optimisation of the energy use in a residential...
TRANSCRIPT
Optimisation of the energy use in a
residential environment
Othman Danoun
Supervisors: Prof. dr. ir. Lieven Vandevelde, dr. ir. Bart Meersman
Counsellor: Dimitar Bozalakov
Master's dissertation submitted in order to obtain the academic degree of Master of Science
in Electromechanical Engineering
Department of Electrical Energy, Systems and Automation
Chair: Prof. dr. ir. Jan Melkebeek
Faculty of Engineering and Architecture
Academic year 2015-2016
Optimisation of the energy use in a
residential environment
Othman Danoun
Supervisors: Prof. dr. ir. Lieven Vandevelde, dr. ir. Bart Meersman
Counsellor: Dimitar Bozalakov
Master's dissertation submitted in order to obtain the academic degree of Master of Science
in Electromechanical Engineering
Department of Electrical Energy, Systems and Automation
Chair: Prof. dr. ir. Jan Melkebeek
Faculty of Engineering and Architecture
Academic year 2015-2016
Preface First of all I want to thank my promoter Bart Meersman for his guidance and good advice during my
master dissertation. His sincere and constructive feedback has been helpful to bring this thesis to a
good end. I also want to thank him for proofreading my dissertation. Secondly, I want to thank
Christof Deckmyn for providing an optimisation example in Matlab. This has helped me a lot in
building my own optimisation model.
Finally, I want to thank my friends and family for the support during my education at the Faculty of
Engineering. I especially want to thank my classmates for the good times inside and outside the
classroom.
Abstract Production and consumption of electricity need to be equal at all times to keep the network
balanced. Conventionally, the production responds to changes in demand but the integration of
renewable energy sources challenges us to think differently. As the integration of renewable sources
grows every day due to the global concern of global warming, the transition towards “generation
drives demand” must be made. A part of the solution is to improve the flexibility of production units
and consumers. Next, storage technology can help to cope with the unbalance between production
and consumption by storing the excess of energy during low consumption and re-use it again during
high consumption and/or low production.
A way to improve the flexibility of consumers is to foresee some kind of control of their consuming
appliances. This can be on a large scale for industrial consumers or on a smaller scale for residential
consumers. This dissertation focuses on residential consumers where the control infrastructure and
logic is located in the residence. So the goal of this work is to optimise the energy use in a residential
environment. We first start with a literature study to obtain a better view of the different
possibilities present. The subject treated here is better known in literature as a Home Energy
Management System (HEMS). As the name says, it manages your energy consumption at home. A
HEMS controls different groups of appliances in a residence. Residential appliances can normally be
divided into four groups, namely the essential appliances, the shiftable appliances, the controllable
appliances and the thermal appliances. Each of these appliances, besides the essential ones because
they provide a necessary value to the members of the house (e.g. lighting), together with a PV –and
storage system form the HEMS.
Secondly an optimisation model for a HEMS is proposed here. An optimisation problem is
characterised by one or more optimisation objectives and need to satisfy a number of constraints.
The optimisation objectives chosen for this dissertation are cost and user’s comfort. There are
several other objectives, like emission and consumption, but the most commonly used ones in
literature is the combination cost and comfort. The constraints are limitations to the operation of the
household appliances that are set due to physical reason or by the residential user. For example the
minimal energy level of the home battery is set to 20% of its capacity such that the anticipated
lifetime is guaranteed.
Finally we end with simulations of three cases and some conclusions drawn from the three case
studies. The last chapter summarises the conclusions and ends with some final remarks based on a
white paper written by the Smart Grid team of Eandis. Eandis is the largest Distribution System
Operator (DSO) of Belgium, so it is interesting to know how they will tackle the problem of a major
change in our energy production and thus distribution.
I
Optimisation of the energy use in a
residential environment Othman Danoun
Supervisors: Prof. dr. ir. Lieven Vandevelde and dr. ir. Bart Meersman
Abstract - Production and consumption of
electricity need to be equal at all times to
keep the network balanced. Conventionally,
the production responds to changes in
demand but the integration of renewable
energy sources challenges us to think
differently. A part of the solution is to
improve the flexibility of residential
consumers. To study the behaviour of
different home appliances and their flexibility
a Home Energy Management System (HEMS)
is proposed in this master dissertation. It
proposes a framework of HEMS including a
grid, PV, a thermal storage tank and a home
energy storage battery. A multi-objective
optimisation algorithm for HEMS is proposed,
which minimizes electricity cost and
maximizes the comfort of the residential user
simultaneously.
Index terms – HEMS, multi-objective
optimisation, optimisation objectives, cost,
user’s comfort, constraints
1. Introduction
Over the past decade renewable energy has
taken a more prominent role in our electrical
energy production. The worldwide concern of
climate change and the limited amount of
primary energy resources left have turned
renewable generation sources such as wind
and solar into an important player on the
energy market. Conventionally, the production
responds to changes in demand but the
integration of renewable energy sources
challenges us to think differently. As the
integration of renewable sources grows every
day, the transition towards “generation drives
demand” must be made.
A way to improve the flexibility of consumers is
to foresee some kind of control of their
consuming appliances. This can be on a large
scale for industrial consumers or on a smaller
scale for residential consumers. This
dissertation focuses on residential consumers
where the control infrastructure and logic is
located in the residence. The subject treated
here is better known in literature as a Home
Energy Management System (HEMS). A HEMS
can consist of four interconnected components
[1]-[4], namely a monitor module, a prediction
module, the control logic unit and the
scheduling module (see Figure 1).
Figure 1: Architecture of a Home Energy
Management System (HEMS).
II
The monitor module monitors the actual
behaviour of the household appliances, the
storage system and the Photovoltaic (PV)
installation. In practice the appliances are
monitored by smart meters that continuously
measure the power consumption over a
certain period. The power production of the PV
is calculated by the prediction module
according to the input data of the monitor
module. These input data are solar radiation
data and environmental variables, such as the
temperature of the PV panels, which can come
from weather forecast or sensors. The
environmental variables are monitored,
because they affect the power output of the PV
panels. The scheduling module contains the
DSM software that computes the optimal
schedule according to the specified
optimisation objectives, e.g. electricity cost and
well-being/comfort, that home owners want to
pursue. After the optimisation procedure, the
control logic unit sends the optimal schedule to
the appliances and directly controls the
shiftable, controllable and thermal appliances.
Dynamic Pricing of electricity is a very
important element in the study of a HEMS. The
variability of electricity generation due to
renewables together with the variable
electricity consumption of households makes it
difficult for the Distribution System Operators
(DSOs) to keep working as before. Of course
this gives uncertainty regarding DSO costs and
revenues. The big challenge in Europe today is
to guarantee the grids stability for any
situation while encouraging the integration of
renewables in a way that is socially and
economically fair. There are several time-based
pricing schemes [5], namely Time of Use (ToU)
scheme, Critical Peak Pricing (CPP) scheme,
Critical Peak Rebate/Peak Time Rebate
(CPR/PTR) scheme, Real Time Pricing (RTP)
scheme and Inclining Block Rate (IBR) scheme.
These pricing schemes are published day
ahead, such that the schedule module can
calculate the optimal schedule for the next
day.
2. System Model
In this section the models for the home
battery, the thermal storage system (Combined
Heat and Power (CHP) unit and thermal
storage tank (TS)), the heat pump (HP) and the
non-interruptible loads, namely the washing
machine (WM), the dishwasher (DW) and the
cloth dryer (CD), will be developed. The
decision vector that is sent by the scheduling
module to the monitor module (see Figure 1) is
written as [4]:
, - (1)
Where
,
- is the power
that is put into or withdrawn from the battery
at every time slot N.
,
- is the
power that is withdrawn or put back into the
grid at every time slot N.
,
- is the working
status of the heat pump at every time slot N.
,
- is the
working status of the combined heat and
power unit at every time slot N.
,
- are the starting
times of the non-interruptible loads, where
WM: Washing machine, DW: Dishwasher, CD:
Cloth dryer.
Note that Pbat and Pgrid are double vectors, SHP
and SCHP are binary vectors, and Tstart is a
discrete vector.
III
2.1 The battery
The HEMS needs to prevent the home energy
storage battery from overcharge and
overdischarge by controlling the State Of
Charge (SOC) within a specified range
[SOCmin,SOCmax], where SOCmin and SOCmax are
the minimum and maximum allowable SOC of
the battery. This constraint is depicted in
equation (2).
The batteries SOC associated with the charge
and discharge of the battery is calculated by
the following equations, respectively [4]:
Where Cbat is the rated battery capacity in kWh,
PBat,tch and PBat,t
disch represent the charging -and
discharging power of the battery at timeslot t,
ηch and ηdisch are the charging -and discharging
efficiency of the battery system.
2.2 The thermal Storage system
As shown in Figure 2 the thermal storage
system consists of the CHP unit and the
thermal storage tank (TS). The CHP unit and
the thermal storage tank will be modeled
together. The HP that satisfies the space
heating demand of the house will be modeled
separately. The natural gas FNG comes in the
ICE of the CHP unit and simultaneously
produces heat QEG and electrical power PCHP.
The heat is recovered in the heat recovery
system and delivered (QCHP) to the thermal
storage tank. This hot water (Qhot water) is then
supplied to the hot water circuit of the
building. The electrical power that the CHP unit
produces is fed to the HP that delivers
Qspace_heating to the building. The excess of
produced electrical power is than supplied to
the grid or battery. If there is less production,
the difference is buffered by the battery, grid
or directly by the PV-system. The user
consumes hot water at approximately 50oC-
60oC. When the temperature of the hot water
exceeds one of these bounds, it gives the
HEMS a signal to start or stop the CHP unit. The
model presented in [4] is used to calculate the
temperature of the hot water inside the
thermal storage tank:
.
/ [
] 0 .
/1
Where and
are the hot water
temperatures (oC) inside the hot water tank in
time slots t and t+1; and
are the
temperatures of the ambient environment and
inlet water in time slot t; C is the equivalent
thermal mass (Wh/K); is the length of a
timeslot in hours; G is the product of the
surface area and thermal resistance of the
thermal storage tank (W/K). , and
are calculated according to the following
equations, respectively:
(W/K)
(K/W)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Figure 2: Schematic diagram of
thermal storage system.
IV
Where is the density of water (1000
kg/m3), is the hot water usage in time slot t
(l/h) and the specific heat capacity of water
(J/kg K). is the rated power of the CHP
unit (kW) and is the status of the CHP
unit in timeslot t (1 = ON, 0 = OFF). and
are the electrical and thermal efficiency of the
CHP unit respectively. In order to ensure the
comfort preference of the user, the CHP unit
should regulate the hot water temperature
within a prespecified range set by the user:
2.3 The Heat Pump (HP)
The model for the space heating/cooling load
developed in [6] is adopted in this dissertation.
In this model, the room temperature is
calculated as
Where and
are the room
temperatures (oC) in time slots t and t+1
respectively; is the length of a time slot;
is the heat gain rate of the house (Wh/h); is
the energy needed to change the temperature
of the air in the room by 1 K (Wh/K); is the
heating capacity of the heat pump unit
(=COP*PHP), PHP is the rated power of the HP
and is the working status of the heat pump
in time slot t (1 = ON, 0 = OFF). Now the room
temperature is expressed as a function of the
working status of the heat pump ( ), the
room temperature can be controlled within a
prespecified range set by the user.
(11)
2.4 The non-interruptible loads
In this dissertation, a washing machine (WM),
dishwasher (DW) and cloth dryer (CD) are
taken as non-interruptible appliances. These
appliances have two statuses, namely ON or
OFF. Once they have turned on, they must
keep working until their task is completed. The
task starting time and the number of time slots
needed for completing the task of each
appliance is set by the user. These appliances
should meet the constraints [4]:
∑
Where * + is the time
slot in which the task of the appliance a is
started; is the number of time slots that
are needed to complete the task of appliance
a.
3. Multi-objective Optimisation of HEMS
3.1 Energy Cost objective.
The overall net energy price over the
scheduling horizon is formulated in equation
(13), which consists of four parts: the first term
represents the overall electricity cost of buying
electricity from the grid, the second term
stands for the overall revenue of selling
electricity to the grid, the third term stands for
the degradation cost of the home energy
storage battery and the last term denotes the
cost of buying natural gas for the CHP unit.
∑ ( )
∑ ( )
∑
∑
(9)
(10)
(12)
(13)
V
(€/kWh) is the electricity price at timeslot
t when the user buys electricity from the grid;
and (€/kWh) is the electricity price at
timeslot t when the user sells electricity to the
grid. Note that ( ) means that
power is withdrawn from the grid and
( ) that power is feed into the
grid. is the battery degradation cost
(€/kWh) and (€/kWh) is the price of
natural gas at timeslot t.
3.2 User’s comfort level objective
In practice a residential user wants to reduce
his energy price without losing comfort. There
are different comfort concerns depending on
the type of home appliance. For example, for
space heating and hot water demand, the user
pays more attention to temperature, however,
for the washing machine (WM), dishwasher
(DW) and cloth dryer (CD), he focuses on when
the tasks of these appliance are completed.
Therefore in [4], a set of comfort level
indicators are proposed based on appliance
type. The comfort cost function is formulated
as
∑
Where A ={HP, TS, WM, DW, CD}; is the
comfort level indicator of appliance a and NA is
the number of controllable appliances in the
HEMS.
3.2.1 The Heat Pump (HP)
To quantify the user’s comfort level under the
operation of the HP, [4] proposes a comfort
level indicator whose definition is based on the
assumption that when the room temperature
is equal to the user setting temperature, the
user is most comfortable. If the room
temperature deviates from the setting value to
a certain extent, the user comfort level will be
decreased. The indicator is calculated as:
∑
Where
*
+
and is determined by
Where is the desired room temperature
set by the user and and
are
the two parameters that are related to the
temperature dead band of the heat pump.
According to the definition of in equation
(15), it is within the range of [0 100], where
zero means the user is most comfortable and
100 the user is least comfortable.
A similar comfort level indicator can be built
for the CHP unit. In this case the hot water
temperature determines the user’s comfort.
3.2.2 The non-interruptible appliances
A user’s comfort level for the non-interruptible
loads (WM, DW, CD) is determined by the
finishing time of these appliances [4].
(
)
(
) (
)
and
specify the valid working
interval for appliance a. If appliance a starts
within the interval ,
-,
where is the delay the user can tolerate
and is the preferred starting time of the
(14)
(15)
VI
user, the user is most satisfied. is the
number of timeslots appliance a need to finish
its task.
3.3 Multi-objective optimisation
In this model, the working status of HP and
CHP in each time slot and the task starting
times of WM, DW and CD are decision
variables. Minimizing the energy cost and
maximizing the user comfort level are the two
objectives of this problem. This model is
transformed into a single objective
optimisation model by the factor α [4]:
( )
( ) ( ) ( ) ( )
Where α is called the user preference factor,
0 ≤ α ≤1, through which the residential user
can take a tradeoff between the energy cost
and the comfort level conveniently.
4. Simulation results
Two cases will be investigated, all based on the
same production -and consumption profile.
The scheduling horizon is 24 hours, and is
divided evenly into 96 timeslots of 15 minutes;
that is Nslot = 96, ∆t = 0.25 h.
In case 0 we want to investigate how the
energy cost changes when we allow some
flexibility of the household appliances. The
results are shown in Table 1.
Case 0 Purchase cost (€)
Battery cost (€)
Sales revenue
(€)
Net cost (€)
No flexibility (case 00)
4.73 2.27 1.84 5.16
Flexibility (case 01)
4.09 2.21 1.57 4.75
From the last column in Table 1 we see that in
the case were we allow some flexibility of the
home appliances the net energy cost is smaller
than the case were there is no or little
operation flexibility of the home appliances
(€ 0.41 difference).
In case 1 the effect of comfort and cost as
objective will be examined. Cost and comfort
are contradictory, which means that the cost
will be larger (smaller) if the comfort level
indicator is lower (higher). Note that a higher
comfort level indicator means that the user’s
comfort is low. The user preference factor α
will be set to 0, 0.5 and 1, where 0.5 means
that cost and comfort are considered equally
important for the user.
From Figure 3 we see that the net energy cost
decreases and the comfort level indicator
increases with increasing α. Important to
remember is that a higher comfort level
indicator indicates that the most comfortable
bounds set by the user are exceeded (within a
reasonable margin) and thus that the user is
less comfortable.
Table 1: Simulation results case 0 (scheduling horizon = 24 h)
Figure 3: The net energy cost and comfort level as a function of the user’s
preference fact or α
VII
5. Conclusions
From case 0 it became clear that allowing
flexibility to the home appliances results in a
lower energy cost. This case is carried out with
a two-level Time ToU pricing scheme that has
an off peak –and standard rate price. The
difference between allowing and not allowing
flexibility resulted in a net energy cost gain
over a day of € 0.41 (see Table 1). This is of the
same magnitude that was obtained in [4].
The user’s preference α is a way for the
residential user to take a tradeoff between
cost and comfort level. The factor α is set to 0,
0.5 and 1, where zero means that only the
comfort level of the user is considered, one
indicates that only the cost objective is
minimized and 0.5 means that cost and
comfort are considered equally. The model is
run for the three values of α. The results
confirm the conclusion given in [4], namely
that the cost –and comfort objective are
contradictory. This means that maximising the
comfort results in a higher cost and vice-versa.
The algorithm behaves pretty much the same
for a three-level ToU -or RTP scheme.
References
[1] Y.Huan, H.Tian,L.Wang., "Demand response for
home energy management system," in Electrical
Power and Energy Systems., 2015, pp. 448-455.
[2] Christos S. Ioakimidis, Luis J.
Oliveira,Konstantinos N. Genikomsakis, Panagiotis
I.Dallas, "Design, architecture and implementation
of a residential energy box management tool in a
smartGrid," Energy, vol. 2013, no. 3 july 2014, pp. 1-
15, 2014.
[3] G. Graditi, M.G.Ippolito, R.Lamedica, A.Piccolo,
A.Ruvio, P.Siano, G.Zizzo, "Innovative control logics
for a rational utilization of electrical loads and air-
conditioning systems in a residential building.,"
Energy and Buildings, vol. 2014, no. 22 May 2015,
pp. 1-17, 2015
[4] Y.Zhang, P.Zeng,S.Li, C.Zang and H.Li., "A Novel
Multiobjective Optimisation Algorithm for Home
Energy Management System in Smart Grid,"
Hindawi, no. 18 January 2015, p. 19, 2015.
[5] Dr. S.C.Breukers, Dr. R.M.Mourik, "The end-
users as starting point for designing dynamic pricing
approaches to change household energy
consumption behaviours," Report for Netbeheer
Nederland, Projectgroep Smart Grids (Pg SG). March
2013.
[6] S. Shao, M. Pipattanasomporn and S. Rahman,
"Development of Physical-Based Demand
Response-Enabled Residential Load Models," vol.
28,no.2, pp. 607–614, 2013.
Table of contents
Chapter 1 ............................................................................................................................................... 1
Introduction ............................................................................................................................................. 1
1.1 Context and Motivation ................................................................................................................ 1
1.2 Outline ......................................................................................................................................... 12
Chapter 2 ............................................................................................................................................. 13
Pricing tariffs ......................................................................................................................................... 13
2.1 Capacity versus Volumetric tariffs ............................................................................................... 13
2.2 Dynamic pricing schemes ............................................................................................................ 17
2.3 Essential components of a dynamic pricing approach ................................................................ 18
2.3.1 Supportive Technology ......................................................................................................... 18
2.3.2 Feedback ............................................................................................................................... 19
2.4 Conclusions .................................................................................................................................. 21
2.5 Research questions...................................................................................................................... 22
Chapter 3 ............................................................................................................................................. 23
Dimensioning ......................................................................................................................................... 23
3.1 The PV-storage system ................................................................................................................ 23
3.1.1 General dimensioning criteria .............................................................................................. 24
3.1.2 Battery types ....................................................................................................................... 26
3.1.3 Sizing the PV-storage system ............................................................................................... 31
3.2 Thermal Storage System .............................................................................................................. 34
3.2.1 Sizing the Thermal Storage and CHP unit ............................................................................. 36
3.2.2 Sizing The Heat Pump ........................................................................................................... 41
Chapter 4 ............................................................................................................................................. 46
Modelling ............................................................................................................................................... 46
4.1 The Battery .................................................................................................................................. 46
4.2 The Thermal Storage System ....................................................................................................... 48
4.2.1 The thermal storage tank and CHP unit ............................................................................... 48
4.2.2 The heat pump ..................................................................................................................... 49
4.3 The non-interruptible loads ........................................................................................................ 50
Chapter 5 ............................................................................................................................................. 51
Optimisation .......................................................................................................................................... 51
5.1 The Cost Objective ....................................................................................................................... 52
5.2 The Comfort Objective ................................................................................................................ 53
5.2.1 Heat Pump (HP) .................................................................................................................... 53
5.2.2 Thermal Storage (TS) ............................................................................................................ 54
5.2.3 The non-interruptible appliances ......................................................................................... 55
5.3 Multi-objective Optimisation Model ........................................................................................... 56
5.3.1 The equality constraints ....................................................................................................... 57
5.3.2 The inequality constraints .................................................................................................... 59
Chapter 6 ............................................................................................................................................. 66
Case studies ........................................................................................................................................... 66
6.1 Case 0 .......................................................................................................................................... 68
6.2 Case 1 .......................................................................................................................................... 73
6.3 Case 2 .......................................................................................................................................... 77
Chapter 7 ............................................................................................................................................. 80
Conclusions ............................................................................................................................................ 80
7.1 Final Remarks .................................................................................................................................. 82
References ............................................................................................................................................. 85
Appendix ................................................................................................................................................ 91
List of Figures
Figure 1: Share in renewable energy in gross electricity consumption (%) of different EU countries
(source: Eurostat 2013). .......................................................................................................................... 1
Figure 2: Architecture of a Home Energy Management System (HEMS). ............................................... 2
Figure 3: Energy distribution in a single-family house of 4 members (source: Eandis). ......................... 4
Figure 4: Wind speed forecast at the residence location [2]. ................................................................. 5
Figure 5: Cost distribution of electricity bill in Belgium (source: Eandis). ............................................... 6
Figure 6: Overview of an external controlled Home Automation Network (HAN) [38]. ......................... 9
Figure 7: Architecture of a Wireless Sensor Network. .......................................................................... 10
Figure 8: Demonstrates the inefficiency of volumetric systems to address capacity [40]. .................. 14
Figure 9: Effect of the switch to capacity tariffs in the Netherlands in 2009 for different Volumes
(source: Liander) [40]. ........................................................................................................................... 16
Figure 10: Impact of automation on peak reduction [42]. .................................................................... 19
Figure 11: Multiple large pricing pilots vs. their percentage reduction in peak load [48]. (w/tech:
pricing scheme used in combination with automated appliances and feedback) ................................ 21
Figure 12: Optimal ratio (production/consumption) as a function of the battery capacity [50]. ......... 24
Figure 13: Optimal storage capacity (top) and Zc, Zs as a function of PV production [50]. ................... 25
Figure 14: Battery capacity versus temperature and charge-rate [50]. ................................................ 26
Figure 15: State Of Charge (SOC) as a function of the Open Circuit Voltage (OCV) for a AGM-battery
[50]. ....................................................................................................................................................... 27
Figure 16: The charging characteristic of a lead-acid (left) -and a Li-ion battery (right) [50]. .............. 28
Figure 17: Calendar lifetime of a Li-ion battery (left) and cycle lifetime for a AGM battery (right) [50].
............................................................................................................................................................... 28
Figure 18: A basic configuration of a micro-CHP unit with thermal storage [58]. ................................ 34
Figure 19: Schematic diagram of CHP-HP system [60]. ......................................................................... 35
Figure 20: Schematic diagram of thermal storage system. ................................................................... 36
Figure 21: Hot water usage profile [61]. ............................................................................................... 38
Figure 22: Temperature of hot water inside thermal storage tank. ..................................................... 39
Figure 23: Thermal power output of CHP unit. ..................................................................................... 40
Figure 24: House floor plan of a single family house [64]. .................................................................... 42
Figure 25: Air Exchange Rates (ACH) for different environments in a single family house [65]. .......... 43
Figure 26: Schematic diagram of a grid-interactive PV-system with DC -coupling [50]. ....................... 47
Figure 27: Schematic diagram that illustrates the three types of children [68]. .................................. 51
Figure 28: Relationship among parameters of HP comfort level indicator [4]. .................................... 54
Figure 29: Relationships among parameters of WM, DW and CD comfort level indicators [4]. .......... 55
Figure 30: Power distribution relationships of the HEMS in smart grid [4]. ......................................... 57
Figure 32: Critical load profile of residential household [72] ................................................................ 66
Figure 31: Forecasted PV power profile [72] ......................................................................................... 66
Figure 33: Electricity pricing schemes [70]- [71] ................................................................................... 67
Figure 34: Forecasted Outdoor temperature [73] ................................................................................ 67
Figure 35: Room temperature band over a day .................................................................................... 68
Figure 36: Simulation results in case of no flexibility of home appliance ............................................. 70
Figure 37: Simulation results in case of flexibility of home appliances ................................................ 70
Figure 38: Simulation results in case of no flexibility of home appliances ........................................... 71
Figure 39: Simulation results in case of flexibility of home appliances ................................................ 72
Figure 40: The net energy cost and comfort level as a function of the user’s preference factor α for
the case with the three-level ToU pricing scheme ................................................................................ 74
Figure 41: Power flow for case with ToU pricing scheme for α = 0.5 ................................................... 75
Figure 42: Power flow for case with RTP scheme for α = 0.5 ................................................................ 75
Figure 43: Power flow for α=0 (ToU-3 level) ......................................................................................... 76
Figure 44: Power flow for α=1 (ToU-3 level) ......................................................................................... 76
Figure 45: Power flow for case without battery degradation cost (ToU-3 level) .................................. 78
Figure 46: Power flow for case with battery degradation cost (ToU-3 level) ....................................... 78
Figure 47: Power flow for case without battery degradation cost (RTP scheme) ................................ 79
Figure 48: Power flow for case with battery degradation cost (RTP scheme) ...................................... 79
Figure 49: Levelized Cost Of Energy (LCOE) for wind turbines [74]. ..................................................... 83
Figure 50: Triangle of efficiency [74]. .................................................................................................... 84
List of Tables
Table 1: Overview of optimisation objectives considered in literature [8]. ............................................ 6
Table 2: Overview of the mathematical optimisation procedures [8]. ................................................... 7
Table 3: Effect of the switch to capacity tariffs in the Netherlands in 2009 (source: Liander) [40]. .... 16
Table 4: Parameter values of some commonly used battery types [50]. ............................................. 29
Table 5: characteristic data of a typical Belgium house [50]. ............................................................... 31
Table 6: Average energy consumption per day for some typical household appliances [50]- [55]. ..... 31
Table 7: Maximum Depth Of Charge (DOD) for different battery types ............................................... 32
Table 8: Correction factor for a specific orientation and roof inclination (source: HESPUL) [56]. ....... 33
Table 9: Summary of the results . .......................................................................................................... 33
Table 10: Commercial available micro-CHP units [58]. ......................................................................... 40
Table 11: The heat transfer coefficients of the different construction items in a house [64]. ............. 41
Table 12: Heating factor for different heat-up times and temperature drops [63]. ............................. 44
Table 13: The calculation of the total heating power for each room of a single family house. ........... 45
Table 14: The dimensioning results of chapter 3. ................................................................................. 45
Table 15: Parameter setting of WM, DW and CD [4] ............................................................................ 68
Table 16: Parameters for a Lithium-iron phosphate (LFP) battery ....................................................... 69
Table 17: Simulation results of case 0 (scheduling horizon = 24 h) ...................................................... 69
Table 18: Simulation results of case 1 (scheduling horizon = 24 h) ...................................................... 73
Table 19: Simulation results of case 2 (scheduling horizon = 24 h) ...................................................... 77
Abbreviations
DSM: Demand Side Management
HEMS: Home Energy Management System
PV: Photovoltaic
HVAC: Heating Ventilation and Air
conditioning
EW: Electrical Water heater
CHP: Combined Heat and Power
TS: Thermal Storage
HP: Heat Pump
WM: Washing machine
DW: Dish Washer
CD: Cloth Dryer
EV: Electrical Vehicle
ADP: Adaptive Dynamic Programming
PSO: Particle Swarm Optimisation
HAN: Home Automation Network
WSN: Wireless Sensor Network
MCU: Micro-Controller Unit
PLC: Power Line Communication
RF: Radio Frequency
DSOs: Distribution System Operators
ToU: Time of Use pricing scheme
CPP: Critical Peak Pricing
CPR/PTR: Critical Peak Rebate/Peak Time
Rebate
RTP: Real Time Pricing
IBR: Inclining Block Rate pricing
IHD: In Home Display
C-Rate: Charge rate of battery
SOC: State Of Charge
DOD: Depth Of Discharge
OCV: Open Circuit Voltage
CCV: Closed Circuit Voltage
FLA: Flooded Lead acid battery
AGM VRLA: Absorbed Glass Mat Valve
Regulated Lead-acid battery
Gel VRLA: Gel Valve Regulated Lead-acid
battery
LFP: Lithium-iron Phosphate battery
NiMH: Nickel- Metal Hydride battery
NiFe: Nickel-iron battery
ICE: Internal Combustion Engine
HRS: Heat Recovery System
MPP: Maximum Power Point
TSO: Transmission System Operator
List of symbols
Zc: Self-consumption
Zs: Self-sufficiency
Wp: Watt peak
Ah: Ampere-hour
A: Ampere
Cap: Useful battery capacity
Cap real: Real battery capacity
CBat: Batteries capacity in kWh
ηch: Charging efficiency of battery
ηdisch: Discharging efficiency of battery
Pmaxch: Max. charging power of battery
Pmaxdisch: Max. discharging power of battery
EPV: Yearly PV production
PPV: Peak power of PV installation
FNG: Fuel energy of Natural Gas
ηe: Electrical efficiency of CHP unit
ηHRS: Thermal efficiency of Heat Recovery
System
ηtot: Overall efficiency of CHP unit
PCHP: Electrical power output of CHP unit
QEG: Heat output of Exhaust Gasses
Qw: Heat loss of Heat Recovery System
Qspace heating: Heat demand for space heating
circuit of building
Qhot water: Heat demand for hot water circuit of
building
Pbat: Power exchange with battery
Pgrid: Power exchange with grid
SHP: Operation status of heat pump
SCHP: Operation status of CHP unit
Tstart: Discrete vector of starting times of non-
interruptible loads
Tambient: Ambient temperature of room where
the thermal storage tank in located
Tin: Cold water inlet heated until the desired
user temperature
U: Heat transfer coefficient
A: Heat-exchanging surface
φT: Transmission losses in a house
φV: Ventilation losses in a house
φRH: Required heating power for a house
Ti: Indoor room temperature
To: Outdoor temperature
HV: Heat transfer coefficient by ventilation
qV: Ventilation flow in a room
ρa: Mass density of air
ca: Specific heat capacity of air
ACH: Air Exchange Rate
Af: Floor area of a room
fRH: Heating factor
COP: Coefficient of performance of the heat
pump
Gt: Heat gain rate of a house
mTS: Mass of water in Thermal Storage tank
C: The equivalent thermal mass
G: Product of the surface area and thermal
resistance of the thermal storage tank
B: Product of hot water mass flow and specific
heat capacity of water ( )
R’: equals
ρwater: The mass density of water
cp: Specific heat capacity of water
τ: Time constant of the thermal storage tank
(R’C)
∆t: Time length of a timeslot in hours
Ft: The hot water flow rate in time slot t
TTSt: Temperature of hot water in thermal
store per timeslot t
TsetTS: User’s set point of hot water
temperature
∆TLTS: Lower temperature limit of thermal
store temperature dead band
∆TUTS: Upper temperature limit of thermal
store temperature dead band
TRoomt: Room temperature per time slot t
TsetRoom: User’s set point of room temperature
∆TLRoom: Lower temperature limit of heat
pump temperature dead band
∆TURoom: Upper temperature limit of heat
pump temperature dead band
PHP: Rated power of heat pump
CHP: Thermal capacity of heat pump (COP*PHP)
Nmina: minimal time slot at which non-
interruptible appliance can start its task
Nstarta: Starting time of non-interruptible
appliance a
Nmaxa: Maximum time slot at which non-
interruptible appliance a finishes its task
Ntaska: Number of time slots needed to finish
task of the non-interruptible appliance a
Pa: Rated power of non-interruptible appliance
a
ctgrid: Purchase cost of electricity at time slot t
ctsell: Selling price of electricity to the grid at
time slot t
cdeg: Degradation cost of the home battery
CC: Capital cost of home battery
LE: Battery life throughput energy in kWh
LN: Theoretical lifetime of battery in number of
cycles
LC: Actual lifetime of battery in number of
cycles
Es: Total energy storage capacity under
reference conditions, namely at 20oC and DOD
of 80%
ηTem: Temperature dependence factor (=LC/LN)
Fcost: Cost function of optimisation problem
Fcomfort: Comfort function of optimisation
problem
Ftotal: Objective function of multi-objective
optimisation problem
NA: Number of controllable appliances in the
optimisation problem
Ia: Comfort level indicator of controllable
appliance a
Nslot: Number of time slots in the scheduling
horizon
tmina: minimal time slot at which non-
interruptible appliance a can start its task
tmaxa: Maximum time slot at which non-
interruptible appliance a finishes its task
tideala: Ideal starting time of non-interruptible
appliance a
tdelaya: Acceptable delay w.r.t. starting time of
the non-interruptible appliance a
tworka: Number of time slots needed to finish
task of the non-interruptible appliance a
α: User’s preference factor (αFcost+(1-α)Fcomfort)
A: Matrix of the inequality constraints (A.x≤b)
b: Column vector of the inequality constraints
Aeq: Matrix of the equality constraints
(Aeq.x=beq)
beq: Column vector of the equality constraints
Ptcl: The consumption of the critical loads in a
house per time slot t
1
Chapter 1
Introduction
1.1 Context and Motivation
Over the past decade renewable energy has taken a more prominent role in our electrical energy
production. The worldwide concern of climate change and the limited amount of primary energy
resources left have turned renewable generation sources such as wind and solar into an important
player on the energy market. Also the 20/20/20 target of the European commission is responsible for
this. The 20/20/20 target stands for 20% of energy production must be renewable, 20% emission
reduction with respect to the emission level of 1990 and 20% increase in energy efficiency. In
Figure 1 you see the share of renewables in different European countries. In the past electricity
always flowed from a centralized power plant to the end users via a transmission and distribution
network. This one-way flow of energy made it possible to maintain the network balance using a
centralized management system. Nowadays the power generation is more decentralized due to the
introduction of renewable energy sources. Today Belgium has over 200 000 electricity generation
systems, scaling from large power plants and large wind turbine parks to small PV installations on the
roof of houses. Only a decade ago Belgium just had a few dozen of electricity generation sites. As a
result, utility companies and network operators will have to change their game plan. Transmission
operators are designing new products for better controlling the grids balance in these changing
market conditions. One of these new products uses cold storage warehouses for flexibility. These
warehouses can be cooled to lower temperatures when there is excess of energy, and turned off for
a couple of hours when there is less energy available without affecting the shelf-life of the stored
products. This gives the transmission operators the flexibility to quickly respond to unbalances in the
grid.
Figure 1: Share in renewable energy in gross electricity consumption (%) of different EU countries (source: Eurostat 2013)
2
To assure the security of supply and deal with the intermittent character of renewables we have to
make our grid smarter and enable an information exchange between users and utility companies.
The main advantage of this information exchange is that we can implement demand side
management (DSM) programs to control the consumption of the users. DSM control can be
advantageous for both users and utility companies. The utility companies will be able to reduce the
peak electricity loads and increase the reliability of their power grid and the users will be rewarded
by a reduction of their electricity bill.
A Home Energy Management System (HEMS) can consist of four interconnected components,
namely a monitor module, a prediction module, the control logic unit and the scheduling module
(see Figure 2). The monitor module monitors the actual behaviour of the household appliances, the
storage system and the Photovoltaic (PV) installation. In practice the appliances are monitored by
smart meters that continuously measure the power consumption over a certain period.
The power production of the PV is calculated by the prediction module according to the input data of
the monitor module. These input data are solar radiation data and environmental variables, such as
the temperature of the PV panels, which can come from weather forecast or sensors. The
environmental variables are monitored, because they affect the power output of the PV panels.
The scheduling module contains the DSM software that computes the optimal schedule according to
the specified optimisation objectives, e.g. electricity cost and well-being/comfort, that home owners
want to pursue. After the optimisation procedure, the control logic unit sends the optimal schedule
to the appliances and directly controls the shiftable, controllable and thermal appliances. These
appliances will be further specified below. Over the past few years HEMS became a very popular
research topic for universities and industries all around the world and therefore we will give a brief
overview of the different approaches in literature.
Figure 2: Architecture of a Home Energy Management System (HEMS)
3
The main idea and approach is for most literature studies the same, but there are some differences
in the used appliances, the optimisation objectives- and procedures. The appliances for HEMS
proposed in literature can be grouped in four categories [1], [2], [3], [4]:
The essential appliances : essential lighting, multimedia,..
The shiftable appliances: dish washer, washing machine, tumble dryer,…
The thermal appliances: Heating Ventilation and Air conditioning (HVAC), electrical water
heater (EW), combined heat and power (CHP) unit, heat pump (HP),…
The battery assisted appliances: electrical vehicle (as buffer and load), electrical battery as
buffer,…
In this dissertation we propose the following loads (see Figure 2):
1) The essential loads: loads that will not be controlled by the HEMS, because they provide a
necessary value to the members of the house
e.g.: essential lighting, television, computer, oven,…
2) The shiftable loads: loads that can be shifted in time, but once it starts its operation it cannot be
interrupted anymore. These appliances can be controlled by the HEMS according to the available
energy and preferential intervals set by the user. The typical loads that can be shifted in time are:
Washing machine (WM)
Dish washer (DW)
Cloth Dryer (CD)
3) The controllable loads: loads that can be controlled within their operation limits according to the
available energy. The two appliances that fit into this description are :
Refrigerator: Temperature control between 4 and 7oC. Above 7oC food items such as meat
,fish and dairy products have a reduced shelf life.
Freezer: The optimal temperature of a freezer is -18oC, because below this temperature the
shelf life of products is not extended. On moments of high production we can reduce the
temperature of the freezer and rely on the thermal inertia of the freezer on moments where
the production is low. When the freezers temperature gets higher than -18oC, this gives us
the input signal to send power to it.
4) The thermal loads: loads that must maintain a certain desired state according to the users set
point. The two thermal loads in a home are :
Space heating/cooling
Water heating
A combined heat and power unit and a heat pump will be used for the heat supply. In literature the
most HEMS use an electrical water heater, but because of the more stringent rules of Europe these
will be probably excluded in the near future. On the 26th of September 2017 the second tier
requirements regarding energy efficiency will come into force. The Energy Labelling scale for
4
water heaters is updated to A+ to F. In 2018 this will again be updated [5]. If we look at the energy
distribution of the different loads in a single-family house (4 members) than the largest consumption
of energy is due to heating (see Figure 3). This means that the control of the thermal appliances in a
home will be the most important ones.
In [1] they use a local battery for storage, whereas in [2] a plugged -in electrical vehicle (EV) is used as
local storage. This subtle difference makes the DSM control a bit more complicated. You have extra
conditions, like preferred battery level of the vehicle and time that the EV is not plugged in, that you
have to take into account. The storage system can be a battery where electrical energy is stored, or a
hot water tank to store thermal energy. This thermal energy can for example be used by a heat pump
for space heating or for hot water usage (e.g. a shower) on moments when the energy production is
low or energy demand is high. Some literature studies also use a micro-wind turbine as power
generation system. In [2] they use this in combination with a PV installation. The prediction of the
power output of a wind turbine is more complicated than for a PV panel, because wind speed data
for a residential location depend on several variables and location characteristics, such as
surrounding buildings and trees, which influences the actual wind energy that the wind turbine can
utilize. This means that relying solely on regional wind data forecast for a region where the residence
is located may result in a wrong prediction of the power output. A method for forecasting the wind
speed at a residential location is given in [6]. In this article they consider a three hour time window
for the forecasted wind speeds and approximate these wind speeds with a discrete probability
function of eight values. These values tell you the wind speed you can expect with a certain
probability of occurrence during a period of three hours. An example of such a wind speed forecast is
given in Figure 4.
There are two optimisation objectives that always come back in literature, namely cost and well-
being/comfort. Cost is sometimes the only objective that is considered [1], but the two objectives are
mostly considered together [2] - [4], [7]. In Table 1 [8] an overview of different objectives
combinations used in literature is given. In some HEMS literature they even consider emissions as an
optimisation objective, but this is less common than the other two.
Figure 3: Energy distribution in a single-family house of 4 members (source: Eandis)
5
The cost objective is described as a function that mostly consists of two terms, namely a term that
describes the electricity that we buy from the grid and a term that describes the electricity that we
sell to the grid [2] - [4]. In [4] they even take the degradation cost of the battery into account. Some
literature studies do not take this into account, but this is a very important parameter, besides the
charging/discharging efficiency, whether or not a profitable situation can be obtained.
Almost all literature studies consider a dynamic pricing for electricity, but in most countries there is
no regulatory framework. According to [9] the introduction of real-time price tariff would create a
flatter demand by ‘’shaving the peaks and filling the valleys’’, which basically means that the peak
loads can be reduced or shifted in time. In chapter 2 a brief overview of different dynamic pricing
tariffs and their potential of reducing peak loads will be given. In [2] they assume that the electricity
has a 5% lower price when supplied to the grid, compared to the buying price in every specific
instance. According to them this ensures two important issues:
1) This allows the utility companies to have a profit margin, allowing them to possibly cover
maintenance cost and unexpected problems of the distribution network .
2) The HEMS creates more value for the installed renewable energy source, by enabling a
higher penetration of renewable energy at the residence. This also reduces the use of the
distribution grid.
When several HEMSs in the grid are synchronized, this might result in undesirable oscillations of the
network frequency and voltage. By desynchronizing the real-time pricing of the HEMSs this
phenomena can be prevented [2]. This is beyond the scope of the present work, because we focus on
the working of a HEMS in a residential environment.
Figure 4: Wind speed forecast at the residence location [2]
6
All HEMS in our literature study assume a limited peak power transfer between grid and home. The
same idea is used in micro grids and the reason for that is:
1) Economical: the user will have a reduced electricity bill, because his cost for distribution and
transportation of electricity will be smaller (see Figure 5 ).
2) The utility companies will have a reduced network operation -and infrastructure cost.
The storage and connection with the grid are the two buffers needed to assure the security of supply
on moments when the energy production is low and/- or the consumption is high. Some literature
studies only assume an electrical buffer (battery), but since heating covers a large piece in our energy
use (see Figure 3) it is a good idea to foresee a thermal buffer where we store hot water.
Objectives References
Cost Well-being Consumption Cost and well-being Cost and consumption pattern Consumption and well-being Cost, consumption, emission Cost, well-being, emission, consumption
[10], [11], [12], [13], [14], [15] [16], [17] [18], [19], [20] [21], [22], [9], [23], [24], [25], [26] [27] [28] [29] [30]
Figure 5: Cost distribution of electricity bill in Belgium (source: Eandis)
Table 1: Overview of optimisation objectives considered in literature [8]
7
Mathematical optimisation References
Linear programming Convex programming Dynamic programming Mixed Integer linear programming Mixed Integer non-linear programming Meta-heuristic search Particle swarm optimisation Tabu search Genetic algorithm Heuristic scheduling Mix of optimisation and heuristics Backtracking-based scheduling State-queuing model Markov decision processes Artificial neural networks
[22], [31] [32] [9], [13] [10], [12], [25], [26], [27], [28] [14] References [21], [33] [23] [20] References [28] [19] [34] [35] [18]
The well-being objective is described in two different ways. In [4], they construct a comfort level
indicator for each controllable device. These indicators are in the range of [0,100], where close to
zero means the device operates within the users preferred limits, and 100 means the device operates
beyond or below the preferred limits of the user, e.g.: heat spacing device, indicator level will be
close to zero if room temperature is around 20oC, and will increase if you deviate from the 20oC.
A very different approach is used in [7] where they use queuing theory to model the well-being
objective. The waiting times of the specified household appliances are modelled according to
queuing theory. Queuing theory mathematically describes a queuing system and tries to relates the
uncertainty in the arrival patterns of its customers (home appliances in our case) to the uncertainty
of how the queuing system performs [36]. It is for everyone clear that the waiting times and the
number of customers in the queuing system depends on how its customers arrive at the queue and
how long they occupy the server (this will depend on the available energy in our case). In literature,
they assume two kind of demands in the model of their queuing system [7]. According to different
delay requirements they categorized it into:
Table 2: Overview of the mathematical optimisation procedures [8]
8
1. Delay-sensitive demands: which are appliances such as a cloth dryer, a washing machine and
a dish washer. These appliances are sensitive to the operation delay, because they directly
affect the users comfort level.
2. Delay-tolerant demands: these are the heating appliances, which are insensitive to the
operation delay due to their thermal inertia.
In [7] they assume that the delay-sensitive demands enter a high-priority queue, while the delay-
tolerant demands are put in a low-priority queue. The high-priority queue gets served first and when
this queue is empty then the low-priority queue is served. The problem they observed with this
approach is that this can cause a significant delay for the delay-tolerant demands. A solution they
proposed is to allow the demands in the low-priority queue to upgrade to the high-priority one with
probability β.
In literature, the problem is formulated as a multi-objective optimisation problem, where the two
optimisation objectives are the boundary conditions that need to be satisfied. Several optimisation
procedures are used in literature, but the two that always come back are the Particle Swarm
Optimisation (PSO) [1], [4] and the Adaptive Dynamic Programming (ADP) algorithm [7].
PSO is a stochastic global optimisation method inspired by the choreography of a bird flock. It
depends on the exchange of information between particles. A particle is a part of the HEMS that we
control, e.g. charging or discharging rates of the battery. In PSO, each particle adjusts its route
towards the positions with their own previous best performance and the best previous performance
of the whole swarm [1]. ADP is an iterative algorithm based on the principle of optimality. It is a very
useful tool for solving optimisation problems. The principle of optimality is, regardless the initial state
and decisions, the remaining decisions must provide an optimal policy [7]. Table 2 [8] gives an
overview of the mathematical optimisation procedures for HEMS scheduling used in literature.
Several projects about HEMSs have been carried out in the last few years. One of them is the
GreenCom project that started in 2012 and ended in 2015. The aim of this project is to balance the
local exchange of energy at the community microgrid level, to avoid affecting the centralized grid
with instability. The GreenCom Energy Management System controls energy consuming devices and
local energy generating and storage devices. The DSM will be based on individual consumer contracts
with attractive tariffs, reward/penalty clauses, etc. GreenCom has conducted its pilot on the island of
FUR, Denmark [37].
9
In the project they implemented a Home Automation Network (HAN) depicted in Figure 6. There are
three options to implement a HAN [38], these are:
Smart Meter Controlled Network
Internal Controller Network
Gateway/External Controlled Network
With the smart meter controlled network the smart meter acts as a central hub for the wireless
network. The data of the appliances will be gathered by the smart meters and sent to the utility
provider via GSM or long distance GSM. The control strategies are then determined by the utility
provider. The internal controlled HANs does not have a connection with the outside world and is
controlled by an internal controller. The control strategies and decisions are implemented in an in
home display (dedicated console/PC) . Within this thesis we will carry out the internal controlled
approach.
The gateway/external controlled network has an connection with the outside world via a dedicated
gateway device, such as an embedded PC/laptop. The data is sent to a backhaul network (web based
service) , which then provides control decisions that can be sent to the HAN via the gateway device.
This approach was carried out in the GreenCom project. For proper monitoring and controlling of the
devices we need a Wireless Sensor Network (WSN) [38]. The power output of the PV installation for
example depends on several factors like irradiation of the sun (W/m2) and temperature of the panels
(affects efficiency). To calculate an accurate power output we need to monitor these factors. The
basic building blocks of a WSN are the sensors, which are the senses of the system, and the
MicroController Unit (MCU) that contains the computational power of the system (see Figure 7).
Other devices where we use sensors for monitoring are the heating system, storage system, etc.
These systems will be further explained in Chapter 3. The antenna is used for wireless
communication with the network. This is one way to communicate with the network and is also
known as RF (Radio Frequency) communication. Another way is PLC (Power Line Communication).
Figure 6: Overview of an external controlled Home Automation Network (HAN) [38]
10
PLC uses the power wiring of the house to transport data from the plugged in devices to the HAN.
The advantage of PLC is that you do not need extra wiring and high speeds can be achieved for data
transfer. PLC is not the best way to transfer data, because the wiring of the house is not designed for
communication. The typical issues are noise on the line, impedance issues leading to attenuation
(weakening of electrical signal), surge protectors ( device that protects against over voltage) blocking
signals, etc. All these issues affect the performance of the network and therefore PLC is not
recommended for communication.
To power a wireless sensor node [38] (see Figure 7) there are three options:
Mains power
The wireless device is continuously powered and allows a greater range and data
throughput. The possibility of this depends on the operational environment. Within the
home environment a part of the network can be powered by mains, but for a ZigBee
network this is essential to ensure a correct operation. ZigBee is a meshed wireless
communication network between devices on a short distance. It is used for sending sensor
data and process monitoring and controlling.
Battery powered
It is the most common method for powering a wireless sensor node and has some pros and
cons. The advantage is that it is flexible in terms of placement and has a life time of 3-5 years
if used properly. The cons are a reduced duty cycle to achieve a long life time, limited to
monitor slow moving events like environmental temperature and of course the maintenance
of the battery.
Energy harvesting
This is a new way of powering a wireless sensor network and is being considered as a way to
permanently power a wireless sensor node. The major advantage is the removal of the
maintenance and battery change. Energy sources for energy scavenging can be light
(indoor/outdoor), temperature differences, vibrations, etc. The source that can be used
Figure 7: Architecture of a Wireless Sensor Network
11
depends on the application and available source. Light is the most energy rich source, and
thus the most frequently used energy source nowadays.
In this thesis, the focus will be on the residential market, because it is the most challenging market to
implement a HEMS in. If we look from the user’s point of view monetary expenses and high comfort
level are the two objectives that they will try to pursue. Due to the DSM program the operation of
certain home appliances will be postponed according to the available energy. Waiting time of the
non-interruptible appliances, the room temperature and the hot water temperature are used to
indicate the user’s comfort level. So if we rephrase, the multi-objective optimisation procedure
simultaneously:
1) minimizes monetary expenses/electricity bill
2) maximizes the comfort level of the residential user
A very important threshold for residential users regarding an implementation of a HEMS, is the
investment cost. The cost of smart appliances, such as a smart washing machine or a smart boiler, is
very high nowadays. Questions that a home owner will ask are:
1) What is the total investment cost?
2) When is the return of investment?
In [3] they implemented a HEMS in a 140 m2 house in Italy. They worked with a fixed electricity cost,
and the yearly saving was budgeted at €193,45. With question 1 and 2 in mind, we need to wonder if
it is necessary to use everything we proposed in the architecture of a HEMS in Figure 2. In Figure 3,
the energy distribution of a single-family home (4 members), we see that the heat consumption is
almost 80% of the total consumption. Washing and drying takes only 4% of the total consumption,
which means that the investment in smart household appliances is high with respect to the amount
of energy you can control. The same holds for refrigeration.
The challenges in smart grids today are the need for a regulatory framework and standardization. In
the last few years Belgium started to provide legal provisions for smart grid concepts, such as
Demand Side Management. Supported by the EU energy policy, a number of definitions of new smart
grid concepts have been included in the national (and regional) laws and codes. Furthermore, the
transmission operators created new balancing products and cost-benefit analyses of smart meters to
allow the participation of loads and generation sources, such as solar panels, to connect to the
distribution grid. These changes show a positive evolution towards DSM participation and regulatory
acceptance. Yet a number of regulatory obstructions, such as pricing tariffs, need to be changed
before DSM can be implemented in a home. Dynamic pricing for residential users is at the moment
limited by Belgian legal provisions. The standardization will have an important impact on interest in
investment and market competitiveness. In the meanwhile there exist very few (inter)national smart
grid standards. Luckily a number of EU standardization institutions (CEN, CENELEC and ETSI) are
developing smart grid standards, e.g. smart utility meters [39].
12
1.2 Outline
In this section we will give a short description of the different chapters. Chapter 2 deals about the
different pricing schemes and their ability to reduce peak demands. The introduction of renewable
energy in our energy generation makes it difficult for the Distribution System Operators (DSOs) to
keep working as before, which gives uncertainty regarding DSO costs and revenues. The big challenge
in Europe today is to guarantee the grids stability for any situation while encouraging the integration
of renewables in a way that is socially and economically fair.
In chapter 3 the thermal -and electrical storage systems are dimensioned. The dimensioning of these
storage systems are important for obtaining an economical working of the HEMS. For example when
the battery is over dimensioned, it will never fully charge and a part of the battery capacity will be
unused. The opposite holds for an under dimensioning of the battery system. In this case the energy
of the PV installation will not be fully exploited. So it is of great importance that all the different
components are correctly tuned to each other.
Chapter 4 describes the models for the home battery, the thermal storage system (CHP unit and
thermal storage tank), the heat pump and the non-interruptible loads of the HEMS. These models are
used in the optimisation part of the HEMS, where the optimal operation schedule of each appliance
is determined. These models are a very basic description of the respective appliances, where the
dynamics are neglected for simplicity.
The different appliances that are modelled in the previous chapter will be used in Chapter 5 to build
the optimisation procedure. The optimisation is based on two optimisation objectives, namely the
overall energy cost and the comfort level of the user. The cost objective will be minimized and the
comfort level of the user will be maximized over the next 24 hours (i.e., next day) based on the
forecasted outdoor temperature, the power output of the PV installation over the next 24 hours and
the forecasted electricity -and natural gas price (CHP unit runs on natural gas). The Genetic Algorithm
(GA) in Matlab is used to solve the optimisation problem.
In chapter 6 three cases are examined to verify the effectiveness of the proposed HEMS model. The
first case is a single objective optimisation (cost only) where the effect of allowing flexibility to the
household appliances on the cost is examined. The second case is a multi-objective optimisation of
cost and comfort level of the user. According to [4] cost and comfort are contradictory, which means
that the cost of energy increases as the user wants more comfort. The cases are simulated for two
types of pricing schemes, namely a Time of Use (ToU) pricing scheme and a Real Time Pricing (RTP)
scheme.
The last chapter, chapter 7, ends with some conclusions based on the simulations done in chapter 6.
The last section of chapter 7 finishes with some final remarks based on a white paper written by the
Smart Grid team of Eandis.
13
Chapter 2
Pricing tariffs
Dynamic Pricing of electricity is a very important element in the study of a HEMS. By the end of 2013
the installed capacity of PV systems and wind turbines in Europe has reached 81 GW1 and 117 GW2
respectively. The variability of electricity generation due to renewables together with the variable
electricity consumption of households makes it difficult for the Distribution System Operators (DSOs)
to keep working as before. Of course this gives uncertainty regarding DSO costs and revenues. The
big challenge in Europe today is to guarantee the grids stability for any situation while encouraging
the integration of renewables in a way that is socially and economically fair. In [40], they propose
several recommendations regarding the design of distribution network tariffs. These
recommendations will be further outlined in this chapter.
2.1 Capacity versus Volumetric tariffs
Volume and capacity are the two factors that determine the consumer’s bill. Depending on the EU
country, the network tariff can be based on:
Volume: the consumers are charged on the total volume of energy they withdraw or feed
into the grid. The measuring unit is watt per hour (Wh, kWh, MWh).
Capacity: the consumers are charged on the maximum amount of energy they withdraw or
feed into the grid at any instance in time. Capacity is measured in watt (W, kW, MW). This
capacity can be:
- Fixed: contractually agreed maximum capacity and corresponding price
- Variable: tariff varies at several moments of a day and aims at shifting the
demand from one period to another. A day can have several peak periods
where the price is set high with respect to the rest of the day (see ToU
pricing below). Note that for a variable capacity tariff a smart meter needs
to be installed.
To make the difference between volumetric –and capacity tariffs clear we will illustrate this with an
example in Figure 8. If the two consumer (case 1 and case 2) would have a volumetric tariff their bill
will be the same, because they consume the same amount of energy (70 kWh). If their network tariff
would be based on the capacity the bill of consumer 1 will be lower than consumer 2, because the
peak load of consumer 1 is lower. In this example it is clear that the capacity based tariff is more fair,
because consumer 2 charges the grid more.
1 EPIA, Global Market Outlook for Photovoltaics 2014-2018, June 2014
2 EWEA, Wind in Power 2013 European Statistics, February 2014
14
In most EU member states the network tariff is mostly heavily based on total volume, with the
exception of some countries like the Netherlands, Finland and Spain where it is capacity based. The
question, “why should we reconsider the distribution network tariffs?” rises. In [40] they give several
grounded reasons to do so:
1. Due to the increased electrification of household appliances (e.g. HP and EV) the peak loads
at certain periods are increasing. Reinforcing the grid because of these few peak periods is
expensive and paid by all consumers. A cheaper solution is to shift a part of this demand at
lower consumption periods. This means that electricity consumption habits of the consumers
have to change. However volumetric tariffs only encourage consumers to reduce their overall
energy consumption, but does not encourage them to limit their immediate consumption to
a certain level.
2. Prosumers (households with PV installation on their roof) are not encouraged to reduce their
injection of energy at peak production times by aligning their own consumption with their
own production. It is very important to consider this when designing a HEMS.
3. In some countries there exists a net metering system where a prosumer receives credits for
at least a portion of the energy he injects into the grid. In Belgium this depends on the power
of your installation and the network tariff per kW for the use of the distribution grid. By
paying this tariff your electricity meter is turned back per kWh when you supply electricity to
the grid. This results in a lower net consumption and thus lower billing basis for the DSOs
while the grid development and maintenance cost, determined by the network capacity, do
not reduce. This and the higher local peak loads due to increased electrification create
revenue uncertainty for the DSO.
Figure 8: Demonstrates the inefficiency of volumetric systems to address capacity [40]
15
4. The connection of distributed energy sources result in higher tariffs paid by non-generators.
Prosumers can compensate this cost due to more self-consumption or net-metering, while
their share of network costs will be transferred to other consumers who cannot invest in
their own PV installation. This can be avoided if the prosumers were able to contribute to
grid stability through ‘’smart contracts’’ with the DSO or a capacity based tariff instead of a
volumetric one.
From this section we remember that [40]:
Grid users should be able to self-consume and self-generate as long as the costs for using the
grid services is reflected in their bill.
Make network tariff more capacity based, and less volumetric, in order to limit the revenue
uncertainty for the DSOs.
To reinforce the recommendation of a more capacity based network tariff, the case study of the
Netherlands will be outlined here [40]. In 2008, the Dutch government decided to switch from a
mixed tariff (combination of volumetric and capacity) to a system completely based on capacity. Two
arguments justified their change:
Network costs are mainly capacity driven and determined by peak loads
New tariffs were considered to introduce a simple billing between DSOs and retailer (only
one bill sent by the retailer, instead of separate bills sent by DSOs and retailers)
To inform and prepare the consumers for the change a large media campaign was organized. Among
other initiatives, like an awareness website created by DSOs and government, consumers with a large
connection and little consumption received a letter inviting them to check their real capacity needs.
In 2009 the capacity tariff was introduced for both electricity (≤ 80A) and gas (≤ 40m3/h). Together
with this change the energy tax was revised in order to compensate for the capacity tariff and avoid
public backlash. The variable component of the energy tax was increased and a fix tax reduction
ensured that households with a standard connection would approximately pay the same bill. The
transition to a new distribution tariff was a smart move of the Dutch government. It was transparent
for consumers, reduced their electricity bill (see Table 3) and reduced the revenue uncertainty for
the DSOs. This tariff change also led to a reduction in administration cost in the energy sector,
because it made it easier for DSOs to bill suppliers.
From this case study we remember that [40]:
Grid users should receive clear and appropriate information before and after new
distribution network tariffs are implemented.
Table 3 gives a detailed calculation of the electricity bill for the mixed tariff (2008) and the purely
capacity based tariff (2009). We indeed see that due to the fixed tax reduction of the Dutch
government the consumers bill reduces. This reduction is less pronounced as the consumption
volume increases (see Figure 9), which can be a motivation for the consumer to lower his
consumption.
16
2008 tariff Volume amount
Connection Fixed part
Variable part
32,64 euro 32,64
18 euro 18
0,0336 €/kWh 2000 kWh 67,20
Tax reduction 0-10 000 kWh
-199 euro -199
0,0752 €/kWh 150,4
€ 69,24
2009 tariff Volume amount
Connection Fixed part
Capacity charge (25A)
16,44 euro 16,44
18 euro 18
115,6 euro 115,6
Tax reduction 0-10 000 kWh
-318,62 euro -318,62
0,1085 €/kWh 2000 kWh 217
€ 48,42
Figure 9: Effect of the switch to capacity tariffs in the Netherlands in 2009 for different Volumes
(source: Liander) [40]
Table 3: Effect of the switch to capacity tariffs in the Netherlands in 2009 (source: Liander) [40]
17
2.2 Dynamic pricing schemes
In [41] they discuss several dynamic pricing tariffs and their ability to reduce peak loads. Each of
these pricing schemes has been examined in several pilot projects in Europe. During the discussion of
each of these pricing schemes we will mention the results of some pricing pilots.
1. Time of Use (ToU) pricing
ToU pricing tariffs changes several times a day and aims at shifting the demand from one
period to another. A day can have several peak periods where the price is set high with
respect to the rest of the day. Four levels of prices can be distinguished (peak, partial peak,
off peak and weekend tariff). Also a varying season tariff is possible. The peak periods of
consumption are fixed (mornings, evenings,..) and communicated in advance to the end
users. It is very important to notice that ToU pricing only shifts the peak load to another
period without reducing the total consumption. Several pilots were conducted in Europe
with ToU pricing- UK, France, Germany, Northern Ireland and Norway. In these pilots a peak
reduction varying from 0 to 12% is realized.
2. Critical Peak Pricing (CPP)
CPP offers lower tariffs during non-peak hours and substantially higher tariffs during peak
hours. The critical periods here are moments that the electricity price is high due to high
consumption (e.g. very cold or hot weather) or when the stability of the grid is jeopardized
(e.g. risk of black-out). The end user and utility company agree a maximum number -and
length of the peak periods, however the peak moments cannot be set in advance, because
these depend on weather conditions. Usually when the utility company expects a critical day
they inform the end users a day in advance. The number of critical days vary from 1-18 days
a year (San Diego Gas & Electric Company 2010). The TEMPO tariff pilot, that was an
experiment of EDF in France from 1989-1996 with 400 000 recruited end users, has used the
combination of CPP and ToU pricing in its program. This program realized an overall national
peak reduction of 4% and achieved a load shifting up to 30% (for a limited number of days
and hours a year. Dynamic pricing in combination with load control achieved a load shifting
of 50% in Sweden.
3. Critical Peak Rebate/Peak Time Rebate (CPR/PTR)
With a CPR scheme the end user is being rewarded if his consumption is less than what the
utility company expects during a few critical peak hours a year. These are usually very hot
summer afternoons or very cold winter evenings. The same as for CPP, the maximum
number and length of the critical periods are agreed upon beforehand. Just like CPP, the
exact timing cannot be predicted, because it depends on market dynamics, but usually the
end users are informed a day in advance of a critical day. In Europe the need of shifting
critical hours is less than in countries where the climate differences are more pronounced.
That’s why CPP and CPR is not used in a large number of pilot projects. CPP and CPR are
useful when there is a flexible load of significant consumption (e.g. air conditioning). Such
loads can be turned off or put in a less consumption mode during the peak moments. The
most CPP and CPR pilots are outside Europe-US, Australia and New Zealand. In these pilots a
load shift of up to 38% is achieved [42].
18
4. Real Time Pricing (RTP)
With the RTP scheme the electricity price is tied to the market price which can change on an
hourly basis. This electricity price is linked to the price on the APX, which is the stock market
for energy. The end users can be encouraged to reduce their consumption during high price
periods by alerting them when the electricity price reaches a certain threshold (e.g. by a text
message alert). To be truly effective smart appliances need to respond automatically to the
RTP scheme. A number of pilots that have not produced robust results show
the following percentages [42], [43]:
13% peak reduction on the basis of 3 European pilots
10% peak reduction on the basis of 12 American pilots
5. Inclining Block Rate (IBR) Pricing
With the IBR pricing scheme the price of electricity increases stepwise with the consumption.
The first block is the cheapest, sometimes even free, but the subsequent blocks are more
expensive as the consumption increases. IBR pricing schemes mainly aims at encouraging a
reduction in consumption. In Belgium the IBR system is in place since 2001, with the aim to
help low income consumers to keep their electricity bills within limits [44]. A few pilots in
Japan and California showed that IBR pricing is the most effective for a group of end users
with a very high consumption.
2.3 Essential components of a dynamic pricing approach
A dynamic pricing approach can have several end goals, like balancing demand and supply, prevent
grid extension or strengthening, achieve end-user energy saving, etc. According to [41] a dynamic
pricing approach consist of three elements:
The pricing scheme
Supportive technology
Feedback
The pricing schemes have been discussed in the section above. In this section we will discuss the
combination of supportive technology and feedback with dynamic pricing schemes.
2.3.1 Supportive Technology
Supportive technology covers a wide range of devices, such as smart meters –and appliances, In
House Displays (IHD), etc. Smart meters and IHD are a must in combination with dynamic pricing
schemes. The smart meter is needed for actual and real-time metering of the energy consumption
patterns. It makes a bi-directional communication possible between end-users and utility or
suppliers, depending on which party is responsible for the roll-out of the meters. The IHD provides in
several forms feedback to the end-user and has been shown that it improves the responsiveness of
the end-users [42]. In the next section we will discuss the different forms of feedback and the
potential to change the energy behaviour of the end-users. More complex technologies, like smart
appliances, can help to support the behaviour change of the end-users. These appliances can be
19
programmed to automatically respond to changes in information. In a HEMS these changes in
information can be price signals, the amount of available energy in your system and the preference
settings of the end-user. The main advantages of automation is that it allows very quick response to
information change and therefore more effective demand shifting /peak load reduction. The state-
wide California pricing pilot and the SMUD pilot in the US are well-known examples of automation in
combination with pricing schemes. In these countries electrical heating/cooling is wide spread and
can be programmed to respond to peak periods. In Figure 10 some results of different pricing pilots
in combination with automation are presented. As can be seen the level of response to pricing
schemes with automation are significantly larger than without automation. These results need of
course to be nuanced, because of the wide spread of electrical heating/cooling and the climate
circumstances in these countries (hot summers and cold winters). Such results will not be feasible in
Europe, since the climate circumstances and the availability of flexible loads, such as air conditioning,
are different. Several studies have shown that the consumer is concerned to hand over control of his
energy demand to another party. In the Netherlands 53% of the respondents said no to utility control
[45]. The same trend is observed in other European countries . A solution for this is the integration of
a HEMS where the demand control unit is located in the end-users home.
2.3.2 Feedback
Feedback is mainly intended to reduce the energy consumption of the end-user and make him more
aware of his consumption pattern with the aim of trying to change his energy consumption
behaviour. Feedback has gained a lot of attention in research and distinguishes three kinds of
feedback: direct, indirect and associative feedback [46]. Direct feedback consists of information that
is available on request and directly responds to changes in the energy consumption. The advantage
of this feedback is that it directly shows the impact of behavioural changes. Indirect feedback is
characterized by a time delay and is suitable to show the effect of changes in the heating
consumption. Associative/unintended feedback results from (associative) learning. For example
Figure 10: Impact of automation on peak reduction [42]
20
when the bill increases after you bought a new device. With the different types of feedback in mind
we will make a distinction between feedback of price changes and feedback of (changes in)
consumption patterns and volumes. In [47], a study on effective feedback to encourage a
behavioural change towards energy consumption reduction is carried out. In this study they conclude
with a number of recommendations. We will mention a few recommendations that are relevant for
this present work:
A smart meter and a user interface are a must. This user interface can be an In-House-
Display, a smartphone App or an ambient technology (e.g. changing light colours).
Feedback must last at least three months, but is preferably permanent. A minimum of three
months is necessary to have the potential to make a ‘new’ behaviour lasting.
The feedback is direct so that the end-user can immediately see the impact of his behaviour
on the energy consumption. This helps to make energy visible and to set priorities with
regard to behaviours that can be changed and how that will affect energy usage.
The feedback is detailed, providing information about devices, spaces, people and functions
(e.g. cooking, heating, …). Pilots showed that the more detailed the feedback, the more it is
effective in changing the energy behaviour.
The feedback is positive, graphical and symbolic. The most appreciated display of feedback is
a combination of graphical and textual information. Graphs are preferred to show historical
feedback. Feedback is positive and not providing too much information about what is not
achieved, but focus on the remaining saving potentials.
These feedback recommendations will be very important to help change the energy behaviour of
end-users. Although the control of the appliances in a HEMS is fully automated, these feedback
recommendations are still useful to give end-users the insight in their energy consumption and learn
them to deal with the fact that we have to consume energy when it is available. Of course this is just
possible to some extent and is the reason why we foresee a buffer in a HEMS.
As mentioned in the part about supportive technology, the combination of pricing schemes with
automated appliances significantly increased the peak reduction (see Figure 10). In [48] they
investigated the combination of a pricing scheme (ToU, CPP, CPR/PTR, RTP) with multiple supportive
technologies and feedback technologies. They noted that the combination generated a higher peak
reduction. The results of multiple large pricing pilots in the US [48] are shown in Figure 11. According
to [41] these results are easily explainable, because with such a wide range of technologies a large
section of different segments (types of households) can be reached. It is logical that with more
automated appliances a higher peak reduction can be reached, however this will not deliver the most
cost-efficient approach. A new Zealand pilot targeted a particular segment (high incomes, high age,
new houses) and found that the ToU pricing scheme worked fine in combination with energy saving
tips and a monthly bill that showed the realized shifts during peak periods per day [49].
21
2.4 Conclusions
In this section we will conclude with some recommendation and lessons regarding distribution
network tariffs and dynamic pricing tariffs.
Recommendations when designing distribution network tariffs [40]:
Grid users should be able to self-consume and self-generate as long as the costs for using the
grid services is reflected in their bill.
Make network tariff more capacity based, and less volumetric, in order to limit the revenue
uncertainty for the DSOs.
Grid users should receive clear and appropriate information before and after new
distribution network tariffs are implemented.
Grid users should receive compensation from the DSOs when adapting their energy
consumption/generation in response to signals (e.g. at peak times).
Figure 11: Multiple large pricing pilots vs. their percentage reduction in peak load [48] (w/tech: pricing scheme used in combination with automated appliances and feedback)
22
Lessons and recommendations regarding dynamic pricing tariffs [41]:
When focusing on load shifting only, this can increase the total consumption. If e.g. the off-
peak price is too low compared to the peak price it can create an increase in consumption.
ToU pricing schemes targets habitual behaviour, while CPR and CPP focuses on conscious and
less frequent behaviour.
People are not motivated by pricing incentives only. Environmental motives, well-being, are
motivators as well.
Load shifting can be achieved without technology by using fridge magnets and calendars that
indicate the peak periods in a day. Additional technology such as an In House Display
increases the response rate.
Combination of pricing schemes with the appropriate technology gives the largest reduction
in peak load and thus the largest response from the end-users.
2.5 Research questions
The topics that will be investigated in this master dissertation are the following ones:
1. What is the effect on the energy cost when we allow flexibility to some house hold
appliances?
2. What is the effect of the cost objective on the comfort objective and vice-versa?
3. What is the effect of different pricing schemes, ToU and RTP, on the performance of the
optimisation algorithm?
4. What is the effect of the battery degradation cost on the charging and discharging behaviour
of the battery?
23
Chapter 3
Dimensioning
In this chapter we will dimension the electrical and thermal storage system of a HEMS. The
dimensioning of these storage systems are important for obtaining an economical working of the
HEMS. For example when we over dimension the battery it will never fully charge and a part of the
battery capacity will be unused. The opposite holds for an under dimensioning of the battery system.
In this case the energy of the PV installation will not be fully exploited. So it is of great importance
that all the different components are correctly tuned to each other.
3.1 The PV-storage system
When an end-user wants to install a PV-storage system, then the question arises: What are the
advantages with respect to a system without storage? Some end-users will do it out of ideological
reasons other for economic reasons. To evaluate the cost advantage two conventional concepts are
proposed in [50], the self-consumption Zc and the self-sufficiency Zs.
Self-consumption Zc
The self-consumption stands for the share of the produced solar power that is instantaneously
consumed. It is defined as the ratio of the own consumed PV energy and the total produced solar
energy. For a classical PV system with no storage the self-consumption is determined by the
instantaneous consumption and thus depends on the load profile. A system with storage will have a
larger self-consumption, because the self-consumed PV energy is the sum of the instantaneous
consumption and the PV energy to charge the battery.
Self-sufficiency Zs
The self-sufficiency is the share of the required energy that instantaneously can be provided. It is
defined as the ratio of the consumed energy that is self-produced and the total consumed energy.
For a classical installation the self-produced energy is equal to the instantaneous consumed PV
energy. When storage is foreseen the energy that originates from the discharge of the battery has to
be taken into account. This means that not only the self-consumption, but also the self-sufficiency
increases when storage is integrated.
The question that arises: Which storage capacity needs to be foreseen such that these parameters
sufficiently increase and is justified with respect to the investment cost? In [50] a dimension method
is proposed that tries to achieve a balance between these aspects. This method will be further
discussed in this chapter.
24
3.1.1 General dimensioning criteria
When dimensioning a PV-storage system the power of the PV panels needs to be determined. In a
classical residential installation the optimal dimensioning is given by setting the annual production
equal to the annual consumption. This is of course a theoretical target, because the production of a
PV installation is influenced by different factors, such as weather, ageing, pollution of the panels, etc.
Also the annual consumption is variable due to different external factors. According to the VREG the
average consumption of a Belgian family (2 parents, 1 child) is 3500 kWh a year. To take all these
influencing factors into account the PV-installation is a bit over dimensioned in practice. Besides this
limitation, there are also limitations on the PV-power that can be installed:
Limitation on number of panels due to limited roof surface
Maximum legal PV power of 5 kW single phase and 10 kW three phase
Available budget
In practice the system efficiency lies around 80%, because there are some conversion losses in the
AC/DC inverter and the battery. This implies that the total energy use increases as the energy
exchange with the battery increases. Figure 12 gives the optimal ratio (PV production/consumption)
as a function of the effective battery capacity. For an effective capacity of 1kWh/MWh consumption
we need to install a PV installation of 1,11 p.u. ,which implies a production that is 11% higher than
the average annual consumption. The 11% higher production is to compensate for the losses,
because the system has a certain efficiency. The power of a PV installation is given in watt peak, Wp.
This is the electrical power output under ideal conditions (1000 W/m2 solar radiation, solar cell
temperature of 25oC). A standard solar panel with an efficiency of 15% and dimensions of 165x100
cm has a peak power of 250 Wp. The production (kWh) depends on the mean solar irradiation of the
sun and the orientation of the panels. The optimal orientation for Belgium is an inclination of 35o and
orientated to the south. The average solar irradiation in Gent lies around 123 W/m2 (58 W/m2 direct
radiation and 66 W/m2 diffuse radiation). It is important to know that an optimal orientation of the
PV installation does not imply a higher self-consumption. A higher self-consumptions can be obtained
by orientating the panels to the southwest, such that the consumption peak in the evening is better
captured by the PV installation.
Figure 12: Optimal ratio (production/consumption) as a function of the battery capacity [50]
25
In [50] the condition for a proper sizing is an equilibrium between the self-consumption Zc and the
self-sufficiency Zs. Based on measurements on 25 houses a method is developed that estimates the
optimal battery capacity. For a new installation the battery size and the power of the PV installation
can be optimally chosen. The boundary condition that is set for such a system is Zc = Zs. As can be
seen on Figure 13 this corresponds to a storage capacity of 0.98 kWh/MWh, an estimated annual
production of 1,11 p.u. and an estimated Zc = Zs = 55%. Implementing the optimal system is not
always possible due to some constraints that are mentioned above. When the peak power of the PV-
installation is constrained due to limited roof surface, an estimation of the annual production has to
be made. An application on PVGIS (Photovoltaic Geographical Information System) can be used to
make an estimation for the annually production
(http://re.jrc.ec.europa.eu/pvgis/apps4/pvest.php?lang=en&map=europe). When the average
production is estimated, we can calculate the ratio production/consumption and read the storage
capacity on Figure 13. To better understand this method, we will illustrate with a small example.
Example [50]:
Assume a house with a mean consumption of 4.5 MWh/year and a PV production of 2.7 MWh/year.
This is equal to a ratio of 2.7/4.5 = 0.6 p.u. From Figure 13, it follows that the optimal capacity is
approximately 0.65 kWh/MWh, which corresponds with a battery capacity of 0.65 x 4.5 = 2.9 kWh, a
Zc = 70% (green line) and a Zs = 40% (blue line).
Figure 13: Optimal storage capacity (top) and Zc, Zs as a function of PV production [50]
26
3.1.2 Battery types
Batteries form an essential part of the PV-storage system, therefore it is important to understand the
characteristics and limitations of a battery. We will discuss some important parameters which needs
to be taken into account when choosing an appropriate battery for your application.
The battery capacity is a measure for the amount of energy that can be stored and is usually
expressed in ampere-hour (Ah). Because the capacity is influenced by some external factors, such as
temperature and discharge current, it is expressed under standard conditions. Usually this is defined
as the product of 20 hours and the current the battery can supply for 20 hours at a standard
temperature of 20oC, while remaining above a specific voltage per cell. For example a battery with a
capacity of 100 Ah can deliver 5 A over a 20-hour period at 20oC. This brings us to another battery
parameter, namely the Charge-rate or C-rate. The C-rate is the speed at which a battery discharges
and is expressed relatively to the battery capacity. When we discharge with a constant current at a
1C rate, the battery will be fully discharged after one hour. In reality there will be some losses, due to
the internal resistance, through which the effective capacity will be lower than 100%. The effect of
temperature and discharge rate on the capacity is illustrated in Figure 14. Observe how the capacity
decreases as the C-rate increases. It seems that a temperature increase has an increase in capacity at
a constant C-rate. This may look as a positive effect, but the lifetime of the battery decreases with
the temperature (see later in Figure 17 left). As a rule of thumb we can say that the lifetime reduces
with a factor two per 10oC higher than 25oC.
The State Of Charge (SOC) and the Depth Of Charge (DOD) are the two parameters that are used to
monitor the performance of a battery. The SOC is a measure for the amount of energy the battery
contains at that moment and is the ratio of the current capacity and the rated capacity of the
battery. Because it is not directly measurable, the SOC is measured indirectly through a voltage
measurement or through other specific parameters of the battery. The DOD indicates the amount of
energy that is already consumed. It is expressed as a percentage of the rated capacity and is linked
with the SOC trough DOD = 100% - SOC. If the battery is completely discharged until the DOD reaches
100%, the lifetime of the battery reduces significantly. A good balance between a useful capacity and
a long lifetime is of great importance. The optimal DOD is however dependent on the battery type.
Some target values for different battery types are given in Table 4.
Figure 14: Battery capacity versus temperature and charge-rate [50]
27
The battery voltage is an important parameter when a battery needs to be charged. There are
several voltages defined for a battery. The nominal voltage is the voltage displayed on the
registration plate of the battery. A battery bank is composed of multiple cells, each with a type
specific voltage. A lead-acid battery cell has a nominal voltage of 2 V, a lithium battery has 3.6 V. By
combining these cells in series, the battery voltage can be tuned to the DC input voltage of the
inverter. The instantaneous voltage of the battery varies around the nominal one and depends on
the SOC.
The Open Circuit Voltage (OCV) is the voltage of the battery in no-load condition. The OCV can give
you an idea of the SOC of the battery. To obtain a proper measurement the battery needs to stabilize
after use. This can take up to 24 hours dependent on the battery type. To illustrate Figure 15 depicts
the relation between the SOC and the OCV of an AGM (Absorbent Glass Mat) battery. In this case the
relation is linear, but this varies depending on the type. The Closed Circuit Voltage (CCV) is the
voltage during charging and discharging mode of the battery. When a current flows through the
battery, there will be a voltage drop over the internal resistance of the battery. The larger the
current, the bigger the voltage drop. This means that the CCV tells us something about the charging
and discharging of the battery. A battery has a minimal allowed voltage, called the Cut-off voltage.
This is the voltage at the terminal of the battery when it is completely discharged. Discharging the
battery below the cut-off voltage often result in damaging the battery and needs to be avoided at all
time.
When charging a battery it is important to set the correct voltage at the terminal of the battery. The
charging happens in a couple of phases and is shown for a lead-acid-and a lithium-ion battery in
Figure 16. Charging a lead-acid battery happens in three stages. In a first stage the battery is charged
with a constant current and is known as the constant current charge. In this stage the battery is
charged until 70 to 80% of its capacity. The voltage increases until the absorption voltage is reached.
When this voltage is reached, the charger switches to the topping charge. In this phase the voltage is
kept constant and the current drops. When the battery is almost fully charged the third phase,
namely the float charge, is initiated. This phase compensates for the loss caused by the self-discharge
of the battery. The charging cycle of a lithium-ion battery (Figure 16-right) is for the two first phases
the same as a lead-acid battery, but with different voltage levels and a smaller tolerance. The cycle
contains now four phases, with an extra phase to charge the battery when the voltage goes below a
certain threshold value.
Figure 15: State Of Charge (SOC) as a function of the Open Circuit Voltage (OCV) for a AGM-battery [50]
28
To guarantee the lifetime of a battery the parameters of the charging cycle needs to be set in the
inverter. The Lifetime of a battery is an important parameter, due to the high investment cost of a
battery system. It is difficult to forecast and depends on the way the battery is used. For batteries
there are two types of lifetime, namely the calendar lifetime and the cycle lifetime. The calendar
lifetime is the expired time before a battery becomes unfunctional, whether the battery was active
or not. This is, among other things influenced by the temperature of the environment the battery
works in. Figure 17 (left) shows the influence of the temperature on the calendar lifetime. On the
other hand the cycle lifetime is the number of cycles a battery can go through until its capacity is
dropped to 80% of its rated capacity. A cycle is defined as one fully charge cycle followed by one fully
discharge cycle. The cycle lifetime given by the constructor is a target value and is strongly
dependent on the DOD that is imposed. A typical characteristic for a AGM battery is given in
Figure 17 (right). A lower DOD results in a higher cycle lifetime, but results in a smaller utilization of
the battery capacity. A tradeoff between lifetime and utilization of the battery has to be made.
Figure 16: The charging characteristic of a lead-acid (left) -and a Li-ion battery (right) [50]
Figure 17: Calendar lifetime of a Li-ion battery (left) and cycle lifetime for a AGM battery (right) [50]
29
To wrap up we will discuss a number of battery types that are qualified to use for residential energy
storage. Table 4 summarizes a number of parameters of different battery types. Note that for the
Nickel iron type battery the cost and cycle lifetime are still uncertain according to [50].
Battery type Power density [W/l]
Energy density [Wh/l]
Cost *€/kWh/cell+
Lifetime [# cycles]
η efficiency
[%]
Weight [kg/l]
DOD [%]
Flooded Lead Acid
85 60 0.1 1100 60-70 2 50
AGM Valve Regulated Lead-acid
110 100 0.44 450 85-95 2 40
Gel Valve Regulated Lead-acid
90 80 0.4 900 90-95 2 70
Lithium-iron phosphate
427 280 0.3 2000 85-92 2.34 80
Nickel-metal hydride
218 200 0.45 1000 66-85 2.75 60
Nickel-iron 77 30 ? ? 65-85 1.3 80
The lead-acid battery is the oldest and most well-known battery type. Due to a low cost this type of
battery is still the most used one. The electrodes are manufactured from lead and lead oxide and are
submerged in an electrolyte consisting of a blend of sulfur acid and distilled water. The lead-acid
batteries can be divided into two categories, namely the Valve Regulated Lead-acid (VRLA) and the
Flooded Lead-acid (FLA). During charging, the electrodes must be completely submerged in the
electrolyte at all times to avoid damage to the battery cells. This requires a regular maintenance of
the battery by adding distilled water, because the electrolyte escapes as gas during the chemical
reaction. Due to the explosiveness of the gas, ventilation needs to be foreseen. This makes the FLA
battery less fit for residential use.
The VRLA battery is developed to eliminate the necessity of adding distilled water and to make it
possible to use a lead-acid cell in every position. The chemical process is not 100% efficient, which
means that in time a VRLA battery will run dry. This indicates that maintenance is still necessary, but
less frequent than for a FLA battery. Within the group of VRLA batteries a distinction based on the
transport medium is made. In this way they distinguish the Gel- and AGM (Absorbent Glass Mat)
VRLA battery. Gel cells add silica dust to the electrolyte, forming a thick gel. This reduces the
movement inside the battery case and makes it possible to use it in every position. Many gel
batteries also use a one-way valve instead of open vents. This helps the internal gasses to recombine
Table 4: Parameter values of some commonly used battery types [50]
30
back into water and reduces the gas leaks. Gel cells must be charged at a lower voltage (C/20) than
flooded or AGM cells to prevent excess gas from damaging the cells. Fast charging them may lead to
permanent damage to the cells. These batteries are sometimes referred to as ‘silicone batteries’.
AGM batteries hold their electrolyte in the glass mats and does not freely flood the plates (which is
the case in a FLA battery). The absorbent glass mats are used to absorb the electrolyte and provides
channels for oxygen gasses which will participate in the recombination reaction to prevent the
escape of hydrogen and oxygen gasses. Due to the physical properties of gelled electrolytes, the
power of a Gel battery declines faster than an AGM battery as the temperature drops below 0oC.
AGM batteries are good for high current -and power applications.
The Lithium-iron phosphate (LFP) battery is commercially available since 2006 and is evolved from
the classical lithium-ion battery. LFP batteries have a high energy density and perform well in the
field of safety, lifetime and power density. These features make the LFP battery ideal for electrical
vehicles, electrical equipment and residential storage. A disadvantage with respect to the FLA is the
higher cost. In several applications, especially solar power, efficiency is of great importance. The
round trip energy efficiency (discharge from 100% to 0% and back to 100% charged) of the average
lead-acid battery lies around 80%. The round trip energy efficiency of a LFP battery is 92%. The
charge process of lead-acid batteries becomes inefficient when the SOC reaches 80%. In contrast a
LFP battery will still achieve 90% efficiency under shallow discharge conditions [51].
The Nickel-metal hydride (NiMH) battery is developed in the eighties with the first application in the
aerospace, because of its high energy density and long lifetime. Just as the VRLA battery, also the
NiMH battery is completely sealed and requires little maintenance. NiMH and NiCad batteries are
one of the most difficult batteries to charge accurately. For Li-ion and lead acid batteries you can
control overcharge by setting a maximum charge voltage. The nickel based batteries do not have a
‘float charge’ voltage, so the charging is based on forcing a current through the battery. The voltage
to do this is not fixed like other battery types. Charging these cells in parallel is difficult, because it is
not sure that each cell or pack of cells has the same impedance, which means that some will take
more current than other cells, even when they are fully charged. This indicates that each string in a
parallel pack will need a separate charging circuit or we need to balance the current in another way.
For example by using resistors that dominate the current control. The charging efficiency of a NiMH
battery is typically 66%, meaning that you need to put 150 Ah into the battery for every 100 Ah you
get out. The faster you charge, the worst this gets. NiMH batteries are sensitive to damage on
overcharge when the charge rate is higher than C/10. Charging a battery of 100 Ah with a C/10 rate,
meaning charging with a current of 10 A, takes 10 hours to fully charge [52].
The Nickel-iron (NiFe) battery is invented by Edison in 1901 and is still often called the ‘Edison
battery’. This type of battery has a good resistant against both overcharge and deep discharge. It is a
very robust battery which is tolerant of abuse and can have a very long lifetime, even if treated bad.
The battery manufacturer Iron Edison sells NiFe batteries of 100 Ah with a cycle life of 11000 cycles
at a DOD of 80% and a lifetime on float of 30 years. The battery storage life is even 85 years. This
type of battery costs 808 $/kWh or 733 €/kWh, but has a very low cost per kWh per cycle (0.07
€/kWh/cycle) due to its high cycle lifetime [53]- [54]. According to [50] the lifetime of NiFe batteries
is not proven yet, but looks promising for residential storage.
31
3.1.3 Sizing the PV-storage system
In this section we will dimension the PV-storage system with the method explained above. In a first
step we will sum up all possible electrical devices in a house and their corresponding daily
consumption to get an idea of the energy demand for each appliance. Furthermore we assume a
house with characteristic data given in Table 5.
The consumption data profile of some appliances are found in [55]. These consumption profiles will
be used in the present work to examine the demand response possibilities of some appliances. The
profiles are not necessary for the sizing of the PV-storage system, but give an idea how much energy
a typical device consumes in a day.
Appliance Energy consumption [Wh] Duration [h] Peak power [W]
Dishwasher 1433 2,2 1180
Washing machine 230 1 1000
Tumble dryer 1358 1 2950
Oven 750 0,5 3000
Micro wave oven 500 0,5 750
Deep fryer 1850 0,75 1800
cooker hood 90 0,75 150
Refrigerator 1497 24 135
Freezer 580 24 100
TV,radio,modem 1830 4 500
Computer 290 3 100
Flat iron 1340 2,5 1200
Coffee set 170 0,25 1100
Vacuum cleaner 1460 1,5 1000
Lighting 1410 4 350
Laptop 210 5 90
Total 14998 24 15405
Average yearly consumption 3.5 MWh
Grid connection 40A, single phase
Favourable roof surface 37 m2
Roof orientation SE
Roof inclination 30o
Table 5: characteristic data of a typical Belgium house [50]
Table 6: Average energy consumption per day for some typical household appliances [50]- [55]
32
We see that the peak power is very high, but the different appliances will never work at the same
moment. This implies that there will be a certain dispersion of the total power. Besides many devices
will rarely work at their peak power. In a next step, we will determine the storage capacity based on
the data given in Table 5. The yearly consumption of 3.5 MWh gives a mean daily consumption of
approximately 10 kWh. Of course this does not correspond with the nearly 15 kWh given in Table 6,
because not all appliances will be used every day. From the research in [50], it seems that
1kWh/MWh is approximately a good estimation for the storage capacity. Note that this corresponds
with the criterion Zc = Zs in Figure 13.
From equation (1) it follows that we need a battery with a useful capacity of 3.5 kWh. Depending on
the type of battery a maximum Depth Of Charge (DOD) will be taken into account. The DOD indicates
how much energy the battery delivers. Another important battery parameter is the State Of Charge
(SOC). This parameter is linked to the DOD as follows DOD = 100%-SOC and is a measure for the
amount of energy the battery possesses at that moment. Note that the DOD and SOC are been
expressed as a percentage of the maximum capacity of the battery. In Table 7 the DOD of different
battery types is given.
Battery type DOD [%]
Flooded Lead Acid (FLA)
50
AGM Valve Regulated Lead-Acid (AGM VRLA)
40
Gel Valve Regulated Lead-Acid (GEL VRLA)
70
Lithium-iron phosphate (LFP)
80
Nickel-metal hydride (NiMH)
60
Nickel-iron (NiFe) 80
The effect of this parameter and other parameters on the lifetime of the battery is already discussed
in the previous section. If we choose the Lithium Iron Phosphate (LFP) battery with a DOD of 80%
than the real battery capacity in kWh is:
(1)
Table 7: Maximum Depth Of Charge (DOD) for different battery types
(2)
33
The capacity of a battery is usually expressed in Ah (Ampere hour). If we assume that the inverter
works at a DC voltage of 48 V, the real battery capacity in Ah is:
In practice a packet LFP batteries with a capacity of 100 Ah can easily be constructed. The last
parameter that needs to be determined is the peak power of our PV installation. As already
mentioned above batteries work with a certain efficiency (charge-and discharge efficiency). From
Figure 13 it follows that the optimal PV production is 11% (1.11 p.u.) higher than the yearly
consumption. Therefore the yearly PV production is estimated at:
The optimal orientated installation in Flanders has a yield of 950 kWh/kWp [50]. Given the SE
orientation and a roof inclination of 30o a correction factor of 0.96 follows from Table 8. This gives us
the following peak power for our PV installation.
Correction factor for a specific orientation and roof inclination
Inclination Orientation
0o 30o 60o 90o
West 0.93 0.90 0.78 0.55
South East 0.93 0.96 0.88 0.66
South 0.93 1 0.91 0.68
South West 0.93 0.96 0.88 0.66
West 0.93 0.90 0.78 0.55
Table 8: Correction factor for a specific orientation and roof inclination (source: HESPUL) [56]
Annually consumption 3.5 MWh
Power of PV panels 4.5 kWp
Estimated annually production 3.9 MWh
Battery type LFP
Capacity of home battery 100 Ah (48 V)
Estimated self-consumption Zc 55%
Estimated self-sufficiency Zs 55%
Table 9: Summary of the results of chapter 3
34
To obtain the desired peak power of 4.3 kWp we use 18 panels of 250 Wp. The total peak power is
then 4.5 kWp. The surface of one panel is 1.6 m2, which mean that the total area is 28,8 m2 ( = 18 x
1.6 m2). This is feasible, because the available roof surface is 37 m2. To wrap up the result are
summarized in Table 9.
3.2 Thermal Storage System
The thermal storage system consists of a Combined Heat and Power (CHP) unit, a Heat pump (HP)
and a hot water tank. A CHP unit is a device that simultaneously produces useful heat and electricity
from a single fuel source. The CHP or co-generation is widely used in large industry due to its efficient
use of fuel. The generated electricity is consumed on-site and the recovered heat is used in an
industrial process or fed into a district heating network [57]. The same concept can be used on a
smaller scale in buildings. These types of CHP units are called micro-CHP units. Micro-CHP systems
are powered by different prime movers, but the most advanced ones are fuel cells, Stirling engine
and internal combustion engine [58]. Studies have been established to assess the performance of an
building-integrated micro-CHP. In [59] such a study is conducted and demonstrated in terms of
primary energy saving and CO2 emission reduction. These two concepts have led to an growing
interest of integrating micro-CHP systems in buildings. The size of an micro-CHP unit can vary from an
electrical output of 1 kWe for a single family house to over 15 kWe for office buildings or small hotels.
For the CHP unit with an Internal Combustion Engine (ICE) as prime mover, the four-stroke spark
ignition engine is typically used. These engines are typically fuelled with natural gas, because of the
lower emissions (CO2) with relation to other fossil fuels such as gasoline or diesel. Approximately
one-third of the fuel energy is converted into mechanical work that drives the generator to produce
electricity. A part of the energy is recovered from the exhaust gasses by the heat recovery system.
Figure 18: A basic configuration of a micro-CHP unit with thermal storage [58]
35
The exhaust gasses have a temperature of 300-600 oC, which indicates that an significant amount of
heat still can be recovered. In general commercially available ICE based micro-CHP units have an
electrical efficiency of 24-30% and an thermal efficiency of 60-70%. This means that an overall
efficiency of 70-80% can be reached. Commercially available micro-CHP systems based on ICEs will be
summarized later in this chapter. A basic configuration of a micro-CHP unit with thermal storage is
shown in Figure 18. The water is drawn below from the tank and first passes the engine cooling heat
exchanger. This is typically a plate heat exchanger that cools the engine. The hot engine cooling
water than passes further to the exhaust gas heat exchanger where the hot gasses further heat up
the water. The thermal buffering is used to improve the thermal efficiency and reduce the duty cycle
of the system. An increasing duty cycle results in a reduced lifetime of the unit and a reduction in fuel
efficiency. The thermal buffering is usually implemented as an insulated tank of which the
temperature is limited to 100oC, because the cooling of the engine must be maintained. The space
heating and domestic hot water circuits run from the thermal storage tank by means of heat
exchangers. Instead of water, the energy storage medium can be a combination of water and a phase
change material (PCM). This improves the storage efficiency, because a smaller volume can be used.
A smaller volume results in smaller heating losses to the environment and thus improves the
efficiency [58].
In [60] a CHP unit is used in combination with a Heat Pump (HP). A CHP-HP system has the potential
to reduce the heat pump energy consumption during cold periods by using the heat recovered from
the PGU (Power Generator Unit), which is another name for CHP unit. This could improve the overall
efficiency and reduce the operational cost. In this article they present the design and feasibility study
of a CHP-HP system in a single family home. The operation cost of a CHP-HP system is compared to
the operation cost of a conventional HP system. Figure 19 represents the basic configuration of the
CHP-HP system proposed in [60]. In this configuration the electrical power to operate the heat pump
is foreseen by the CHP unit. The recovered heat from the CHP unit is used to provide additional heat
to the space heating -and hot water circuit of the building. This recovered heat is first used to provide
the space heating demand and the remaining is used to provide hot water for the building. An
auxiliary electrical heater and a gas fired boiler is foreseen to provide additional energy, when the
recovered heat is not sufficient.
Figure 19: Schematic diagram of CHP-HP system [60]
36
3.2.1 Sizing the Thermal Storage and CHP unit
Based on the two concepts shown above, we propose the following configuration (see Figure 20) of
the thermal storage system we will use in this work. A dimensioning of the different components will
be conducted based on space heating demand -and hot water consumption data.
The natural gas FNG comes in the ICE of the CHP unit and simultaneously produces heat QEG and
electrical power PCHP. The heat is recovered in the heat recovery system and delivered (QCHP) to the
thermal storage tank. This hot water (Qhot water) is then supplied to the hot water circuit of the
building. The electrical power that the CHP unit produces is fed to the HP that delivers Qspace heating to
the building. The excess of produced electrical power is than supplied to the grid or battery. If there
is less production, the difference is buffered by the battery, grid or directly by the PV-system. To size
the CHP unit we must notice that the hot water demand is the restricting factor in this problem,
because hot water can only be replenished by the CHP unit. This is a choice we make to keep the
control logic of our HEMS simple, because there also exist heat pumps that can heat up water. The
useful heat QCHP can be written as:
( )
A CHP unit has an electrical efficiency ηe, which means that the fuel energy FNG and electrical energy
PCHP are coupled with the following equation:
The heat recovery system has a thermal efficiency ηHRS, this implies that Qw can be written as:
(3)
(4)
Figure 20: Schematic diagram of thermal storage system
37
( )
If we insert equation (4) and (5) into equation (3) we obtain the relation between the electrical -and
thermal output of a CHP unit.
( )
( )
, ( )-
From these equations we can also derive the total efficiency of a CHP unit.
( )
The next step is to look at the daily hot water consumption profile. According to [61] the historically
most widely used hot water profile has been the so called ‘ASHRAE draw profile’. In [61] they made a
review of the source data (Perlman,1985). From this review two important considerations have been
revealed: first, all monitored data is from Canadian residences, and second, typically houses with two
adults and two children where a cloth washer and dishwasher are present. This data is given as a
percentage of the total daily hot water consumption (see Figure 21). So if we multiply this data with
the daily consumption of hot water, we obtain a consumption profile in liter per hour. A single family
house (4 members) consumes approximately 150 l/day according to [62]. This corresponds to an
energy of:
( )
(5)
(6)
(7)
38
We have assumed that cold water at 16 oC comes in and is heated to the desired 60 oC. We need 5.11
kWh of energy to cover the heat demand in a day. If we assume that we have a time available of 1
hour to heat the water in the storage tank to the desired temperature, than we need a CHP unit with
a thermal power output of approximately 5 kW. In the next section we will sum a number a
commercially available micro-CHP units. To see if the 5 kWth is a feasible power for the given hot
water profile a small model is made in Simulink. The following first-order differential equation, which
represents the thermal energy flow in the thermal storage system, is implemented in Simulink.
( )
The hot water tank is modeled as a closed system, where heat is extracted through a heat exchanger
(see Figure 18). We assume that the CHP unit works at its nominal power when turned on, so
dynamical behaviour is neglected in this simple model. The first term at the right part of equation (8)
represents the input heat energy from the CHP unit, the second term is the heat extracted by the hot
water circuit of the building, and the last term represents the heat losses to the environment. We
assume that the tank is placed in an non-heated room with an ambient temperature Tambient of 15oC.
The heat losses to the environment can then be written as:
( )
The overall heat transfer coefficient UTS is given by [58]:
Where,
(8)
Figure 21: Hot water usage profile [61]
39
Is the radiation term. The convective heat transfer coefficient for the thermal storage were taken as
free convection in water and air. The conductivity coefficient and thicknesses L were taken from [58].
This gives a global heat transfer coefficient of:
We assume a thermal storage V that is equal to the average daily hot water consumption, which is
150 liters in our case. The typical height of a storage tank is 1.2 m. If we assume the tank has a
cylindrical form, this gives us a surface ATS of:
√
Now that all parameters are known the simulation in Simulink is run. Figure 22 shows the
temperature of the hot water inside the tank. We have started with an initial water temperature of
16 oC inside the tank. As can be seen on Figure 22 this takes approximately one hour until the water
temperature reaches 60oC. After that the temperature is controlled within the desired temperature
range of 50-60oC. From these simulations we can conclude that the chosen CHP unit with a thermal
output of 5 kW is a feasible choice to cover for the hot water demand profile given in Figure 21.
Figure 23 shows the thermal output power of the CHP unit. The first on period is the largest, due to
the cold water at the start. To finish the part of the CHP unit, we will show a number of commercial
available CHP units and their characteristics. This information is shown in Table 10.
Figure 22: Temperature of hot water inside thermal storage tank
40
CHP unit ECOWILL, freewatt,
ecoPOWER 1.0
ecoPOWER 3.0/4.7
SenerTec Dachs
ZuhauseKraftwerk
Yanmar
Engine Honda GE160EV
Marathon engine
Dachs Volkswagen CNG 2.0
Miller Cycle gas engine
Cylinders 1-cylinder 1-cylinder 1-cylinder 4-cylinder 3-cylinder
Electrical efficiency
26.3% 24.7% 27% 33.5% 31.5%
Thermal efficiency
65.7% 65.7% 61% (72%) 56.4% 53.5%
Overall efficiency
92.0% 88.9% 88% (99%) 90% 85%
Electrical/ Thermal output
1.0 kW/2.5 kW
1.3–4.7 kW/4.0-12.5
kW
5.5 kW/ 12.5 kW (14.8kW)
19 kW/32 kW 10 kW/16.8 kW
Voltage 230 V, 50/60 Hz, single
phase
400 V, 50 Hz, 3 ~
230/400 V, 50
Hz, 3 ~
400 V / 50 Hz, 3 ~
240/120 ACV, 60 Hz, single
phase
Sound level < 52 dB (A) < 52 dB (A) 52–56 dB (A) < 50 dB (A) 56 dB (A)
Maintenance interval
6000 hours 4000 hours 3500 hours 5000 hours 10000 hours
Table 10: Commercial available micro-CHP units [58]
Figure 23: Thermal power output of CHP unit
41
3.2.2 Sizing The Heat Pump
To determine the power of a heat pump, the heat losses of the building need to be determined. The
calculation of these losses are stipulated in the norm NBN EN 12831 [B22] and- specific for Belgium
applications- in the NBN B 62-003 [B2] norm. For the calculation of the heat losses, the worst case
scenario is assumed. This means that the heat installation needs to deliver the heat demand at the
lowest possible outdoor temperature. In [63] they propose that the total thermal heating power
(ΦHL) that need to be installed is a sum of the transmission losses (ΦT), the natural-and/or mechanical
ventilation losses (ΦV ) and a term that takes the required heating power into account (ΦRH):
( )
With i the number of heated rooms. In [64] a heat load calculation of a single family house using the
CLTD/GLF (Cooling Load Temperature Difference/Gain-Loss Factor) method is carried out. These
calculations are based on the house with a floor plan shown in Figure 24. The heat transfer
coefficients U of the roof-and wall construction, the doors and the windows are shown in Table 11.
For each room we have to calculate the heat loss by transmission. This typically consist of different
terms, depending on whether the room has a window, partition wall or outer wall. The general
formula for the heat transfer loss by transmission is given by:
(W)
Where U is the heat transfer coefficient and A the heat exchanging surface. For an outer wall the
temperature difference is Ti – To, where Ti is the indoor temperature and To is the outdoor
temperature. For the inner/partition wall we use a temperature difference of Ti – Ta, where Ta is the
adjacent room temperature. For simplicity we assume that adjacent rooms that are not heated have
a temperature of 15oC. This is an assumption we make to make the calculations a bit easier. For
clarity we will do the calculation for the living room. The calculation for the other rooms are similar
and are summarized in Table 13.
(9)
Item U [W/m2K]
Roof construction Conventional roof-attic-ceiling combination
0.28
Wall construction Brick, insulation, gypsum wallboard Partition wall
0.34
0. 4
Doors Wood, solid core
1.82
Windows Clear double-pane glass in wood frames 3 mm thick The window glass has a 600 mm overhang at the top. Assume closed, medium-colour venetian blinds.
2.84
Table 11: The heat transfer coefficients of the different construction items in a house [64]
42
The living room has one outer wall and two partition walls. The outer west wall has a window and
door with the dimensions given in Figure 24. The heat loss through this wall is given by:
( ) ( ) ( ( ))
In the introduction we have mentioned that the heat losses are calculated for the worst case
scenario. According to the NBN B 62-003 [B2] norm the temperature of the worst case scenario that
need to be taken into account in the region East-Flanders is -8oC. The average indoor temperature is
assumed to be 20oC. The garage is typically a room that is not heated. As mentioned above, this
room has a temperature of 15oC. Now the heat losses through the partition walls can be calculated.
For the partition wall between living room and garage we obtain:
( )
( ) ( )
The heat losses through the door, the window and the roof still needs to be determined. These
values are respectively :
( ) ( ( ))
( ) ( ( ))
( ) ( ( ))
If we add everything up we become the transmission heat loss of the living room in equation (10).
Figure 24: House floor plan of a single family house [64]
1 2
3
43
The formula that is used to calculate the ventilation loss in a room is shown in the equation below
[63]. The ventilation flow is determined by the nature of the ventilation systems, where we make a
distinction between natural ventilation and forced ventilation. The ventilation guidelines for the
different areas in a house are found in a source book called the Residential Exposure Assessment.
The table from this book is for convenience shown in Figure 25.
( ) ( ) (W)
Where:
HV is the heat transfer coefficient by ventilation (W/K)
qV is the ventilation flow in the room (m3/h)
ρa is the mass density of air (1.2 kg/m3)
ca is the specific heat capacity of air (0.278 Wh/kg.K)
The ventilation flow is calculated with the Air exchange Rate (ACH) values given in Figure 25. The
general formula is given by
Where V is the volume of the living area. The living room falls in the class of the living areas and
requires a ventilation rate of 0.35 h-1 [65]. This corresponds to a ventilation flow of
( )
If we fill in all the parameters we obtain a ventilation loss of:
( ( ))
(10)
(11)
Figure 25: Air Exchange Rates (ACH) for different environments in a single family house [65]
44
The last term of equation (9), namely the required heating power φRH, still need to be determined. In
[63] they propose the next formula to calculate φRH
(W)
Where:
Af is the floor area of the heated room (in m2)
fRH (W/m2) is the heating factor, which depends on the heat-up time and the adopted
temperature drop of the room temperature when the heating system is turned off. In
Table 12 a number of target values for fRH are given.
If we assume the heat-up time is one hour with an temperature drop of 2K (fRH = 22), the required
heating power for the living room is
( )
This gives us for the living room a total thermal heating power of:
Heat-up time (h)
Residential
1 K 2 K 3 K
Heating factor fRH (W/K)
1 11 22 45
2 6 11 22
3 4 9 16
4 2 7 13
The same approach is used for the other rooms. The calculations for these rooms are carried out in
Excel and summarized in Table 13. From the calculations it follows that the total required thermal
heating power that need to be installed is approximately 8.2 kWth. If we use a heat pump with a
Coefficient Of Performance (COP) of 3 we need a heat pump with a rated power of 2.7 kWe.
A last thing that needs to be determined is the heat gain rate Gt (Wh/h) of the house. The heat gain
rate is the rate the house looses heat to the outside through the outside walls, doors, windows and
roof. It actually expresses how good or bad the house is isolated. The expression for Gt is given by the
equation below and is a function of the inside-and outside temperature at every instant t.
( ) (Wh/h)
Table 12: Heating factor for different heat-up times and temperature drops [63]
45
G0 (W/K) describes the amount of heat loss per K difference between inside -and outside
temperature. This parameter depends on the heat transfer coefficients U and areas A of the outer
parts of the house. The calculation for G0 is carried out in Excel, but for clarity this parameter will be
determined for the living room. In equation (10) and (11) the transmission and ventilation loss of the
living room is determined for the case of an inside temperature of 20oC and an outside temperature
of -8oC. As already mentioned G0 is the amount of heat lost per K difference between the indoor -and
outdoor temperature. This gives us for the living room a G0 factor of:
( )
The same approach is used for the other heated rooms, which are carried out in Excel. This gives us a
total heat gain rate Gt of:
( ) (Wh/h)
This parameter will be later used in the modeling part of the heat pump in chapter 4. To wrap up the
result of this chapter are summarized in the table below.
Battery capacity CBat 4.8 kWh
Rated power of PV panels 4.5 kWp
Rated power of CHP unit 5 kWth
Volume of thermal storage 150 l
Rated power of heat pump 2.7 kWe
Environment ( ) ( ) ( ) ( )
Living room (fRH=22,ACH =0.35)
632 345.6 980 1957.6
Kitchen (fRH=22,qV = 25 l/s)
447.3 840.7 496.5 1784.5
Bedroom 1+2 (fRH=11,ACH =0.35)
686.5 239.5 335.8 1261.8
Bedroom 3+bath (fRH=11,ACH = 0.35)
958.4 472.5 662.5 2093.4
Utility (fRH=6, qV = 7.5 l/s)
640.6 252.2 183.2 1076
Total 3364.8 2150.5 2657.9 8172.3
Table 13: The calculation of the total heating power for each room of a single family house
(12)
Table 14: The dimensioning results of chapter 3
46
Chapter 4
Modelling
In this chapter the models for the home battery, the thermal storage system (CHP unit and thermal
storage), the heat pump and the non-interruptible loads, namely the washing machine, the
dishwasher and the cloth dryer, will be determined. Also the operating constraints of each appliance
will be described in this chapter. The decision vector that is sent by the scheduling module to the
monitor module (see Figure 2) is written as [4]:
, -
Where
,
- is the power that is feed into or withdrawn from the
battery at every time slot N.
,
- is the power that is withdrawn or feed into the grid at
every time slot N.
,
- is the working status of the heat pump at every time slot N.
,
- is the working status of the combined heat and power
unit at every time slot N.
,
- are the starting times of the non-interruptible loads, where
WM: Washing machine, DW: Dishwasher, CD: Cloth dryer.
Note that Pbat and Pgrid are double vectors, SHP and SCHP are binary vectors, and Tstart is a discrete
vector.
4.1 The Battery
The HEMS needs to prevent the home energy storage battery from overcharge and overdischarge by
controlling the SOC within a specified range [SOCmin,SOCmax], where SOCmin and SOCmax are the
minimum and maximum allowable SOC of the battery. This constraint is depicted in the equation
below.
The batteries SOC associated with the charge and discharge of the battery is calculated by the
following equations, respectively [4]:
(13)
47
Where Cbat is the rated battery capacity in kWh, PBat,tch and PBat,t
disch represent the charging -and
discharging power of the battery at timeslot t, ηch and ηdisch are the charging -and discharging
efficiency of the battery system. Note that this also includes the efficiency of the inverter (see
Figure 26). For safety the charging and discharging powers should be controlled such that it is smaller
or equal to the maximum allowable values. This results in two other constraints that need to be
fulfilled:
A schematic diagram of a grid-interactive PV- system with battery is shown in Figure 26. The battery
and PV-system are coupled on the same DC-bus, such that only one inverter is needed instead of
two. The coupling with the battery is via a Maximum Power Point (MPP) controller that controls the
working point such that the PV panels work at their maximum power and delivers the right DC-
voltage at the exit. In practice the control logic such that the above constraints are satisfied is
implemented in the inverter.
(14)
(15)
(16)
Figure 26: Schematic diagram of a grid-interactive PV-system with DC -coupling [50]
48
4.2 The Thermal Storage System
As shown in Figure 20 the thermal storage system consists of the CHP unit, the thermal storage tank
and the heat pump. The CHP unit and the thermal storage tank will be modeled together. The heat
pump that satisfies the space heating demand of the house will be modeled separately.
4.2.1 The thermal storage tank and CHP unit
The thermal storage tank and CHP unit are used to deliver the hot water demand of the house. It is
clear that in this problem the temperature of the hot water in the thermal storage need to be
controlled. The user consumes hot water at approximately 50oC-60oC. When the temperature of the
hot water exceeds one of these bounds, it gives the HEMS a signal to start or stop the CHP unit. The
first order differential equations represented in equation (8) will be used to model the temperature
inside the thermal storage tank. For convenience equation (8) is repeated here.
( )
The equation will be rewritten in a more convenient form, such that it can later be used in the
optimisation part in Chapter 5. We first start by rewriting equation (8) [66]:
( ) ( )
( ) ( )
( )
By integrating both sides one gets:
∫( )
Where
If we solve equation (17) we obtain the temperature inside the thermal storage tank at every instant
in time [4]:
(
* [
] [ (
*]
(17)
(18)
49
Where and
are the hot water temperatures (oC) inside the hot water tank in time slots t and
t+1; and
are the temperatures of the ambient environment and inlet water in time slot
t; C is the equivalent thermal mass (Wh/K); is the length of a timeslot in hours; G is the product of
the surface area and thermal resistance of the thermal storage tank (W/K). , and are
calculated according to the following equations, respectively:
(W/K)
(K/W)
(kW)
Where is the density of water (1000 kg/m3), is the hot water flow rate in time slot t (l/h)
and the specific heat capacity of water (J/kg K). is the rated power of the CHP unit (kW) and
is the status of the CHP unit in timeslot t (1 = ON, 0 = OFF). In the model of the CHP unit we
assume that the CHP unit works at its rated power when turned on. This is done for simplicity, such
that we do not have to take the dynamic behaviour into account. Note that in equation (21) the
relation between the electrical –and thermal output of a CHP unit is used (see equation (6)). In order
to ensure the comfort preference of the user, the CHP unit should regulate the hot water
temperature within a prespecified range set by the user:
4.2.2 The heat pump
The model for the space heating/cooling load developed in [67] is adopted in this thesis. In this
model, the room temperature is calculated as
Where and
are the room temperatures (oC) in time slots t and t+1 respectively; is
the length of a time slot; is the heat gain rate of the house given in equation (12); is the energy
needed to change the temperature of the air in the room by 1 K (Wh/K); is the heating capacity
of the heat pump unit (=COP*PHP) and is the working status of the heat pump in time slot t (1 =
ON, 0 = OFF). To change the air in the room by 1 K, the energy required (Wh/K) is calculated as
(Wh/K)
Where is the specific heat capacity of air for a typical room condition and is the volume
of the house. The specific heat capacity of air is 1.012 J/(g K). This can also be written as
Because 1 g of air is 1/1290 m3 and 1 J is 1/3600 Wh. This results in a of
(19)
(20)
(21)
(22)
(23)
50
( )
See Figure 24 for Vhouse. Now the room temperature is expressed as a function of the working status
of the heat pump ( ), the room temperature can be controlled within a prespecified range set by
the user.
Where and
stands for the minimum and maximum room temperatures.
4.3 The non-interruptible loads
In this dissertation, a washing machine (WM), dishwasher (DW) and cloth dryer (CD) are taken as
non-interruptible appliances. These appliances have two statuses, namely ON or OFF. Once they have
turned on, they must keep working until their task is completed. The task starting time and the
number of time slots needed for completing the task of each appliance is set by the user. These
appliances should meet the constraints [4]:
∑
Where * + is the time slot in which the task of appliance a is started;
is
the number of time slots that are needed to complete the task of appliance a. The first constraint of
equation (25) expresses that the task should be finished within a prespecified time interval set by the
user. The second constraint expresses the non-interruptible nature of the appliance, namely once it
starts they must complete their entire cycle. The last constraint expresses the number of time slots
each appliance need to complete its task. For simplicity it is assumed that the appliances work at
their rated power during operation. The power of appliance a in time slot t is calculated as
Where Pa is the rated power of appliance a.
(24)
(25)
(26)
51
Chapter 5
Optimisation
The different appliances that are modelled in the previous chapter will be used here to build the
optimisation procedure. The optimisation is based on two optimisation objectives, namely the overall
energy cost and the comfort level of the user. The energy cost is minimized and the comfort of the
user is maximized over the next 24 hours (i.e., next day) based on the forecasted outdoor
temperature, the power output of the PV installation over the next 24 hours and the forecasted
electricity -and natural gas price (CHP unit runs on natural gas). The Genetic Algorithm (GA) [68] in
Matlab will be used to carry out the optimisation. From chapter 1 it became clear that the Particle
Swarm Optimisation (PSO) is a frequent used optimisation algorithm in literature. Matlab also has
this algorithm, but it can only be used for unconstraint problems. In this work an algorithm that
solves constraint problems with both continuous and discrete variables is needed. The GA is a
method for both solving constraint and unconstraint problems based on natural selection that
mimics the biological evolution. At each time step the algorithm randomly selects individuals from
the population and uses them as parents to produce the children for the next generation. There are
three types of children [68]:
Elite children are individuals in the current population with the best fitness values. These
individuals automatically survive the next generation.
Crossover children are individuals that are created by combining the vectors of a pair of
parents.
Mutation children are created by introducing random changes or mutations to a single
parent.
Figure 27: Schematic diagram that illustrates the three types of children [68]
52
5.1 The Cost Objective
The overall net energy price over the scheduling horizon is formulated in equation (27), which
consists of four terms: the first term represents the overall electricity cost of buying electricity from
the grid, the second term stands for the overall revenue of selling electricity to the grid, the third
term stands for the degradation cost of the home energy storage battery and the last term denotes
the cost of buying natural gas for the CHP unit.
∑ ( )
∑ ( )
∑
∑
(€/kWh) is the electricity price at timeslot t when the user buys electricity from the grid; and
(€/kWh) is the electricity price at timeslot t when the user sells electricity to the grid. Note that
( ) means that power is withdrawn from the grid and ( ) that
power is feed into the grid. is the battery degradation cost (€/kWh) and (€/kWh) is the
price of natural gas at timeslot t. The term
is the power flow of natural gas in kW (see equation
(4), FNG); is the working status of the CHP unit in timeslot t and is the length of a timeslot in
hours.
A detailed description and calculation method of the batteries degradation cost can be found in [69].
For convenience the most important formulas are shown here. The degradation cost cdeg of a battery
can be calculated by
Where Cc is the battery capital cost in € and LE is the battery life throughput energy in kWh. Battery
lifetime is expressed in number of cycles, measured at a reference DOD (Depth Of Discharge) and a
reference temperature. In order to reflect the impact of temperature on lifetime, LE is defined in [69]
as follows:
Where LN is the battery lifetime in number of cycles at a reference condition, i.e., ambient
temperature T = 20oC, DOD = 80%; Es is the total energy storage of the battery under reference
condition in kWh; ηTem is the temperature dependence factor and is defined as the ratio of the actual
lifetime LC to the reference one LN, as shown below.
(27)
(28)
53
5.2 The Comfort Objective
In practice a residential user want to reduce his energy price without losing comfort. There are
different comfort concerns depending on the type of home appliance. For example, for space heating
and hot water demand, the user pays more attention to temperature, however, for the washing
machine, dishwasher and cloth dryer, he focuses on when the task of these appliances are
completed. Therefore in [4], a set of comfort level indicators are proposed based on appliance type.
The comfort cost function is formulated as
∑
Where A ={HP, TS, WM, DW, CD}; is the comfort level indicator of appliance a and NA is the
number of controllable appliances in the HEMS.
5.2.1 Heat Pump (HP)
To quantify the user’s comfort level under the operation of the HP, [4] proposes a comfort level
indicator whose definition is based on the assumption that when the room temperature is equal to
the user setting temperature, the user is most comfortable. If the room temperature deviates from
the setting value to a certain extent, the user comfort level will be decreased. The indicator is
calculated as
∑
Where *
+ and
is determined by
Where is the desired room temperature set by the user and
and are the two
parameters that are related to the temperature dead band of the heat pump. According to the
definition of in equation (30), it is within the range of [0 100], where zero means the user is most
comfortable and 100 the user is least comfortable. The comfort level indicator is visualised in
Figure 28. The indoor temperature horizon is divided into three zones:
A comfortable zone: lies within [
-
A tolerable zone: is larger than
, but smaller than or
is smaller than
, but larger than
(29)
(30)
54
An intolerable zone: is larger than or smaller than
.
5.2.2 Thermal Storage (TS)
The definition of comfort level indicator is similar to that of the heat pump and is calculated as
∑
Where *
+ and
is determined by
and are user setting parameters whose meanings are similar to the parameter of
the heat pump comfort level indicator. Just as also lies within the range of [0 100] and can be
divided into three zones.
(31)
Figure 28: Relationship among parameters of HP comfort level indicator [4]
55
5.2.3 The non-interruptible appliances
A user’s comfort level for the non-interruptible loads (WM, DW, CD) is determined by the finishing
time of these appliances. The user will be most comfortable if the appliance starts on the moment
the user wants. In reality the user can tolerate some delay, which makes the non-interruptible
appliances flexible loads. Of course this delay is limited, which means that if the delay is too long, the
user will be unsatisfied. The comfort level indicators for WM, DW and CD share the same expression
and are calculated as
(
)
(
) (
)
Where * +, namely the Washing machine, Dish washer and Cloth dryer.
and
specify the valid working interval for appliance a. If appliance a starts within the
interval ,
-, where
is the delay the user can tolerate and is the
preferred starting time of the user, the user is most satisfied. is the number of timeslots
appliance a need to finish its task. The comfort level of the user decreases as the appliance starting
time is beyond the delay set by the user. The relationship between these parameters is depicted in
Figure 29. Also this indicator is divided into three zones, namely the comfortable zone, the tolerable
zone and the invalid zone. The comfort level indicator Ia is in the range [0 100] too and has the same
meaning as the comfort level indicator of the HP and TS.
(32)
Figure 29: Relationships among parameters of WM, DW and CD comfort level indicators [4]
56
5.3 Multi-objective Optimisation Model
The multi-objective optimisation model for the HEMS is formulated as
( ) ( ) ( ) ( ) ( ) ( )
In this model, the working status of HP and CHP in each time slot and the task starting times of WM,
DW and CD are decision variables. Minimizing the energy cost and maximizing the user comfort level
(which corresponds to minimizing the comfort level indicators Fcomfort) are the two objectives of this
problem. For implementation reasons the multi-objective optimisation model proposed in (33) is
transformed into a single objective optimisation model by weighting method [4], which means that
we can take a tradeoff between the cost -and comfort objective by changing α.
( )
( ) ( ) ( ) ( ) ( ) ( )
Where α is called the user preference factor, 0 ≤ α ≤1, through which the residential user can take a
tradeoff between the energy cost and the comfort level conveniently. The reason why we transform
the model to a single objective optimisation model, is because the genetic algorithm (GA) in Matlab
cannot work with integer variables for multi-objective problems. The syntax for the GA in Matlab is
as follows:
( )
Where fitnessfcn is the objective function (= Ftotal); nvars is the dimension of fitnessfcn (= 7); A*x ≤ b
are the inequality constraints; Aeq*x = beq are the equality constraints; LB and UB are the lower-and
upper bounds of the variables; nonlcon are the nonlinear constraints (not applicable here); IntCon is
a vector with the integer variables and options changes the default optimisation parameters by the
values given in options. Because GA does not accept equality constraints for problems with integer
variables, the equality constraints have to be transformed into inequality constraints as follows:
The Aeq matrix and beq vector are than included in the A matrix and b vector respectively. The syntax
now looks as follows:
( , - , - , - )
(33)
(34)
57
5.3.1 The equality constraints
The equality constraint expresses that production (PV panels, CHP unit and grid) equals consumption
at any moment in time. The control vector of the HEMS is shown below (see introduction of
chapter 4).
, -
Where Pbat and Pgrid are continuous variables and SHP,SCHP and Tstart are integer variables. As reference
we assume that Pgrid >0 when power is withdrawn from it and Pgrid <0 when power is feed into the
grid. For the battery we take an opposite reference, namely Pbat <0 means that the battery delivers
power to the loads or the grid and Pbat>0 indicates that the battery is being charged. With these
references in mind we can write out the equality constraint as follows:
∑
Where A = {WM, DW, CD} and Ptcl is the critical load power, which is not controlled by the HEMS. If
we put all the control variables on the left side of the equality sign and the known ones on the right
side we can write (36) in the form Aeq.x = beq.
∑
The variable is a discrete vector. This vector represents the starting times of the non-
interruptible loads and is determined in the first time slot of the scheduling horizon. The equation
above can then be rewritten as:
(35)
(36)
Figure 30: Power distribution relationships of the HEMS in smart grid [4]
58
∑
If we use the definitions given in chapter 4, the control variables can be written as:
∑
Where St are the working statuses (ON/OFF) of the corresponding appliances. The right hand side of
equation (37) can be written in matrix form, such that we obtain the form Aeq.x = beq. For one
timeslot we obtain the following coefficient row aeq for the right hand side and column vector beq for
the left hand side of equation (37).
, -
[
∑
]
The scheduling horizon consists of Nslot time slots, which means that Aeq has Nslot rows and 7*Nslot
columns; beq will be a column vector with size Nslot. This gives us the following matrices Aeq and beq:
[
]
[
∑
∑
∑
]
Aeq is a diagonal matrix with the coefficients of equation (37) on the main diagonal. This matrix can
be interpreted as follows: each row represents a timeslot t, and each column represents the control
variables of the corresponding time slot t. For example if we look to the first row of Aeq, the first
element of length 7 represents the control variables of timeslot 1, the second column of the second
timeslot and so on. Because at each time slot t, the consumption needs to be covered, we only get
the coefficients of (37) on the main diagonal.
(37)
(38)
59
5.3.2 The inequality constraints
The inequality constraints depicted in chapter 4 describe the operating constraints of each home
appliance. Some constraints are imposed by the residential user, for example max. and min. room
temperature, others are due to operational restrictions. For example the SOC of the battery must
maintain a minimum charge of 20% in order to obtain an acceptable lifetime of the home battery.
The home battery
We will first start by writing out the inequality matrix Abat of the battery. Equation (13) depicts the
charging constraint of the battery. For convenience this constraint is repeated here.
is calculated in two ways, depending on whether the battery is in charging -or discharging state.
The matrix will be written out for the charging state only. The discharging state is done in a similar
way. The charging state of the home storage battery is depicted in equation (14) and repeated
below.
If we put equation (14) into (13) we get
The equation is multiplied with the batteries capacity Cbat, such that we obtain the energy in the
battery at every timeslot t. The inequality constraint needs to be in the form A.x ≤b. This means that
the second equation from above needs to be multiplied with -1.
Equation (39) in matrix form looks as follows:
[
(
)
(
)
]
(39)
(40)
60
The top part of matrix (40) represents the first equation of (39), the bottom part represents the
second one. In (39) one can see that the present energy content of the battery is depending on the
energy content of the previous time step. This effect can be seen in the matrix , where at each
time step (each row in (40) represents a time slot) a zero row is replaced by ( ). The
same holds for the lower part of (40). In this way a lower triangle matrix is obtained with dimension
2Nslot x 7Nslot. The b vector of (39) is a 2Nslot column vector and is shown in (41).
[ (
)
(
)
]
The same matrix can be built for the discharging state of the battery, with the difference that the
matrix has row elements equal to (
). The b vector is the same as in (41). In block
matrix representation the A -and b matrix for the battery looks as follows:
[
]
[
]
The HP
The constraint of the HP operation is determined by the room temperature set by the residential
user. This constraint is depicted in (24) and repeated here.
The room temperature TtRoom is modeled in chapter 4 by equation (23):
Combining these two equations results in the following inequality constraints:
(41)
(42)
61
if the equations above are further written out with (12) we get:
(
*
(
*
If we write (43) in the form A.x ≤ b we obtain:
[
(
.
/
.
/
.
/
)
(
.
/
.
/
.
/
)
]
[
(
(
*
(
*
(
*
(
*
∑(
*
(
*
∑ (
*
)
(
(
*
(
*
(
*
(
*
∑(
*
(
*
∑ (
*
)
]
(43)
(44)
62
To understand how the matrices in (44) are obtained the first equation in (43) will be written out for
first two time slots. T0Room is the initial room temperature, which is considered known.
(
*
(
*
(
*
[(
*
]
The last equation is obtained by evaluating T1Room in the second equation. The first row of matrix A
and b in (44) is obtained by rewriting equation T1Room. Note that each decision vector Xi, i = 0,…,Nslot-1,
is a column vector of length 7.
(
*
↔ 0
1 [
]
(
*
The same approach is applied on equation T2Room, we get:
(
*
(
*
(
*
↔ 0 .
/
1
[
]
.
/
.
/
63
The CHP unit
Similar to the HP, the CHP unit should maintain the hot water temperature within a prespecified
range set by the user (see (22)). This constraint is for convenience repeated here.
The temperature of the hot water inside the thermal storage tank is modeled in (18):
(
* [
] [ (
*]
Combining the two expressions results in the inequality constraints that limits the operation
boundary of the CHP unit. These constraints are expressed by
(
* [
] [ (
*]
(
* [
] [ (
*]
To know how to build up the matrices, the first equation in (45) will be written out for the first two
time slots. T0TS is the initial temperature of the hot water. For the first timeslot we get:
(
* [
] [ (
*]
↔
.
/ [
] 0 .
/1
For reducing the writing of symbols, we have replaced (
) with q0.
↔ 0 .
/1
.
/ 0 .
/1
↔ 0 ( .
/) 1 [
]
(
* [ (
*]
For the second timeslot we obtain the following expression:
(
* [
] [ (
*]
Substituting T1TS into the equation above results in:
,
(
) ( )-
(
) ( )
↔ (
*
(
* ,
(
)-
( )
(45)
64
↔ [ (
*
(
* ( ) ( ) ] [
]
,
(
)-
(
)
A pattern can be seen in these equations, namely in the second time step, the coefficients of the
previous time step are multiplied with
of the next time step. Rewriting equations (45) in the
form A.x ≤ b results in the following matrices. The top part of represents the first equation of
(45) and the bottom part the second equation. The same holds for .
[
(
(
)
(
)
(
.
/)
(
)
∏
(
)
)
(
(
)
(
)
(
.
/)
(
)
∏
(
)
)
]
[
(
,
(
)-
,
(
)-
(
)
,
(
)- ∏
(
)
(
)
)
(
,
(
)-
,
(
)-
(
)
,
(
)- ∏
(
)
(
)
)
]
(46)
65
The non-interruptible loads
For the non-interruptible loads it is important that the corresponding appliance carries out its job
within a prespecified time interval set by the user. The constraints of these type of appliances are
expressed in (25). It is important to keep in mind that the starting times of the non-interruptible
loads are discrete quantities and are determined in the first time slot of the scheduling horizon. To
satisfy the residential user, the appliance’s task should start within the time interval (see (25)):
Where * +, namely the Washing machine, Dish washer and Cloth dryer.
The starting times of the non-interruptible appliances are determined by the intlinprog solver
in Matlab. The syntax for this solver looks as follows:
intlinprog(fa,nvars,[],[],Aeqa,beq
a,LB,UB)
Where Aeqa and beq
a are represented in (47)
[
(
)
(
)
]
Where and
represent the start and end position of the ones in matrix .
The cost function fa is a column vector that is the product of the rated power of appliance a (Pa in
kW) and the price scheme of electricity (€/kWh). The price scheme is adapted, where the price is set
to zero when there is enough PV production (e.g. 2 kW). In this way the solver will prefer to schedule
the appliance during PV production. The second term in intlinprog specifies the number of variables,
which is the number of timeslots (Nslot) in our case. Note that each variable in (47) is binary, where
zero means OFF and one means ON. The third and fourth term are meant for the inequality
equations, but this is not applicable here. The equality equation (47) expresses that appliance a can
start its operation between time slot , ,
- and must take timeslots to finish its
operation. The last two terms in intlinprog represent the lower-and upper bound of each variable.
Because each variable is binary (1/0) the lower bound vector LB is a zero vector and the upper bound
vector UB is a unit vector. The intlinprog solver returns a binary vector of length Nslot with exact
successive ones. The position of the first one element determines the start time of the non-
interruptible appliance a.
(47)
66
Chapter 6
Case studies
In this chapter simulations are performed to examine the effectiveness of the proposed HEMS
strategy. The simulation programs are coded in Matlab and are run on a Windows 10 (64 bit) Intel®
Core™ i3-5005U CPU @ 2 GHz computer with an 8 GB memory. Three cases will be investigated, all
using the same production -and consumption profile. The scheduling horizon is 24 hours, and is
divided evenly into 96 timeslots of 15 minutes; that is Nslot = 96, ∆t = 0.25 h. The data of the used
pricing schemes, namely Real Time Pricing (RTP) scheme and Time of Use pricing (ToU) scheme, are
from [70] and [71] respectively. The RTP pricing scheme data is taken from the Belpex site. Belpex is a
power exchange for anonymous, cleared, short term trading in electricity and is composed of three
market segments, namely Belpex DAM, Belpex CIM and Belpex SRM. For more information the
reader is referred to the website of Belpex (www.Belpex.be). The ToU pricing scheme is found in an
Canadian master thesis [71]. We will use two type of ToU pricing schemes. One has two pricing
levels, namely an Off-Peak price from 12 pm – 7 am and a standard price from 8 am - 12 pm. The
second is a three-level ToU scheme with an Off-Peak price from 12 pm – 8 am, a standard price from
12 am – 4 pm and an On-Peak price from 8 am – 12 am and from 4 pm – 12 pm. A PV production
profile [72] and critical load profile [72] is hard to find online and is provided by my promoter Bart
Meersman and counsellor Dimitar Bozalakov. The forecasted outdoor temperature is taken from a
website named AccuWeateher.com [73]. This was the temperature in Gent in the month March. The
pricing schemes, PV production profile, critical load profile and outdoor temperature are shown in
Figure 31- Figure 34 respectively.
Figure 32: Forecasted PV power profile [72] Figure 31: Critical load profile of residential household [72]
67
Figure 33: Electricity pricing schemes [70]- [71]
Figure 34: Forecasted Outdoor temperature [73]
68
6.1 Case 0
In case 0 we want to investigate how the energy cost changes when we allow some flexibility to the
household appliances. In a first simulation (case 00) the room temperature boundary is set to
,
- , where Tset is the user’s temperature set point (Tsetroom = 20oC), the
temperature in the Thermal Storage (TS) water tank is limited to ,
-, with
equal to 55oC. The starting times of the non-interruptible appliances are not regulated and are
set equal to the ideal starting times , (see Table 15). In a second simulation (case 01) the
temperature boundaries are selected wider. For the TS tank the temperature of the hot water is
regulated within the temperature range of [50oC, 60oC]. The boundaries of the room temperature
depend on the moment of the day and are represented in Figure 35. In the night we allow a
temperature band of 2oC around the temperature set point of the user. From 6 am – 8 am the
temperature band is reduced to 0.5oC, because than the residential user is awake. During the day,
the temperature band is elected wider again, because no one is at home normally. When the
residential user comes home in the evening, the temperature band is narrowed again. The non-
interruptible loads are controlled within a minimum and maximum time slot as explained in section
4.3 of chapter 4. These boundaries for the washing machine (WM), the dishwasher (DW) and the
cloth dryer (CD) are taken from [4]. For convenience of the reader the values are shown in Table 15.
Parameter Washing machine (WM)
Dish washer (DW)
Cloth dryer (CD)
(time slot) 37 37 77
(time slot) 76 84 96
(time slot) 48 56 80
(time slot) 4 4 8
(time slot) 4 4 4
Rated power (kW) 0.5 1 4
Table 15: Parameter setting of WM, DW and CD [4]
Figure 35: Room temperature band over a day
69
Each of the cases explained above, namely case 00 and case 01, are run five times. After each run the
net energy cost, the battery degradation cost, the revenue of selling electricity to the grid and the
purchase cost are noted. The battery degradation cost is calculated according to equation (28) for a
LFP (Lithium-iron phosphate ) battery. This type of battery has the following specification given in
Table 16, where LN is the battery life cycle time, Es is the energy capacity of the battery, ηTem is the
temperature dependence factor, DOD is the depth of charge and CC is the investment cost of the
battery.
LN 2000 cycles
Es 4,8 kWh
ηTem 0,9
DOD 0,8
CC € 2212
If we fill in these values in (28) we obtain an battery degradation cost of € 0.32/kWh. The selling price
of electricity to the grid determines the revenue the HEMS makes. According to [2] the selling price
of electricity should be 5% lower when supplied to the grid, compared to the buying price in every
specific instance. This ensures that the utility companies have a profit margin and the HEMS creates
more value for the installed renewable energy source, by enabling a higher penetration of renewable
energy at the residence. The simulations are done with the two-level ToU pricing scheme shown in
Figure 33. The results of the simulations are shown in Table 17. Note that these values are average
values of the five runs. Std. stands for the standard deviation between the five runs.
Case 0 Purchase cost (€)
Std.
Battery cost (€)
Std.
Sales revenue
(€)
Std.
Net cost (€)
Std.
No flexibility (case 00)
4.73 0.12 2.27 0.11 1.84 0.06 5.16 0.08
Flexibility (case 01) 4.09 0.18 2.21 0.17 1.57 0.09 4.75 0.18
From the last column in Table 17 we see that in the case where we allow some flexibility of the home
appliances the net energy cost is smaller than the case were there is no or little operation flexibility
of the home appliances (€ 0.41 difference). The purchase cost and sales revenue is larger in case 00.
This can be explained by Figure 36 and Figure 37. In these figures we see that the power flow with
the grid is a bit larger in the case of no flexibility. Note that the maximum power exchange with the
grid is limited between 3 kW for feeding power into the grid and 4 kW for extracting power from the
grid. The reason for that is because we want to limit the power exchange with the grid, such that the
own produced PV energy is used as much as possible. The battery degradation cost is a bit smaller in
the case where flexibility is allowed. The power exchange of the battery and grid for the two cases
are shown in Figure 36 and Figure 37 respectively. The lower graph on these figures depict the state
of charge of the battery as a percentage of the energy content of the battery (=4.8 kWh).
Table 16: Parameters for a Lithium-iron phosphate (LFP) battery
Table 17: Simulation results of case 0 (scheduling horizon = 24 h)
70
Figure 36: Simulation results in case of no flexibility of home appliance
Figure 37: Simulation results in case of flexibility of home appliances
71
The state of charge of the battery reaches about 90% of its capacity in both cases. In the first part of
the day (0-12 h), the battery is barely being charged, because there is no PV power available. When
the sun starts to shine (see blue and green graph on figure PGrid), the battery is being charged. Note
that at the same moment the excess PV power is feed into the grid. The maximum charging- and
discharging power of the battery is assumed to be 2 kW, as indicated by the green dotted line on
graph PBattery. On this graph we see that during PV production the battery charges with a power
smaller than 2 kW. A reason for this phenomenon is because a part of the PV power is instantaneous
consumed. The batteries state of charge (SOC) is regulated between 20% and 100%. The 20% SOC is
because the Depth of Discharge (DOD) of an LFP battery is set to 80% (DOD = 1-SOC) such that the
expected life time is not jeopardized. In order to force the algorithm to charge the battery during PV
production, we set the battery degradation cost negative. In this way the algorithm thinks it sells the
energy to the battery and makes profit. When there is no PV production, the degradation cost is set
to its normal positive value. The use of a battery degradation cost is of great importance to obtain a
profitable situation, such that the battery is being charged or discharged in a reasonable way. From
simulations without a battery degradation cost the algorithm bought electricity from the grid to
charge the battery and discharged in the next time slot to the loads, such that the net electricity cost
is reduced (see case 2). Of course this is not a profitable situation.
Figure 38: Simulation results in case of no flexibility of home appliances
72
In Figure 38 and Figure 39 the electrical power outputs of the Heat Pump (HP) and the Combined
Heat and Power unit (CHP) are depicted in the top graphs. Note that these appliances have two
working conditions, namely ON or OFF. When they are ON they work on their rated power, which is
2.7 kWe for the HP and 1.7 kWe for the CHP unit. The course of the room temperature and the hot
water temperature inside the Thermal Storage tank (TS) are shown in the lower graphs. The
temperature boundaries are depicted as green dotted lines. The first thing that stands out is that the
temperature boundaries are sometimes crossed. This is due to the fact that the HP and CHP unit are
modeled in a way where they only have two working conditions. So it is possible that at a certain
timeslot the temperature is within bounds, for example close to the upper bound. When the
algorithm decides to turn on the appliance (HP or CHP unit) it outputs heat at rated power such that
the temperature in the next timeslot exceeds the upper temperature bound. It is clear that in the
case of flexibility the CHP unit is less active than in the other case. This results of course in less fuel
consumption and thus in a lower energy cost. The influence of the HP is not very clear, because the
temperature band is much smaller, but we see that it is less active when the temperature band is
wider. A last remark that can be made is that the Genetic Algorithm used in Matlab is not very robust
for the constrained problem in this thesis. The results between different runs of the same case are
always different, which makes it hard to draw good conclusions. Nevertheless, based on the standard
deviation given in Table 17, this case (case 0) shows that allowing flexibility in the operation of the
home appliances results in a decrease of the net energy cost.
Figure 39: Simulation results in case of flexibility of home appliances
73
6.2 Case 1
In case 1 the effect of comfort and cost as an objective will be examined. As explained in the
introduction chapter cost and comfort are contradictory, which means that the cost will be larger
(smaller) if the comfort level is higher (lower). This effect will be investigated in this case for the
three level Time of Use (ToU) pricing scheme and the Real time Pricing (RTP) scheme. The RTP
scheme only contains the price of electricity per kWh, without distribution cost, taxes, etc. As
explained in chapter 1 (see Figure 5) the price on your electricity bill consists of several parts (energy,
distribution cost, transport cost, taxes,…) and energy is only a third of the total cost. So to be a bit
realistic, the RTP scheme will be added with a cost that represents the grid compensation
(=distribution cost, transport cost, taxes, etc.). This grid compensation will be calculated by taking the
maximum electricity price in the RTP scheme and divide it by a third. The maximum electricity price
of the RTP scheme is about 4 cents per kWh (see Figure 33), which results in a grid compensation
price of 12 cents per kWh. The residential user can make a tradeoff by changing the user preference
factor α. α equal to one means that the algorithm only minimizes the cost objective, whereas α equal
to zeros indicates that the algorithm will only take the comfort objective into account. In case 1 the
user preference factor α will be set to 0, 0.5 and 1, where 0.5 means that cost and comfort are
considered equally important for the user. As mentioned in the comfort objective section (5.2) of
chapter 5, the comfort level indicators are defined as a function that increases when the control
quantity, temperature for heat pump (HP) and CHP unit and starting time for the non-interruptible
appliances, deviates from the user’s set point. The margins in this case are chosen small, with a
for the HP and a
for the CHP unit. If the HP and
CHP unit control the room temperature and hot water temperature respectively within ,
with the room temperature -and hot water temperature set point respectively, the comfort level
indicator will be at its minimal value. Therefore, the smaller α, the smaller the operation limits for the
thermal appliances and thus the less flexible they are. The results of the simulations are shown in
Table 18 and Figure 40.
Three –level Time of Use pricing scheme
α Purchase
cost (€)
Std.
Battery cost (€)
Std.
Sales revenue
(€)
Std.
Net cost (€)
Std.
Comfort level
indicator
Std.
0 4,85 0.01 1,92 0.05 2,13 0.04 4,63 0.06 6,3 1.7
0.5 4,33 0.21 2,25 0.21 2 0.18 4,57 0.18 16.8 5.3
1 4,12 0.13 2 0.14 1,91 0.17 4,21 0.21 24,1 6
Real Time Pricing scheme
α Purchase
cost (€)
Std.
Battery cost (€)
Std.
Sales revenue
(€)
Std.
Net cost (€)
Std.
Comfort level
indicator
Std.
0 5.47 0.04 2.15 0.4 2.11 0.06 5.51 0.41 4.8 2.4
0.5 5 0.09 2.14 0.15 1.98 0.08 5.17 0.17 22 6.9
1 4.9 0.08 2 0.12 1.8 0.15 5.09 0.2 28.9 5.6
Table 18: Simulation results of case 1 (scheduling horizon = 24 h)
74
From Table 18 or Figure 40 we see that the net energy cost decreases and the comfort level indicator
increases with increasing α. Important to remember is that a higher comfort level indicator indicates
that the most comfortable bounds set by the user are exceeded (within a reasonable margin) and
thus that the user is less comfortable. This indicator is defined within the range of [0 100], where
zero means that the user is most comfortable and 100 indicates that the user is least comfortable. As
mentioned before the user preference factor α is a way the residential user can take a tradeoff
between cost and comfort level. A higher α means that the consumer finds cost more important than
comfort and vice-versa. The sales revenue decreases slightly with increasing α. The net cost is larger
for the case with the RTP scheme. This is probably due to the fact that the difference between the
maximum and minimum electricity price is smaller in the case for the RTP scheme than for the ToU
pricing scheme. The battery cost and sales revenue are from the same magnitude in both cases
(around € 2). This indicates that the algorithm controls the battery and exchange with the grid in a
similar way for both cases (see Figure 41 and Figure 42). The only significant difference between the
ToU -and RTP scheme is the purchase cost, which is smaller in the case with the ToU pricing scheme.
Note that the results given in Table 18 are average values of five different simulation runs. The blue
and red lines of Figure 40 are not fitting lines, but are just drawn to indicate the rising and falling
trend of the net cost and comfort level indicator respectively. The influence of the user’s preference
factor α is shown in Figure 43 (α=0) and Figure 44 (α=1). In these figures we see that the power that is
put back in the grid is smaller when α equals to one. This is also confirmed by looking to the sales
revenue, which is a bit smaller for α = 1. There is some variation between the different simulation
runs for a fixed α. This is because the genetic algorithm (GA) is stochastic, which means that it makes
random choices of the initial population it starts calculating with. Normally you can choose a fix initial
population in GA, but this options is not available for problems with discrete variables.
Figure 40: The net energy cost and comfort level as a function of the user’s preference factor α for the case with the three-level ToU pricing scheme
75
Figure 42: Power flow for case with RTP scheme for α = 0.5
Figure 41: Power flow for case with ToU pricing scheme for α = 0.5
76
Figure 43: Power flow for α=0 (ToU-3 level)
Figure 44: Power flow for α=1 (ToU-3 level)
77
6.3 Case 2
In this case we will investigate the influence of the battery degradation cost on the operation of the
optimisation model. We will do this for a user preference factor α of 1 and look how the different
parts (purchase cost and sales revenue) of the cost change. This will be done for a ToU pricing
scheme and a RTP pricing scheme, such that we can compare it with the results of the previous case.
According to [4] the battery degradation cost is a way to assure a profitable situation of the battery
performance. The results of the simulations are shown in Table 19.
Time of Use (ToU) – 3 level
Case Purchase cost
(€) Std.
σ Sales revenue
(€) Std.
σ
Without battery degradation cost
4.31 0.02 1.87 0.1
With battery degradation cost
4.12 0.13 1,91 0.17
Real Time Pricing scheme
Case Purchase cost
(€) Std.
σ Sales revenue
(€) Std.
σ
Without battery degradation cost
4.83 0.09 1.83 0.1
With battery degradation cost
4.9 0.08 1.8 0.15
If the purchase cost in this case is compared with the previous case were a battery degradation cost
is foreseen, a smaller purchase cost is obtained in the latter one. The difference is 0.19 euro (€ 4.31-
€ 4.12) for the case with the ToU pricing scheme. The difference is small, but nevertheless present.
The difference in purchase cost is less pronounced for the RTP scheme (€ 4.83 ↔ € 4.9). Based on
the standard deviation we can say that the influence of the battery degradation cost is negligible for
the case with the RTP scheme. The sales revenue are of the same order in both cases. The effect of
the battery degradation cost for the case with the ToU pricing scheme can be seen on Figure 45 and
Figure 46. On the bottom graph of Figure 45 (the State of Charge) we see that at the first half of the
day (0-12 h) the battery charges and discharges continuously, while in the case with a degradation
cost (bottom graph Figure 46) the battery shows less charging and discharging cycles. These charging
and discharging cycles are not good for the batteries lifetime and need to be avoided. So we can
conclude that the implementation of a battery degradation cost results in a more profitable situation
with less charging and discharging cycles of the battery. The model shows a different behaviour for
the RTP scheme as can be seen on Figure 47. When we do not foresee a degradation cost the battery
charges in the night during the low price period and discharges when the price rises again. This
behaviour is clearly visible from 0-10 h on Figure 47. We do not see this behaviour when a battery
degradation cost is present in the model (see Figure 48).
Table 19: Simulation results of case 2 (scheduling horizon = 24 h)
78
Figure 46: Power flow for case with battery degradation cost (ToU-3 level)
Figure 45: Power flow for case without battery degradation cost (ToU-3 level)
79
Figure 48: Power flow for case with battery degradation cost (RTP scheme)
Figure 47: Power flow for case without battery degradation cost (RTP scheme)
80
Chapter 7
Conclusions
To study the behaviour of different home appliances and their flexibility a Home Energy
Management System (HEMS) is proposed in this thesis. It proposes a framework of HEMS including a
grid, PV, a thermal storage tank and a home energy storage battery. A multi-objective optimisation
algorithm for HEMS is proposed, which minimizes electricity cost and maximizes the comfort of the
residential user simultaneously. The algorithm controls the operation of schedulable home
appliances, such as a washing machine (WM), a dish washer (DW) and a cloth dryer (CD). It also
controls two heating appliances, namely a heat pump (HP) and a combined heat and power (CHP)
unit along with the power distribution among the grid and the home battery. These appliances are
controlled according to the electricity price, the price of natural gas (NG), forecasted power output of
the PV, forecasted outdoor temperature and user preferences. The residential user has the ability of
selling electricity to the grid for revenue. In order to quantify the user’s comfort level, a set of
comfort level indicators proposed by [4] are used. These comfort level indicators are proposed based
on the home appliance’s characteristics and user’s preferences. At the begin of this present work a
number of research questions were asked to investigate the influence of some parameters. These
questions are repeated here for the convenience of the reader (see section 2.5).
The topics that will be investigated in this thesis are the following ones:
1. What is the effect on the energy cost when we allow flexibility to some house hold
appliances?
2. What is the effect of the cost objective on the comfort objective and vice-versa?
3. What is the effect of different pricing schemes, Time of Use (ToU) and Real time Pricing
(RTP), on the performance of the optimisation algorithm?
4. What is the effect of the battery degradation cost on the charging and discharging behaviour
of the battery?
The first question is investigated in case 0. From this case it became clear that allowing flexibility to
the home appliances results in a lower energy cost. This case is carried out with a two-level ToU
pricing scheme that has an off peak –and standard rate price. The difference between allowing and
not allowing flexibility results in a net energy cost gain of € 0.41 (see Table 17). This is of the same
magnitude that was obtained in [4].
The second and third question are studied in case 1. The user’s preference α is a way for the
residential user to take a tradeoff between cost and comfort level. The factor α is set to 0, 0.5 and 1,
where zero means that only the comfort level of the user is considered, one indicates that only the
cost objective is minimized and 0.5 means that cost and comfort are considered equally. The model is
run for the three values of α. The results confirm the conclusion given in [4], namely that the cost –
and comfort objective are contradictory. This means that maximising the comfort results in a higher
cost and vice-versa. The algorithm behaves pretty much the same for a three-level ToU -or RTP
81
scheme if we look at the power distribution among grid and battery. There is a slight difference in
the energy cost, where the algorithm obtains a smaller cost for the ToU scheme. This effect can be
explained by the fact that the price difference between min. and max. is larger in the ToU scheme,
which means that the algorithm has the opportunity to buy cheap electricity during the off peak
period.
In the last case we examined the effect of the battery degradation cost on the performance of the
algorithm. This is carried out for two pricing schemes, namely three-level ToU -and RTP scheme. For
the ToU pricing scheme we saw that the presence of a battery degradation cost resulted in less
charging and discharging cycles of the battery (see Figure 45-Figure 46), which is of course a
profitable situation w.r.t. the batteries lifetime. The presence of a degradation cost also resulted in a
smaller purchase cost, and thus a smaller net energy cost. The algorithm shows a completely
different behaviour for the RTP scheme. With the presence of a degradation cost the battery barely
charges in the first part of the day (see Figure 48, 0-10 h) and even shows small charging –and
discharging cycles. When the degradation cost is omitted the power distribution of the battery
changes. During the first part of the day (0-10 h) we now see that the battery charges when the
electricity price is low and discharges to the loads when the price rises again (see Figure 47, 0- 10 h).
Because of the small difference between the low price -and high price period, the effect on the
energy cost is small. As can be seen on Table 19 the average purchase cost is € 4.83 when there is no
degradation cost and € 4.9 when a degradation cost is presence. Note that these are average values
of five different runs, where in one of the simulation runs without a degradation cost the purchase
cost was € 4.73. This again confirms that the genetic algorithm is not robust for the type of
optimisation problem we are solving here.
To end with the next section discusses the vision of Eandis about the transition to green energy.
Eandis has published a white paper called ‘The Journey to Green Energy’ [74] where they tell their
vision about the energy transition. Eandis is the largest Distribution System Operator (DSO) of
Belgium, so it is interesting to know how they will tackle the problem of a major change in our energy
production and thus distribution.
82
7.1 Final Remarks
The motive of adapting our energy generation towards green is the worldwide concern of global
warming. Adapting our energy generation towards green is one way to deal with the global warming
problem, but changing our consumption behaviour is a crucial step in this transition. According to
Eandis the transition towards a ‘’green’’ landscape is a gradual process where the biggest challenge
will be to act in the right way at the right time to facilitate this transition. ‘’Flexibility’’ in consumption
and production will play an important role, but adding them into our energy landscape will only be
possible if the right policies, control systems and market mechanisms evolve at the same pace [74].
The journey towards green energy will result mostly in decentralized electricity generation, which
means that the solution also has to be found in a decentralized manner. Every element (demand,
generation and storage) need to participate to keep the system balanced. Nowadays “ demand drives
generation”, but as we want to evolve towards a green energy landscape the shift towards
“generation drives demand” must be introduced. There are many solutions to achieve this and new
solutions will emerge, but focusing on only one solution is not a good idea. The impact of industrial
consumers is of course larger than in the residential domain, but Eandis believes that the residential
domain will be needed in the future. For now the focus will be on the industrial consumers while
preparing for the change in behaviour of residential customers. One of the reason for this, is because
certain pilot projects such as Address, Ecogrid and Linear have shown that the financial benefits for
residential users are currently very low. Eandis believes that a socio-cultural change will be
necessary, but hopes that the necessary technology (storage and load shifting automation) will allow
to achieve the necessary flexibility with minimized impact. A way to achieve this demand-shifting is
via financial signals, for example dynamic pricing schemes, or controlled by a technical signal that is
centrally or locally created based on the situation of the market or network [74].
The change towards a green energy production does not imply that we should only allow renewable
generation in our network, because we will still need for example the gas fired plants filling the gaps
between generation and demand. This until other technology, e.g. storage, can take over. As a result
of the intermittency of renewable generation we need to provide additional flexible power to fill the
gap between the available renewable generation and the demand. This fact makes renewable energy
such as wind and solar expensive, because the cost of producing green energy is the sum of the wind
turbines and solar panels + the cost of the necessary balancing power to maintain the network
balanced. If we look at Figure 49 we see that the integration cost of wind energy is as large as the
generation cost of it. One of the effects of decentralized generation is that the energy will circulate in
two directions in the network. The first challenge is controlling the voltage and guaranteeing the
security of the network with a highly bi-directional load flow. The second challenge will be to bring
the generation electrically as close as possible to the demand. This will reduce the amount of
investment required and the losses in the network. Working on the right network architecture is
crucial to realize a financial benefit, reduce network losses and allow green energy to be more
present in the network. The way to do this is to maximize the integration of local generation on
demand feeders. This will allow power to flow through the shortest way to the consumer and results
in fewer losses and voltage problems [74].
83
Tariff schemes are a way to handle the “pull” type of demand response by just sending new pricing
tariffs and hope the consumer will react [74]. The most dynamic model used is to communicate the
electricity prices day-ahead to customers. The danger of pricing/tariff schemes is an overreaction
that results in a heavy unbalance or congestion problem, as there is not always a feedback loop. Each
control model will only converge if you adjust your actions based on the obtained results. This means
that based on the situation of the network, you should be able to adjust the tariff schemes
dynamically. Several pilot projects, such as Linear and Ecogrid, showed that the financial result of
residential demand response is so low nowadays that the business opportunity is still negative.
However Eandis believes that the journey to green will continue and at some point the residential
demand will need to contribute to balance the system. Think of the increasing electrification of
house hold appliances, such as a heat pump and an electrical vehicle [74].
The start of using balancing power on distribution networks will need a great number of participants,
because of the smaller impact than large industrial consumers. This implies that an aggregator will
have to contract a lot more customers to ensure the continuous availability of flexibility resources he
is selling to a Transmission System Operator (TSO) like Elia for example. As the number of clients
increase it is difficult to have a deterministic view on what is happening on the network when the
aggregator acts for balancing purposes. Residential and smaller consumers will at some point in the
future start to participate in the market. However it will not be possible to determine their actions in
a deterministic way nor control them directly. Reactions from the market will come in a stochastic
base, not only in volume but also in reaction time. Control systems will have to act in a total different
way, where central control will have to make room for local automation systems that react to local
signals, perhaps together with global broadcasted information. This will become more important and
even crucial [74].
Figure 49: Levelized Cost Of Energy (LCOE) for wind turbines [74]
84
When looking to an energy problem we have to consider three types of efficiency that compose the
global efficiency. These three types of efficiencies are the energy efficiency, the financial efficiency
and the green efficiency. In most cases these efficiencies are contradictory, which means that it is
hard to focus on one of them without restricting the other two (see Figure 50). The first type of
efficiency, namely energy efficiency, focuses on using as little energy as possible. For example to
maximize the efficiency of heating -or refrigeration processes, it is a matter of finding the right
balance between adding energy to and keep energy inside your process. This does not necessary
mean consuming when there is excess of green energy or during low price periods. To obtain a
maximal financial efficiency we need to consume electricity at the lowest possible cost. This does not
imply that consuming during low price periods that the consumption will be less. When we wait
longer, for example until a low price period, before turning on heating or cooling systems, this will
probably result in a lower energy efficiency because the system will have consumed more before
reaching the desired temperature. Finally green efficiency will demand to maximize the use of green
energy such that the amount of CO2 emissions reduce. Trying to maximize green energy will need
some kind of storage and balancing power to keep the system balanced. As known this costs a lot of
money, which will result in a lower financial efficiency [74].
We will end with some interesting features related to demand response. On low voltage feeders a
congestion normally leads to a voltage problem. Too much local production will lead to over-voltage
and too much demand results in an under-voltage. Acting based on a local voltage measurement may
be a solution for congestion problems. Controllers can start appliances that absorb the excess of
produced energy, which otherwise should be curtailed. This not only solves the voltage problem, but
also increases the energy efficiency. Another example can be an electrical vehicle charger that
reduce its current based on the voltage drop it causes. Besides of maintaining the voltage level, also
the frequency is an important parameter for maintaining a balanced network. The use of local
frequency relays on heavy consuming appliances, such as a water boiler and heat pump, can resolve
frequency problems. If the trigger to shut off is set to a frequency just above the trigger for
frequency containment reserves (49.8 Hz), this will cut off a large part of the demand that may
prevent a further drop or at least help limit the impact of the frequency drop. The remaining problem
is to decide how long they have to be cut off and how we need to start them up again. The restart
may cause a problem, because when for example all boilers are restarted at the same moment this
may cause another frequency drop. A sequential start up can be a solution [74].
Figure 50: Triangle of efficiency [74]
85
References
[1] Y.Huan, H.Tian,L.Wang., "Demand response for home energy management system.," in Electrical
Power and Energy Systems., 2015, pp. 448-455.
[2] Christos S. Ioakimidis, Luis J. Oliveira,Konstantinos N. Genikomsakis, Panagiotis I.Dallas.., "Design,
architecture and implementation of a residential energy box management tool in a smartGrid,"
Energy, vol. 2013, no. 3 july 2014, pp. 1-15, 2014.
[3] G. Graditi, M.G.Ippolito, R.Lamedica, A.Piccolo, A.Ruvio, P.Siano, G.Zizzo, "Innovative control logics
for a rational utilization of electrical loads and air-conditioning systems in a residential building.,"
Energy and Buildings, vol. 2014, no. 22 May 2015, pp. 1-17, 2015.
[4] Y.Zhang, P.Zeng,S.Li, C.Zang and H.Li., "A Novel Multiobjective Optimisation Algorithm for Home
Energy Management System in Smart Grid," Hindawi, no. 18 January 2015, p. 19, 2015.
[5] European commission- Europe, "GuideLinesSpaceWaterheaters," 2013.
[6] Ioakimidis CS, Oliveira LJ, Genikomsakis KN., "Wind power forecasting in a residential location as
part of the energy box management decision tool," IEEE Trans Ind Informat.
[7] Y.Liu, C. Yuen, R.Yu, Y.Zhang, S. Xie., "Queuing-Based Energy Consumption Management for
Heterogeneous Residential Demands in Smart Grid.," IEEE, no. May 4,2015, p. 10, 2015.
[8] M. Beaudin, H.Zareipour, "Home energy management system: A review of modelling and
complexity.," in Renewable and Sustainable Energy Reviews., 2015, pp. 318-335.
[9] Livengood D , Larson R, "The energy box: locally automated optimal control of residential electricity
usage," Serv Sci, pp. 1-16, 2009.
[10] Molderink A, Bakker V, Bosman M, Hurink J, Smit G., "Management and control of domestic smart
grid technology," IEEE Trans Smart Grid, 2010.
[11] Clastres C, Pham T, Wurtz F, Bacha S.," Ancillary services and optimal household energy
management with photovoltaic production," 2010, pp. 55–64.
[12] Pham T, Clastres C, Wurtz F, Bacha S, Zamai E, "Optimal household energy management and
economic analysis: from sizing to operation scheduling ," Adv Appl Mech Eng Technol, 2008.
[13] Kim TT, Poor HV, "Scheduling power consumption with price uncertainty ," IEEE Trans Smart Grid ,
2011, pp. 519–27.
86
[14] Barbato A, Capone A, Chen Lin, Martignon F, Paris S, "A power scheduling game for reducing the
peak demand of residential users, " IEEE online conference on green communications (GreenCom),
October 2013, pp. 137–42.
[15] Galus MD, Andersson G., "Demand management of grid connected plug-in hybrid electric vehicles
(phev)," ENERGY , pp. 1–8, November 2008.
[16] Ha D, de Lamotte F, Huynh Q,.," Real-time dynamic multilevel optimization for demand-side load
management, " IEEE international conference on industrial engineering and engineering
management, December 2007, pp. 945–9.
[17] Adika CO, Wang Lingfeng," Autonomous appliance scheduling for household energy management,"
IEEE Trans Smart Grid, March 2014, pp. 673–82.
[18] Matallanas E, Castillo-Cagigal M, Gutierrez A, Monasterio-Huelin F,Caamano-Martin E, Masa D, et
al., "Neural network controller for active demand-side management with PV energy in the
residential sector," Appl Energy, 2012, pp.90-7.
[19] Junghoon Lee, Hye-Jin Kim, Gyung-Leen Park, Mikyung Kang.," Energy consumption scheduler for
demand response systems in the smart grid, " J Inf Sci Eng, 2012, pp. 955–69.
[20] Yao L, Chang W, Yen R, "An iterative deepening genetic algorithm for scheduling of direct load
control," IEEE Trans Power Syst , 2005, pp. 1414–21.
[21] Pedrasa M, Spooner T, MacGill I., "The value of accurate forecasts and a probabilistic method for
robust scheduling of residential distributed energy resources," in In: PMAPS 2010., 2010.
[22] Ilic M, Black J, Watz J., "Potential benefits of implementing load control," in In: Power engineering
society winter meeting. New York: IEEE, vol. 1, 2002, pp. 177–82.
[23] Ha D, Ploix S, Zamai E, Jacomino M. Tabu," search for the optimization of household energy
consumption "IEEE international conference on information reuse and integration, 2006, pp. 86–
92.
[24] Yu Zhe, Jia Liyan, Murphy-Hoye MC, Pratt A, Piccioli EG, Tong Lang," Modeling and stochastic
control for home energy management, " IEEE Trans Smart Grid, December 2013, pp. 2244–55.
[25] Beaudin M, Kiani A, Zareipour H, Schellenberg A," Residential energy management using a two-
horizon algorithm, " IEEE Trans Smart Grid , 2013.
[26] Dehnad A, Shakouri H, "A novel model of intelligent electrical load management by goal
programming for smart houses, respecting consumer preferences," in Energy Power Eng 2013., pp.
5(10):622–7.
[27] Molderink A, Bakker V, Bosman M, Hurink J, Smit J.," Domestic Energy management methodology
for optimizing efficiency in smart grids, ": In: IEEE conference on power technology, 2009.
87
[28] Hassan Naveed Ul, Pasha Muhammad Adeel, Yuen Chau, Huang Shisheng,Wang Xiumin, "Impact of
schedulingflexibility on demand profile flatness and user inconvenience in residential smart grid
system.," in Energies., 2013, pp. 6608–35.
[29] Bozchalui MC, Hashmi SA, Hassen H, Canizares CA, Bhattacharya K., "Optimal operation of
residential energy hubs in smart grids, " IEEE Trans Smart Grid, December 2012, pp. 1755–66.
[30] Sianaki O, Masoum M., "A multi-agent intelligent decision making support system for home energy
management in smart grid: a fuzzy topsis approach," in Multiagent Grid Syst. 181–95,
2013;9(January (3)).
[31] Conejo AJ, Morales JM, Baringo L, "Real-time demand response model," Trans Smart Grid,
2010;1(December (3)), pp. 236–42.
[32] Mohsenian-Rad A, Wong V, Jatskevich J, Schober R, "Optimal and autonomous incentive-based
energy consumption scheduling algorithm for smart grid.," in In: Innovative smart grid technologies
(ISGT)., January 2010., pp. 1-6.
[33] Pedrasa M, Spooner T, MacGill I.," Coordinated scheduling of residentia distributed energy
resources to optimize smart home energy services, " IEEE Trans Smart Grid, 2010;1(September (2)),
pp. 134–43.
[34] Lu Ning, Chassin DP, "A state queueing model of thermostatically controlled appliances," In:
Proceedings of IEEE PES power systems conference and exposition, 2004.
[35] Misra S, Mondal A, Banik S, Khatua M, Bera S, Obaidat MS, "Residential energy management in
smart grid: a Markov decision process-based approach," Green computing and communications
(GreenCom)., pp. p.1152–7., August 2013, IEEE and internet of things (iThings/CPSCom). IEEE
international conference on and IEEE cyberphysical and social computing.
[36] Dieter Fiems, Joris Walraevens, Queuing analysis and Simulation, 2015.
[37] Dr Maurizio A. Spirito. (2011) GreenCom myGrid; Energy Efficient and Interoperable Smart Energy
Systems for Local Communities, Project Coordinator : Istituto Superiore Mario Boella ( ISMB) Italy,
European Commission FP7 ICT-2011.6.1 Smart Energy Grids.
[38] Liam Moore. (28-03-2013, Version 0.5) D5.1 Home Appliance, Energy Generation and Storage
Analysis Review Report.
[39] Ronnie Belmans, Bart Beusen, Bart Boesmans, Wim Cardinaels, Bert Claessens, Sven Claessens, Paul
Coomans, Reinhilde D'hulst, Wim De Meyer, Jan Degraeve, Chris Develder, Bemjamin Dupont, Wim
Foubert, Jan Gordebeke, Felix Hoornaert, Sandro lacovella, "Demand response for Families," Genk,
2014.
[40] EDSO for smart grids, "Adapting distribution network tariffs to a decentralised energy future,"
September 2015.
88
[41] Dr. S.C.Breukers, Dr. R.M.Mourik, "The end-users as starting point for designing dynamic pricing
approaches to change household energy consumption behaviours," Report for Netbeheer
Nederland, Projectgroep Smart Grids (Pg SG). March 2013.
[42] Stromback, J., Dromacque, C., Yassin, M.H., "The potential of smart meter enabled programs to
increase energy and systems efficiency: a mass pilot comparison," Helsinki: VaasaETT, Global
Energy Think Tank., 2011.
[43] Frontier Economics & Sustainability First, "Demand Side Response in the domestic sector - a
literature review of major trials," August 2012, Undertaken by Frontier Economics and
Sustainability First, for the UK Department of Energy and Climate Change.
[44] Commissie voor de Regulering van de Elektriciteit en het Gas (CREG)., "de haalbaarheid van de
invoering van een progressieve prijszetting van elektriciteit in België," Brussel, 2010.
[45] Accenture, "Engaging the New Energy Consumer, Accenture perspective- operational imperatives
for energy efficiency.," 2010.
[46] S. Darby, "The Effectiveness of Feedback on Energy Consumption. A Review for Defra of the
literature on metering, billing and direct displays," 2006.
[47] Mourik, R.M., "Zonder slimme meter geen effectieve energiebesparing, maar de slimme meter
alleen is niet genoeg - Een desk research naar de Effectiviteit van Energiegerelateerde feedback
met of zonder slimme meter - Particulieren en kleinzakelijke doelgroep ," Liander, Mei 2011.
[48] Faruqui, A.; Sergici, S. , "Household response to dynamic pricing of electricity- a survey of the
experimental evidence. ," 2009.
[49] Thorsnes, F. Williams, J., Lawson, R, "Consumer responses to time varying prices for electricity, "
Energy Policy 49, 2012, pp. 552–561.
[50] Lemcko, UGent,iwT and KU Leuven, "Praktische gids voor het implementeren van PV-
batterijsystemen," Mei 2015.
[51] victron energy. (2016, March) 12,8 Volt Lithium-Iron-Phosphate Batteries. datasheet. [Online],
'Accessed at (March 20,2016) via' https://www.victronenergy.com/upload/documents/Datasheet-
12,8-Volt-lithium-iron-phosphate-batteries-EN.pdf'
[52] Powerstream. (2015, November) NiMH Battery Charging Basics. [Online], 'Accessed at (March 20,
2016) via' http://www.powerstream.com/NiB.htm
[53] (2016, March) Iron Edison. [Online], 'Accessed at (March 20, 2016) via'
http://ironedison.com/images/products/Iron%20Edison/Import%20Customer%20Price%20List%20-
%20Iron%20Edison%202014.pdf
89
[54] Encell technology. (2016, March) Iron Edison. [Online], 'Accessed at (March 20, 2016) via'
https://ironedison.com/images/products/Encell/EncellDataSheet100.pdf
[55] P. Manisa,K.Murat,R.Saifur,T.Yonael, "Load profiles of selected major household appliances and
their demand response oppurtunities," IEEE, 15 october 2015.
[56] brussels instituut voor milieubeheer, "FOTOVOLTAÏSCHE ZONNE-ENERGIE FACTOREN DIE DE
PRODUCTIE BEÏNVLOEDEN ," November 2010.
[57] Jeremy Harrison. (2014, 1 December) Micro combined heat and power. [Online], 'Accessed at (April
16,2016) via' http://www.microchap.info/INDEX.HTM
[58] SONJA KALLIO, "MODELLING OF A COMBINED HEAT AND POWER SYSTEM," Master dissertation, 5th
August 2012.
[59] V. Dorer, Andreas Weber, "Energy and CO2 emissions performance assessment of residential
micro-generation systems with dynamic whole-building simulation programs," 2009.
[60] Heejin Cho *, Riasat Sarwar, Pedro J. Mago, Rogelio Luck, "Design and feasibility study of combined
heat and power systems integrated with heat pump," in Applied Thermal Engineering. Mississippi
State University: Elsevier, December 2015, pp. 155-165.
[61] Philip Fairey,Danny Parker, "A review of hot water draw profiles used in performance analysis of
residential domestic hot water systems," Florida Solar Energy Center, 2004.
[62] "Measurement of Domestic Hot Water Consumption in Dwellings," Department for Environment,
Food and Rural Affairs (Defra), 2008.
[63] "Ontwerp en dimensionering van centrale-verwarmings installaties met warm water (H4),"
Wetenschappelijk en Technisch Centrum voor het Bouwbedrijf, 14, April, 2013.
[64] "Cooling load calculation of a single family house using CLTD/GLF method," ASHRAE Fundamentals
2001 Chapter 28.
[65] Jeffrey Driver, Scott R. Baker, David McCallum, Residential Exposure Assessment: A Sourcebook.:
Springer Science & Business Media, 6 dec. 2012.
[66] K. Elamari, L.A.C. Lopes, R. Tonkoski, "Using Electric Water Heaters (EWHs) for Power Balancing and
Frequency Control in PV-Diesel Hybrid Mini-Grids," Department of Electrical and Computer
Engineering, Concordia University, Montreal, Canada, 2011.
[67] S. Shao, M. Pipattanasomporn and S. Rahman, "Development of Physical-Based Demand Response-
Enabled Residential Load Models," vol. 28,no.2, pp. 607–614, 2013.
[68] The MathWorks, Inc., How the Genetic Algorithm Works, © 1994-2014, Help function in matlab.
90
[69] Chengke Zhou, Member, IEEE, Kejun Qian, Malcolm Allan, and Wenjun Zhou, "Modeling of the Cost
of EV Battery Wear Due to V2G Application in Power Systems," IEEE TRANSACTIONS ON ENERGY
CONVERSION, vol. 26, no.4, pp. 1041-1050, December 2011.
[70] (2016, March) Belpex. [Online], 'Accessed at (March 24,2016) via https://www.belpex.be/
[71]
Mohamed Abdussalam Bregaw, "Modeling, Simulation and Optimisation of Residential and
Commercial Energy Systems," Dalhousie University, Halifax, Nova Scotia, master's thesis
August,2013.
[72] Barry Rand," Plasmon Resonance for IMproving the Absorption of solar cells (PRIMA)," Deliverable
1.2 – Report on the evaluation of the potential broadband device performance based on scattering
and near field effects, Increase, p.12, 2009.
[73] (2016, March) AccuWeather.com. [Online], 'Accessed at (March 22,2016) via
http://www.accuweather.com/nl/be/ghent/30438/hourly-weather-forecast/30438
[74] Smart Grid program team of Eandis managed by Patrick Reyniers, "The Journey to Green Energy,"
p. 49, November 2013.
91
Appendix
Matlab code tic
clc;
clear all;
%%
%%%%%%%%%%%%%%%%%%
% Initialisation %
%%%%%%%%%%%%%%%%%%
disp('Initialisation...');
%% Parameters input
type='single'; % single objective optimisation
timesteps = 96; % # of timesteps
alpha = 1; % users's preference factor -> alpha*Fcost+(1-alpha)*Fcomfort
N_slot = 96; % 96 timeslots of 15 min
% dpunitstotal = 4; % Total number of controllable appliances:(1)Battery % (2) Grid (3) Heat Pump (HP) % (4) Combined Heat
and Power Unit (CHPU) % (5) Washing machine (WM) % (6) Dishwasher (DW) %(7) Cloth dryer (CD)
dpname = ['battery,','grid,','HVAC,','CHPU,','WM','DW','CD'];
dpinclude = [1,2,3,4,5,6,7]; % Appliances that are included
dpunits = length(dpinclude); % effective number of appliances
ngen = dpunits * timesteps; % number of variables
92
%battery
nu_bat = 0.90*0.97; %efficiency battery*efficiency inverter
dt = 0.25; % 15 min = 1/4 h
Emax = 4.8*4*dt; % maximum storage capacity of battery 4.8 kWh
E0 = 0.2*Emax; % minimum energy content of battery
E_init = 0.2*Emax; % energy in battery @ t=0
P_max = 2; %Max. charging power of battery [kW]
P_min = 2; %Max. discharging power of battery [kW]
L_N = 2000; % cycle lifetime of battery [# cycles]
DOD = 0.8; %Depth of discharge of battery = max. amount of energy the battery can deliver as a percentage of Emax
n_Tem = 0.9; %actual lifetime/reference lifetime
C_c = 2212; %battery capital cost [euro]
C_d = C_c/(L_N*Emax*n_Tem*DOD); %battery degradation cost [euro/kWh]
data = load('DATA'); % loading PV -and load profile, pricing schemes
% prices
%daprce = data.RTP(1:timesteps) + max(data.RTP(1:timesteps))*3;
daprce = data.ToU_2(1:timesteps); % Buying price of electricity [euro/kWh]
revenue = 0.95*daprce; % Selling price of electricity to the grid [euro/kWh]
NG = data.NG_max(1:timesteps); %Buying price scheme of natural gas [euro/kWh]
degcost = C_d*[ones(40,1);-ones(28,1);ones(28,1)]; % Degradation cost scheme of battery [euro/kWh]
% Input
P_PV = data.P_PV(1:timesteps)./max(data.P_PV)*4.5; %PV production [kW];4.5 kWp
P_cl = data.P_critic1(1:timesteps); % Critical load profile [kW]
Ft = data.Ft(1:timesteps); %hot water usage profile [l/h];
T_outdoor = data.T_outdoor(1:timesteps); % outdoor temperature in deg. C
y=load('case01'); %Loading results of previous simulation run
T_room = y.T_room; %Room temperature used to calculate comfort cost function of HP
T_w = y.T_water; %Hot water temperature used to calculate comfort cost function of Thermal storage tank
93
% limits
lb1 = [-Emax -5 0 0 1 1 1]; lb2 = [-P_min -3 0 0 0 0 0]; %lower bound limits
ub1 = [10 10 1 1 96 96 96]; ub2 = [P_max 4 1 1 0 0 0]; %upper bound limits
lb = horzcat(lb1,Horzcat(timesteps-1,lb2)); %makes lower bound row of length 7*timesteps
ub = horzcat(ub1,Horzcat(timesteps-1,ub2)); %makes upper bound row of length 7*timesteps
%CHPU
P_CHPU = 1.7; %rated power of CHPU [kW]
nu_e = 0.247; % electrical efficiency of CHPU
nu_HX = 0.657; % thermal efficiency of heat recovery system
nu_CHPU = nu_e/(1-nu_e)*1/nu_HX; %P_CHPU = nu_CHPU*Q_CHPU
T_TS_max = (60-2.5)*(alpha==0)+(60-1.25)*(alpha==0.5)+60*(alpha==1); % max. temp. of thermal storage
T_TS_min = (50+2.5)*(alpha==0)+(50+3.75)*(alpha==0.5)+50*(alpha==1); % min. temp. of thermal storage
T_env = 15; % environment temperature of thermal store
T_in = 16; % inlet temperature of water
G_TS = 0.50588*1.754; %ratio of the surface area of tank to thermal resistance of the tank U_TS*A_TS [W/K]
C = 150*4186/3600; %equivalent thermal mass m_TS*c_p [Wh/K]
Bt = Ft*4186/3600; %[W/K]
Rt = (Bt+G_TS).^-1;%[K/W]
Q_CHPU = 1/nu_CHPU*P_CHPU*1000; % Thermal power of CHPU [W]
T_set_w = 55;% user set point of hot water temperature
dT_L = 2.5;dT_U=2.5;% Temperature band of thermal store
Tw_m = T_TS_min;Tw_mm = T_TS_max;
tau = Rt*C;
G = G_TS;
dT_m = max(T_set_w-Tw_m,Tw_mm-T_set_w);
f_t = exp(-dt*tau.^-1);
T_n = T_in;
94
%HP
P_HVAC = 2.7; % rated power of HP [kW]
COP = 2;%COP of HP
Q_HVAC = COP*P_HVAC*1000;
T_room_mm = 21;%max room temperature [C]
T_room_m = 19; % min room temperature in [C]
T_set_r = 20; % user set point of room temperature [C]
T_room_max = ([(T_set_r+2)*ones(24,1);
(T_set_r+0.5)*ones(8,1);(T_set_r+2)*ones(36,1);(T_set_r+0.5)*ones(20,1);(T_set_r+2)*ones(8,1)])*(alpha==1)+
(T_room_mm*ones(t mesteps,1))*(alpha==0)+((T_room_mm+0.5)*ones(timesteps,1))*(alpha==0.5); % Upper bound of room temperature
T_room_min = ([(T_set_r-2)*ones(24,1); (T_set_r-0.5)*ones(8,1);(T_set_r-2)*ones(36,1);(T_set_r-0.5)*ones(20,1);
(T_set_r-2)*ones(8,1)])*(alpha==1)+(T_room_m*ones(timesteps,1))*(alpha==0)+((T_room_m-0.5)*ones(timesteps,1))*(alpha==0.5); %
Lower bound of room temperature
DT_m = max(T_set_r-T_room_m,T_room_mm-T_set_r);
DT_L = 1;DT_U=1; % Temperature band of HP [C]
G0 = 195.7; %Gt = G0*(Tt_outdoor-Tt_room)-> heat gain rate of the house [Wh/h.K]
dc = 4*216; %energy to change temp of air by 1 degree Celsius [Wh/K]
T_room_init = 20; %initial room temperature [C]
Gt = G0*(T_outdoor-T_room); % heat gain rate of the house [Wh/h]
%WM
N_task_WM = 4; % # of time slots needed to finish job
N_max_WM = 76; % Time slot at which job needs to be finished
N_min_WM = 37;% Time slot at which job may start WM
P_WM = 0.5; % rated power of WM
P =
[daprce(1:40);daprce(41)/2;daprce(42)/4;zeros(20,1);daprce(63)*0.2;daprce(64)*0.4;daprce(65)*0.6;daprce(66)*0.8;daprce(67:96]
;% Electricity price->made zero during PV production
95
f1 = P_WM*P; % Electricity cost function for WM
Aeq1 = [zeros(1,N_min_WM),ones(1,N_max_WM-N_task_WM-N_min_WM),zeros(1,timesteps-(N_min_WM+N_max_WM-N_task_WM-N_min_WM))];
beq1 = N_task_WM;
LB = zeros(timesteps,1);
UB = ones(timesteps,1);
U_WM = intlinprog(f1,timesteps,[],[],Aeq1,beq1,LB,UB);% Determine optimal starting time WM
N_start_WM = find(U_WM,1,'first')*(alpha==1)+48*(alpha==0); %Starting time of WM
%DW
N_task_DW = 4; % # of time slots needed to finish job
N_max_DW = 84;% Time slot at which job needs to be finished
N_min_DW = 37;% Time slot at which job may start DW
P_DW = 1; % Rated power of DW
f2 = P_DW*P;% Electricity cost function for DW
Aeq2= [zeros(1,N_min_DW),ones(1,N_max_DW-N_task_DW-N_min_DW),zeros(1,timesteps-(N_min_DW+N_max_DW-N_task_DW-N_min_DW))];
beq2 = N_task_DW;
U_DW = intlinprog(f2,timesteps,[],[],Aeq2,beq2,LB,UB);% Determine optimal starting time DW
N_start_DW = find(U_DW,1,'first')*(alpha==1)+56*(alpha==0);%Starting time of DW
%CD
N_task_CD = 8; % # of time slots needed to finish job
N_max_CD = 93;% Time slot at which job needs to be finished
N_min_CD = 77;% Time slot at which job may start CD
P_CD = 4; % Rated power of CD
f3 = P_CD*P;% Electricity cost function for CD
Aeq3= [zeros(1,N_min_CD),ones(1,N_max_CD-N_task_CD-N_min_CD),zeros(1,timesteps-(N_min_CD+N_max_CD-N_task_CD-N_min_CD))];
beq3 = N_task_CD;
96
U_CD = intlinprog(f3,timesteps,[],[],Aeq3,beq3,LB,UB);% Determine optimal starting time CD
N_start_CD = find(U_CD,1,'first')*(alpha==1)+80*(alpha==0);%Starting time of CD
%% Construction of cost- and comfort functions
DPcostfunctions=DPcostfunction(dpinclude);
Degcost = degcostfcn(dpinclude);
Rev_cost = revenuecostfcn(dpinclude);
Comfortfunctions = Comfortfunction(dpinclude);
NonIntfcn = NonIntcomfortfcn(dpinclude);
HP_comfort = HP_comfortfcn(dpinclude);
CHP_comfort = CHP_comfortfcn(dpinclude);
[Fcost,Fcomfort]=Parametratie(timesteps,dpunits,DPcostfunctions,Comfortfunctions);
F_deg = Parametratie_bat(timesteps,dpunits,Degcost);
F_revenue = Parametratie_bat(timesteps,dpunits,Rev_cost);
F_nonIntcomfort = Parametratie_comfort(timesteps,dpunits,NonIntfcn);
F_HP = Parametratie_comfort(timesteps,dpunits,HP_comfort);
F_CHP = Parametratie_comfort(timesteps,dpunits,CHP_comfort);
Ydeg = str2func(strcat('@(x,degcost)',F_deg));
Yrev = str2func(strcat('@(x,revenue)',F_revenue));
Ycost=str2func(strcat('@(x,revenue,daprce,degcost,NG,nu_e,P_CHPU)',Fcost));
Ycomfort=str2func(strcat('@(x,T_set_r,T_room_m,T_room_mm,DT_m,DT_L,DT_U,T_room,Gt,dc,Q_HVAC,N_slot,T_set_w,dT_L,dT_U,Tw_m,Tw_
mm,f_t,G,Bt,Rt,dT_m,T_w,T_env,T_n,Q_CHPU)',Fcomfort));
YnonInt = str2func(strcat('@(x)',F_nonIntcomfort));
Y_HP = str2func(strcat('@(x,T_set_r,T_room_m,T_room_mm,DT_m,DT_L,DT_U,T_room,Gt,dc,Q_HVAC,N_slot)',F_HP));
Y_CHP = str2func(strcat('@(x,N_slot,T_set_w,dT_L,dT_U,Tw_m,Tw_mm,f_t,G,Bt,Rt,dT_m,T_w,T_env,T_n,Q_CHPU)',F_CHP));
%% Construction of equality vector beq
beq1 = b_vector_eq(timesteps,P_PV,P_cl,P_WM,U_WM,P_DW,U_DW,P_CD,U_CD);
beq = [beq1;-beq1];
97
%% Construction of equality matrix Aeq
veq =[1 -1 P_HVAC -P_CHPU 0 0 0];
Aeq1 = A_matrix_eq(timesteps,dpunits,veq);
Aeq = [Aeq1;-Aeq1];
%% Construction of inequality matrix A
% battery
%charge
v_ch =[dt*nu_bat 0 0 0 0 0 0];
A_ch_max = A_matrix_ch_max(timesteps,dpunits,v_ch);
A_ch_min = -A_ch_max;
A_ch = [A_ch_max;A_ch_min];
%discharge
v_disch =[dt/nu_bat 0 0 0 0 0 0];
A_disch_max = A_matrix_ch_max(timesteps,dpunits,v_disch);
A_disch_min = -A_disch_max;
A_disch = [A_disch_max;A_disch_min];
A_bat = [A_ch;A_disch];
%HVAC
cst1 = (1-G0*dt/dc);
v_HVAC = [0 0 dt/(dc)*Q_HVAC 0 0 0 0];
A_HVAC_max = A_matrix_HVAC(timesteps,dpunits,cst1,v_HVAC);
A_HVAC_min = -A_HVAC_max;
A_HVAC = [A_HVAC_max; A_HVAC_min];
%CHPU+Thermal storage
a = Rt.*(1-exp(-(Rt*C).^-1*dt))*Q_CHPU;
b = exp(-(Rt*C).^-1*dt);
98
A_CHPU_max = A_matrix_CHP(timesteps,dpunits,a,b);
A_CHPU_min = -A_CHPU_max;
A_CHPU = [A_CHPU_max;A_CHPU_min];
%WM
v_WM = [0 0 0 0 1 0 0];
A_WM_max = A_matrix_nonInt(timesteps,dpunits,v_WM);
A_WM = [A_WM_max;-A_WM_max];
%DW
v_DW = [0 0 0 0 0 1 0];
A_DW_max = A_matrix_nonInt(timesteps,dpunits,v_DW);
A_DW = [A_DW_max;-A_DW_max];
%CD
v_CD = [0 0 0 0 0 0 1];
A_CD_max = A_matrix_nonInt(timesteps,dpunits,v_CD);
A_CD = [A_CD_max;-A_CD_max];
%Total A
A = [A_bat;A_HVAC;A_CHPU;Aeq;A_WM;A_DW;A_CD];
%% Construction of inequality vector b
% battery
b_ch = [E_init;Emax*ones(timesteps-1,1);-E_init;-E0*ones(timesteps-1,1)];
b_disch = b_ch;
b_bat = [b_ch;b_disch];
99
%HVAC
b_HVAC_max = b_vector_HVAC(timesteps,T_room_max,T_room_init,G0,cst1,dt,dc,T_outdoor);
b_HVAC_min = b_vector_HVAC_min(timesteps,T_room_min,T_room_init,G0,cst1,dt,dc,T_outdoor);
b_HVAC = [b_HVAC_max;b_HVAC_min];
%CHPU+Thermal storage
c = G_TS*Rt*T_env + Bt.*Rt*T_in;
d = (1-exp(-(Rt*C).^-1*dt));
b_CHPU_max = b_vector_CHP_max(timesteps,T_TS_max,T_set_w,b,c,d);
b_CHPU_min = b_vector_CHP_min(timesteps,T_TS_min,T_set_w,b,c,d);
b_CHPU = [b_CHPU_max;b_CHPU_min];
%WM
b_WM_max = [N_max_WM-N_task_WM;zeros(timesteps-1,1)];
b_WM_min = [-N_min_WM;zeros(timesteps-1,1)];
b_WM = [b_WM_max;b_WM_min];
%DW
b_DW_max = [N_max_DW-N_task_DW;zeros(timesteps-1,1)];
b_DW_min = [-N_min_DW;zeros(timesteps-1,1)];
b_DW = [b_DW_max;b_DW_min];
%CD
b_CD_max = [N_max_CD-N_task_CD;zeros(timesteps-1,1)];
b_CD_min = [-N_min_CD;zeros(timesteps-1,1)];
b_CD = [b_CD_max;b_CD_min];
%Total b
b = [b_bat;b_HVAC;b_CHPU;beq;b_WM;b_DW;b_CD];
100
%%
%%%%%%%%%%%%%%%%%%%%%%%
% DISPATCH ALGORITHME %
%%%%%%%%%%%%%%%%%%%%%%%
disp('Optimalisation...');
if(isequal(type,'single'))
disp('Single objective optimisation...');
optionsGA = gaoptimset('TolFun',1e-4,'TolCon',1e-4,'PopulationSize',200, 'Generations', ngen*200,'Display','iter');
[x,fval,exitflag] = ga(@(x)MultiObj(x,alpha, daprce, revenue, degcost, NG, nu_e, P_CHPU, T_set_r, T_room_m, T_room_mm,
DT_m, DT_L, DT_U, T_room, Gt, dc, Q_HVAC, N_slot, T_set_w, dT_L, dT_U, Tw_m, Tw_mm, f_t, G, Bt, Rt, dT_m, T_w, Q_CHPU, T_env,
T_n, Ycost, Ycomfort), ngen, A, b, [], [], lb, ub, [], Binvar(timesteps,[3 4 5 6 7]), optionsGA);
else
disp('Error type!');
end
%%
%%%%%%%%%%%%%%%%%%%%%%%
% RESULTS %
%%%%%%%%%%%%%%%%%%%%%%%
disp('Processing results...');
if(isequal(type,'single'))
disp('Single objective results...');
fvaltot=(fval(:,1));
[minimum, ind]=min(fvaltot);
for j=1:ngen
xopt(j)=x(ind,j);
end
for i=1:1
fval_opt(i) = fval(ind,i);
end
else
101
disp('Error type!');
end
toc
% Putting simulation results in a table
xopt_new = round(xopt*10.^1)/10.^1;
setpoints = reshape(xopt_new, dpunits, [])';
SOC = cumsum(setpoints(:,1)*dt);% Calculating State Of Charge of battery
T_room = RoomTempConstr(setpoints(:,3),T_room_init,G0,dc,Q_HVAC,T_outdoor);% Calculating room temperature profile
T_water = WaterTemp(setpoints(:,4),T_set_w,G_TS,Bt,Rt,C,T_in,T_env,Q_CHPU);% Calculating hot water temperature profile
results = [(1:1:timesteps)' round((setpoints(:,1)-setpoints(:,2)+P_HVAC*setpoints(:,3)-P_CHPU*setpoints(:,4)-beq1)*10)/10
setpoints(:,1), setpoints(:,2), setpoints(:,3)*P_HVAC, T_room, setpoints(:,4)*P_CHPU, T_water, SOC,
[N_start_WM;setpoints(2:timesteps,5)],P_WM*U_WM,(1:1:timesteps)',[N_start_DW;setpoints(2:timesteps,6)],
P_DW*U_DW,[N_start_CD;setpoints(2:timesteps,7)], P_CD*U_CD, daprce, revenue(1:timesteps), degcost, NG];
dataset = mat2dataset(results);
dataset(:,:);
dataset.Properties.VarNames = {'Timeslot','equality_test', 'P_Bat','P_Grid','P_HVAC','T_room','P_CHPU','T_water',
'SOC','N_start_WM','P_WM','Timeslot1','N_start_DW','P_DW','N_start_CD','P_CD', 'Buying', 'Selling','degcost','NG'}
% Printing results on screen
fval %fitness value of cost function
disp('Results')
%energy cost
fcost_net = Ycost(x,revenue,daprce,degcost,NG,nu_e,P_CHPU) %net energy cost
f_deg = Ydeg(x,degcost) %battery degradation cost
f_rev = -Yrev(x,revenue) %Sales revenue
f_purchase = fcost_net-f_deg+f_rev %Purchase cost of energy
102
%comfort cost of each appliance
fcomfort =
1/5*(Ycomfort(x,T_set_r,T_room_m,T_room_mm,DT_m,DT_L,DT_U,T_room,Gt,dc,Q_HVAC,N_slot,T_set_w,dT_L,dT_U,Tw_m,Tw_mm,f_t,G,Bt,Rt
,dT_m,T_water,T_env,T_n,Q_CHPU)- YnonInt(x) + YnonInt([0 0 0 0 N_start_WM N_start_DW N_start_CD, zeros(1,ngen-7)])) %Total
comfort cost
f_HP = Y_HP(x,T_set_r,T_room_m,T_room_mm,DT_m,DT_L,DT_U,T_room,Gt,dc,Q_HVAC,N_slot)%comfort cost of heat pump
f_CHP = Y_CHP(x,N_slot,T_set_w,dT_L,dT_U,Tw_m,Tw_mm,f_t,G,Bt,Rt,dT_m,T_water,T_env,T_n,Q_CHPU)%comfort cost of CHP unit
f_nonInt = YnonInt([0 0 0 0 N_start_WM N_start_DW N_start_CD, zeros(1,ngen-7)]) %comfort cost of non-interruptible appliances
fcomfort_test = 1/5*(f_HP + f_CHP + f_nonInt) %Total comfort cost test
save('case01'); % saving the results