internet bubble: the impact on the development path of the telecommunications sector
Wolter Lemstra, 2006, Delft University of Technology, the Netherlands.
13. Multi-agent model predictive control with applications to power networks
Rudy Negenborn, 2007, Delft University of Technology, the Netherlands.
14. Dynamic bi-Level optimal toll design approach for dynamic traffic networks
Dusica Joksimovic, 2007, Delft University of Technology, the Netherlands.
15. Intertwining uncertainty analysis and decision-making about drinking water
infrastructure
Machtelt Meijer, 2007, Delft University of Technology, the Netherlands.
16. The new EU approach to sector regulation in the network infrastructure industries
Richard Cawley, 2007, Delft University of Technology, the Netherlands.
17. A functional legal design for reliable electricity supply: how technology affects
law
Hamilcar Knops, 2008, Delft University of Technology, the Netherlands, and
Leiden University, the Netherlands.
18. Improving real-time train dispatching: models, algorithms and applications
Andrea D’Ariano, 2008, Delft University of Technology, the Netherlands.
19. Exploratory modeling and analysis: a promising method to deal with deep
uncertainty
Datu Buyung Agusdinata, 2008, Delft University of Technology, the Netherlands.
20. Characterization of complex networks: application to robustness analysis
Almerima Jamakovic, 2008, Delft University of Technology, the Netherlands.
21. Shedding light on the black hole: the roll-out of broadband access networks
by private operators
Marieke Fijnvandraat, 2008, Delft University of Technology, the Netherlands.
22. On stackelberg and inverse stackelberg games & their applications in the optimal toll design problem, the energy markets liberalization problem, and in
the theory of incentives
Katerina Stankova, 2009, Delft University of Technology, the Netherlands.
23. On the conceptual design of large-scale process & energy infrastructure systems: integrating flexibility, reliability, availability, maintainability and economics (FRAME) performance metrics
Austine Ajah, 2009, Delft University of Technology, the Netherlands.
NGInfra PhD Thesis Series on Infrastructures
199
24. Comprehensive models for security analysis of critical infrastructure as complex systems
Fei Xue, 2009, Politecnico di Torino, Italy.
25. Towards a single European electricity market: a structured approach for regulatory mode decision-making
Hanneke de Jong, 2009, Delft University of Technology, the Netherlands.
26. Co-evolutionary process for modeling large scale socio-technical systems evolution
Igor Nikolić, 2009, Delft University of Technology, the Netherlands.
27. Regulation in splendid isolation: a framework to promote effective and efficient
performance of the electricity industry in small isolated monopoly systems
Steven Martina, 2009, Delft University of Technology, the Netherlands.
28. Reliability-based dynamic network design with stochastic networks
Hao Li, 2009, Delft University of Technology, the Netherlands.
29. Competing public values
Bauke Steenhuisen, 2009, Delft University of Technology, the Netherlands.
30. Innovative contracting practices in the road sector: cross-national lessons in
dealing with opportunistic behaviour
Mónica Altamirano, 2009, Delft University of Technology, the Netherlands.
31. Reliability in urban public transport network assessment and design
Shahram Tahmasseby, 2009, Delft University of Technology, the Netherlands.
32. Capturing socio-technical systems with agent-based modelling
Koen van Dam, 2009, Delft University of Technology, the Netherlands.
33. Road incidents and network dynamics: effects on driving behaviour and traffic
congestion
Victor Knoop, 2009, Delft University of Technology, the Netherlands.
34. Governing mobile service innovation in co-evolving value networks
Mark de Reuver, 2009, Delft University of Technology, the Netherlands.
35. Modelling risk control measures in railways
Jaap van den Top, 2009, Delft University of Technology, the Netherlands.
36. Smart heat and power: utilizing the flexibility of micro cogeneration
Michiel Houwing, 2010, Delft University of Technology, the Netherlands.
37. Architecture-driven integration of modeling languages for the design of
software-intensive systems
Michel dos Santos Soares, 2010, Delft University of Technology, the Netherlands.
200
NGInfra PhD Thesis Series on Infrastructures
38. Modernization of electricity networks: Exploring the interrelations between
institutions and technology
Martijn Jonker, 2010, Delft University of Technology, the Netherlands.
39. Experiencing complexity: A gaming approach for understanding infrastructure
Geertje Bekebrede, 2010, Delft University of Technology, the Netherlands.
40. Epidemics in Networks: Modeling, Optimization and Security Games
Jasmina Omi, 2010, Delft University of Technology, the Netherlands.
41. Designing Robust Road Networks: A general method applied to the Netherlands
Maaike Snelder, 2010, Delft University of Technology, the Netherlands.
42. Simulating Energy Transitions
Emile Chappin, 2011, Delft University of Technology, the Netherlands.
43. De ingeslagen weg. Een dynamisch onderzoek naar de dynamiek van de uitbesteding van onderhoud in de civiele infrastructuur
Rob Schoenmaker, 2011, Delft University of Technology, the Netherlands.
44. Safety Management and Risk Modeling in Aviation: the challenge of quantifying management influences
Pei-Hui Lin, 2011, Delft University of Technology, the Netherlands.
45. Transportation modelling for large-scale evacuations
Adam J. Pel, 201,1 Delft University of Technology, the Netherlands.
46. Clearing the road for ISA Implementation?: Applying Adaptive Policymaking
for the Implementation of Intelligent Speed Adaptation
Jan-Willem van der Pas, 2011, Delft University of Technology, the Netherlands.
47. Design and decision-making for multinational electricity balancing markets
Reinier van der Veen, 2012, Delft University of Technology, the Netherlands.
48. Understanding socio-technical change. A system-network-agent approach
Catherine Chiong Meza, 2012, Delft University of Technology, the Netherlands.
49. National design and multi-national integration of balancing markets
Alireza Abbasy, 2012, Delft University of Technology, the Netherlands.
50. Regulation of Gas Infrastructure Expansion
Jeroen de Joode, 2012, Delft University of Technology, the Netherlands.
51. Governance Structures of Free/Open Source Software Development. Examining the role of modular product design as a governance mechanism in the
FreeBSD Project
George Dafermos, 2012, Delft University of Technology, the Netherlands.
52. Making Sense of Open Data From Raw Data to Actionable Insight
Chris Davis, 2012, Delft University of Technology, the Netherlands.
NGInfra PhD Thesis Series on Infrastructures
201
53. Intermodal Barge Transport: Network Design, Nodes and Competitiveness
Rob Konings, 2009, Delft University of Technology, Trail Research School, the
Netherlands.
54. Handling Disruptions in Supply Chains: An integrated Framework and an
Agent-based Model
Behzad Behdani, 2013, Delft University of Technology, the Netherlands.
55. Images of cooperation a methodological exploration in energy networks
Andreas Ligtvoet, 2013, Delft University of Technology, the Netherlands.
56. Robustness and Optimization of Complex Networks: Spectral analysis, Modeling and Algorithms
Dajie Liu, 2013, Delft University of Technology, the Netherlands.
57. Wegen door Brussel: Staatssteun en publieke belangen in de vervoersector
Nienke Saanen, 2013, Delft University of Technology, the Netherlands.
58. The Flexible Port
Poonam Taneja, 2013, Delft University of Technology, the Netherlands.
59. Transit-Oriented Development in China; How can it be planned in complex
urban systems?
Rui Mu, 2013, Delft University of Technology, the Netherlands.
60. Cross Culture Work: Practices of Collaboration in the Panama Canal Expansion Program
Karen Smits, 2013, VU University Amsterdam, the Netherlands.
61. Structuring Socio-technical Complexity; Modelling Agent Systems Using Institutional Analysis
Amineh Ghorbani, 2013, Delft University of Technology, the Netherlands.
62. Towards Playful Organzations; How online gamers organize themselves (and
what other organizations van learn from them)
Harald Warmelink, 2013, Delft University of Technology, the Netherlands.
63. Electricity without borders; The need for cross-border transmission
Carlo Brancucci Martı́nez-Anido, 2013, Delft University of Technology, The
Netherlands.
64. The Power of Electric Vehicles; Exploring the Value of Flexible Electricity
Demand in a Multi-actor Context
Remco Verzijlbergh, 2013, Delft University of Technology, The Netherlands.
Order information: [email protected]
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement