Improving the heating efficiency of detached houses by instant messaging CLARA LYNN

Improving the heating efficiency of detached houses by instant messaging CLARA LYNN
Improving the heating efficiency of
detached houses by instant messaging
CLARA LYNN
Master’s Degree Project
Stockholm, Sweden 2014
XR-EE-RT 2014:019
Abstract
House heating is responsible for substantial energy consumption in industrialized countries. However, the climate control in detached houses is often sub
optimally done. These climate controllers are usually composed by subsystems
that have limited information about the state of the building. Improvements
usually requires large investments from the house owner, unless new sensor networks technologies are adopted.
In this Master thesis, how to improve the efficiency of house heating using
instant messaging (IM) is investigated. Devices in an inhabited house have been
configured to be connected by an IM client. This allows to perform automatic
control through a simple software. The dynamical model of the heating and
cooling system is studied and the relevant sensor measurements are identified.
The possible reduction in heat consumption is quantified by simulations. It is
shown that the method is applicable, and that it can result in a reduction of the
heat consumption in detached houses, even by using a small subset of potential
sensor measurements.
Contents
1 Background
1.1 Domestic heating . . . . . . . . . . . . . .
1.2 The Intelligent EnergiAnvändning-project
1.3 Related work . . . . . . . . . . . . . . . .
1.3.1 Setback schedules . . . . . . . . . .
1.3.2 Reactive control . . . . . . . . . .
1.3.3 Predictive control . . . . . . . . . .
1.3.4 Implementation challenges . . . . .
1.4 Problem formulation . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
2
2
2
3
3
4
5
2 Theory
2.1 Linear Time-Invariant systems .
2.1.1 Discrete LTI-systems . . .
2.2 Model of heat flows in buildings .
2.2.1 Calculation of parameters
2.3 Sensor networks . . . . . . . . . .
2.3.1 XMPP . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
6
7
7
9
11
11
3 Methods
3.1 Implementation . . . . . . .
3.1.1 About the house . .
3.1.2 Scenario description
3.1.3 Implemented control
3.2 Simulation . . . . . . . . . .
3.2.1 Model . . . . . . . .
3.2.2 Data . . . . . . . . .
3.2.3 Control strategies . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12
12
12
14
16
16
17
19
19
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Results
21
4.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5 Discussion
24
5.1 Potential for energy savings . . . . . . . . . . . . . . . . . . . . . 24
5.2 Potential market . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3 Potential problems . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6 Conclusions and further works
iii
25
Chapter 1
Background
1.1 Domestic heating
In 2012, the households of Sweden consumed 86 TWh of energy, corresponding
to 23% of the total energy consumption of the country [1]. Of this, 58 TWh, i.e.,
67%, was spent on heating of houses and domestic hot water [2]. While other
major energy consuming sectors such as shipping and industrial production always strive to reduce their energy consumption and thus their costs, households
are often not as prone to invest in improved energy efficiency due to the size of
the investments.
There are about 2 000 000 detached houses used as primary residence in
Sweden [3]. Both houses and heat systems are generally made to have a long
lifetime. As an example, more than half of Sweden’s detached houses were built
before 1971, and only about a fifth of all the detached houses in Sweden had any
kind of improvement or changes in their heating system during made 2012. As
older houses generally have a significantly higher heat consumption than newer
ones [2], one can easily understand that many house owners could gain quite by
investments in more efficient heating systems.
One of the deficiencies with older heating systems is that their regulation
often is based on the momentary outdoor temperature only, not taking indoor
temperature or weather forecasts into account. The regulation is done using
a heating curve, i.e., a linear mapping from outdoor temperature to radiator
temperature.
Furthermore, when the heating system in a house has been partly improved,
for example by adding insulation or changing from natural ventilation to a forced
ventilation with heat recycling, the primary heating system is seldom adjusted,
leading to an overproduction of heat. This is partly due to the house owners
reluctance to pay for a professional to adjust their systems, and partly due to
the lack of shared interface between heating, ventilation and air conditioning
(HVAC) systems from different manufacturers. This makes it hard to tune the
different systems to each other.
In Tabel 1.1 it can be seen how the detached houses in Sweden were heated
in 2012. The number of houses with installed heat pumps is increasing steadily
[2]. Most new heat pumps has indoor sensors as an option, but according to
Gunnar Forsberg, product marketing manager for heat pumps at Viessmann,
1
Method of heating
Biofuel and electric heating
Hydronic electric energy
Electric heating
District heating
Heat pumps (earth/water)
Biofuel
Heat pumps (earth/water)
and biofuel
Heat pumps (earth/water)
and electric heating
Oil
Oil and electric heating
Oil and biofuel
Oil, biofuel and electric heating
Other methods
Number of thousands of houses
406
234
233
231
209
192
83
48
17
12
10
4
182
Table 1.1: The heating method of the detached houses of Sweden in 2012 [2].
Air/air heat pumps are included in the electric heating, and air/water heat
pumps are sorted under hydronic electric heating.
the house owners are often reluctant to take such an extra cost when installing
new systems, despite the possibility to save energy later on.
1.2 The Intelligent EnergiAnvändning-project
This master thesis has been done at Sustainable Innovation, as a part of the
IEA-project. IEA is an acronym for Intelligent EnergiAnvändning (intelligent
energy services). It is a Vinnova funded project run in cooperation with several
companies within real estate, home alarms, house heating, ventilation systems
etc.
The objective of the project is to create an industrial cooperation for platform standardization of services in domestic and business premises. The intention is to enable appliances from different companies to communicate and share
information. This information can then be used to improve the energy efficiency
of the appliances.
1.3 Related work
1.3.1 Setback schedules
Much thought has been put to improving the efficiency of HVAC systems. Setback schedules, i.e., that the setpoint temperature is decreased when the building is unoccupied or the occupants are sleeping is a well established strategy [4],
2
especially in non-residential buildings. E.g., in [5], a building with fixed occupation hours and setback schedule is considered, and artificial neural networks
are used to determine the optimal time to start the preheating of the building.
In Sweden, Scypho, a similar control system for residential buildings has been
launched to the consumer market recently [6]. Setback schedules are however
not very well suited for controlling the HVAC systems of residential buildings,
since most people are unaware of the exact occupational patterns of their homes
[7], [4]. On the contrary, the importance of dynamical learning of occupational
patterns is stressed by Aiello et al. [8].
1.3.2 Reactive control
To improve the efficiency of residential HVAC systems reactive regulation has
been suggested [4]. Reactive regulation follows the same logic as prescheduled
setback ditto, that buildings need not be fully heated when the inhabitants are
asleep or away, but instead of using a predefined schedule, sensors of some sort
are used to detect the activities of the residents.
Since buildings constitute quite slow thermal systems [9], [10], it is an intrinsic property of this strategy is that it causes discomfort for the residents
who will return to a cold home. Furthermore, the faster heating methods that
are available to reduce this discomfort are more wasteful than slower ones that
could be used to preheat a building, or keep the temperature constant [4].
1.3.3 Predictive control
Predictive control is the most successful strategy so far [10], [11], [8]. Optimally,
the control of HVAC systems should be predictive in several aspects: occupational patterns and heating load, which is turn depends on the dynamics of the
building, weather conditions and activities of the residents [11], [10].
Ma et al. [12] suggests using Model Predictive Control (MPC) to include
weather forecasts in the control of a chilling system at a university campus, and
have shown promising results in simulations, compared to the current, manual
control.
Parisio et al. [11] includes both weather and occupancy predictions, as well
as a detailed model of the building dynamics. A Stochastic Model Predictive
Control (SMPC) has been simulated, with promising results. In the suggested
method, no presumptions about the distribution of the forecast errors prediction
is needed. Instead, copulas are used to learn the distribution of the errors from
statistics. The method does however require close knowledge about the physical
qualities of the modeled building, which might make it costly to implement.
Lu et al. [4] and Gao et al. [7] suggests using simple motion and door sensors to track occupancy patterns. Three states are considered: away, asleep and
home and active. Gao et al. is focused on assessing the potential gain, performing simulations from existing data. Lu et al. on the othe hand presents results
from both simulations and implementations. Data collected from the sensors
are used to establish occupational patterns, which are then used to predict when
the residents will return to the building, and preheat the building to this time.
To determine the current activity of the residents, the data collected from the
sensors are processed online, using a Hidden Markov Model.
3
In [13], the control of the ventilation of a multi usage university building
is considered. Here, no fixed setback schedule is used, but there exists a distinct occupancy pattern. Using observations from wireless cameras, multivariate
Gaussian and agent based models are used to predict the occupancy of the rooms
of the building. It is stated that agent based models have the drawback that the
predictions need to be done offline, while multivariate Gaussian based models
can update their predictions online using Bayesian inference and knowledge of
the current occupancy. Simulations show promising results for the suggested
methods.
1.3.4 Implementation challenges
All the predictive approaches discussed above does require a sensor networks of
some sort to be installed in the building that is to be controlled. In [14], common
obstacles and challenges connected to deploying large-scale sensor systems in
residential buildings are presented. Large-scale deployments in terms of the
number of nodes, the number of building and duration in time are discussed.
When the number of nodes in a building grows, wall plugs grow a scarce
resource. To some extent this can be handled through battery powered sensors,
but as mentioned in [15], this requires the sensors to be very power efficient,
especially as the time scale increases. Furthermore, the aesthetics of the sensors
grows important as the number of them increases, as the sensors otherwise
tend to dominate the visual landscape of the room. Moreover, the plethora of
communication protocols and interfaces places high demands on the architecture
of the sensor systems. In [14], the issue is stressed, stating that ”Deploying a
dozen different COTS products in a single home is akin to maintaining a dozen
sensor networks simultaneously, each with independent failures”.1 This can be
handled by using custom made sensors and actuators, at an increased cost.
When the number of building grows, the work required to collect ground
truth of the residents activities, to be used in the learning process necessary for
prediction of occupational patterns, naturally grows with it. Similarly, the work
required to deploy and maintain the sensor systems increases. Furthermore, this
work must be prepared carefully, since the time slot when the residents are at
home and awake is limited.
With increasing time duration the risk of failures in the sensor system increases. This requires the system to be stable in the event of failures of the
subsystems, which in turn requires a well thought out system architecture.
The control strategies previously in this section are all, with the exception of
Scypho, in the research phase, and proof of concepts rather than commercially
viable systems. However, smart thermostats are slowly emerging to the consumer market. Nest Labs has launched a self learning thermostat in the US,
Canada and the UK [16], [17]. However, the difference in standards for HVAC
systems in different countries has impeded a wider launch of the system [16].
In Sweden, the start up Ngenics are launching a similar product [18], [19]. So
far, the thermostat is only available for a limited number of heat pump models,
due to differences in interface, communication protocols etc. [18].
Another impediment is the lack of consumer interest. The reduction in cost
1 Where
COTS is a acronym for Commercial Of The Shelf.
4
and carbon emission per household is seemingly too small to suffice as initiative
for most homeowners [4]. Instead, improved comfort seems to be the key factor
to motivate an investment in the HVAC system in detached homes, according
to representatives of Ngenics as well as the heat pump manufacturers CTC
Enertech and Viessmann, and Aiello et al. [8].
1.4 Problem formulation
As mentioned in section 1.1, heating of domestic buildings is responsible for a
substantial part of the energy consumption in industrialized countries. As have
been noted in section 1.3, there has been plenty of research done about how the
HVAC systems can be optimized. The majority of the proposals does however
require big investments, such as a new heat pump or a specially designed sensor
system if applied to existing buildings, since the cheap COTS products suggested
by Lu et al. [4] and Gao et al. [7] are not feasible in large scale deployment
[14]. According to Joachim Lindborg, project manager of the IEA-project, the
lack of suitable interfaces and common communication protocols is the main
obstacle for home automation in already established buildings.
This master thesis aims to investigate a realizable and affordable way of
improving the HVAC systems of inhabited, detached houses. More specifically,
the method should require a minimal investment and be applicable in with
HVAC sub systems from different manufacturers. Moreover, the method should
not require the user to preschedule the settings of the HVAC system, nor require
any modeling of the thermodynamical properties of the specific house. The
strategy of choice is to enable already installed devices to communicate via an
open source chat. No additional sensors will be installed, since the house alarm
already collects the necessary data. To explore the benefit and the realizability
of this strategy, simulations as well as implementations in an inhabited house is
performed.
5
Chapter 2
Theory
2.1 Linear Time-Invariant systems
In this section the key properties of linear time-invariant systems (LTI-systems)
is introduced.
A common way to model a physical systems is state-space models. In the
T
state-space description the state vector, x(t) = x1 (t) . . . xn (t) , contains
information about the current state of the system that will affect future outputs
of the system. In the following equations, u(t) denotes the input to the system
and A(t), B(t), C(t) and D(t) represents the dynamics of the modeled system:
ẋ(t) = A(t)x(t) + B(t)u(t)
y(t) = C(t)x(t) + D(t)u(t).
The LTI systems are an especially tractable class of state-space models.
These have constant system matrices and can be formulated as
ẋ(t) = Ax(t) + Bu(t)
,
y(t) = Cx(t) + Du(t)
where A, B, C and D are constant matrices. The LTI system above has the
solution
Z t
x(t) = eA(t−t0 ) x(t0 ) +
eA(t−τ ) Bu(τ )dτ,
t0
and thus, the state of an undisturbed LTI system can be computed at each point
in time after t0 , given the initial state x0 . Furthermore, the key properties of
the modeled system, such as stability, observability and controllability can be
deduced from the matrices A, B and C.
A system is said to be BIBO stable if a bounded input results in a bounded
output. In the LTI-case, the system is stable if and only if all the eigenvalues
of A have negative real parts.
A system is said to be completely controllable if it can be transfered from
any initial state to any arbitrary state in finite time. LTI-systems
are completely
controllable if and only if the reachability matrix C = B, AB, A2 B, . . . , An−1 B
has full row rank.
6
A system is observable if all changes in input and state is observable in the
output. LTI-systems are completely observable if and only if the observability
matrix


C
 CA 


1 

O =  CA 
 .. 
 . 
CAn−1
has full row rank [20].
2.1.1 Discrete LTI-systems
Since LTI-systems can be solved for any time instance, given their initial state,
their discrete counterpart can be determined as well. Thereby, the discrete
system equations can be readily attained. With a fixed sampling interval T and
an input that is fixed between the sampling instances, it is given by
RT
x(T (k + 1)) = eAT x(T k) + 0 eA(T −τ ) dτ Bu(k) = Γx(k) + Φu(k)
y(k) = Cx(k) + Du(k)
The properties of the discrete system is determined by the matrices Γ, Φ and
C. The discrete system is stable if the eigenvalues of Γ is situated inside the
unit circle. The conditions for controllability and observability are the same as
in the continuous case [20]
2.2 Model of heat flows in buildings
In [11] a physical model of the heat flows in a building is introduced. The parameters used in the following are described in Table 2.1. The main restrictions
of the model are:
• No infiltrations are considered, i.e., all airflows in and out of the building
are considered to be controlled.
• The zone is well mixed, i.e., one does not have to take into account heat
flows within the airmass of the zone.
• The thermal effects of vapor production are ignored.
The indoor temperature is calculated using energy balances of zones, i.e.,
rooms or otherwise delimited spaces:
room
mair,zone cpa dTdt
= Qvent + Qint + Σj Qwall,j + Σj Qwin,j + Qheating ,
(2.1)
where
Qvent
Qint
Qheating
ṁvent ∆Tvent
=
=
mair,zone
=
cNpeople
= Arad hrad (Tmr − Troom ),
for a zone with n walls.
7
ṁvent (Tair,sa −Troom )
mair,zone
(2.2)
The walls of the zone are modeled as a system with two capacitances and
three resistances, yielding the following equations for the temperature on the
inside and the outside of the walls:
j
dTwall,o
dt
j
dTwall,i
dt
j
j
ho Ajwall (Tee
−Twall,o
)+
=
(T
j
j
−T
)
wall,i
wall,o
j
R
wall
C j /2
j
hi Ajwall (Troom −Twall,i
)+
=
(T
j
j
−T
)
wall,o
wall,i
j
R
wall
C j /2
(2.3)
,
j
where C j is the thermal capacity of the wall and Tee
is the equivalent temperature. The equivalent temperature is such that it accounts for the different
radiation heat exchange due to the orientation of the external walls:
Tee,j = Tamb +
aI j
.
αe
With the substitution
ṁvent (Tair,sa − Troom ) =
(∆Th − ∆Tc ) = uh − uc ,
(2.4)
where ∆Th and ∆Tc are nonnegative variables representing the temperature
difference through the heating and cooling coils, and uh and uc multiplied by
cpa model the portion of the ventilation heat flow due to heating and cooling
respectively, the system can be formulated as a LTI-system:
ẋ(t) = Ax(t) + Bu(t) + Eω(t)
y(t) = Cx(t)
(2.5)
where

Troom
1
 Twall,i 


 .. 
 . 
 n 

x(t) = 
 Twall,i

T 1

 wall,o 
 . 
 .. 

(2.6a)
n
Twall,o
is the state vector,


uh
u(t) =  uc 
∆Th,rad
(2.6b)
is the input vector and


Tamb
w =  Ij 
Npeople
8
(2.6c)
is the disturbance vector. A is an 2n + 1 × 2n + 1 matrix.
A=

1
(−Σj hi Ajwall
 mair,zone cpa
hi A1wall


C 1 /2






















− Σj Rj1 )
win
hi A1wall
mair,zone cpa
hi A1wall
− C 1 /2 − R11
wall
...
0
...
hi An
wall
mair,zone cpa
0
0
1
......
..
.
..
.
..
0
..
.
.
hi An
wall
C n /2
0
...
0
..
.
...
0
1
0
......
R1
wall
C 1 /2
..
.
..
.
0
..
.
0
0
B is a 2n + 1 × 3 matrix.



B=

..
..
.
−
..
1
−
ho A1wall
C 1 /2
......
0
..
.
1
0
Rn
wall
C n /2
1
mair,zone
0
..
.
1
− mair,zone
0
..
.
0
0
0
0

......
0

























.
0
..
0
Rn
wall
C n /2
.
E is a 2n + 1 × n + 2 matrix.

G1 A1win
1
1
n
mair,zone cpa Σj=1 Rj
mair,zone cpa
win


0
0

..
..


.
.



0
0

E=
1
ho A1wall αae
h
A
o
wall


C 1 /2
C 1 /2

..

.
0


..
..

.
.

ho An
wall
C n /2
−
0
.
..
0
..
.
0
hi An
wall
C n /2
0
R1
wall
C 1 /2
C 1 /2
......
−
1
R1
wall
C 1 /2
...
.
...
0
......
0
0
..
.
..
.
..
.
..
.
1
Rn
wall
C n /2
−
ho An
wall
C n /2
−
1
Rn
wall
C n /2
Arad hrad 
mair,zone cpa
0
..
.




0
...
...
Gn An
win
mair,zone cpa
c
mair,zone cpa
0
..
.
..
.
0
..
.
..
.
..
.
0
..
.
..
.
..
.
0
..
.
..
.
..
.
.
0
0
..
.
..
.
..
.
..
.
..
.
...
...
a
ho An
wall αe
n
C /2
0
0
..










.








Together, these matrices model the dynamics of heat flows in a building.
Their parameters is described in the following section.
2.2.1 Calculation of parameters
Many of the parameter introduced in table 2.1 are pure material constants.
However, some of them are determined by several other parameters, or by current building standard. These parameters, and the methods to calculate them,
will be presented below.
The heat resistance of a wall composed by n layers of homogeneous materials
can be calculated as
Rwall = hi +
d1
dn
+ ... +
+ ho ,
λ1
λn
(2.7)
where dj denotes the thickness of layer j, and λj denotes its thermal resistance.
9
Table 2.1: Parameters used in section 2.2
αe
a
Arad
Aj wall
Aj win
c
Cj
cpa
Gj
hi
ho
hrad
Ij
mair,zone
ṁvent
Npeople
Rj wall
Rj win
Tair,sa
Tamb
Tee,j
W
m2 K
m2
m2
m2
W
J
K
J
kgK
W
m2 K
W
m2 K
W
m2 K
W
m2
kg
kg
s
K
W
K
W
K
K
K
Tj i
Tj o
Tmr
∆Th,rad
uc
K
K
K
K
uh
kgK
s
kgK
s
External heat transfer coefficient. Identical to ho .
Absorption factor for shortwave radiation.
Emission area of the radiators.
Wall area of the j:th surface.
Window area of the j:th surface.
A constant related to equipment and occupants activity.
Thermal capacity of the wall of the j:th surface.
Specific heat of dry air.
Solar gain heat coefficient (SHGC) of the windows
on the j:th surface.
Indoor heat transfer coefficient.
Outdoor heat transfer coefficient. Identical to αe .
Heat transfer coefficient of the radiators.
Solar radiation on the j:th surface.
Air mass in the room.
Ventilation mass flow.
Number of persons in the room.
Thermal resistance of the wall on the j:th surface.
Thermal resistance of the window on the j:th surface.
Supply air temperature.
Outdoor temperature.
The equivalent temperature, i.e., the outdoor temperature adjusted to account for heat radiation from
the outside wall j.
Indoor surface temperature of the j:th wall.
Outdoor surface temperature of the j:th wall.
Mean radiant temperature of the radiators.
The difference between Tmr and Troom
Input to the ventilation. uc cpa represents the ventilation heat flow due to cooling.
Input to the ventilation. uh cpa represents the ventilation heat flow due to heating.
10
The heat resistance of a layer composed of several material can be approximated with two methods, the λ-method and the U-value method. These methods yields a lower and an upper bound on the resistance of the layer. To estimate
the actual value, the mean of these two values are calculated [9].
In the λ-method, the proportion of different material in the wall is calculated,
and a compound thermal conductivity is calculated by weighting the conductivity of the different materials with their proportion of the wall and adding them
together.
In the U-value method, the wall is divided into sectors with homogeneous
material for which the thermal resistance is calculated. Then the inverse of the
thermal resistance is calculated by weighting the inverse of the resistances of
the sectors with the sectors proportion of the wall, and adding the together.
The window area, Awin , of a detached house usually accounts for 20% of
the wall area [9]. The radiators are shaped to cover the area underneath the
windows [22].
2.3 Sensor networks
Wireless sensor networks (WSN) consists of networks and actuators that have
been connected through a wireless network. On account of the increased availability of cheap sensors and micro processors, the use of WSNs in Internet of
Things (IoT) applications such as house automation is emerging. However, the
nature of WSNs poses some boundaries on their application. The main restriction is due the limited resources of the components [15]. The components
generally posses a rather small processing power, and they are often powered
by batteries. This requires computations to be kept simple and efficient, and
to preferably be done as seldom as possible. Furthermore, the network will always be exposed to time delays and package drops. This requires systems being
controlled by the network to be stable under such disturbances.
2.3.1 XMPP
eXtensible Messaging and Presence Protocol (XMPP) is a XML-based open
protocol for ”presence, instant messaging, and real-time communication” [23].
One of the more well known applications of XMPP is Google Talk. A key feature
of XMPP is that anyone can set up a server and start conversations with users
on other servers. This makes the protocol suitable for usage in IoT applications,
since it provides an apt platform for machine to machine communication.
Several IoT extensions have been accepted in the XMPP- standard [24].
These extension makes it possible to set up networks through instant messaging
(IM), wherein sensors can report their state and users or automated scripts can
control actuators in sensor networks by text messages.
11
Chapter 3
Methods
Since the project was divided in to two parts, an implementation part and a
simulation part, this chapter is divided accordingly.
3.1 Implementation
In this section, the implementation that was carried out in an inhabited house
is described. The section starts by introducing the properties of the house
in question. Thereafter, the control that was intended to be implemented is
described.
As often, the physical realities did not allow for all of the intended actions
to be put into work. Therefore, the aforementioned sections are followed by one
about the control that actually was implemented.
3.1.1 About the house
The house is situated in Solna, a suburb of Stockholm. It is a two story wooden
house from the 1930s with unfurnished basement and attic. The living area
sum to 150 m2 and other areas sum to 75 m2 . As most houses from that time,
the house was deliberately built drafty, to allow for natural ventilation. Due to
governmental restrictions, no extra isolation has been added.
The ventilation of the house has been enhanced by a forced ventilation with
rotary heat exchange. The ventilation system is a Systemair VR700 DCV system, with three settings: normal, high and minimal speed, according to Sofia
Rask, product manager at Systemair. It will be assumed that the normal speed
fulfills the requirements of 0.35 l inlet air per second and square meter floor
area and that the minimal speed supplies the minimum flow 0.1 l/(sm2 )[25].
According to Systemair, the system is able to reuse 80% of the heat in the
outlet air.
The house is heated by a heat pump of model Optimum G2 from Thermia,
which also heats tap water. The heat distribution is hydronic, with radiators
supplemented by underfloor heating in parts of the house.
To control the heat production, the heat pump uses as heating curve, i.e.,
a linear mapping from outdoor temperature to a suitable temperature of the
water flowing into the radiator system. An example of a heating curve can be
12
Figure 3.1: A typical heating curve from Thermia. The outdoor temperature
(x-axis) is mapped to a flow temperature (y-axis).
Figure 3.2: The RPi that is connected to the ventilation system.
seen in Figure 3.1. The heating curve can be adjusted to suite the building, but
this has not been done by the current house owners. For additional heating, a
stove has been installed on the ground floor.
The house is protected by an alarm from Verisure. According to the house
owner, as well as my own observations, this alarm has three settings: disarmed,
armed away and armed home. The latter one is a shell alarm, which is used at
night time. The alarm is equipped with a temperature sensor and a CO2 sensor.
Communication system
As a part of the IEA-project, a set of Raspberry Pi:s (RPi) have been connected
to the heat pump, the ventilation and the alarm, according to Joachim Lindborg,
the project manager of the IEA-project. The RPi:s are connected to the devices
by wire, and they are powered by wire. The RPi:s connected to the ventilation
system can be seen in Figure 3.2 and 3.3. They are connected to Internet, and
are logged in on a XMPP-based chat server.
Through the chat, other users can ask the RPi:s to report the state of the
device they are connected to. As an example, the RPi connected to the alarm
can report whether the alarm is turned on and the temperature in the room is,
13
Figure 3.3: The ventilation system, with the connection to the RPi.
while the one connected to the ventilation can report the ventilation speed and
the temperature of the different air flows in the ventilation system.
3.1.2 Scenario description
The control is performed by a Python script, using fuzzy logics. The governing
principle is that the status and the measurement of the indoor temperature from
the alarm is used to decide appropriate settings for the speed of the ventilation
and the setpoint temperature of the heat pump. A schematic illustration of the
information flows can be seen in Figure 3.4.
This principle is realized by a script, which logs on to an IM server and
starts a conversation with the RPI:s connected to the heat pump, the alarm
and the ventilation. The script first asks the RPi connected to the alarm for
the alarm status and its measurement of the indoor temperature. Based on
this information, the script then sends control messages to the RPi:s connected
to the ventilation and the heatpump. They, in turn, sets the control values
of the unit that they are connected to. The script starts this little round of
conversation at regular intervals, defined by the user when starting the script.
As mentioned in the previous section, the alarm has three possible statuses.
The control performed by the script is based on this status, e.g., when the alarm
status is armed away, the speed of the ventilation is lowered and the setpoint
temperature of the heat pump is decreased, compared to when the alarm status
is disarmed. The indoor temperature as measured by the alarm is also used in
the regulation, to avoid overheating of the house. The different states of the
system can be seen in Figure 3.5.
One should note that the states shown in Figure 3.5 was selected based on
the preferences of the inhabitants of the house involved in the implementation.
Other settings, such as decreased ventilation at nighttime or deeper temperature
setbacks could easily be implemented.
The script can easily be run by the users of the house. The users can also
change the parameter setting in the scripts while it is running. They can either
do this by IM or through a GUI set up in the IEA-project. Since the script
14
Figure 3.4: The information flows of the control. The control script makes a
request for the status and the indoor temperature from the RPi connected to
the alarm. The RPi reads out these values from the alarm and send an answer
with the requested values to the script. Based on this information, the control
script sends a setpoint temperature to the RPi connected to the heat pump and
and a ventilation speed to the RPi connected to the ventilation unit.
Figure 3.5: The logics behind the control. When nobody is home, the ventilation
is decreased and the setpoint of the heat pump is lowered. When the shell
protection is activated, the ventilation is normal but the setpoint of the heat
pump is lowered. If the indoor temperature is more than two degrees over the
setpoint of the state, the setpoint of the heat pump is lowered by two degrees.
15
Figure 3.6: The user can change the value above if they wish to change the
setpoint temperature of the heat pump when the alarm is of.
is online, it is possible for IM-friends of the script to send messages to it. For
example, if the inhabitants would be away for a longer time, they might want to
set the setpoint temperature of the heat pump to 15o C. Then this can either be
done by the user sending the message tempAway = 15 to the script through any
XMPP-based chat server. Else, the users can reach the GUI through a website,
and change the parameters by typing the desired value in a box like the one in
Figure 3.6.
3.1.3 Implemented control
Due to hardware constraints, the control presented above could not be fully
implemented. It was found that the current interface between the heat pump
and the RPi did not support setting the setpoint temperature. Instead, it
did only support turning on and off the heat production and the hot water
production. Since it is unclear how the heat pump might be affected by being
turned on and off frequently, it was decided to omit the heat pump from the
control.
With the reduced control, the conversation between the RPi:s is even duller.
At regular time instances, the script asks the RPi connected to the alarm for
the alarm status. If the alarm activated, the script sends a message to the
ventilation to go on minimal speed. If the alarm is disarmed or the shell alarm
is activated, the scripts sends a message to the ventilation to go on normal
speed. The logics of the implemented control is illustrated in Figure 3.7
3.2 Simulation
The aim of the simulations was to asses th gain of the control of the HVAC
system used in the implementation. To this end, a model of the heat and air
flow in a similar house was defined, so that the heating used with the current
control could be estimated, as well as the heat consumption with a control
similar of that of the implementation.
16
Figure 3.7: The implemented control. The governing principle is that when
nobody is home, the ventilation can be reduced.
3.2.1 Model
To model the heat flow in a house the model suggested in [11], i.e. equation 2.5,
was used, with some adjustments.
As mentioned in section 2.2, this model has several limitations. To simplify
the model further, no account has been taken to the heat flow caused by solar
radiation. Furthermore, the building was considered as a single zone, with
uniform temperature inside the building.
The parameter values was chosen in accordance to an example of a one family
dwelling with good insulation in [9]. This models a two storey house with a
total area of 150 m2 , wooden exterior walls, tiled roof over an unfurnished attic
and a concrete foundation on top of insulation and gravel. The corresponding
parameter values can be seen in Table 3.1.
Validation of the model
To validate the model, the parameters in Tabel 3.1 was estimated using structured parametrization with the System Identification Toolbox from Matlab. The
model presented above was set up as an initial guess, and the parameters that
were to be estimated was set as free parameters.
To fit into the framework of the parameter estimation, the ambient temperature was regarded as an input, rather then an disturbance, and corresponding
changes was done in the structure of the A and B matrices.
For the validation, data from the house in Solna was used. The available
data was the outdoor and the indoor temperature, the temperature of the supply
air from the ventilation system and the flow temperature. The data used was
from the nights between the 16:th and the 17:th and 17:th and 18:th of May.
Samples from every second minute from 8:44 PM to 8:36 AM was feed in to the
estimation.
However, the fact that only one of the states in the model, the indoor temperature, had been measured meant that the major part of the state evolution
was unobservable. Due to this, the estimation was unable to adjust the parameter values. Therefore, the values presented in Table 2.1 were the ones used in
the simulations.
17
Table 3.1: Parameter values used in the simulations.
αe
a
Arad
A1,2 wall
A3,4 wall
A1,2 win
A3,4 win
c
C1,2
C3,4
cpa
G1,2,3,4
hi,wall
hi,floor
hi,ceiling
ho,wall
ho,floor
ho,ceiling
hrad
mair,zone
R1,2 wall
R2,3 wall
R1,2 win
R3,4 win
25 mW
2K
0.4
12.6 m2
36 m2
48 m2
33.6 m2
33.6 m2
W
2 422.3 kJ
K
3 229.7 kJ
K
J
1000 kgK
0.4
7.7 mW
2K
5.9 mW
2K
7.7 mW
2K
25 mW
2K
25 mW
2K
25 mW
2K
58 mW
2K
432 kg
K
0.15 W
K
0.11 W
K
7.2 W
K
9.6 W
18
Stability of the model
The continuous model has all its eigenvalues in the left half plane, and can
therefore be concluded to be stable. The model was discretized using the trapezoidal rule with an time step of two minutes. The discretized model has all its
eigenvalues within the unit circle is thus stable.
Model of the heating system
The temperature of the heating system, ∆Trad was modeled using a heating
curve. I.e., a linear function with the outdoor temperature as input and the
setpoint temperature as parameter was defined. The curve was adjusted so as
to result in reasonable indoor temperatures with the used building model.
It should be noted that this heating curve results in rather low values of
∆Trad , since the model is of a well insulated house without any infiltration.
Model of the ventilation system
To model the temperature of the inflowing air, the efficiency of the heat recovery
of the ventilation in the house in the implementation, 80%, was used.
3.2.2 Data
The temperature data used in the simulations was retrieved from the KTH-ACL
HVAC Testbed 1 . The data used was from two subsequents days and nights in
early February 2014. Data samples from every two minutes was used. A plot of
the outdoor temperature can be seen in Figure 3.8
The occupation data and the alarm status was set up from a hypothetical
scenario. In this scenario, the inhabitants of the house is a family with two
adults and two children. The family wakes up at 6 AM and turn of the shell
alarm. They leave home at 8 AM and turn on the alarm. At 4 PM the children
and one adults returns home and at 6 PM the second adult joins them. At 10
PM the shell alarm is turned on.
To avoid transient effects, constant data was added before the actual simulation data. As the system has a settling time of almost 24 hours, a day and a
nights worth of extra data was added.
3.2.3 Control strategies
Two control strategies were implemented. A ”smart” control similar to the
intended control in the implementation was simulated. With this control, the
temperature setpoint was 20 when the alarm was disarmed, 18 when the shell
alarm was turned on and 17 when the alarm was fully activated.
A ”dumb” control with constant setpoint and ventilation speed was simulated for reference.
1 hvac.ee.kth.se
19
Figure 3.8: The outdoor temperature used in the simulations. On the y-axis,
the temperature is i given in Celsius and on the x-axis the time is given in steps
of 120 seconds.
20
Chapter 4
Results
This section describes the results from the implementation and the simulation.
4.1 Implementation
The implementation that was performed, i.e.,control of the ventilation, worked
well. The script was allowed to run for three days and nights, and no disruptions
occurred. The inhabitants of the house did not report any changes in the indoor
air quality, and are positive towards further tests, preferably involving their heat
pump.
4.2 Simulation
The resulting indoor temperature with the two controllers can be seen in Figure
4.1 and 4.2.
The differences between the radiator temperature and the room temperature
can be seen in Figure 4.3 and 4.4. One might note that the shape in the plot
with the smart controllers resembles the step response from an RC-circuit. This
is perfectly reasonable, since the walls of the building is modeled as a systems
of two capacitances and three resistances.
Since the heat consumption in the building is a function not only of the
radiator temperature, but also of the mass of water in circulation in the heating
system, no exact quantification of the reduction in heat consumption can be
done. However, ∆Trad can be compared, as an indirect measurement of the
heat emitted by the radiators.
The ratio between the accumulated ∆Trad from the smart control and the
reference control is 0.78, indicating that a significant reduction in heat consumption can be achieved.
Furthermore, the ratio between the accumulated uc with the smart control
and the reference control is 0.78, indicating that reduction in heat consumption
might largely be caused by the reduction in ventilation.
21
Figure 4.1: Results from the simulation with the reference control. The simulation starts at midnight. On the y-axis, the temperature is i given in Celsius
and on the x-axis the time is given in steps of 120 seconds.
Figure 4.2: Results from the simulation with the smart control. The simulations
starts at midnight. On the y-axis, the temperature is i given in Celsius and on
the x-axis the time is given in steps of 120 seconds.
22
Figure 4.3: ∆Trad with the reference controller. On the y-axis, the simulated
temperature difference between the air and the radiator is i given in Celsius,
and on the x-axis the time is given in steps of 120 seconds.
Figure 4.4: ∆Trad with the smart controller. On the y-axis, the simulated
temperature difference between the air and the radiator is i given in Celsius,
and on the x-axis the time is given in steps of 120 seconds.
23
Chapter 5
Discussion
5.1 Potential for energy savings
In the simulations, a 20% reduction of accumulated ∆Trad could be observed.
This suggests that an energy saving of up to 20% could be achieved with the
proposed control. However, since the simulations have been performed with an
unvalidated model and over a quite short time period, the actual potential for
energy savings could be different.
5.2 Potential market
With the increased interest in energy efficiency in our society, the interest for
optimized heating of domestic premises is growing. Even though the difficulties
with the heat pump in the implementation might seem disheartening, the heat
pump companies in the IEA-project are eager to let their pump communicate
with other appliances.
5.3 Potential problems
As was apparent in the work with the implementations, creating an interface
with home appliances might be an obstacle. A less obvious problem is that frequent changes in the setpoint of heat pumps might affect the pumps negatively.
It could increase the wear on the pump, or prevent it from operating optimally.
Lastly, sending messages about the alarm status of you house over Internet
could be unwise. If the chat is not encrypted properly, the channel could be
hacked, exposing the information to criminals.
24
Chapter 6
Conclusions and further works
The suggested method was applicable, and resulted in energy savings. Since the
method was not tested during the heating season, and the heat pump could not
be involved, and no reliable method to quantify the energy savings was found,
it is hard to say how well the method stands up against other ones.
To further evaluate the applicability of the method the following is suggested:
• Run the control (with heatpump) in several houses, during a full heating
season.
• Investigate the possibilities to improve the control using MPC, preferably
with an machine learning algorithm to estimate the dynamics and the
occupancy patterns of the modeled house.
• Include weather forecasts and the influence of solar radiation in the control
strategy.
25
Bibliography
[1] Energimyndigheten, Energitillförsel och energianvändning i Sverige 2012,
TWh,
https://www.energimyndigheten.se/Global/Ny
[2] Energimyndigheten, Energistatistik för småhus 2012
http://www.energimyndigheten.se/Global/Press/Pressmeddelanden/Energistatistik
[3] Statistiska Centralbyrån Bostadsbestånd (kalkylerat)
http://www.scb.se/sv /Hitta-statistik/Statistik-efteramne/Boende-byggande-och-bebyggelse/Bostadsbyggande-ochombyggnad/Bostadsbestand-kalkylerat/87469/2012A01/Kalkyleratbostadsbestand-2012/ [20140217]
[4] J. Lu, T. Sookoor, V. Srinivasan, G. Gao, B. Holben, J. Stankovic, E.
Field, K. Whitehouse, The smart thermostat: using occupancy sensors
to save energy in homes, proc. 8th ACM Conference on Embedded Networked Sensor Systems, pp 211-224, 2010
[5] I.-H. Yang, M.-S. Yeo, K.-W. Kim, Application of artificial neural network
to predict the optimal start time for heating system in building, Energy
Conversion and Management, vol. 44, no. 17, pp 2791-2809, 2003.
[6] http://www.nyteknik.se/nyheter/innovation/forskning utveckling/article3805815.ece
[20140905].
[7] G. Gao, K. Whitehouse, The self-programming thermostat: optimizing setback schedules based on home occupancy patterns, proc. 1st ACM Workshop on Embedded Sensing Systems for Energy-Efficiency in Buildings,
pp 67-72, 2009.
[8] T. A. Nguyen, M. Aiello, Energy intelligent buildings based on user activity: A survey, Energy and Buildings, vol 56, pp 244-257, 2013.
[9] B.-Å. Petersson, Tillämpad Byggnadsfysik. Studentlittertur, 2009.
[10] F. Oldewurtel, A. Parisio, C. N. Jones, D. Gyalistras, M. Gwerder, V.
Stauch, B. Lehmann, M. Morari, Use of model predictive control and
weather forecasts for energy efficient building climate control, Energy and
Buildings, vol. 45, pp15-27, 2012.
26
[11] A. Parisio, M. Molinari, D. Varagnolo, K. H. Johansson, A Scenariobased Predictive Control Approach to Building HVAC Management Systems, proc. 2013 IEEE International Conference on Automation Science
and Engineering, Aug. 2013, pp.428-435
[12] Y. Ma, F. Borrelli, B. Hencey, A. Packard, S. Bortoff, Model predictive
control of thermal energy storage in building cooling systems, in 48th IEEE
conference on Decision and Control and 28th Chinese Control Conference,
2009.
[13] V. L. Erickson, Y. Lin, A. Kamthe, R. Brahme, A. Surana, A. E. Cerpa,
M. D. Sohn, S. Narayanan, Energy efficient building environment control
strategies using real-time occupancy measurements, proc. 1th ACM Workshop on Embedded Sensing Systems for Energy-Efficiency in Buildings,
pp19-24, 2009.
[14] T. W. Hnat, V. Srinivasan, J. Lu, T. I. Sookor, R. Dawson, K. Whitehouse,
The hitchhiker’s guide to successful residential sensing deployments, proc.
9th ACM Conference on Embedded Networked Sensor Systems, pp 232245, 2011.
[15] C. Fischione, P. Park, P Di Marco, K. H. Johansson Design Principles of
Wireless Sensor Networks Protocols for Control Applications in Wireless
Network Based Control. Springer, pp 203-237, 2011.
[16] https://nest.com/blog/2014/04/02/the-uk-just-got-a-little-more-comfy/
[20140905]
[17] https://nest.com/blog/2012/05/29/nest-thermostat-is-coming-to-canada/
[20140905]
[18] http://www.ngenic.se/ [20140905]
[19] http://www.nyteknik.se/nyheter/innovation/forskning utveckling/article3804554.ece
[20140905]
[20] A. Lindquist, J. Sand, An introduction to mathematical systems theory,
KTH, 2010.
[21] V. L. Ericksson et al., Energy efficient building environment control using real-time occupancy measurements, in BuildSys2009, November 2009,
pp19-24.
[22] T.-G. Malmström, Installationsteknik - Introduktion, Samordning med
byggnaden, Värme, Ventilation. Kungl. Tekniska högskolan Institutionen
för byggvetenskap, 2004.
[23] http://xmpp.org/about-xmpp/xsf/ [20140613]
[24] http://xmpp.org/xmpp-protocols/xmpp-extensions/ [20140613]
[25]
”Boverkets byggregler” 6:251 http://www.boverket.se/Global/byggao-forvalta-ny/dokument/regler-om-byggande/boverkets-byggreglerbbr/bbr20/Avsnitt-6-BBR-20.pdf [20140214]
27
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