Content uploaded by Nadeem Javaid
Author content
All content in this area was uploaded by Nadeem Javaid on Jul 16, 2016
Content may be subject to copyright.
Performance Evaluation of Heuristic Algorithms in
Smart Grids
By
Sahar Rahim
CIIT/SP14-REE-004/ISB
MS Thesis
In
Electrical Engineering
COMSATS Institute of Information Technology
Islamabad – Pakistan
Fall, 2015
Performance Evaluation of Heuristic Algorithms in
Smart Grids
A Thesis Presented to
COMSATS Institute of Information Technology, Islamabad
In partial fulfillment
of the requirement for the degree of
MS (Electrical Engineering)
By
Sahar Rahim
CIIT/SP14-REE-004/ISB
Fall, 2015
ii
Performance Evaluation of Heuristic Algorithms in
Smart Grids
A Graduate Thesis submitted to Department of Electrical Engineering as partial
fulfillment of the requirement for the award of Degree of M.S (Electrical Engineering).
Name
Registration Number
Sahar Rahim
CIIT/SP14-REE-004/ISB
Supervisor
Prof. Dr. Shahid A. Khan,
Professor,
Department of Electrical Engineering,
COMSATS Institute of Information Technology (CIIT),
Islamabad Campus,
January 2016.
Co-Supervisor
Dr. Nadeem Javaid,
Associate Professor,
Department of Computer Science,
COMSATS Institute of Information Technology (CIIT),
Islamabad Campus,
January 2016. iii
Final Approval
This thesis titled
Performance Evaluation of Heuristic Algorithms in
Smart Grids
By
Sahar Rahim
CIIT/Sp14-REE-004/ISB
has been approved
For the COMSATS Institute of Information Technology, Islamabad
External Examiner:
Prof. Dr. Muhammad Sher,
Dean, Faculty of Basic and Applied Sciences
International Islamic University, Islamabad
Supervisor:
Prof. Dr. Shahid A. Khan
Department of Electrical Engineering,
CIIT, Islamabad
Co-Supervisor:
Dr. Nadeem Javaid
Associate Professor, Department of Computer Science,
CIIT, Islamabad
HoD:
Prof. Dr. Shahid A. Khan
Department of Electrical Engineering,
CIIT, Islamabad
iv
Declaration
I Ms. Sahar Rahim, CIIT/SP14-REE-004/ISB, hereby declare that I have
produced the work presented in this thesis, during the scheduled period of
study. I also declare that I have not taken any material from any source
except referred to wherever due that amount of plagiarism is within acceptable
range. If a violation of HEC rules on research has occurred in this thesis, I shall
be liable to punishable action under the plagiarism rules of the HEC.
Date:
Signature of the student:
Sahar Rahim
CIIT/SP14-REE-004/ISB
v
Certificate
It is certified that Sahar Rahim CIIT/SP14-REE-004/ISB has carried out all the
work related to this thesis under my supervision at the Department of Electrical
Engineering, COMSATS Institute of Information Technology, Islamabad and the
work fulfills the requirements for the award of the MS degree.
Date:
Supervisor:
Prof. Dr. Shahid A. Khan,
Professor
Co-Supervisor:
Dr. Nadeem Javaid,
Associate Professor
Head of Department:
Prof. Dr. Shahid A. Khan
Department of Electrical Engineering.
vi
DEDICATION
To Almighty Allah, My Family
&
COMSATS Institute of Information Technology
vii
ACKNOWLEDGMENT
Alhamdulillah. With the Bounty, Mercy and Blessing of ALLAH, this dissertation is completed.
Allah guided me in many ways to successfully finalize my efforts. I have received innumerable
support from various people. I would like to mention few words for adequately capture all my
gratitude. First of all, I would like to express my heartiest appreciation to my supervisor, Prof.
Dr. Shahid A. Khan, for his continuous and inspiring guide and co-supervisor, Dr. Nadeem
Javaid, for his patience, support, encouragement, insightful criticism and guidance from
foundation to concluding level. It is impossible for me to complete this dissertation within due
time without their guidance and utmost efforts. I would also like to appreciate my dissertation
committee, Prof. Dr. Junaid Mughal, Dr. Qadeer Ul Hassan and Dr. Mustafa Shakir for their
critical comments and valuable time.
I want to take it as opportunity to heartily thank my best friend and all my fellows who assist me
in all my harsh time and during precarious phases of this dissertation. Finally, I like to thank my
father, whose support, trust and tremendous care makes me able to work hard and my mother,
whose hands always raised to pray for my success. I would like to extend my thanks to my
sisters and brother for their love, friendly attitude and motivate me with encouraging words.
Sahar Rahim
CIIT/SP14-REE-004/ISB
viii
ABSTRACT
Performance Evaluation of Heuristic Algorithm in
Smart Grids
Smart grid (SG) is evolutionary idea in which all components of conventional power grid are
modernized with the advance integration of information technology, sensors and autonomous
system. The bi-directional flow of information and power in SG with optimal integration of
renewable energy sources encourage customers to participate in energy management schemes
and demand response. Meanwhile, innovation components of grid such as transmission, demand
side management and demand response are develop as modernized applications with lots of
benefits as well as challenges in SG. Therefore, we explore potential solution to the interesting
and challenging problems of SG. In this dissertation, we comparatively evaluate the performance
of home energy management controller which is designed on the basis of heuristic algorithms;
genetic algorithm (GA), binary particle swarm optimization (BPSO) and ant colony optimization
(ACO). In this regard, we introduce a generic architecture for demand side management (DSM)
which integrates residential area domain with smart area domain via wide area network. In
addition, problem formulation is carried via multiple knapsack problems. For energy pricing,
combined model of time of use tariff and inclined block rates is used. Simulation results show
that all designed models for energy management act significantly to achieve our objectives and
proven as a cost-effective solution to increase sustainability of SG. GA based energy
management controller performs more efficiently than BPSO based energy management
controller and ACO based energy management controller in terms of electricity bill reduction,
peak to average ratio minimization and user comfort level maximization and its execution time is
also less than other two models.
ix
TABLE OF CONTENTS
List of Figures x
List of Tables xi
1 Introduction 1
2 Related Work and Motivation 6
2.1 Motivation ................................ 11
3 Proposed Model 13
3.1 Energy consumption model ....................... 15
3.2 Load categorization ........................... 16
3.2.1 Fixed appliances ........................ 17
3.2.2 Shiftable appliances ....................... 17
3.2.3 Elastic appliances ........................ 18
3.3 Energy price model ........................... 19
3.4 Local energy generation ........................ 20
3.5 Energy storage ............................. 21
3.6 Residential users ............................ 22
3.6.1 Passive users .......................... 22
3.6.2 Semi-active users ........................ 22
3.6.3 Active users ........................... 23
3.7 Problem formulation .......................... 23
3.8 PAR ................................... 24
3.9 Waiting time .............................. 25
3.10 Objective function ........................... 27
4 Heuristic algorithms 29
4.1 GA .................................... 30
4.2 BPSO .................................. 32
4.3 ACO ................................... 33
5 Simulations and results 35
5.1 Energy consumption pattern and Electricity bill reduction ..... 39
5.2 PAR ................................... 43
5.3 Waiting time .............................. 44
5.4 Parametric tuning for all models .................... 45
viii
LIST OF FIGURES
1.1 SG .................................... 2
3.1 Components of DSM .......................... 14
3.2 EMC model for residential users .................... 15
3.3 DSM functional diagram ........................ 16
3.4 End users classification ......................... 23
3.5 Parameters of appliance ........................ 25
3.6 Range of operation time ........................ 26
3.7 Waiting time .............................. 27
5.1 TOU Tariff Model ........................... 36
5.2 Energy consumption (kWh) ...................... 40
5.3 Electricity bill (cent) .......................... 41
5.4 Electricity bill reduction per day .................... 41
5.5 Electricity bill per day ......................... 42
5.6 Electricity bill (cent/day) for 50 users ................. 42
5.7 PAR curve ................................ 43
5.8 Possible trade off between electricity cost and waiting time ..... 44
x
LIST OF TABLES
1.1 Brief comparison between traditional grid and SG .......... 3
2.1 Summarized related work ........................ 10
4.1 Key points of GA-EMC ........................ 32
4.2 Key points of BPSO-EMC ....................... 33
4.3 Key points of ACO-EMC ........................ 34
5.1 Parameters of Fixed Appliances .................... 37
5.2 Parameters of Shiftable Appliances .................. 37
5.3 Parameters of Elastic Appliances ................... 37
5.4 GA parametric list ........................... 38
5.5 BPSO parametric list .......................... 38
5.6 ACO parametric list .......................... 38
5.7 Execution time ............................. 39
5.8 Summarized results ........................... 45
5.9 Parameter Evaluation for GA ..................... 47
5.10 Parameter Evaluation for BPSO .................... 48
5.11 Parameter Evaluation for ACO .................... 50
xi
1
INTRODUCTION
1
Chapter 1 Introduction 2
Traditional electrical power system is inadequate to meet modern power grid chal-
lenges such as reliability, stability, robustness, etc. [1]. Thus, a new infrastructure
is needed to smartly meet these challenges and reduce pressure on global envi-
ronment. In this regard, smart grid (SG) integrates communication technologies,
computational abilities, control systems and sensors with existing grid and enables
two way flow of information between utility and end users. SG has modernized all
sections of present electrical system as shown in fig. 1.1.
Figure 1.1: SG
Specifically, the consumers are now prosumers because they have the ability to sell
their generated electricity to the utility. Thus, renewable energy sources (RESs)
(e.g., solar, wind, etc.) play a vital role in the concept of SG. Utilities are always
interested in increasing their profit and reducing of peak to average load. On the
other hand, prosumers wish to reduce their electricity bills without compromising
their comfort level. Main aims of SG are to enhance efficiency, sustainability,
capacity and customer engagement [2]. Some of the major differences between
traditional power grid and SG are summarized in table. 1.1
Chapter 1 Introduction 3
One of the important aspects of SG is demand side management (DSM) which is
the best way to maintain balance between demand and supply. Two main func-
tions of DSM are load management and demand response (DR). Load management
focuses on the improvement of energy efficiency to avoid major distress and black-
out. The benefits of load management are numerous such as reduced number of
peak power plants, efficient energy consumption, electricity bill reduction and im-
proved performance of the power grid in term of reliability and flexibility [3]. DR
is a responsive action taken by a customer against dynamic price models. It offers
many financial and operational benefits for electricity utilities, end user, and grid
operations. The highly volatile nature of load may threaten the integrity of grid
within seconds. Therefore, DR is important to tackle these uncertainties, as it
provides flexibility at relatively low rates [4].
The common objectives of SG are electricity bill reduction, minimization of aggre-
gated power consumption and minimization of both electricity bill and aggregated
power. To achieve these objectives, many DSM techniques and algorithms are
Table 1.1: Brief comparison between traditional grid and SG
Infrastructures Traditional grid SG
Power system
Centralized generation
Uni-directional power and
information flow
(utility to consumer)
Low storage capacity
Distributed generation
Bi-directional power and
information flow
(utility to (from) consumer)
Grid energy storage capacity
Information technology
Aged metering system
No monitoring system
Lack of management units
Advanced metering system
Phasor management unit
Information management unit
Communication system Wired technology Wired and wireless technologies
Energy sources system Non- renewable sources
(mainly fossil fuel)
Both non-renewable and
renewable sources
(photovoltaic panels (PV), wind
turbine, plug-in electric
vehicles, etc)
Power losses Wastage of electricity due
to limited power storage
Efficient use of electricity
minimizes power losses
Chapter 1 Introduction 4
proposed in the previous years. For example, in [5-7], integer linear programming,
mixed integer linear programming and mixed integer non-linear programming are
used for electricity cost minimization. Similarly, in [8], authors use convex pro-
gramming for relatively large number of users to reduce their electricity bills. In
[9, 10], integer linear programming and mixed integer linear programming are used
to optimally schedule the appliances to minimize the aggregated power consump-
tion. [11] uses mixed integer linear programming to reduce both electricity bills
and aggregated power consumption. However, these techniques can not tackle
large number of different household appliances having unpredictable, non-linear
and complex energy consumption patterns due to randomness in human behavior.
In this dissertation, heuristic optimization techniques are used due to its ex-
ceptional characteristics: flexibility for specified constraints, ease of implemen-
tation, low computational complexity and low computational time [12]. Earlier
researchers demonstrated the benefits of different heuristic techniques for their de-
signed objectives. [13] uses evolutionary algorithm (a heuristic approach) to mini-
mize electricity cost for all types of sectors (residential, commercial and industrial).
We have chosen three popular heuristic optimization techniques: genetic algorithm
(GA), binary particle swarm optimization (BPSO) and ant colony optimization
(ACO) to evaluate our designed objective function due to their self-organization,
self-optimization, self- protection, self-healing and decentralized control system
[12]. In the literature, various works have been done to enhance the efficiency
of DSM using these heuristic optimization techniques. For example, authors in
[14, 15, 16] presents different models to reduce utility electricity bills using GA.
Electricity cost reduction for end users is achieved in [17, 18] with particle swarm
optimization (PSO) technique. Signaled PSO (a heuristic approach) is used to
reduce aggregated load and execution time with absolute error for number of con-
sumers in [19]. An efficient self-optimizing system for DSM is proposed in [20] to
optimally schedule load using ACO technique and [21, 22] investigates congestion
problem in SG through DR by applying ACO to minimize cost and maximize user
Chapter 1 Introduction 5
comfort level. Although mentioned work performed well for their designed objec-
tives but some critical issues are ignored.
To attain electricity cost minimization objective, they ignored user comfort level
and their electricity pricing model is not compatible with real scenarios and pro-
vides comprehensive analysis about the effectiveness of different heuristic algo-
rithms in home energy management. We first design an energy management
controller (EMC) model for single and multiple homes using multiple knapsack
problem (MKP) and then apply three heuristic algorithms (GA, BPSO and ACO)
to get feasible solution for designed objective function. To calculate electricity
bills, we use time of use (TOU) tariff model with inclined block rate (IBR), so
that peak formation is avoided. Performance evaluation for GA, BPSO and ACO
on the basis of designed EMC via simulations is done in terms of energy consump-
tion pattern, electricity bill, peak to average ratio (PAR), user comfort level, and
execution time.
2
RELATED WORK AND
MOTIVATION
6
Chapter 2 Related Work and Motivation 7
Many researchers around the world worked to optimally schedule smart appliances.
In this regard, some of the papers are discussed as follow:
In [23], authors investigate the problem of household appliance scheduling to en-
hance energy efficiency of electrical grid and provide benefits to end users. They
proposed a solution that optimally schedules a set of appliances. To minimize
customer electricity bills and maintain energy consumption within a limit, they
use day-ahead variable peak pricing model and map their problem by using MKP.
By limiting the energy demand within certain capacity, problem of load shedding
can be removed. Results show that this model effectively reduces utility electricity
bills while keeping power consumption within pre-defined limits. Another model
of home energy management controller for residential users is proposed in [24].
Objective function is formulated by knapsack problem and dynamic programming
approach is used to solve problem and to set consumer preferences for each ap-
pliance. These priorities were the value of appliances that are used to schedule
the appliance to satisfy their operational time constraints to avoid peak formation
and to reduce electricity cost.
In [13], authors present an efficient model of DSM that reduces PAR and elec-
tricity bills for residential, industrial and commercial users. Scheduling problem
is formulated as a minimization problem and then problem is evaluated by using
heuristic evolutionary approach. Heuristic algorithms show better results because
of their flexible nature that allow the implementation of individual load pattern
in order to minimize inconvenience. Proposed model is beneficial for both utilities
and customers in a way that PAR reduction causes minimization in the number
of peak power plants while incentive based model helps consumer to reduce their
electricity bills. Simulation results show that the proposed DSM strategy achieves
significant savings, while reducing the peak load demand of the smart grid.
In [14], authors discuss an efficient architecture for energy management system
by using home area network (HAN) for residential users. They combine real-time
pricing (RTP) tariff model with the IBR because when only the RTP is adopted,
Chapter 2 Related Work and Motivation 8
there is a risk that most of the appliances operate during the hours of lowest elec-
tricity price that cause peak formation. To strengthen the stability of electricity
system, peak formation must be avoided. To solve these issues in an optimized
way, objective function is formulated. As this kind of optimization problem is non-
linear, therefore they use GA to optimize their problem. Simulation results shows
that proposed model is very effective to reduce PAR and electricity cost. Another
DSM model is proposed in [15] for residential users to reduce PAR and electricity
bill minimization. GA is used to get optimal start time of each appliance in each
time slot while satisfying its operational constraints. There is a tradeoff between
electricity cost and waiting time. When waiting time of an appliance is zero, its
electricity cost is increased and vice versa. Combined model of RTP with IBR is
used to avoid peak formation. Simulations are carried out for single and multiple
users. Results show the effectiveness of proposed DSM model for both single and
multiple user scenarios.
An efficient home energy management scheme is proposed in [17] to schedule large
number of interruptible load in the time period of 16 hours. Binary particle swarm
optimization (BPSO), which is an extended version of PSO is used to achieve the
scheduling objective. The objective is to minimize the electricity bills while satis-
fying the operational constraints and minimize the frequency to interruptions. The
effectiveness of proposed approach is improved by dividing the swarm into number
of subswarms. The scheduling technique proven as useful scheme for a relatively
challenging scheduling task, and have potential advantages in scheduling widely
varied and technically complex interruptible loads. In [18], real-time model for op-
timal power usage of household appliances is proposed. BPSO algorithm is used
to solve the formulated problem by encouraging the participation of both utilities
and consumers. By considering the features of the appliances and living habits
of customers, the appliances are divided into three categories. Results show the
significant performance of proposed scheme for load shifting, energy saving and
cost reduction.
Chapter 2 Related Work and Motivation 9
An efficient heuristic approach is presented in [25] for scheduling of smart appli-
ances in residential area. The proposed algorithm is evaluated by comparing the
electricity cost and computational time with an exact algorithm. Variable energy
price model is used for scheduling of appliances. Hourly prices for electricity, the
operation start times of set of appliances are optimized to reduce cost of energy
consumption while satisfying the operational and peak power constraints. Re-
sults show that electricity cost obtained by heuristic algorithm is within 5% of the
optimal cost of exact algorithm whereas computational time is reduced by expo-
nential factor. In [26], a home energy management model is designed in which
each appliance is operated according to its schedule within predefined time lim-
its. The objective is to reduce utility electricity bills while satisfying operational
constraints. GA is used to evaluate the objective function and get optimal start
operational time for each appliance to reduce electricity bill and avoid peak for-
mation. It is a comparative study in which GA based energy management model
results are compared with simulated annealing and greed method. Simulation re-
sults show that GA acts efficiently to reduce cost while minimizing power usage
at any instant of time than others.
An adaptive energy management model for DSM in residential area is described
in [20]. Authors aim to optimize the use of distributed RESs to reduce utility
electricity bill. They use ACO as an optimization algorithm to schedule shiftable
load and MAPE-K feedback loop as a predictive model to deal with the inter-
mittent nature of RESs. Results show that proposed cost-effective self-optimizing
model is able to adapt sudden changes in environmental conditions and optimize
the power usage of residential users. Authors in [21] proposed an efficient scheme
to manage congestion problem in SG through DR. Their objective is to optimally
schedule different generation resources to minimize cost and maximize customer
satisfaction. ACO is used as optimization technique to provide benefit to con-
sumers and fuzzy satisfying technique to choose the most feasible solution from
the set of pareto optimal solution. Results show that proposed scheme is effective
Chapter 2 Related Work and Motivation 10
Table 2.1: Summarized related work
No. Ref. Techniques Objective (s) Achievement (s) Deficiency (ies)
1 [23] MKP
Electricity consumption
and bill
reduction
Efficiently limits power
consumption to reduce
electricity bills
User comfort level
and integration of
renewable
energy resources
2 [24]
MKP
+
Dynamic
Programming
Reduction in electricity
bills and avoid
peak formation
Sets priorities to satisfy their
operational time constraints
with user preference
Integration of renewable
energy resources
3 [13]
Heuristic
evolutionary
approach
Electricity bills
reduction for
residential, commercial and
industrial user
Beneficial model for both
utilities and customers in
a way to PAR reduction and
peak load minimization
User comfort level
and integration of
renewable
energy resources
4 [14]
GA
+
RTP and IBR
Avoid peak formation
and electricity bill
minimization
Effectively reduce PAR
and electricity cost
Congestion problem, user
comfort level and
integration of renewable
energy resources
5 [15]
Kp
+
GA
+
RTP and IBR
Utility electricity cost
minimization and peak
formation reduction
Effective model for both
single and multiple users
Congestion problem,user
comfort level and
integration of renewable
energy resources
6 [17] BPSO
Minimize electricity
bills and frequency
to improve
Acts potentially to
achieve designed
objectives
Congestion problem,user
comfort level and
integration of renewable
energy resources
7 [18] BPSO
Energy saving and
electricity cost
reduction
Encourage utility and
consumer participation to
maintain balance
between demand and
supply
User comfort level
and integration of renewable
energy resources
8 [25]
Exact
algorithm
+
Heuristic
algorithm
Compare electricity
cost reduction
and computational
time for both algorithms
Electricity cost obtained
from heuristic algorithm is 5%
optimized than exact algorithm
User comfort level
and integration of renewable
energy resources
9 [26]
GA
+
Simulated
annealing
+
Greedy
method
Comparative study for
electricity cost reduction
GA acts effectively
than other to two
algorithm
User comfort level
and integration of renewable
energy resources
10 [20]
ACO
+
MAPE-K
Deals intermittent
nature of RESs to
reduce electricity bills
Cost-effective and
self-optimized model
User comfort level
and peak formation
problem
11 [21]
ACO
+
Fuzzy
techniques
Congestion problem
with electricity cost
minimization
Effective model for
both generation and
demand management
Integration of renewable
energy resources
12 [22] ACO Congestion control with
cost reduction
Improved model to minimize
electricity cost
User comfort level
and integration of renewable
energy resources
both for generation selection and demand management in the most economical
way. In [22], authors investigate congestion cost model in real-time power grid
system. They built a congestion factor to control opening of both generation and
load sides. Non-linear programming is used to formulate real time congestion
problem and cost is minimized on the basis of an optimization algorithm (ACO).
Results show that improved model can significantly minimize electricity cost. All
the techniques, objectives, achievement (s) and deficiency (ies) mentioned above
is summarized in table. 2.1.
Chapter 2 Related Work and Motivation 11
2.1 Motivation
In SG, optimization of energy consumption schedules and user cost minimization
are two difficult tasks due to randomness in the energy consumption patterns
of end users. In literature, an efficient home energy management controller to
reduce utilities electricity bills and PAR is still an issue. Mostly, user comfort
level is neglected while reducing electricity bills. Typically, the target of DSM is
to efficiently manage the energy schedules such that electricity price is minimized
while maximizing user comfort level. In SG, optimization problems are as follow:
•Minimize the electricity bill.
•Minimize the aggregated power consumption.
•Minimize both electricity bill and aggregated power consumption.
•Minimize PAR.
•Maximize user comfort.
•Efficient integration of RESs.
Many strategies have been proposed in the previous years to effectively tackle
these mentioned problems. Authors in [5-7], present three different techniques:
integer linear programming, mixed integer linear programming and integer non-
linear programming for electricity bill reduction. However, integration of RES,
user comfort and power consumption minimization problems are ignored in these
models. Similarly, [8] uses convex programming to deal with large number of
consumer for electricity bill reduction. Results show that this technique gives ef-
fective solution however, at the cost of increased computational time. The linear
programming based scheme in [9] is effective for residential areas. However, lack
of RES integration and non-adaptability with dynamic pricing model are its ma-
jor drawbacks. In [11], both electricity bill minimization and aggregated power
minimization problems are investigated using mixed integer linear programming
for dual optimization functions. Proposed scheme is implemented for single home
Chapter 2 Related Work and Motivation 12
but it does not deal with the flexibility of power usage patterns and human be-
haviors. Thus, to resolve these issues in previously proposed schemes for DSM,
we proposed an efficient model using MKP to formulate an objective function for
residential area. To evaluate our objective function, heuristic optimization tech-
niques are used due to its ability to deal with large and complex scenarios within
less computational time and less computational complexity [12].
We have considered three heuristic techniques: GA, BPSO and ACO to achieve
our objectives and compare their results. Despite of great efforts in literature for
DSM strategy using these heuristic algorithms, there is still a room for improve-
ment to make system compatible with growing demand of power. As in [14], GA
based DSM model is presented to reduce electricity bills for residential area while
ignoring user comfort level. Authors used RTP tariff model to calculate electricity
bills which is a major drawback of this model because real time data transmis-
sion causes great chance for data loss that cause discomfort both for utility and
customer. Authors in [17], proposed a model for DSM using PSO techniques to
get objective of electricity bill reduction without considering user comfort max-
imization. Whereas, in [20], authors used ACO and feedback prediction model
to enhance efficiency of DSM. They proposed an economical model that optimize
power usage without considering user satisfaction parameters.
We apply these heuristic optimization techniques in a novel way to enhance the
efficiency of DSM. In our focused scenario, household appliances are classified into
three categories and problem is formulated by using MKP. TOU tariff model is
used to calculate electricity bills for end users and to get feasible solution for de-
signed objective function; we used GA, BPSO and ACO. Also our proposed model
significantly integrates RES energy with grid power to deals with issues keeping
in view the interest of both players (utilities and consumers).
3
PROPOSED MODEL
13
Chapter 3 Proposed Model 14
In SG, DSM enables more efficient and reliable grid operations. Its two main
functions are energy management and demand side control activities for end users.
In residential area, every smart home is equipped with EMCs and smart meters
to make stable and reliable bi-directional communication between utilities and
customers. All elements, such as electrical appliances, sensors, local generation
and energy storage systems (ESSs) give their information to EMC through HAN
and EMC controls scheduling of appliances. After collecting all information, EMC
sends it to SG domain through WAN. There are various wireless solutions for
communication links between the smart meters and the EMCs such as ZigBee,
Z-Wave, Wi-Fi, or a wired (HomePlug) protocol [1]. Simple architecture of DSM
is shown in fig. 3.1.
Sensors
Distributed RESs
Smart devices
Residential area
domain
ESSs
EMCs
HAN
SG domain
Distribution
Operation
Market
Service provider
Customers
WAN
Two way communication
One way communication
Figure 3.1: Components of DSM
In residential area based DSM, we consider Nsmart homes and Msmart appli-
ances as shown in fig. 3.2. In this model, all smart homes have smart metering
system and EMC. End users change their energy usage according to incentive
based schemes offered by utilities.
Conceptual DSM diagram of our proposed scheme is shown in fig. 3.3. In each
home, consumer inputs different parameters of appliances to appliances scheduler
and then appliance manager gives signal to various appliances about their on/off
Chapter 3 Proposed Model 15
Smart
meter EMC
Smart
meter EMC Smart
meter EMC
Smart
meter EMC
Power distribu on
ulity
User N+1
User N+2 User N
User 1
Two way communica on
One way communica on
Figure 3.2: EMC model for residential users
status. For electricity pricing model, TOU tariff is used to calculate electricity bill
against the energy consumption cost per day.
3.1 Energy consumption model
Let A={a1, a2, a3, . . . , am}be the set of appliances such that a1,a2,a3,···,amare
number of appliances that belong to each category. If t∈T={1,2,3,··· ,24 }
denotes the scheduling horizon, then hourly energy consumption demand of a
appliance is given as,
Ea(t) = {Ea
t1+Ea
t2+Ea
t3+. . . +Ea
t24 }(3.1)
where, Ea
t1,Ea
t2,Ea
t3,···,Ea
t24 denotes energy consumption demand of each appliance
in the respective time slots. The per day total energy consumption demand for all
Chapter 3 Proposed Model 16
Appliance 1
Appliance 2
Appliance M
User
Appliance
manager
Appliance
scheduler
Master control
TOU pricing
genera on
Aggrega on
process
Network
interface
Hourly projected
energy demand
Next day TOU
pricing
User energy
uliza on
New data pricing
Appliance
status
Appliance
schedule
Heuris c
techniques
System se ng
and parameters
Database manager
Historical appliance data
a a
a
a
a
a
User Interface
Fixed device
Shi able device
Elas c device
HAN
a
a
a
Two way communica on
One way communica on
Figure 3.3: DSM functional diagram
appliances is calculated as follows,
ET=
24
X
t=1 A
X
a=1
E(i,t)(3.2)
In order to design the optimization model for home energy management, we have
categorized the load according to the characteristic of appliances and life style of
end users as discussed in the following section.
3.2 Load categorization
We classify appliances into three categories; fixed, shiftable and elastic appliances
according to their power consumption pattern and time of use [27]. Detail of all
these categories is given as follow:
Chapter 3 Proposed Model 17
3.2.1 Fixed appliances
These are also called regular appliances because their usage or length of operation
can not be modified. For example, lights, fans, clothes iron, microwave oven,
toaster, tv, etc. We represent fixed appliances by Fed and its power consumption
as ν. If each fixed appliance fed ∈Fed has power rating ρfed , then total power
consumption in each time slot is calculated as,
ν(t) = X
fed∈Fed
24
X
t=1
ρfed (t)×χfed (t)(3.3)
where, χfed (t) is the state of each fixed appliance in particular time slot and it is
given as,
χfed (t) =
1 if appliance is ON
0 if appliance is OFF
(3.4)
3.2.2 Shiftable appliances
These are also called burst load because these are manageable and can be shifted
in time without altering their load profile. For example, washing machine, dish
washer, clothes dyer, etc. We denote shiftable appliances by Sed and their power
consumption by ∆. Each shiftable appliance is characterized by its length of
operation which is denoted as τsed and it is pre-defined by end users each day.
Consumers set start time and end time for each shiftable appliance as,
αsed ≤τsed ≤βsed (3.5)
where, αsed and βsed are the start and end times of a shiftable appliance that are
set by end consumer. If each shiftable appliance sed ∈Sed has power rating factor
Chapter 3 Proposed Model 18
ρsed , then the total power consumption is calculated as,
∆(t) = X
sed∈Sed
24
X
t=1
ρsed (t)×χsed (t)(3.6)
where, χsed (t) is the state of each shiftable appliance in particular time slot and it
is given as,
χsed (t) =
1 if appliance is ON
0 if appliance is OFF
(3.7)
3.2.3 Elastic appliances
These are also called interruptible appliances because these are fully controllable
in terms of both usage time and power consumption profile. For example, air
conditioner, refrigerator, water heater, space heater, etc. We represent elastic ap-
pliances by Eed and its power consumption is denoted by κ. Each elastic appliance
eed ∈Eed has power rating ρeed, power quantity factor λeed, length of operation
τeed , start time αeed and end time βeed . These attributes are set by the consumer,
such that,
αeed ≤τeed ≤βeed (3.8)
Power consumption of each elastic appliance ζeed is calculated as follows,
ζeed =λeed ×ρeed ∀eed ∈Eed (3.9)
The total power consumption is calculated as,
κ(t) = X
eed∈Eed
24
X
t=1
ρeed (t)×χeed (t)(3.10)
Chapter 3 Proposed Model 19
where, λis used to vary the power quantity use in the predefined range and χsed (t)
is the state of each elastic appliance in particular time slot given as,
χeed (t) =
1 if appliance is ON
0 if appliance is OFF
(3.11)
3.3 Energy price model
A number of tariff models are available to define electric energy prices for a day or
for short time duration. Among these, TOU tariff model is defined for electricity
prices depend on the time of day and are pre-defined in advance. Critical peak
pricing (CPP) is a variant of TOU in which price is considerably raised in case of
emergency situations (e.g. high demand). RTP based electricity prices can change
as often as hourly, reflecting the utility cost of supplying energy to customers at
that specific time. In our model, we use TOU with power dependent tariff known as
inclined block tariff or IBR. The energy price at time tis an increasing, piecewise,
linear function of the total energy demand. As E(t) is the total power consumption
of all appliances in a home at each time slot tand it is calculated as,
E(t) =
24
X
t=1 ν(t) + ∆(t) + κ(t)(3.12)
To calculate electricity bills, energy price for each unit consumed in each time slot
is represented by C(t) and according to IBR model, it is designed as,
C(t) =
C1(t) 0 ≤E(t)≤Eth1(t)
C2(t)Eth1(t)≤E(t)≤Eth2(t)
C3(t)Eth2(t)< E(t)
(3.13)
Chapter 3 Proposed Model 20
where, Eth1(t) and Eth2(t) are two power consumption thresholds and C1,C2and
C3are costs for these particular cases.
3.4 Local energy generation
The residential users can also use distributed RESs such as PV panels, wind tur-
bines, electric vehicles, etc. The distributed RESs are used to fulfill the energy
demand locally or to charge the storage devices. Assume that each home is fitted
with PV panel that is capable of generating 50% of total grid power. In this case,
consumer becomes prosumer because he/she can generate its own energy. They
can also sell the generated RESs energy back to the grid depending upon their
agreement with the utility. PV panel generates solar power depending on solar
radiations and total estimated radiation varies for every month. Solar power out-
put depends on radiation amount, direction of panels and transfer efficiency. The
generated energy in each time slot is characterized as Ψr
tand it is calculated by
the following expression [28].
Ψr(t) = 10 ×1
√2Πσexp −(t−µ)2
2σ2(3.14)
Where, µis the mean of distribution and σis the variance. The hourly RESs
energy must be greater than zero during day time. The daily energy supply from
renewable system (PV panel) installed by the users is denoted by Θ and calculated
as,
Θ(t) =
24
X
t=1
Ψr(t) (3.15)
Let Θmax be the maximum generation capacity of PV panel then available energy
must be in following range,
0≤Ψr(t)≤Θmax ∀t∈T(3.16)
Chapter 3 Proposed Model 21
If locally generated power is greater than total power demand for appliances in each
time slot, then the total power is in negative sign which means that the generated
power can be sold back to grid or it can be stored to reduce energy usage during
peak hours. In order to be eligible for participation in some agreement with grid to
sell negative power back to grid, user must oblige to meet a specific power capacity
Ψrmin ;
Θmax(t)≥Ψrmin (t)∀t∈T(3.17)
3.5 Energy storage
When the energy generation exceeds consumption, it is stored. The stored energy
is used at high peak hours or it can be used in night, when solar energy is not
present. We model ESS with a set Band for each battery b∈B, we introduce a
binary variable χbnthat shows charging and discharging status of all batteries. A
binary variable χbnis defined for each time slot,
χbn(t) =
1 Charging
0 Discharging
(3.18)
Charging and discharging rates of a battery are represented by non-negative vari-
ables as rc
bnand rd
bn, respectively. Such variables are bounded by the following
constraints [27],
rc
bn< rc,max
b×χbn∀b∈B(3.19)
rd
bn< rd,max
b×(1 −χbn)∀b∈B(3.20)
where, rc,max
band rd,max
bare maximum capacity of charging and discharging rates,
respectively. There are energy losses during charging and discharging in each
battery and its efficiency rate is between 0 <ℵc<1 and 0 <ℵd<1 for charging
Chapter 3 Proposed Model 22
and discharging. A binary variables shows that both these operations cannot
occur at the same time. Despite the benefits of ESSs, their cost may limit their
applicability in real scenarios.
3.6 Residential users
We design our model for three types of users in residential area such as passive,
semi-active and active users. We define these categories as,
3.6.1 Passive users
They only consume electrical energy of the grid and does not generate or store
electrical energy. They can only shift there load from high peak to low peak and
reduce their electricity bills. The set of passive users is represented by P. The
energy consumption profile for each user is calculated by the following equation:
Ei∈p(t) =
24
X
t=1
Ei(t) (3.21)
where, i∈Pconsumes electrical energy in time slot t.
3.6.2 Semi-active users
They have RESs such as solar panels and wind turbines. They consume energy
both from power grid and RES to reduce their electricity bills. The set of semi-
active users is represented by S. The energy consumption profile for t∈Tis
calculated as,
Es∈S(t) =
24
X
t=1 Es(t)−Θs(t)(3.22)
where, s∈Sbelongs to set of semi-active users and Θs(t)is the solar panel
generated energy.
Chapter 3 Proposed Model 23
3.6.3 Active users
They take energy from RES and store it in storage devices such as batteries as
well as also take electrical energy from grid to fulfill their need. The set of active
users is represented by A. The energy consumption profile for t∈Tis calculated
by the following equation:
Ea∈A(t) =
24
X
t=1 Ea(t)−Θa(t)±Ba(t)(3.23)
where, a∈Abelongs to set of active users and Θa(t) is the solar panel generated
energy and Ba(t) is the energy stored in batteries. The batteries are charged from
RES (not from grid). If Bais positive, it means battery is charging and if negative,
battery is discharging. The conceptual diagram of all types of users is shown in
fig. 3.4.
SG
Smart
meter EMC
Grid power
+
RES
Grid power
Grid power
+ RES +
ESS
Passive user
Semi-
active
user
Active user
Two way communication
One way communication
Figure 3.4: End users classification
3.7 Problem formulation
In this work, main objectives are to reduce consumer cost by optimizing the energy
consumption patterns of appliances to maximize the comfort level of end user
and to maintain balance between demand and supply. Here, we formulate our
Chapter 3 Proposed Model 24
scheduling problem by using MKP. MKP is a resource allocation problem that
consists of Mresources (capacities) and set of Nobjects [29]. We take jnumber
of knapsacks, and map our scheduling problem in MKP as follows:
•We consider jnumber of knapsacks as power capacities in each time slot.
•Number of appliances as number of objects.
•The weight of each object as the energy consumed by appliances in each
time slot. Note that it is independent of t.
•The value of object in a specific time slot is the cost of power consumption
of the appliance in that time slot.
•The value of binary variable χcan be 0 or 1 depending on the state of
electrical appliance.
With the help of this model customer’s electricity bill can be controlled, and for
utility side, it is also beneficial because it ensures that the grid is not over stressed.
As the total power consumption for all types of appliances should not exceed
maximum power capacity in each hour denoted as γ(t), we introduce constraint
which limits the power consumption and depends on load profile and its states.
Constraints show that power consumption is predefined,
24
X
t=1 E(t)×χ(t)≤γ(t)(3.24)
Here, γ(t) is the power capacity in each hour that is available from grid and
χ(t)∈[0,1] denotes the states of appliances. Total power consumption in each
hour must be limited to this capacity factor.
3.8 PAR
It is beneficial for the utility and consumer to reduce PAR so that power supply
and demand balance can be maintained. We have defined PAR for single user as
the ratio of peak load and average load in each time slot. It is represented as φ
Chapter 3 Proposed Model 25
and its mathematical form is as follow,
φ(t) = maxa∈A(E(t))
1
TP24
t=1(E(t)(3.25)
Therefore, for nnumber of users, the PAR is written as,
φn(t) = maxa∈A(E(t, n))
1
TPN
n=1(P24
t=1 E(t, n)) (3.26)
3.9 Waiting time
It is necessary for residents to set some parameters for each shiftable and elastic
appliance. In scheduling problem, we omit fixed appliances because these appli-
ances do not play any role in energy management system and must run with first
priority. We assume start time αaand end time βafor each schedulable appli-
ance such that αa< βa. The operation time interval (OTI) for each appliance is
the time in which it performs its functionality. Let τabe the length of operation
(LOT) of an appliances that is required to complete the task. These parameters
are needed to set by the resident via user interface and then this information is
sent to EMC. Assumed that βa−αamust be greater than or equal to τa, we define
operation start time by ηa. As, we already know αa,βa,τaand χafor each ap-
pliance but ηais unknown. Once we get ηa, we can calculate power consumption
pattern. Relationship between all these parameters is shown in fig. 3.5. Now for
α
β
τ
ƞ
Figure 3.5: Parameters of appliance
each appliance, there exists a group of parameters comprising the OTI, LOT, and
power consumption values per unit time. ηamust be greater than or equal to αa
Chapter 3 Proposed Model 26
and less than or equal to βa−αa. Therefore, range of ηais given by,
ηa∈[αa, βa−τa] (3.27)
The range of ηais shown in fig. 3.6. Usually, residents want to finish their work
α
β
α
β
τ
τ
Range of ƞa
ƞa=βa-τa
Figure 3.6: Range of operation time
as soon as possible. Therefore user comfort depends upon waiting time reduction
and cost minimization. There is a trade off between cost and waiting time. When
we minimize cost, customer compromises on waiting time and when waiting time
reduces customer pays huge cost. Mathematically, waiting time is represented as
ϕand for each schedulable appliance and is given as,
ϕa=ηa−αa
βa−τa−αa
(3.28)
If an appliance operates at a later time, the later the appliance operates the larger
the waiting time will be. The smallest and the largest values of ϕare set between
0 and 1. Assume that for washing machine, a resident sets the parameter OTI as
[αw, βw] and LOT as τw. If it starts working at its starting time that is αwthen
its ϕwis zero and if it start working at latest time such that βw−lwthen its ϕw
would be one, as shown in fig. 3.7.
Chapter 3 Proposed Model 27
α
β
α
β
τ
τ
ƞ=βa-τa
ƞa=αa
αa
βa
τa
βa-τa-αa
ƞa=αa
Figure 3.7: Waiting time
3.10 Objective function
The overall objective function of our scheduling problem is to minimize electricity
bill with optimal use of power from grid and to minimize waiting time (to avoid
frustration of end users). Additionally, optimal integration of RESs is also a key
point to reduce green house gas (GHG) emission. We formulate our objective
function as an optimization function and is modeled as,
min
24
X
t=1 a1
A
X
a=1
(∆a(t)×Υa(t))+a2ϕa(t) (3.29)
s.t:
αssd ≤τssd ≤βssd (3.29a)
αsed ≤τsed ≤βsed (3.29b)
ηa∈[αa, βa−la] (3.29c)
ϕa≤5 (3.29d)
0≤Ψr
t≤Θmax ∀t∈T(3.29e)
Chapter 3 Proposed Model 28
rc
bn< rc,max
b×χbn∀b∈B(3.29f)
rd
bn< rd,max
b×χbn∀b∈B(3.29g)
24
X
t=1
(∆a(t)×χa(t)≤γa(t)) (3.29h)
χa(t)∈[0,1] (3.29i)
where, Υais the electricity cost in each time slot that must be minimized while
keeping waiting time of shiftable appliances minimized. a1and a2are weights of
two parts of objective function and their values are a1, a2∈[0,1] or a1+a2= 1.
It shows that either a1or a2would be 0 or 1. In this work, our major concern is
with electricity cost reduction with maximizing comfort level of end users. For this
purposed model, we assume waiting time of each shiftable appliance not greater
than 5, if operation start time of an appliance is greater than our assumption then
utility pays penalty.
4
HEURISTIC ALGORITHMS
29
Chapter 4 Heuristic algorithms 30
Due to highly volatile load behavior of residential users and intermittent nature
of RESs, we consider our defined problem as non-linear optimization function and
traditional optimization techniques in [5-11] can not handle the complexity of our
proposed model due to their non-flexible nature. Therefore, we apply heuristic al-
gorithms (GA, BPSO and ACO) to solve our designed MKP. These algorithms are
similar due to population based search methods. They move from one population
to another population in number of iterations with improvement using a combina-
tion of deterministic and probabilistic rules. In the following sections, we discuss
some of the latest research works of GA, BPSO and ACO as an optimization
solutions.
4.1 GA
It is most suitable for complex non-linear models where location of the global opti-
mum is a difficult task. Due to its probabilistic nature for development of solution,
GA does not guarantee optimality even when it may be reached [30]. As in [16],
a combined model of RESs and ESSs is proposed to minimize electricity bills for
residential, commercial and industrial areas. Authors use probabilistic model to
design their optimization function and optimal solution is obtained from GA tech-
nique. RTP is used for electricity pricing model. Results show favorable effects
for designed model but in our proposed model, we used GA in more promising
manner to achieve our objectives: electricity cost reduction, PAR minimization,
maximizing user comfort and optimal integration of RESs. In comparison to [16],
we used MKP to balance the demand and supply capacity model and to optimize
our objective function, GA is used. Furthermore, combined TOU with IBR pric-
ing model is used instead of RTP because in real time pricing model, chances of
data loss are increased due to congestion problem. Another model to improve the
efficiency of DSM is proposed in [14]. Authors investigate electricity cost minimiza-
tion and peak formation problem to make DSM efficient. They defined appliance
Chapter 4 Heuristic algorithms 31
scheduling problem for power consumption as optimization problem. GA is used
to get optimal solution subject to cost minimization and PAR reduction and for
electricity pricing model, they used RTP tariff model with IBR. In comparison to
[14], our proposed solution is more effective due to its unique implementation. In
our model, grid capacity is predefined by using MKP to maintain balance between
generation and demand curve. To give compromising results for user satisfac-
tion, we used GA in more appropriate manner. However, they use RTP with IBR
which is not suitable for electricity bill calculation due to increased information
loss chances during high data transfer rates and we use TOU tariff model with
IBR in which date loss problem is diminished. Whereas, we used PV panel to deal
with greenhouse gas emission problem that they totally ignored in their proposed
scheme. Authors in [15], proposed GA based home energy management controller
for single home in residential area. RTP is used for electricity bill calculation.
Results show effectiveness of this model but compared to our model, they ignore
some parameters of DSM model. Detailed GA based energy management con-
troller (GA-EMC) model is shown in algorithm 1 (refer to appendix (7.1)), which
is improved form of algorithm in [15]. Objective function (refer eq. 29) and its
constraints (refer eq. 29a to eq. 29i) are used to find feasible solution. Users input
initial parameters (αa,βa,τaand ρa) for all appliances whereas we treat ηas vari-
able quantity. GA creates a random population initially that consists of certain
number of chromosomes represented by binary string as ON/OFF status of each
appliance. Each chromosome is evaluated using eq. 29. TOU with IBR is embed-
ded as electricity pricing scheme and PV panels model is used as distributed RES
to achieve our objectives. Key modifications that we have implemented in GA al-
gorithm [15] to achieve our objectives and its expected outcome (s) are mentioned
in table. 4.1.
Chapter 4 Heuristic algorithms 32
Table 4.1: Key points of GA-EMC
Enhancement mode Expected Outcome (s)
MKP capacity factor
(refer eq. 29a to 29i)
Limit energy consumption
within certain range
Combined model of
TOU and IBR
(steps 8, 9, ... ,14)
Reduce electricity bills
PAR reduction
Integration of RES
energy model
(steps 22, 23,..., 27)
Reduce peak power plants
Reduction in green house gas emission
Maximize consumer participation
Further minimized electricity cost
4.2 BPSO
BPSO becomes the prominent evolutionary approach to solve global optimization
problems due to its ability to handle non-differential, non-linear multimodal func-
tion, parallel behavior, ease of implementation and good convergence properties
[31]. Recently, many researches have proposed to make efficient energy manage-
ment controller using BPSO. In this regard, a real-time appliance usage model is
proposed in [17]. Authors use BPSO technique to achieve their objectives; electric-
ity bill reduction, peak shaving, valley filling and demand curve smoothing. They
categorized domestic appliances on the basis of characteristics of appliances and
living habits of end users. TOU tariff pricing model is considered as a billing model
for electricity cost calculation. In our work, we categorized household appliances
on the basis of time of usage and appliance power consumption patterns. Com-
bined pricing model, TOU with IBR is used to avoid peak formation and overflow
problem. To make system more efficient, we used PV panels to reduce electricity
bills and avoid environmental pollution that is totally ignored in [17]. Similarly, in
[32], another model is proposed for energy management in DSM based on BPSO.
Main objective of this work is to minimize electricity cost for residential area by
scheduling shiftable load. They ignore user comfort level while investigating DR
program and use TOU pricing model to calculate electricity bills for end users.
Chapter 4 Heuristic algorithms 33
However, in our model, we formulate our objection function by MKP techniques
and BPSO is used to evaluate our designed optimization function. We used TOU
with IBR to avoid peak formation. Thus, our proposed model gives more signif-
icant solution for electricity bill minimization, PAR reduction and user comfort
with optimal integration of RESs. All steps of our proposed model is shown in al-
gorithm. 2 (refer appendix(7.2)). Compared to [31], we modified BPSO according
to customer needs. Feasible operation time ηis calculated by evaluating objec-
tive function (refer eq. 29) and its constraints (refer eq. 29a to eq. 29i). Each
particle in the generation is represented by a binary string denoted as states of
appliances. These particles are updated by individual velocity and particle posi-
tion as in [31]. Our proposed model is applicable for single and multiple homes in
residential areas. In table. 4.2, enhancement points and its expected outcomes for
BPSO algorithms are mentioned.
Table 4.2: Key points of BPSO-EMC
Enhancement mode Expected Outcome (s)
MKP capacity factor
(refer eq. 29a to eq. 29i)
Balance between demand
and supply
Combined model of TOU and IBR
(steps 26 to 37)
Reduce electricity bills
PAR reduction
Integration of RES energy model
(steps 39 to 44)
Reduce peak power plants
Reduction in green house gas emission
Maximize consumer participation
Further minimized electricity cost
4.3 ACO
ACO is a meta-heuristic optimization approach that is used to solve discrete com-
binatorial optimization problems. It has unique properties of self-healing, self-
protection and self-organization [20]. In literature, ACO is used for DSM in many
ways. For-example, authors in [22], investigate congestion management and cost
minimization problems. They formulate their focused problem as a non-linear
Chapter 4 Heuristic algorithms 34
Table 4.3: Key points of ACO-EMC
Enhancement mode Expected Outcome (s)
MKP capacity factor
(refer eq. 29a to eq. 29i)
Balance between demand
and supply
Model of TOU and IBR
(steps 9 to 15)
Reduce electricity bills
and PAR reduction
Integration of RES model
(steps 25 to 30)
Reduce peak power plants
Reduction in greenhouse gas emission
Maximize consumer participation
Minimized electricity cost
programming problem and electricity bill minimization is achieved using ACO. To
our knowledge, ACO implementation in residential area is not done before. In our
work, we use ACO to evaluate the designed optimization function to get optimized
schedules for home appliances. Our scheme gives novel idea to implement ACO as
optimization tool for DSM in residential area. In [33], linear programming is used
to designed the optimization function. Refer to [34], we modified its algorithm for
our designed scenario. Algorithm. 3 (refer appendix(7.3)) gives detailed view of
ACO based EMC (ACO-EMC) model. ACO is used to evaluate objective function
(refer eq. 29) and its constraints (refer eq. 29a to eq. 29i) to get feasible oper-
ational time for all appliances. Our proposed model is applicable for single and
multiple homes in residential areas. Major modifications and possible outcomes
in ACO algorithm in contrast to [34] are given in table. 4.3
5
SIMULATIONS AND RESULTS
35
Chapter 5 Simulations and results 36
To evaluate different performance metrics of EMC, we conduct extensive simu-
lations in MATLAB. In these simulations, we compare our objectives: electricity
bill reduction, energy consumption pattern, PAR reduction, user comfort level and
optimal integration of RESs (PV panels) by using different heuristic algorithms;
GA, BPSO and ACO. Subject to fair comparison, we used TOU tariff model of
Jemena Electricity Networks (VIC) Ltd [35, 36] for residential area with IBR. Ac-
cording to this model, time horizon of 24 hours is divided into three periods as
shown in fig. 5.1. Peak hours are from 3 PM to 9 PM in local time weekdays;
shoulder hours are 7 AM to 3 PM and 9 PM to 10 PM in local time weekdays
and 7 AM to 10 PM in local time weekends while off peak hours are 10 PM to 7
AM local time all days. Price rate for the peak hours, shoulder peak hours and off
peak hours are 15 cent/kWh, 9 cent/kWh and 4 cent/kWh. The purpose of using
dynamic pricing model instead of fixed pricing schemes is to enable customers to
make informed decisions that can be beneficial for them in terms both of electric-
ity bill reduction and comfort level. These TOU tariff model prices are readily
available to customers having advance metering infrastructure.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
2
4
6
8
10
12
14
16
Time (hours)
Cost (cent/kWh)
TOU prices
Figure 5.1: TOU Tariff Model
For simulations, we design a model for residential area in which each home is
equipped with 10 smart appliances and 4 end users. These appliances are further
characterized into fixed, shiftable and elastic appliances. Different parameters of
all appliances can be defined through end users directly or can be obtained from
Chapter 5 Simulations and results 37
learning algorithms. In this work, different parameters are obtained from users in
advance. Appliances with their parametric values that are used in simulations are
shown in table. 5.1, table. 5.2, and table. 5.3, respectively. In table. 5.1, fixed ap-
pliance has only ρaparameter measured in kWh because these are non-manageable
appliances and do not play any role in load scheduling problem. Whereas, other
Table 5.1: Parameters of Fixed Appliances
Appliances ρa(kWh)
Lighting 0.6
Fans 0.75
Clothes iron 1.5
Microwave oven 1.18
Toaster 0.5
Coffee maker 0.8
two categories of appliances; shiftable and elastic appliances are known as schedu-
lable appliances. As, in table. 5.2, the parameters for shiftable appliances are αa,
βa,ξa,ϕaand ρaare kWh. ϕais the unique parameter in shiftable appliance
because these appliances can be interruptible during its length of use. For elastic
appliances, the parameters are αa,βaand ρain kWh are shown in table. 5.3.
Table 5.2: Parameters of Shiftable Appliances
Appliances αa
(hours)
βa
(hours)
ϕa
(hours)
ρa
(kWh)
Washing machine 8 16 5 0.78
Dish washer 7 12 5 3.60
Clothes dyer 6 18 5 4.40
Table 5.3: Parameters of Elastic Appliances
Appliances αa
(hours)
βa
(hours)
ρa
(kWh)
Air conditioner 6 24 1.44
Refrigerator 6 24 0.73
Water heater 6 24 4.45
Space heater 6 24 1.50
Chapter 5 Simulations and results 38
To evaluate the performance of GA-EMC, BPSO-EMC and ACO-EMC, it is re-
quired to set important parameters of these heuristic algorithms. Parameters of
GA-EMC, BPSO-EMC and ACO-EMC are given in table. 5.4, table. 5.5 and ta-
ble. 5.6, respectively.
Table 5.4: GA parametric list
Parameters Values
Population size 10
Selection Roulette wheel
Elite count 2
Crossover 0.8%
Mutation 0.2%
Stopping criteria Max. generation
Max. generation 800
Table 5.5: BPSO parametric list
Parameters Values
Swarm size 200
Max. velocity 4 m/s
Min. velocity -4 m/s
Local pull (c1) 2 N
Global pull (c2) 2 N
Initial momentum weight 1.0 Ns
Final momentum weight 0.4 Ns
Stopping criteria Max. iteration
Max. iteration 600
When we apply these algorithms on our designed objective function, execution
time is different depending on some characteristics. Execution time of an algorithm
Table 5.6: ACO parametric list
Parameters Values
Ant quantity 10
Pheromone intensity factor 2
Visibility intensity factor 6
Evaporation rate 5
Trail decay factor 0.5
Stopping criteria Max. iteration
Max. iteration 600
Chapter 5 Simulations and results 39
is the time in which an algorithm completes its functionality. BPSO executes in
more time than GA and unscheduled EMC and ACO takes more time to complete
its functionality than BPSO, GA and unscheduled EMCs. Execution time for all
models is shown in table. 5.7.
Table 5.7: Execution time
Execution time Values (seconds)
Without EMC 0.0983
GA-EMC 1.0191
PSO-EMC 24.1933
ACO-EMC 77.7434
5.1 Energy consumption pattern and Electricity
bill reduction
Let the knapsack capacity of power grid is 20 kWh for each time slot per day. To
integrate distributed energy sources, we have considered PV panel as a source of
renewable energy and batteries as storage system. We use solar panel for power
generation which meets 50% of the total load demand. Each smart house is
equipped with 1-kW PV arrays. This translates 500 W to 10 kW energy gen-
eration capacities. The purpose of integrating RESs with GA-EMC, BPSO-EMC
and ACO-EMC is to reducing greenhouse gas and to give further advantage to
end users by minimizing their electricity bills. Energy consumption pattern using
GA-EMC, BPSO-EMC and ACO-EMC without or with RES is shown in fig. 5.2.
It is shown in fig. 5.2 that maximum energy consumption value are limited to
19.4250 kWh, 18.6750 kWh, 19.4250 kWh, 19.4250 kWh, 18.6450 kWh, 18.8250
kWh and 18.2450 kWh for without EMC, GA-EMC, BPSO-EMC, ACO-EMC,
GA-EMC (RES), BPSO-EMC (RES) and ACO-EMC (RES), respectively. It is
concluded that energy consumption pattern of all models are under predefined
knapsack capacity of grid. It is important to notice that GA-EMC acts slightly
Chapter 5 Simulations and results 40
better than BPSO-EMC and ACO-EMC whereas, ACO-EMC (RES) performed
well than others by reducing maximum energy consumption value. During high
energy consumption hours, consumers use energy from RES and ESS to further
minimize utility electricity bills. Results show that electricity consumption and
loses can be further optimally reduced when consumers smartly handle their elec-
tricity usage and accomplish their energy needs by using energy in an intelligent
way. The maximum value of electricity bill in unscheduled model is 266 cent as
2 4 6 8 10 12 14 16 18 20 22 24
0
5
10
15
20
Time (hours)
Energy consumption (kWh)
Without EMC
With GA−EMC
With BPSO−EMC
With ACO−EMC
With GA−EMC(RES)
With BPSO−EMC(RES)
With ACO−EMC(RES)
Figure 5.2: Energy consumption (kWh)
shown in fig. 5.3. It is reduced to 81 cent in the case of GA-EMC while it is
reduced from 266 cents to 98 cent in BPSO-EMC and to 114 cent in ACO-EMC.
During peak hours (16-22), sufficient electricity cost reduction is shown for all
designed models (GA-EMC, BPSO-EMC and ACO-EMC). GA-EMC acts more
effectively than BPSO-EMC and ACO-EMC in achieving our designed objective
of electricity cost reduction due to its unique parameters (crossover and mutation)
and BPSO-EMC acts slightly better than ACO-EMC due to its characteristics of
local and global exploration. When we integrate RES with these models electricity
bills is further reduced due to consumer participation. GA-EMC, BPSO-EMC and
ACO-EMC with RES models show maximum values at 75 cent, 90 cent and 98
cent as in fig. 5.3. Now, total electricity bill reduction per day of a single home for
all models is shown in fig. 5.4. Electricity bill reduction in the case of GA-EMC,
Chapter 5 Simulations and results 41
2 4 6 8 10 12 14 16 18 20 22 24
0
50
100
150
200
250
300
Time (hours)
Electricity bill (cent)
Without EMC
With GA−EMC
With BPSO−EMC
With ACO−EMC
With GA−EMC(RES)
With BPSO−EMC(RES)
With ACO−EMC(RES)
Figure 5.3: Electricity bill (cent)
BPSO-EMC ans ACO-EMC is 48.79%, 40.43% and 28.26% respectively. This
shows that GA-EMC is more cost-efficient than BPSO-EMC and ACO-EMC. The
Figure 5.4: Electricity bill reduction per day
total electricity bill reduction per day for all designed models with RES is shown
in fig. 5.5. Here, it is clear that GA-EMC performs more effectively than BPSO-
EMC and ACO-EMC with the integrated RES model. In the case of GA-EMC
with RES, the per day electricity bill is 752 cents, whereas, in case of BPSO-EMC
with RES, it is 940 cents and in ACO-EMC, it is 1046 cents. Therefore, electricity
bill reduction using GA-EMC with RES is 65%, BPSO-EMC with RES is 57%
and ACO-EMC with RES is 52%. All above results are for single home but what
Chapter 5 Simulations and results 42
Figure 5.5: Electricity bill per day
happens if we increase number of homes. To see the effect, we have considered 50
homes for which energy consumption and electricity cost reduction are measured
in a particular day. From fig. 5.6, it is verified that our designed models achieved
significant results. As these controllers designed to optimize starting time of all
appliances while satisfying constraints of objective function in 24 hours so that
residential users gets benefic by reducing their electricity bills and utilities get
advantage by keeping demand under power capacity of power grid.
5 10 15 20 25 30 35 40 45 50
500
1000
1500
2000
2500
Number of homes
Electricity bill (cent/day)
Without EMC
With GA−EMC
With BPSO−EMC
With ACO−EMC
Figure 5.6: Electricity bill (cent/day) for 50 users
Chapter 5 Simulations and results 43
5.2 PAR
Performance of all the designed models (GA-EMC, BPSO-EMC and ACO-EMC)
with respect to PAR reduction is shown in fig. 5.7. It shows that PAR is signif-
icantly reduced for GA-EMC, BPSO-EMC and ACO-EMC as compared to the
unscheduled case because these are designed to avoid peak formation in any hour
of a day. Results prove that our proposed models effectively tackle the peak for-
mation problem. PAR curves for GA-EMC, BPSO-EMC and ACO-EMC describe
that power consumption of appliances are optimally distributed in 24 hours with-
out creating peak in peak hours (16-22) of a day. BPSO-EMC has high PAR than
ACO-EMC and GA-EMC and GA-EMC is more effective in PAR reduction due
to its ability to generate new population of more feasible solution using crossover
and mutation. Peak formation is a major drawback in traditional electric power
system as it causes customer to pay high electricity bills and utility suffers high
demand that causes blackout or load shedding. We have used combined model of
TOU and IBR for electricity billing to avoid peak formation via giving information
to consumers. The performance of these algorithms in our scenario is improved
due to power capacity factor that cause utilities to fulfill the demand of customers
and gives chance end user to reduce electricity bills.
2 4 6 8 10 12 14 16 18 20 22 24
0
50
100
150
200
250
Time (hours)
PAR curve
Without EMC
With GA−EMC
With BPSO−EMC
With ACO−EMC
Figure 5.7: PAR curve
Chapter 5 Simulations and results 44
5.3 Waiting time
User comfort is related to both electricity bill and waiting time of an appliance. In
order to achieve lower electricity bills, smart users must operate their appliances
according to optimal schedule of EMC. During scheduling horizon of shiftable
appliances, operational time is not fixed due to price variation in dynamic pricing
models. Generally, it is observed that electricity cost reduction and waiting time
show inverse relationship. By applying waiting time constraints (refer eqs. 29c and
29d) on the objective function (refer eq. 29), we have enhanced the performance
of EMC in terms of user comfort and electricity bill reduction. In fig. 5.8, it
is shown that electricity bill is high if rate of waiting time is zero and it is low
with increase in rate of waiting time for all models. Performance of GA-EMC
is much better than other due to minimize effect of tradeoff. The purpose of
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
5
10
15
20
25
30
35
Waiting time rate
Electricity cost (cent)
Without EMC
With GA−EMC
With BPSO−EMC
With ACO−EMC
Figure 5.8: Possible trade off between electricity cost and waiting time
scheduling algorithm to delay the operation of any appliance is to optimize the
system according to the designed objective function. When energy consumption
of an appliance is more than the power capacity of a particular hour or during
high peak hour, appliance scheduler shifts the appliance to another time slot. By
ignoring waiting time factor, optimized scheduling can not be achieved. On the
other hand, our proposed scheme gives an effective solution. Results of all models
are significant to achieve our goals, however, GA-EMC based model is proven
Chapter 5 Simulations and results 45
more effective than the BPSO-EMC and ACO-EMC models. All the results are
summarized in table. 5.8.
Table 5.8: Summarized results
Cases Total load (kWh/day) Total cost (cent/day) PAR reduction Cost reduction (%)
without RESs
Cost reduction (%)
with RESs
Without EMC 258 2201 244.6747 - -
GA-EMC 258 1127 81.8808 48.79 65
BPSO-EMC 258 1311 95.2281 40.43 57
ACO-EMC 258 1579 127.5380 28.26 52
5.4 Parametric tuning for all models
In this section, tuning effects for different parameters of three heuristic techniques
(GA, BPSO and ACO) on our designed model is discussed in detail. Results are
summarized in table. 5.9, table. 5.10 and table. 5.11.
Table. 5.9, shows how effectively power consumption, electricity bill with RES,
PAR reduction, electricity bill with RES and execution time values are changed
when parameters of GA altered. In our work, population size, maximum gener-
ation, crossover and mutation parameters are analyzed at different values while
keeping others constant as defined in table IV. We consider three values of popu-
lation size (200, 1000 and 2000), four values for maximum generation (2000, 1500,
1000 and 800) and three values for each crossover (1, 0.8 and 0.6) and mutation
(0, 0.2 and 0.4). After evaluating our designed objective function for each value
of considered parameters, we get results that are summarized here. Maximum
value of scheduled load is in the range of 12.7480 kWh to 18.6750 kWh that is less
than 19.4550 kWh which is maximum value unscheduled load and also in limits of
assumed power grid capacity of 20 kWh whereas, maximum value of electricity bill
Chapter 5 Simulations and results 46
is in the range of 55 cent to 84 cent that is optimized results than unscheduled cost
that is 266 cent. For PAR reduction, its values are change between 38.4030 and
215.6530. In the way, electricity cost per day without using RES is 1.125×103cent
that is reduced cost than 2.201×103cent and when we integrate RES with our de-
signed model than electricity cost reduction is between 5.57×102cent and 1.05×102
cent. Execution time is greatly effected when population size and maximum gener-
ation is changed. As, it slightly increase with increase in number of population size
and generation. It is clearly notice from our tables that performance of GA-EMC
is more significant than other two models due to its evolutionary nature. As in
GA-EMC, population is randomly generated depending upon nature of problem
whereas, selection is done using designed objective function and roulette wheel cri-
teria, moreover, crossover and mutation plays key role to generate new population
which is fitter than the older one. Thus, GA-EMC gives more optimized results
than other two model within less time. Performance evaluation for different
parameters of BPSO-EMC is summarized in table. 5.10. For analy-
sis, we consider three different values for swarm size (10, 20 and 40),
three values for maximum iteration (1800, 1500 and 600) and four
values for each local pull factor c1(0, 1, 2 and 3) and global pull
factor c2(4, 3, 2 and 1) and remaining parameters are mentioned in
table V. It is proven from results that BPSO-EMC performs well but
not as good as GA-EMC. Maximum optimized energy consumption
values for BPSO-EMC model are between 15.1543 kWh and 19.3106
kWh that obey power capacity of grid whereas, maximum value for
Chapter 5 Simulations and results 47
Table 5.9: Parameter Evaluation for GA
Pop.
size Max. gen Cros.
(%)
Mut.
(%)
Max. energy
(kWh)
Max. cost
(cent) Max. PAR Cost (cent/day)
without RES
Cost (cent/day)
with RES
Exe. time
(sec.)
200
2000
1500
1000
800
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
13.0650
17.1450
18.6450
17.1450
18.6450
17.1450
17.1450
17.1450
18.6450
12.7450
18.6750
18.6450
57
74
81
74
81
74
74
74
81
55
82
81
44.9548
85.9937
104.1863
81.5619
81.7315
115.0993
58.9935
71.0060
178.6347
38.4030
81.8808
81.4787
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.051 ×103
7.52 ×102
8.02 ×102
6.03 ×102
9.65 ×102
6.05 ×102
7.529 ×103
7.529 ×102
8.83 ×102
6.26 ×102
7.529 ×102
7.20 ×102
2.4326
2.3215
2.2022
1.8583
1.7596
2.0181
1.2520
1.2366
1.2059
1.0403
1.0191
1.0019
1000
2000
1500
1000
800
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
17.1450
18.6450
18.6450
17.1450
18.6450
18.6450
18.6450
18.6450
18.6450
17.1450
18.6450
18.6450
75
84
81
81
75
81
81
81
81
75
81
81
82.0499
215.6530
158.4000
66.4213
136.2192
124.1276
81.7315
110.8996
158.4000
75.1562
94.7048
158.4000
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
7.53 ×102
6.28 ×102
7.17 ×102
7.20 ×102
6.78 ×102
5.57 ×102
8.83 ×102
6.39 ×102
7.06 ×102
9.03 ×102
9.65 ×102
8.87 ×102
3.7997
3.6981
3.0158
2.2721
2.1938
2.3219
1.5643
1.5487
1.4883
1.3744
1.3384
1.3195
2000
2000
1500
1000
800
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
1
0.8
0.6
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
0
0.2
0.4
18.6450
17.9250
18.6450
17.9250
18.6450
18.6450
17.1450
18.6450
18.6450
17.1450
18.8450
18.6450
81
78
81
78
81
81
74
81
81
75
81.48
81.48
81.7315
120.3357
156.7356
90.4732
137.4787
212.5796
94.4845
110.8936
136.2192
74.3415
93.5172
127.3058
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
1.127 ×103
7.20 ×102
7.29 ×102
6.35 ×102
8.14 ×102
8.02 ×102
8.02 ×102
7.53 ×102
6.39 ×102
6.25 ×102
6.78 ×102
8.84 ×102
5.62 ×102
3.2209
3.1092
3.0691
2.6474
2.6172
3.0554
2.1044
2.0605
2.0118
1.8535
1.8292
1.7825
cost reduction is in the range of 95 cent to 113 cent for different values
of parameters. These variations are due to local pull and global pull
parameter that directly affects optimization phenomena to evaluate
objective function. Similarly, in the case of PAR reduction in which
values are reduced as compared to unscheduled model. Electricity
cost reduction for a particular day without RESs is 1.311 ×103cent
and with RES it is furthered. Execution time is highly increases with
increase in the number of swarm particles due to execution of step in
BPSO for each individual. As, GA is different from BPSO and ACO
Chapter 5 Simulations and results 48
Table 5.10: Parameter Evaluation for BPSO
Swarm.
size
Max.
iter
Loc. pull
(c1)
glob. pull
(c2)
Max. energy
(kWh)
Max. cost
(cent) Max. PAR Cost (cent/day)
without RES
Cost (cent/day)
with RES
Exe. time
(sec.)
10
1800
1200
600
0
1
2
3
0
1
2
3
0
1
2
3
4
3
2
1
4
3
2
1
4
3
2
1
18.5684
17.2285
17.2783
17.9422
17.5230
19.3106
15.1543
18.4918
18.8250
19.1240
18.6250
18.8250
105
105
103
105
101
99
96
101
105
101
99
105
95.6426
95.1082
135.1126
110.824
136.4045
158.1255
180.9130
105.5971
133.3874
201.4372
95.2281
186.9246
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
8.45 ×102
9.40 ×102
9.41 ×102
8.67 ×102
9.39 ×102
8.52 ×102
1.03 ×102
9.40 ×102
8.56 ×102
1.037 ×103
9.40 ×102
9.33 ×102
47.6986
47.8157
48.4773
48.1125
47.1692
47.1188
57.7714
48.0297
24.0676
27.5235
24.1933
24.1033
20
1800
1200
600
0
1
2
3
0
1
2
3
0
1
2
3
4
3
2
1
4
3
2
1
4
3
2
1
17.4665
18.6450
17.7227
16.8228
18.6257
19.1387
16.8228
16.4068
18.4999
18.7275
17.2525
17.3655
101
101
96
96
104
111
101
101
101
100
96
95
120.5838
209.2746
190.8179
181.8585
203.6886
149.9150
122.7205
216.3967
131.4721
205.5694
182.0045
131.0050
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
9.41 ×102
8.55 ×102
9.42 ×102
9.39 ×102
7.77 ×102
7.72 ×102
8.36 ×102
9.35 ×102
8.62 ×102
7.77 ×102
8.57 ×102
9.36 ×102
142.0157
144.9287
143.9084
143.2447
96.9877
96.5602
97.2115
98.1963
45.6568
44.9692
45.8085
45.1651
40
1800
1200
600
0
1
2
3
0
1
2
3
0
1
2
3
4
3
2
1
4
3
2
1
4
3
2
1
17.4665
18.3676
17.7227
18.9652
17.6810
19.1658
17.5402
18.3967
18.4999
18.7275
17.2525
16.3655
111
102
104
113
111
111
113
102
96
111
113
111
156.9369
99.0604
166.8758
154.7958
100.4732
97.4787
100.5796
186.1581
150.8515
114.8936
156.8815
186.3058
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
1.311 ×103
7.41 ×102
6.10 ×102
6.08 ×102
8.54 ×102
6.10 ×102
6.43 ×102
6.57 ×102
8.50 ×102
6.06 ×102
5.24 ×102
6.89 ×102
8.57 ×102
287.0919
306.9691
286.2626
472.8304
289.2429
439.8685
286.8129
196.6268
95.8205
95.7764
95.2590
95.5248
due to its unique parameters (crossover and mutation) due to which
it complete its functionality within less time than BPSO and ACO.
While in BPSO, local and global exploration enables the algorithm
(refer appendix (b)) to give feasible solution while obeying defined
constraints (eq. 29a to eq. 29i). Both global exploration and local
exploration is affected by updating particle’s velocity and position.
From table. 5.11, simulation results for tuning of different parame-
ters are shown. In the case of ACO-EMC model, we have considered
two values of ant population (10 and 20), three number of iterations
(2000, 1500 and 600), three values for visibility intensity factor (6,
Chapter 5 Simulations and results 49
10 and 15) and two trial decay values (0.5 and 1). ACO-EMC acts
better than unscheduled models but its performance is not as good
as other two models (GA-EMC and BPSO-EMC). Maximum opti-
mized value after evaluating our objective function is in the range
of 17.5064 kWh to 19.4250 kWh that is under predefined power grid
capacity and electricity bill reduction is between 111 cent and 127
cent that is much less than 266.3492 cent for the cost of unscheduled
model. PAR reduction is in the range of 95.5807 to 215.6530. In the
case of ACO-EMC, electricity bill reduction is 1.58 ×103cent and
with RES it is more reduced to 1.00 ×102cent. Execution time of
ACO-EMC is very high than others due to its pheromone update for
each ant and number of step for designed algorithm (refer appendix
(c)) is repeated till convergence to feasible solution.
Chapter 5 Simulations and results 50
Table 5.11: Parameter Evaluation for ACO
Ant
pop.
Max.
iter.
Vis. int.
(β1)
Tra. dec.
factor
Max. energy
(kWh)
Max. cost
(kWh) Max. PAR Cost (cent/day)
without RES
Cost (cent/day)
with RES
Exe. time
(sec.)
10
2000
1500
600
6
10
15
6
10
15
6
10
15
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
19.4250
18.6450
19.4250
18.6450
18.3640
18.6450
17.9250
19.4250
18.6450
19.4250
17.9850
19.4250
18.9750
18.2450
18.2450
19.4250
18.2450
18.6750
127
114
119
122
114
114
117
127
122
127
118
127
114
115
112
119
106
115
122.3898
137.3747
132.6316
127.6530
95.5807
127.5206
122.3898
132.6316
215.6530
132.6316
122.7994
132.6316
127.5380
124.5747
127.5253
132.6312
215.6530
127.5655
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.578 ×103
1.414 ×103
1.465 ×103
9.64 ×102
9.99 ×102
1.464 ×103
9.90 ×102
9.26 ×102
6.67 ×102
9.13 ×102
1.025 ×103
8.266 ×102
8.99 ×102
1.175 ×103
6.62 ×102
1.046 ×103
7.01 ×102
6.59 ×102
9.44 ×102
363.8474
282.1364
311.5912
253.6215
292.9011
283.5336
208.0962
201.0578
211.9933
203.0963
220.3764
211.6243
77.4729
111.5901
78.5933
75.3730
77.7434
76.6911
20
2000
1500
600
6
10
15
6
10
15
6
10
15
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
0.5
1
19.4250
19.4250
18.6450
17.5046
18.6450
19.4250
19.4250
18.6450
19.4250
17.5046
18.6450
19.4250
18.6450
19.4250
18.6450
17.5064
18.6450
19.4250
119
127
122
130
128
127
119
122
111
132
127
111
122
119
104
132
122
132
132.6316
132.6316
215.6341
127.3058
211.0265
132.6316
132.6316
127.3018
132.6316
127.3065
211.0265
132.6316
161.8373
132.6316
211.6541
205.1154
127.5107
211.6541
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
1.579 ×103
6.10 ×102
5.58 ×102
6.58 ×102
7.65 ×102
7.07 ×102
6.45 ×102
7.00 ×102
7.82 ×102
1.14 ×102
7.65 ×102
7.65 ×102
7.85 ×102
6.64 ×102
6.72 ×102
9.49 ×102
9.55 ×102
7.38 ×102
6.64 ×102
1582.6952
1434.3018
1525.2417
1404.2934
1621.2112
1455.0140
1125.8606
1387.2139
1185.5512
1125.3564
1303.6541
1103.6541
622.5746
521.5444
419.7896
411.8112
417.8833
401.9631
6
CONCLUSION AND FUTURE
WORK
51
Chapter 6 Conclusion and Future Work 52
In this dissertation, we have presented an efficient DSM model for
residential energy management system in order to avoid peak for-
mations while decreasing the utilities electricity bill by preserving
user comfort level within acceptable limits. We evaluated our de-
signed objective function by using three heuristic algorithms (GA,
BPSO and ACO) and used combine pricing models, TOU tariff and
IBR model for electricity bill calculation. From the results, it is
clearly justified that our proposed model works more efficiently with
GA-EMC than BPSO-EMC and ACO-EMC in term of electricity
bill reduction, minimizing PAR while considering user satisfaction.
GA-EMC executes with less execution time than others as GA-
EMC>BPSO-EMC>ACO-EMC. Additionally, results of extended
model for multiple users are shown to validate our work in terms of
relative scalability.
In future, we will focus on human behavior to achieve comfort level
of consumer and to minimize frustration cost and improve security
and privacy issues between end user and utility. We will also work on
the different optimization methods so that more accurate data trans-
formation is achieved with in less execution time and computational
complexity.
Appendix
Appendix (a): GA-EMC
Algorithm 1 GA-EMC algorithm
1: Initialize all parameters (αa,βa,τa,ρa)
2: For all users n ∈N do
3: For all appliances a ∈A do
4: For all time slots t ∈T do
5: Randomly generate population represents the patterns of appliances.
6: for p=1:popsize do
7: Each individual evaluate the fitness function using eq. 29
8: F=fitnessf unction
9: if (F(p)< F (p−1))&&(E(t)< γ(t)) then
10: F(p) = F(p)
11: if ϕausing(28) ≥5&&(η≤τa)then
12: start an appliance else wait till low peak hours
13: else
14: F(p) = F(p−1)
15: Pattern(1,:)=popnew(1,p)
16: if AppP atterna== 1 then
17: τa=τa−1
Generate new population Select pair a, b by Roulette selection criteria
18: if Pc> rand then
19: crossover(a, b)
20:
21: if Pm> rand then
22: mutate(a, b)
23: popnew(popsize,N) Repeat until stopping criteria
24: if E(t) is high then
25: Θ(t) energy used
26: else
27: E(t)
28: Return best individuals
Appendix
Appendix (b): BPSO-EMC
Algorithm 2 BPSO-EMC algorithm
1: Initialize all parameters (αa,βa,τa,ρa)
2: For all users n ∈N do
3: For all appliances a ∈A do
4: Initialize particles velocities to 0 and individual best to current population
5: for S=1:swarmsize do
6: Each particle evaluate the fitness function using eq. 3.29 F=fitnessf unction
7: if (F(S)< F (S−1))&&(E(t)< γ(t)) then
8: Fbest =F(S)
9: if ϕausing(28) ≥5&&(η≤τa)then
10: start an appliance
11: else
12: wait till low peak hours
13: else
14: F(S) = F(S−1)
15: Pattern1(1,:)=swarmpop(1,S)
16: if AppP atterni== 1 then
17: τa=τa−1
18: Initialize global best to Fbest
19: Each individual update velocity and position of each particle refer [31]
20: For Each bit
21: if rand < 1
1+e−vthen
22: bit ←1
23: else
24: bit ←0
Repeat until maximum iteration reached
25: if E(t)is high then
26: Θ(t) energy is used
27: else
28: E(t) used
Appendix
Appendix (c): ACO-EMC
Algorithm 3 ACO-EMC algorithm
1: Initialize all parameters (αa,βa,τa,ρa) For all users n ∈N do For all appliances
a∈A do For all time slots t ∈T do
2: for An=1:antpop do
3: Each ant evaluate the fitness function using eq. 29
4: F=fitnessf unction
5: if (F(An)< F (An −1))&&(E(t)< γ(t)) then
6: Fbest =F(An)
7: if ϕausing(28) ≥5&&(η≤τa)then
8: start an appliance
9: else
10: wait till low peak hours
11: else
12: F(An) = F(An −1)
13: Pattern2(1,:)=antpop(1,An)
14: if P attern2An== 1 then
15: τa=τa−1
16: Local update pheromone for each ant refer [34]
17: Choose best solution so far
18: Global update pheromone for each ant refer [34]
19: Repeat until maximum iteration reached
20:
21: if E(t) is high then
22: Θ(t) energy is used
23: else
24: E(t) used
BIBLIOGRAPHY
[1] V. Gungor, D. Sahin, T. Kocak, S. Ergut, C. buccella, C. Cecati
and G. Hancke, “Smart Grid Technologies: Communication Tech-
nologies and Standards”, Industrial Informatics, IEEE Transac-
tion, Vol. 7, No. 4, pp. 529-539, Nov. 2011.
[2] M. Hashmi, S. Hanninen, and K. Maki, “Survey of Smart
Grid Concepts, Architectures and Technological Demonstrations
Worldwide”, IEEE PES Conference, Medellin, Oct. 2011.
[3] L. Gelazanskas, and K. A. A. Gamage, “Demand Side Manage-
ment in Smart Grid: A Eeview and Proposals Forfuture Direc-
tion”, Sustainable Cities and Society, Vol. 11, pp. 22–30, Feb.
2014.
[4] P. Siano, “Demand Response and Smart Grids — A Survey”, Re-
newable and Sustainable Energy Reviews, Vol. 30, pp. 461–478,
Feb. 2014.
[5] A. Molderink, V. Bakker, M. G. Bosman, J. L. Hurink, and G.
56
BIBLIOGRAPHY 57
J. Smit, “Domestic Energy Management Methodology for Opti-
mizing Efficiency in Smart Grids”, IEEE PowerTech, Bucharest,
pp. 1-7, 28 Jun(2008) - 2 Jul (2009)
[6] T. Sousa, H. Morais, Z. Vale, P. Faria, and J. Soares, “Intelligent
Energy Resource Management Considering Vehicle-to-Grid: A
Simulated Annealing Approach”, IEEE Transaction Smart Grid,
Vol. 3, No. 1, pp. 535-542, Mar. 2012.
[7] J. Soares, T. Sousa, H. Morais, Z. Vale, and P. Faria, “An Op-
timal Scheduling Problem in Distribution Networks Considering
V2G”, in Proc. IEEE CIASG, Paris, 11-15 Apr. 2011.
[8] K. M. Tsui and S. C. Chan, “Demand Response Optimization
for Smart Home Scheduling Under Real-Time Pricing”, IEEE
Transaction Smart Grid, Vol. 3, No. 4, pp. 1812-1821, Dec. 2012.
[9] Z. Zhu, J. Tang, S. Lambotharan, W. H. Chin, and Z. Fan, “An
Integer Linear Programming Based Optimization for Home De-
mand Side Management in Smart Grid”, in Proc. IEEE PES
ISGT, Washington, DC, 16-20 Jan. 2012.
[10] P. O. Kriett and M. Salani, “Optimal Control of a Residential
Microgrid”,8th World Energy System Conference, Vol. 82, No. 1,
pp. 321-330, Jun. 2012.
BIBLIOGRAPHY 58
[11] J. Wang, Z. Sun, Y. Zhou, and J. Dai, “Optimal Dispatching
Model of Smart Home Energy Management System”, in Proc.
IEEE PES ISGT ASIA, Tianjin, 21-24 May 2012.
[12] D. Maringer, “Portfolio Management with Heuristic Optimiza-
tion”, Springer, 2004.
[13] T. Logenthiran, D. Srinivasan, and T. Z. Shun, “Demand Side
Management in Smart Grid using Heuristic Optimization”, IEEE
Transaction Smart Grid, Vol. 3, No. 3, pp. 1244-1252, Sept. 2012.
[14] Z. Zhao, W. C. Lee, Y. Shin and K. Song, “An Optimal Power
Scheduling Method for Demand Response in Home Energy Man-
agement System”, IEEE Transaction Smart Grid, Vol. 4, No. 3,
pp. 1390-1400, Sept 2013.
[15] M. A. Khan, N. Javaid, A. Mahmood, Z. A. Khan and N. Al-
rajeh, “A Generic Demand-Side Management Model for Smart
Grid”, International Journal of Energy Research, Vol. 39, No. 7,
pp. 954–964, Jun. 2015.
[16] A. Arabali, M. Ghofrani, M. Etezadi-Amoli, M. S. Fadali, and
Y. Baghzouz, “Genetic-Algorithm-Based Optimization Approach
for Energy Management”,IEEE Transaction on Power Delivery,
Vol. 28, No. 1, Jan 2013.
BIBLIOGRAPHY 59
[17] Y. Zhou, Y. Chen, G. Xu and Q. Zhang, “Home Energy Man-
agement with PSO in Smart Grid”,Industrial Electronics (ISIE),
IEEE 23rd International Symposium on, Istanbul, pp. 1666 -
1670, Jun. 2014.
[18] M. A. A. Pedrasa, T. D. Spooner and I. F. MacGill, “Scheduling
of Demand Side Resources Using Binary Particle Swarm Opti-
mization”, IEEE Transactions on Power Systems, Vol. 24, No. 3,
pp. 1173 - 1181, Aug. 2009.
[19] J. Soares, M. Silva, T. Sousa, Z. Vale, and H. Morais, “Dis-
tributed Energy Resource Short-Term Scheduling using Signaled
Particle Swarm Optimization”, 8th World Energy System Con-
ference, Vol. 42, No. 1, pp. 466-476, Jun. 2012.
[20] T. Dethlefs, T. Preisler and W. Renz, “Ant-Colony based
Self-Optimization for Demand-Side-Management”, Conference:
SmartER Europe, Essen, Feb. 2015.
[21] J. Hazra, K. Das and D. P. Seetharam, “Smart Grid Congestion
Management through Demand Response”, 2012 IEEE Third In-
ternational Conference on Smart Grid Communications (Smart-
GridComm), Tainan, pp. 109 - 114, 5 - 8 Nov. 2012.
[22] B. Liu, J. Kang, N. Jiang and Y. Jing, “Cost Control of
the Transmission Congestion Management in Electricity Systems
BIBLIOGRAPHY 60
based on Ant Colony Algorithm”, Energy and Power Engineer-
ing, Vol. 3, No. 1, pp. 17 - 23, Feb. 2011.
[23] N. Kumaraguruparan, H. Sivaramakrishnan and S. Sapatnekar,
“Residential task scheduling under dynamic pricing using the
multiple knapsack method”, IEEE Conference, Washington, DC,
pp. 1-6, Jan. 2012.
[24] O. A. Sianaki, O. Hussian and A.R. Tabesh, “A Knapsack Prob-
lem Approach for Achieving Efficient Energy Consumption in
Smart Grid for End-user Life Style”,IEEE conference, Waltham,
MA, Sept. 2010.
[25] C. Ogwumike, M. Short and M. Denai, “Near-Optimal Schedul-
ing of Residential Smart Home Appliances using Heuristic Ap-
proach”, IEEE International Conference on Industrial Technol-
ogy, Seville, Spain, 2015.
[26] H. Miao, X. Huang and G. Chen, “A Genetic Evolutionary Task
Scheduling Method for Energy Efficiency in Smart Homes”, In-
ternational Journal of Electrical Engineering, Vol. 7, No. 5, pp.
1827-6660, 2012.
[27] A. Barbato and A. Capone, “Optimization Models and Methods
for Demand-Side Management of Residential Users: A Survey”,
Sept. 2014
BIBLIOGRAPHY 61
[28] C. Monteiro, T. Santos, L. A. F. Jimenez, I. J. R. Rosado and
M. S. T. Olarte, “Short-Term Power Forecasting Model for Pho-
tovoltaic Plants based on Historical Similarity”, Energies, May
2013.
[29] H. Kellerer, U. Pferschy, and D. Pisinger, “Knapsack Problems”.
Springer, 2004.
[30] S. Mardle and S. Pascoe, “An Overview of Genetic Algorithms
for the Solution of Optimisation Problems”, CHEER, Vol. 13,
1999.
[31] H. M. Lugo-Cordero, A. Fuentes-Rivera, R. K. Guha, and E.
I. Ortiz-Rivera, “Particle Swarm Optimization for Load Balanc-
ing in Green Smart Homes”, Evolutionary Computation (CEC),
IEEE Congress , pp. 715 - 720, 2011
[32] T. Remani, E.A. Jasmin and I. T. P. Ahamed, “Load Scheduling
with Maximum Demand Using Binary Particle Swarm Optimiza-
tion”, IEEE International Conference on Technological Advance-
ments in Power and Energy, Kollam, 24-26 Jun. 2015.
[33] I. K. Yu and Y. H song, “A Novel Short-term Generation
Scheduling Technique of Thermal Units using Ant Colony Search
Algorithms”, International Journal of Electrical Power and En-
ergy Systems, Vol. 23, No. 6, pp. 471–479, Aug. 2001.
BIBLIOGRAPHY 62
[34] S. Rahim, S. A. Khan, N. Javaid, N. Shaheen, Z. Iqbal and
G. Rehman, “Towards Multiple Knapsack Problem Approach
for Home Energy Management in Smart Grid”, 8th IEEE In-
ternational Conference on Network-Based Information Systems,
Taipei, Taiwan, 2015.
[35] “Jemena Electricity Networks (VIC) Ltd - Network Tariffs For
The 2015 Calendar Year (Exclusive of GST)”
