Flexibility Provisions from Energy

Storage Systems and Loads in Smart

Grid

by

Omar Alrumayh

A thesis
presented to the University of Waterloo
in fulfillment of the
thesis requirement for degree of
Doctor of Philosophy
in
Electrical and Computer Engineering

Waterloo, Ontario, Canada, 2021

© Omar Alrumayh 2021

 Examining Committee Membership

The following served on the Examining Committee for this thesis. The decision of the
Examining Committee is by majority vote.

External Examiner: Bala Venkatesh, Professor,

Dept. of Electrical and Computer Engineering,
Ryerson University

Supervisor: Kankar Bhattacharya, Professor,

Dept. of Electrical and Computer Engineering,
University of Waterloo

Supervisor: Steven Wong, Adjunct Associate Professor,

Dept. of Electrical and Computer Engineering,
University of Waterloo,
Home Affiliation: Natural Resources Canada

Internal Member: Magdy Salama, Professor,

Dept. of Electrical and Computer Engineering,
University of Waterloo

Internal Member: Ramadan El-Shatshat, Lecturer,

Dept. of Electrical and Computer Engineering,
University of Waterloo

Internal-External Member: Ali Elkamel, Professor,

Dept. of Chemical Engineering,
University of Waterloo

ii
 Author’s Declaration

I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis,
including any required final revisions, as accepted by my examiners.

I understand that my thesis may be made electronically available to the public.

iii
 Abstract

Electric power systems are experiencing a movement toward increasing the share of
renewable energy sources (RESs), while having to cope with the retirement of
conventional generating units to facilitate an eco-friendly system. However, the
uncertainty and variability associated with RESs and the demand, call for additional
sources of flexibility. Residential, commercial, and industrial loads are a potential source
of flexibility in power systems. In addition, recent deployments of energy storage
systems (ESSs) can contribute significantly to power system flexibility. Therefore, the
effective management of flexible sources can lead to an improved power system operation.
This thesis investigates options for capturing the flexibility of residential loads and
ESSs in a power distribution system. A two-stage optimization framework is developed
wherein multiple home energy management systems (HEMSs) simultaneously optimize
their respective energy consumption patterns, and determines their flexibility provisions,
which are communicated to the local distribution company (LDC). A flexibility evaluation
approach is developed to estimate the residential energy hub (REH) flexibilities at each
bus in the distribution system. Intra-hour flexibility indices are calculated to represent the
REHs’ willingness to alter their consumptions. Different clusters of residential customers
are considered, classified by their ownership of photovoltaic (PV) panels and ESSs, and
their preferred objectives. The LDC aggregates the controllable demand profiles and the
flexibilities of each HEMS to optimize its operational performance and hence determines
peak reduction signals that are sent to the HEMSs. Studies are carried out considering
a 33-bus distribution system coordinating with 1,295 houses connected at different buses,
with varying customer preferences and objectives, to demonstrate the applicability of the
proposed scheme.
ESSs can improve the energy management in distribution systems, especially with the
increasing penetration of HEMSs that schedule household appliances and render them as
smart loads. A large number of uncoordinated HEMSs can result in significant changes to
the aggregated load profile of the distribution system. Therefore, a new framework and
mathematical model for integrating ESSs in the distribution grid is proposed to minimize
the operation cost of the LDC and to alleviate the impact of uncoordinated HEMS

iv
operation on the distribution grid. A novel neural network (NN) based
state-of-health (SoH) estimator for a lithium-ion (Li-ion) battery based ESS is proposed,
which is incorporated within the LDC’s planning problem. The results show that the
proposed estimation model is an accurate estimation of the SoH of the ESS. Also, the
LDC’s ESS investment plan decisions are compared considering the proposed SoH of the
ESS vis-á-vis a linear degradation model, and when degradation of ESS is not considered
in planning.
The third research problem addressed in the thesis investigates the ESS’s role in
providing the LDC with flexibility services. A novel flexibility service framework is
developed based on the battery energy storage system (BESS)s’ capability in providing
different levels of charge rate (C-Rate). This work proposes a cooperative game theory
based approach to determine the allocation of monetary benefits among flexible BESSs.
The proposed model ensures a fair distribution of monetary gains among the coalition
members and proposes a novel flexibility pricing scheme.

v
 Acknowledgements

First and foremost, I shall praise and thank Allah Almighty for giving me the strength,
patience, knowledge, ability and opportunity to undertake this research journey and to
persevere the successful completion of this chapter of my life.
Then, I would like to take this opportunity to express my sincere gratitude and
appreciation to Prof. Kankar Bhattacharya and Dr. Steven Wong, for the invaluable
guidance, continuous support and encouragement which they provided throughout my
graduate studies at the University of Waterloo. Without their knowledge, enthusiasm,
availability for discussions at any time and, most importantly, patience, I could not have
made my way through this journey. They will continue to be my role models in their
exceptional professionalism and commitment to a high standard of quality throughout
my academic career.
My appreciation and thanks are also extended to Prof. Ali Elkamel, Prof. Magdy
Salama, and Prof. Ramadan El-Shatshat from the University of Waterloo, for serving on
my doctoral Advisory Committee, and for their insightful comments and feedback. I am
also very thankful to Prof. Bala Venkatesh from Ryerson University, for serving as the
external thesis examiner and for his careful reading, wise comments and insightful
observations. Furthermore, I would like to thank Prof. Stacey Acker for chairing my
thesis examination committee.
With deep appreciation, I would to thank Qassim University, Saudi Arabia, for granting
me the scholarship to pursue my graduate studies. I would like to thank the Ministry of
Education, Saudi Arabia, and the Saudi Arabian Cultural Bureau in Canada for their
support. I also acknowledge the partial funding from Natural Sciences and Engineering
Council of Canada (NSERC) through the NSERC Energy Storage Technology (NEST)
Network.
My appreciation also goes to all the administrative staff at the University of Waterloo
and in the Department of Electrical and Computer Engineering for their assistance and
cooperation throughout my PhD studies.
I would like to express my heartfelt gratitude to mother Ruqayyah Alodhayb, my father

vi
Nasser Alrumayh, my brothers, and my sisters all of whom have constantly given me endless
love, support, encouragement, and sincere prayers.
My deepest appreciation goes to my lovely wife Hajar and my son Walied for their
endless love, unfailing understanding, support and encouragement during my PhD journey.
They tolerated the stressful times and shared both the ups and downs of my PhD studies.
Their precious love, care, and patience made my journey toward my thesis completion
possible.
I would like to thank all my colleagues in the Power and Energy Systems Group at the
University of Waterloo, and to the lab mates in the the Electricity Market Simulation and
Optimization Laboratory (EMSOL) for their valuable discussions and creating a pleasant
and friendly working environment. Special thanks to my colleagues and officemates Talal
Alharbi, Hisham Alharbi, Nitin Padmanabhan, and Muhammad Umar Azam.
Finally, this thesis would not have been possible without the help, support and
guidance of so many friends who shared happiness and tough times with me throughout
various stages of this study and for making my stay memorable. Among them, Abdulaziz
Almutairi, Nizar Alsharif, Badr Lami, Rayed Alyousef, Hisham Alabduljabbar, Abdulaziz
Alaskar, Omar and Abdullah Bin Humaid, Majed Alotaibi, Nawaf Almutairi, Walied
Alharbi, Omar Alarafj, Yasser Assolami, Abdulrahman Alorinan, and Dawood Alsaedi.

vii
 Dedication

To my parents
To my wife and my son
To my Brothers and Sisters
To my supervisors

viii
Table of Contents

List of Tables xiii

List of Figures xv

List of Abbreviations xvii

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Flexibility of Residential Loads for Demand Response Provisions . . 5
1.2.2 Planning for Energy Storage Systems in a Distribution Grid . . . . 9
1.2.3 Service Provisions from Energy Storage Systems and Sharing of the
Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Background Review 15
2.1 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 REH and Mathematical Model of HEMSs . . . . . . . . . . . . . . . . . . 19

ix
 2.2.1 HEMS Objective Functions . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Operational Constraints of REH Appliances . . . . . . . . . . . . . 21
2.3 Energy Storage Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.1 Energy Storage System Characteristics . . . . . . . . . . . . . . . . 29
2.4 Cycle Counting Using RCA . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Artificial Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6 Cooperative Game Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3 Flexibility of Residential Loads for Demand Response Provisions in

Smart Grid 40
3.1 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Proposed Coordination Framework and Mathematical Models . . . . . . . 44
3.2.1 Residential Energy Hub and HEMS (Stage: Ia, Ib, Ic) . . . . . . . . 45
3.2.2 LDC Operations Model (Stage-II) . . . . . . . . . . . . . . . . . . . 49
3.2.3 Coordination of Multiple HEMS and LDC (Stage-III) . . . . . . . . 51
3.3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.1 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.2 HEMS Level Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3.3 LDC Level Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.4 Computational Aspects . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4 Inclusion of Battery SoH Estimation in Smart Distribution Planning

with Energy Storage Systems 64
4.1 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

x
 4.2 Proposed Planning Framework and Mathematical Models . . . . . . . . . . 68
4.2.1 REH Operations Model . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.2 LDC Operations Model . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.3 RainFlow Counting Algorithm . . . . . . . . . . . . . . . . . . . . . 71
4.2.4 Degradation Model of ESS . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.5 Proposed NN Based ESS SoH Estimator . . . . . . . . . . . . . . . 73
4.2.6 LDC Planning Model . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

5 Cooperative Operation of Battery Energy Storage Systems Participating

in Flexibility Services Provisions 91
5.1 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Proposed Flexibility Services Framework . . . . . . . . . . . . . . . . . . . 94
5.2.1 Flexibility Services: Definitions . . . . . . . . . . . . . . . . . . . . 94
5.2.2 BESS Degradation Cost Function . . . . . . . . . . . . . . . . . . . 96
5.2.3 LDC Operations Considering BESS Flexibility Services . . . . . . . 98
5.2.4 Allocation of Cost Savings . . . . . . . . . . . . . . . . . . . . . . . 102
5.3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.1 Calculation of Shapley Value of BESS Investors . . . . . . . . . . . 104
5.3.2 Operational Performance of BESS Flexibility Services in Different
Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

xi
6 Conclusions 116
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

References 121

xii
List of Tables

1.1 Some Energy Storage Projects in Ontario, Canada [11] . . . . . . . . . . . 4

1.2 Summary of Literature Review on Flexibility of Residential Loads . . . . . 8

2.1 Parameters of the HEMS Model . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2 Rainflow Counting Algorithm Result . . . . . . . . . . . . . . . . . . . . . 35

F LEX
3.1 Energy Consumption and Cost of REH #400 per Day, for Different Pj,t 56
3.2 LDC Load Profile Characteristic and Power Losses . . . . . . . . . . . . . 60
3.3 Model Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.1 Tuned Weights and Biases of the NN . . . . . . . . . . . . . . . . . . . . . 80

4.2 ESS Optimal Plan (Considering Scenario-1 Load Profile). . . . . . . . . . . 81
4.3 ESS Optimal Plan Using the Proposed Approach (case-1) for Different Load
Profile Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 NPV of Operation and Installation Costs for Various Cases and Scenarios
($1000) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5.1 Flexibility Service Definition . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5.2 Total Savings in a BESS Coalition for LF Service . . . . . . . . . . . . . . 106
5.3 Marginal Contributions and Shapley Value Calculation for BESS B-2 . . . 107

xiii
5.4 Marginal Contributions and Shapley Value Calculation for BESS B-23 . . . 107
5.5 Marginal Contributions and Shapley Value Calculation for BESS B-30 . . . 108
5.6 Grand Coalition Savings for a given Flexibility Service . . . . . . . . . . . 108
5.7 Allocation of Saving of BESS B-2 (800 kW/200 kWh) . . . . . . . . . . . . 109
5.8 Allocation of Saving of BESS B-23 (2400 kW/600 kWh) . . . . . . . . . . 109
5.9 Allocation of Saving of BESS B-30 (3200 kW/800 kWh) . . . . . . . . . . 109
5.10 Share of Savings Among BESSs . . . . . . . . . . . . . . . . . . . . . . . . 110

xiv
List of Figures

2.1 Typical cycles to failure vs. DoD curve of Li-ion batteries [72]. . . . . . . . 32
2.2 Peak filtering of an ESS operation profile. . . . . . . . . . . . . . . . . . . 33
2.3 Valley filtering of an ESS operation profile. . . . . . . . . . . . . . . . . . . 34
2.4 NN structure with a single hidden layer and output. . . . . . . . . . . . . . 35

3.1 Schematic of a Residential Energy Hub. . . . . . . . . . . . . . . . . . . . . 45

3.2 Required levels and stages of the proposed coordination scheme. . . . . . 46
3.3 Schematic of coordination between HEMS and local distribution companies
(LDC). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Coordination of multiple HEMS and LDC. . . . . . . . . . . . . . . . . . . 54
3.5 Load and flexibility profiles at residential and grid levels. . . . . . . . . . . 56
3.6 Optimal operation of selected appliance (dishwasher) at house #400. . . . 57
3.7 Temperature profile inside house #400, refrigerator, and water heater. . . . 58
3.8 Optimal load profile for House hj at different PRR requests. . . . . . . . . 59
3.9 Total load profile of LDC. . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.10 Average voltage at each bus at different PRR requests. . . . . . . . . . . . 61

4.1 Proposed ESS planning framework considering smart REHs and battery SoH. 69
4.2 Structure of the proposed NN-based SoH estimator. . . . . . . . . . . . . . 74

xv
4.3 Residential load profile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4 Training, validation, testing histogram of NN-based SoH estimator. . . . . 83
4.5 Total distribution system demand and power drawn from substation for
Case-1 (Proposed Approach) over the planning period. . . . . . . . . . . . 85
4.6 Comparison of SoH profiles of ESS located at Bus-8 over a day. . . . . . . 87
4.7 Comparison of SoH profiles of ESS located at Bus-8 over the plan period. . 87
4.8 Estimated SoH profiles of optimal ESS at Bus-23 for various scenarios. . . 88
4.9 Total demand, power imported from external grid, and ESS operation over
one-day of year-10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.10 System voltage profiles for different case studies, for Scenario-1 and -2,
during peak demand in the terminal year. . . . . . . . . . . . . . . . . . . 90

5.1 The proposed flexibility services classifications based on change in SoC. . . 95

5.2 Number of cycles of BESS versus depth of discharge (DoD) at different C-
Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3 Degradation cost of BESS for different flexibility services. . . . . . . . . . . 98
5.4 Load profile of the LDC over a day. . . . . . . . . . . . . . . . . . . . . . . 105
5.5 Hourly Electricity Price (HEP) profile over a day. . . . . . . . . . . . . . . 105
5.6 SOC of BESSs providing different flexibility services in Scenario SC-1. . . . 111
5.7 SOC of BESSs for different flexibility services in Scenario SC-2. . . . . . . 112
5.8 SOC of BESSs for different services in Scenario SC-3. . . . . . . . . . . . . 113

xvi
List of Abbreviations

BESS battery energy storage system

C-Rate charge rate
DR demand response
DER distributed energy resource
DG distributed generator
DoD depth of discharge
ESS energy storage system
EMS energy management system
EV electric vehicle
FERC Federal Energy Regulatory Commission
HEMS home energy management system
HVAC heating, ventilation and air-conditioning
HF high flexibility
IESO Independent Electricity System Operator
LDC local distribution companies
Li-ion lithium-ion
LF low flexibility
MILP mixed integer linear programming

xvii
MINLP mixed integer non-linear programming
MF moderate flexibility
NaS sodium sulphur technologies
NPV net present value
NN neural network
O&M operation and maintenance cost
PEV plug-in electric vehicle
PV photovoltaic
PRR peak reduction requests
RCA Rainflow Counting Algorithm
RES renewable energy source
REH residential energy hub
SOC State of Charge
SoH state-of-health
TOU time-of-use

xviii
Chapter 1

Introduction

1.1 Motivation

With continuously increasing demand for energy, globally, and which is expected to grow
by 12% from its 2019 levels by the year 2030 [1], there has been serious concerns on
the long-term adequacy, availability and supply of energy. Traditional fossil fuel-based
energy sources are detrimental to the environment and are not the favoured options by
governments for new capacity. Thus, it is necessary to investigate other eco-friendly energy
resources and solutions to meet the demand in the long-run.
To address this issue, governments have introduced new strategies, policies and dynamic
pricing tools in order to reduce the peak demand and hence defer capacity investment costs.
The Peaksaver Plus [2] and the Global Adjustment Program in Ontario, Canada [3] are
examples of policies to reduce the peak demand of the local distribution companies (LDC);
while the time-of-use (TOU) tariff applicable in Ontario is a pricing instrument seeking
the same objective of lowering the peak demand. The development of demand response
(DR) programs and the smart use of customers’ flexibility can help the LDC enhance grid
reliability and operational efficiency.
According to the US Federal Energy Regulatory Commission (FERC), DR is defined
as [4], “Changes in electric usage by end-use customers from their normal consumption

1
patterns in response to changes in the price of electricity over time, or to incentive payments
designed to induce lower electricity use at times of high wholesale market prices or when
system reliability is jeopardized ”. On the other hand, flexibility is the ability of the power
system to respond to fast changes in supply and demand, and in the balance of the two [5].
Increased flexibility needs refer to increased fast response requirements from generators
and loads, though not increasing the overall capacity.
In 2018, the peak demand in Ontario was 23,240 MW; this lower than expected level
was partially attributed to the deployment of various DR programs in the province. The
estimates reveal that around 7% of the demand reduction was from DR programs such as
Industrial Conservation Initiative, capacity-based DR, and the Peaksaver Plus residential
DR program [2].
Effective implementation of DR programs relies on the availability of smart grid
infrastructure, where deployment of two-way communication technologies between the
LDC and end-users is crucial. Smart meters pass information to the LDC on customers’
consumption levels and their readiness to participate in DR programs, by altering their
consumption patterns based on price signals received from the LDC. The LDC can hence
achieve a considerable degree of flexible operation by coordinating the controllable loads
of retail customers, individually.
The effective participation of residential customers in DR programs can be achieved
using a home energy management system (HEMS) which is a residential controller that
determines the optimal ON/OFF decisions of the household appliances/devices based on
the customer’s objectives and preferences. According to a study by Hydro One [6], the
display of electricity rates in real time, such as TOU rates, can increase load shifting on
typical days from 3.7% to 5.5%, while on hot days, up to 8.5%, and bring about 13%
reduction in energy consumption. The concept of energy hubs has been proposed in the
last decade [7] where various energy system activities are performed within a hub, namely,
energy production, conversion, storage, and consumption of different energy carriers [8].
The integration of renewable energy sources (RESs), energy storage systems (ESSs), and
smart loads at the residential customer’s end have transformed these loads to a residential
energy hub (REH).

2
 The increasing transformation of residential loads to REHs can provide powerful
capabilities to benefit both customers and the LDC. Therefore, there is a need to assess
the flexibility of REHs to provide DR services and evaluate their impact on consumption
and distribution grid operation.
Besides home energy management, ESS can also provide flexibility, energy and capacity
resource, and other benefits to the LDC such as loss reduction, peak shaving, increased
RES integration, islanded operation support. In recent years, several ESS projects with
varying sizes and locations have been commissioned by the Independent Electricity System
Operator (IESO) in Ontario, Canada, to provide services such as load leveling, emergency
reserve, and frequency regulation. For example, IESO York Region Non-Wires Alternatives
Demonstration Project investigates the potential of high penetration of distributed energy
resources (DERs) as an alternative for traditional infrastructure to meet local demand
and electricity market needs without introducing any negative impact on system reliability
[9]. In USA, the FERC Order 2222 [10] has been recently put in place that creates an
opportunity for ESSs and REHs to indirectly participate in electricity markets through
local aggregators and hence promote competition.
The diversity of ESS technologies can provide different capabilities which add value
to their contributions in power system applications. There are many ESS projects under
development or in the operational stage. In 2014, the IESO of Ontario, Canada, initiated
a competitive procurement framework for 50 MW energy storage to provide capacity and
ancillary services to the grid. Two consecutive phases were considered for the energy
storage deployed; the first phase started in mid of 2014 by selecting five companies to
provide ancillary services to the power grid with 34 MW capacity; the second phase offered
10-year contracts to five companies with 16.75 MW capacity in six separate energy storage
projects.
By Spring 2018, seven energy storage facility projects have been commissioned and
became operational. Two out of the seven are assigned to provide reactive power support
and voltage control service, while the others are responsible for regulation services. As of
December 2020, three out of the six energy storage projects from the second phase have
started commercial operation, while the rest are underway. Table 1.1 shows examples
of energy storage projects in Ontario, Canada, at the distribution and transmission grid

3
level [11].

Table 1.1: Some Energy Storage Projects in Ontario, Canada [11]

Proponent Location Technology Capacity Objective
Milton, Ontario
Baseload Power Corp. at distribution grid Flow Battery 4 hours, 2 MW Grid support and arbitrage

Newmarket, Ontario 4 hours, 2 MW Peak shaving and on-going grid

Ameresco Canada Inc. at distribution Solid Battery Two relability and stability
grid facilities with increase in RES
SunEdison Canada Flow Battery 3 projects with total
Origination LP. IESO grid (VRB) 4 hours, 20 MW N/A

NextEra Canada Development Solid Battery

and Acquisitions, Inc. Distribution grid (Li-ion) 4 hours, 2 MW Peak shaving

Compressed Air
NRStor Inc. IESO grid Energy Storage 7 MWh N/A

One of the fastest growing ESSs is lithium-ion (Li-ion) batteries, which are a
commercial and mature technology used for small-scale and large-scale grid applications.
Li-ion batteries represent around 10% of the total installed ESS capacity in Canada [12],
and their penetration is expected to increase with the decline in its capital cost [13].
Consequently, there is a need to properly size and site ESSs in an operational-planning
framework in order to minimize the total cost of investment for their owner and increase
the benefit from their flexible operation. However, it should be noted that the lifetime of
Li-ion batteries are highly impacted by their cyclic operation and calendar aging.
Although the short-term charging / discharging decisions would accrue financial benefits
to the ESS owner, these would negatively impact the asset lifetime [14]. Neglecting the
degradation of Li-ion batteries can lead to over-estimation of its available capacity and
the expected financial benefit from the asset. In order to capture the inter-temporal
changes in the state-of-health (SoH) of battery energy storage system (BESS), there is a
need to model the degradation of Li-ion batteries as a function of its charging /
discharging actions and its calendar aging, which can be incorporated within power
system operation and planning problems.
The ownership of BESS can highly impact its operation, limiting the owners’ financial
benefit. BESSs are well-known for their prominent role in power systems such as energy
arbitrage, peak demand reduction, and ancillary services. To unlock the ultimate benefits
of BESS, the investors can transfer the right of BESS control to another entity, such as

4
an aggregator or the LDC, which requires access to more flexible resources. The LDC, for
example, may need high flexibility in operation due to its exposure to volatility in demand,
electricity prices, or intermittent generation. Also, the LDC does not have to own the BESS
facilities in order to increase its flexible resources. The development of financial benefit
allocation frameworks would encourage investors to participate in flexibility provisions
and maximize the overall payoff. Therefore, there is a need to evaluate the contribution
of individual BESSs each seeking to maximize the overall system benefit, which can hence
be modeled as a cooperative game.

1.2 Literature Review

This section reviews the relevant literature pertaining to the topics and issues discussed in
this thesis including operation, planning, and incentive design for ESSs and flexibility of
residential loads for DR provisions in smart grid.

1.2.1 Flexibility of Residential Loads for Demand Response

Provisions

In recent years, several researchers have reported their works on the mathematical modeling
of an HEMS that optimizes the operation of household appliances [15–22].
In particular, a mixed integer linear programming (MILP) model was proposed in [22],
optimizing the household appliance schedule, for incorporating into automated residential
energy controllers. The mathematical model included most of the household appliances in
addition to ESS and photovoltaic (PV) panels. The model parameters were determined
from practical tests, and the model was implemented on households in Ontario, Canada;
the results showed a significant reduction in both energy cost and peak demand while
maintaining the customer preferences. The paper [22] forms the basis for the research on
flexibility of residential loads presented in this thesis.
Some studies have dwelled upon coordinating the residential loads to provide an
aggregated DR service to the utility while maximizing the aggregator’s benefit.

5
 In [23], an incentive-based energy consumption scheduling problem was formulated,
while considering the interaction amongst households and the LDC, to minimize the total
energy cost and the peak-to-average ratio of residential loads. Since the work considered
only shiftable loads (i.e. dishwasher, washer, and dryer), minimization of energy
consumption of the household appliances was not considered.
In [24], a decentralized HEMS was formulated as a multi-stage stochastic optimization
problem to minimize the aggregator’s cost by balancing the residential demand with
available generation in real-time; four deferrable loads with fixed load profiles were
considered. However, neither [23] or [24] considered the modeling of controllable loads
with flexible thermal or electrical capacity such as heating, ventilation and
air-conditioning (HVAC) and ESS, where energy consumption minimization can be
achieved by operating the device within the customer’s comfort zone.
In [25], a cooperative energy management system (EMS) was proposed to coordinate
the operation of DERs of a neighborhood comprising residential prosumers operating in
isolated mode. RESs and ESSs were considered, where the aim of the EMS was to reduce
the mismatch between supply and demand. However, the paper considered the residential
loads as fixed, without taking into account the inherent flexible operation capabilities of
the household appliances.
In [26], a centralized coordination approach was proposed to achieve a pre-defined load
profile by applying a day-ahead load shifting technique on residential, commercial, and
industrial loads. The results indicated that the proposed load shifting approach reduced
the system peak load and operation cost. Nevertheless, the paper did not investigate the
impact of demand management on grid operations, such as bus voltage levels and system
losses.
Distributed heat pumps, which represented a number of households’ controllable loads,
and BESS were coordinated in [27] to provide power smoothing services to mitigate the
impact of power fluctuations of the integrated RESs on the connection line between a
microgrid and main grid.
In [28], a decentralized stochastic optimization model was proposed for controlling the
energy consumption of several houses including flexible and non-flexible loads, ESS, and

6
RES to minimize the total energy cost of the neighborhood using a real-time scheduling
algorithm. The paper considered a case study, where a neighborhood with a few houses
was coordinated to minimize the total cost. The paper assumed that the RES generation
surplus would be curtailed without considering a more economic option of selling the
surplus to the LDC. In addition, the work did not address the impact of the decentralized
approach on distribution grid operation with multiple neighborhoods.
In [29], a decentralized coordination of multiple prosumers was proposed which aimed
to maximize the aggregator’s benefit while minimizing the reverse flow of energy at the
service point, using a DR program. Other works [30, 31] proposed heuristic approaches to
coordinate the DR of residential loads to minimize the cost of each customer, flatten the
system load profile, and address the issue of possible peak rebound that might result
from uncoordinated DR. The DR of residential loads was achieved by controlling a small
set of household appliances- dishwasher, washer, dryer, and plug-in electric
vehicles (PEVs). Furthermore, the paper investigated the impact of coordinated DR on
distribution operation, and an improvement in the voltage profiles and reduction in
system losses was noted. However, the work did not consider some of the basic flexible
household appliances such as HVAC and stove.
Coordination of a small number of residential HEMSs, which comprised controllable
appliances in addition to PEVs, ESSs, and distributed generation, to minimize the total
neighborhood energy procurement cost while mitigating any unfair use of the distribution
transformer, was proposed in [32]. In [33], a two-stage nodal pricing scheme was proposed to
motivate the residential customers to participate in DR programs considering distribution
system operational aspects; however, power system losses were not considered, and several
household appliances were not taken into account in the HEMS as possible flexible loads.
In recent times, flexibility provisions have become a very important topic in power
systems as a result of the increasing penetration of RESs into the power system, which calls
for increased number of flexible resources such as controllable loads and ESSs. Therefore,
these flexible resources need to be managed wisely to achieve customer-level and grid-level
benefits. For example, flexible loads can provide DR services to the LDC. According to [34],
uncoordinated DR might affect the distribution grid operation negatively, whereas proper
control of DR can eliminate the adverse impact of such modifications in the demand-side

7
behavior. Another study stated that strategic operation of ESSs in electricity markets
can limit the potential flexibility benefits of such a system [35]. The work [36] proposed
a decomposition approach to estimate the controllable and non-controllable residential
loads within a total forecasted load of a distribution system, which can be used in DR
applications.
In [37], A multi-objective mixed integer non-linear programming (MINLP) based HEMS
was proposed considering the customer’s energy savings and comfort level, wherein each
residential customer was equipped with controllable appliances and ESSs in a residential
microgrid.
Table 1.2: Summary of Literature Review on Flexibility of Residential Loads
REH Model Grid Operation
Paper Water and
Heater Fridge Stove HVAC Lighting Dishwasher Washer Dryer Pool ESS RES PEV Constraints
[21] C F F C NC C C C NC C NC NC NC
[23] F F F F F C C C NC NC NC C Linear feeder model
[24] NC NC NC NC NC C C C NC NC NC C NC
[25] F F F F F F F F NC C C NC NC
[26] NC NC C NC NC C C C NC NC NC NC C
Tie-line between
[27] NC NC NC C NC NC NC NC NC C C NC microgrid and
main grid
[28] Residential loads represented as shiftable and non-shiftable loads C C NC NC
[29] Interruptible, uninterruptible, and thremostat-controlled load C C NC
[30, 31] NC NC NC NC NC C C C NC NC NC C C
[32] F F F F F C C C NC C C C NC
[33] C F F C F C C C NC C NC C Linear feeder model
[34] C C C C C C C C NC NC NC NC C
[38] C NC NC C NC NC NC NC NC NC NC NC NC
Proposed work C C C C C C C C C C C C 30-bus system
Considered as an optimization variable (C), Fixed Load (F), Not considered (NC)

In [38], an electric storage space heater of a residential customer was modeled to

optimize its operation considering the flexibility of heating loads while minimizing the
total operation cost; on the other hand, the retailer maximized its profit by optimally
determining the electricity prices while minimizing the supply-demand imbalance in the

8
system. This was achieved by encouraging the flexible residential customers to alter their
consumption behavior in response to the announced prices by the retailer. The study
investigated the impact of different pricing schemes on both customers and retailer costs.
It was noted that the TOU scheme was the best option for the residential loads.
However, the retailer model did not consider the distribution grid constraints or the
associated power losses, which affects the total cost of the retailer.
Table 1.2 presents an overview of the literature review pertaining to flexibility of
residential loads and their ability to provide DR services. Many researchers have simply
considered dishwasher, washer and dryer as flexible loads, while the other households
appliances were considered fixed loads [23, 24, 30]. Other works have mainly investigated
the DR provisions from residential thermal loads [33, 38]. Moreover, to the best of the
author’s knowledge, very few commercial HEMS solutions exist in the market today that
integrate home automation, local generation, and energy storage with the external power
system.
The integration of PEVs, ESSs, and/or RESs were considered by a few researchers,
which noted the significant changes in the residential load profiles and customers’ monthly
bills. Very little research has been carried out to explore the impact of residential loads’
flexibility and DR provisions on distribution system operations [26,34]. Hence, the inherent
benefits of managing flexible resources to facilitate the integration of RESs and ESSs calls
for in-depth research on this subject.

1.2.2 Planning for Energy Storage Systems in a Distribution

Grid

There is a growing body of literature addressing the issue of integrating ESSs into the
distribution grid to assist the LDC in managing grid operations.
The technical and economic advantages include, increasing the operating margins to
facilitate RESs and electric vehicles (EVs) penetration [39] and deferring upgrades to the
distribution grid. The authors in [40] proposed a multi-objective optimization approach
to optimally site distributed generators (DGs), ESSs and RESs. An optimal planning

9
approach was proposed in [41] and [42] to determine the location, rated energy and power
capacity of BESSs to mitigate the impact of uncertainties associated with renewables-based
DGs. In [43], a genetic algorithm based planning framework was proposed to optimally size
and site ESSs in a distribution system grid in order to minimize system losses, defer system
upgrades, and maximize the profit from energy arbitrage. Another work [44] proposed a
stochastic approach to coordinate the planning of ESSs and RESs while considering DR.
In [45], a chance-constrained approach was proposed for ESSs planning in a distribution
grid. However, none of the above works have considered the lifetime of the ESSs nor the
impact of degradation on the asset’s operation.
Several researchers have considered degradation of ESSs [46–51], but for different
studies. In [46], an operational planning scheme was proposed to coordinate wind farms
with ESSs to mitigate the impact of wind forecast errors and extend the lifetime of
BESSs by reducing frequent charge / discharge operations.
A multi-stage approach was proposed in [47] to optimally plan for transmission
expansion and ESS deployments considering a linear degradation rate of BESSs. In [49],
a decomposition approach was proposed to optimally determine the size and year of
installation of ESSs in a microgrid while assuming an annual degradation factor of the
batteries.
A BESS sizing model was proposed in [49] to optimally determine the size and the
number of units required to minimize the microgrid operation cost and unserved energy
cost. In [50], a planning model for isolated microgrids was proposed considering re-purposed
EV batteries, assuming a linear change in the ESS’s SoH due to calendar and cycle aging.
In [51], the participation of ESSs in frequency regulation in the PJM market was assessed
considering the impact of different depth of discharge (DoD) levels on the SoH of various
types of Li-ion batteries.
It is noted that researchers have examined several simple approaches to consider the
SoH of BESSs in their optimization problems. However, none of the above works have
captured the change in BESSs’ SoH due to calendar and cycle aging, and the issue of
under- or over-estimating BESS capacity resulting in over- or under-capacity planning in
the distribution system, being left un-addressed [47, 48, 50].

10
 Some of the works in the literature have reported the development of approaches to
estimate the lifetime of Li-ion based BESS. The proposed methods can be classified as:
model-based, data-driven, and fusion-based methods [52]. In [53], an SoH estimation
approach based on the equivalent internal resistance of Li-ion BESS was proposed to
observe the change in the battery SoH in real-time. Another work [54] proposed to
estimate the SoH of Li-ion batteries using the data obtained from the battery EMS such
as current, voltage and temperature. In [51], a semi-empirical Li-ion battery degradation
model was proposed to overcome the issue of limited observation data availability, from
operating the ESS for specific applications.
Researchers have proposed a battery life assessment approach that can be integrated
in different stationary BESS applications in power system problems. In [51], an off-line
battery lifetime assessment was proposed, based on the Rainflow Counting
Algorithm (RCA), which requires the history of BESS operations. Such an approach
requires the history of operating BESS as an input. However, since distribution system
planning problems with BESSs must simultaneously compute the battery State of
Charge (SOC) profiles based on BESS operational decisions and the optimal BESS plan,
RCA-based degradation models [51] cannot be implemented directly within these models
because they require historical data of BESS operations. Therefore, there is a need to
develop an estimation approach of BESS’s SoH in order to tackle the issue of under- or
over-estimating their degraded capacity.

1.2.3 Service Provisions from Energy Storage Systems and

Sharing of the Benefits

This subsection discusses different incentive schemes for DR and flexibility provisions
procured from ESSs in the smart grid.
Dynamic pricing approach is proposed in [55] to indirectly alter the consumption of
residential customers and the operation of their ESS units. In [56], a non-cooperative
game is proposed to capture the competitive behavior between residential loads and ESSs
to minimizing their electricity cost individually. After that, a non-cooperative

11
Stackelberg game is proposed to model the interactions between the LDC, who
determines the electricity prices, and the residential customers, who respond to the signal
by changing demand, to minimize the peak-to-average power ratio.
A Stackelberg game approach is proposed in [57], which possesses individual
rationality and incentive compatibility, to model an auction market environment for joint
ESS ownership sharing between several facility managers and residential communities.
The work aims to exploit the full potential benefits of ESSs owned jointly by different
entities and provide various services to a third party. In [58], a Stackelberg game is
introduced to maximize the revenue of the shared ESS owner while the electricity retailer
aims to minimize the total cost of residential customers. A bargaining based incentive
approach is proposed in [59] to encourage joint energy trading between microgrids, each
equipped with ESS, and hence to maximize the total obtained benefit.
Another work in [60] proposes a cooperative game theory-based approach to manage
the operation of several ESSs and hence minimize the coalition energy cost. The
proposed method shows an improvement in the LDC operation by flattening the load
profile and reducing the reverse power flow. A coalition game theory-based energy
management problem is proposed in [61] to maximize the benefit of a coalition of local
energy communities equipped with flexible loads, ESSs, and RESs. The grand coalition
effectively increases the total payoffs and improves the overall load profile.
It is noted from the above discussions that most of the works proposed monetary
benefits to encourage ESSs to participate in system operations and efficient use of available
resources. However, none of the works examined the participation of ESSs based on their
marginal contribution to a coalition considering their physical characteristics such as the
rate of change in SOC, which significantly impacts the ESS lifetime.

1.3 Research Objectives

The main objectives of the research presented in this thesis include evaluating the flexibility
provisions from residential customers through REHs and ESSs and provide a flexibility
service to the grid operator. The research objectives are listed below:

12
• Augment the mathematical model of the residential customer’s HEMS reported in
[22] to represent it as an REH considering detailed characteristics of ESSs and PV
units. The developed model will take into account the detailed interactions, data and
information exchanges between different entities within the REH as well as between
the REH and the LDC.

• Propose a novel concept of REH flexibility index to quantify the flexibility resource
available from a household. Thereafter, develop a coordination scheme that can take
into account distinctly individual objectives of each REH as well as that of the LDC,
seeking to enhance grid operational efficiency, and hence create an aggregated DR
service for the system. This will involve a novel LDC operations model, which will
determine the optimal peak reduction requests from the individual REHs.

• Propose a novel neural network (NN) based degradation model to estimate the SoH
of Li-ion batteries of an ESS in a distribution system. The data to be used for NN
training shall comprise SOC profiles of ESS obtained considering different LDC load
mix including a set of simulated REH operations extended over the plan period.

• The developed NN-based SoH model of the ESS will be incorporated into a planning
model to determine the optimal energy capacity, power rating, location and year of
installation or replacement of ESSs in the distribution system while internalizing the
ESS battery capacity degradation due to cycling and aging effects.

• Propose a novel BESS flexibility provision framework based on their capability of

providing different charge rate (C-Rate) levels. This framework will allow individual
investors to participate in flexibility provisions as a service to the LDC. A cooperative
game theory-based approach will be applied to maximize the total system payoff and
hence distribute it fairly among all the BESS participants based on their marginal
contributions to the total system saving using the Shapley value criterion.

13
1.4 Outline of the Thesis

The rest of the thesis is organized as follows: Chapter-2 presents background on HEMS
and its mathematical model, and the concept of energy hubs, followed by a discussion on
ESS and their applications in smart grids. In Chapter-3, the proposed residential loads’
flexibility evaluation and a novel coordination framework for DR provisions is presented,
along with results of different case studies. In Chapter-4, a novel NN based SoH estimator
for a Li-ion battery-based ESS is proposed, which is then incorporated within the LDC’s
planning model to determine optimal ESS plans. Chapter-5 proposes a cooperative game
theory-based approach to distribute the total savings, accrued by the LDC, among BESS
owners for their participation in flexibility provisions in a distribution grid. Finally, in
Chapter-6 the conclusions and contributions of this thesis are presented, and the scope for
future research are outlined.

14
 Chapter 2

Background Review

This chapter presents a brief review of the topics and tools relevant to the research
presented in this thesis. A detailed discussion of the HEMS and the REH, and the
associated mathematical models are presented. Thereafter, BESS characteristics and
their applications in smart grids are briefly discussed, followed by a brief review of the
RCA, and the basic features of a NN model.

2.1 Nomenclature
Indices and Sets
hj Index for household, hj ∈ H
i Index for household appliances, i ∈ A
i = {fridge (fr), water heater (WH), stove (STV), dishwasher (DW), washer
(W ), dryer (Dry), lighting (LI), pool (Pool), energy storage system (ESS),
PV panel battery (BP V )}
j,k Index for bus, {j, k} ∈ N
t Index for time, t ∈ T

15
Parameters
ALt Activity level inside a house
MW,DRY Maximum allowed time gap between operation of washer and dryer
Eimin , Eimax Minimum and maximum energy level of i ∈ {ESS, BPV}, kWh
EOTi , LOTi Early / late operation time of appliance i ∈ {DW,W,DRY,PMP,STV}
Gj,k Conductance of feeder between bus j and k
gi Heat rate of device i ∈ {HT,WH}
HW Ut Hot water usage at time t
Kt Price elasticity associated with the illumination inside a house
Lout min
t , Lt Natural / minimum illumination at time t
MUTi , MDTi Minimum uptime / downtime of appliance,
i ∈ {AC,HT,DW,W,DRY,PMP,STV}
M STi Maximum successive operation time of appliance,
i ∈ {DW,W,DRY,PMP,STV}
PCmax
i
, PDmax
i
Maximum charging/discharging power of i ∈ {B,ESS}, kW
PD 0 Connected load at time t, kW
Phmax
j ,t
House hj maximum demand at time t, kW
Pbhmax
j ,t
Updated house hj maximum demand at time t, kW
Pi Rated power of appliance i, kW
PLI Rated power of the lighting system LI, kW
PP V,hj ,t PV generation at time t, kW
ROTi Required operation time of appliance i ∈ {DW,W,DRY,PMP,STV}, min
Yj,k Magnitude of admittance matrix element, p.u

16
αi , βi Cooling/Warming effect associated with ON / OFF state of device,
◦
i ∈ {AC,HT,FR,WH}, C /min
η1 , η2 Charging/discharging efficiency of ESS, %
ρGAS Gas price, $/m3
ρTOU
t Time of use price, $/kW h
ρSEL Residential customers electricity selling price to the LDC, $/kW h
ρi Effect of the difference between inside and outside temperature on the inside
temperature of device i ∈ {AC,HT,FR,WH}
γi Cooling/Warming effect associated with OFF state of device i ∈ {FR,WH}
τ Length of the time interval, minutes
◦
Θmin max
in , Θin Minimum and maximum allowable temperature inside a house, C
◦
Θset
i Preferred temperature setting of i ∈ {AC,HT,FR,WH}, C
◦
Θout,t Ambient temperature, C

Variables
Ei,t Energy level of i ∈ {ESS, BPV} at time t, kWh
Lt Illumination level at time t, kW
PCBPV ,t , PDBPV ,t Total charging / discharging power from BPV at time t
H LDC
Pi,t , Pi,t Discharging power from device i ∈ {ESS, BPV} to house / LDC at time t,
kW
i
PLDC,t Charging power drawn by device i ∈ {ESS, BPV} from LDC at time t, kW
H
PLDC,t Delivered power by LDC to meet the house demand at time t, kW
BPV
PLDC,t Total charging power drawn by BPV at time t, kW
PPBPV
V,t Generated power by PV solar panel used to charge the PV battery (BPV)
at time, kW

17
PPHV,t Generated power by PV solar panel consumed by household appliances at
time t, kW
PPLDC
V,t Generated power by PV solar panel sold to LDC at time t, kW
SCi ,t Binary charging status of i ∈ {ESS, BPV} at time t, ON/OFF
SDi ,t Binary discharging status of i ∈ {ESS, BPV} at time t, ON/OFF
Si,t Binary status of device i at time t, ON/OFF
Ui,t Binary status denoting start up of device i at time t, ON/OFF
Di,t Binary status denoting shut down of device i at time t, ON/OFF
Vj,t Voltage at bus j at time t, p.u
δj,t Voltage angle at bus j at time t, radians
Θin,t Temperature inside a house

18
2.2 REH and Mathematical Model of HEMSs

The REH is a residential building with different energy activities that take place in order
to produce, consume, and / or store electricity. To optimize various activities of an REH,
HEMSs are in place; these are residential controllers that carry out scheduling of the REH
including all house appliances, the ESS, PV panel battery, and the power interchanges with
the external grid, considering the customer’s preferences and objectives. The objective of
the HEMS is to minimize the energy consumption, energy cost, or maximize the comfort
level of the REH customer. In doing so, the HEMS helps the household to better evaluate,
visualize, and manage its energy consumption by providing updated information on its
energy use and the current energy price. According to [62], a study by Hydro One shows
that “Providing real-time displays to customers on Regulated Price Plan (RPP) and Time-
of-Use (TOU) rates increased the load shifting impacts on typical days from 3.7% to 5.5%
and on hot days (over 30°C) to 8.5%. In addition, real-time feedback to the HEMS can
decrease the overall energy consumption up to 13% ”.
The HEMS requires the REH customer to update its preferences, such as the minimum
and maximum allowable temperature setting inside the house, in order to autonomously
control the appliances. Two-way communication infrastructure is required to facilitate
such autonomous control of the appliances / devices by the HEMS; HomePlug, ZigBee,
and Wi-Fi are the three common open communication protocols [63].
Several mathematical models have been proposed to optimize the household appliances’
and devices’ operations; the HEMS mathematical model formulated in [64, 65] extends the
work reported in [22]. A brief description of the HEMS model is presented next.

2.2.1 HEMS Objective Functions

The objective functions of the HEMS, for a given REH, that have been considered, are:

19
 Minimize Cost:
"
X n X X o
J1 = ρTOU
t Pi Si,t + PLI Lt + i
PLDC t
+ PD0
t∈T i∈A i∈{BPV,ESS}
i6∈{LI,ESS,BP V }
#
X n X o
+ ρGAS gi Si,t − ρSEL PDLDC
i,t
LDC
+ PPV,t (2.1)
i∈{HT,W H} i∈{BPV,ESS}

In (2.1), the terms within the first curly brackets denotes the total power drawn from the
grid to supply the following: household appliances, lighting system, the charging power
drawn by the ESS and the PV panel battery, and the uncontrollable load. The total gas
consumption is represented by the term associated with ρGAS , where the price of gas is
assumed to be a fixed rate, equivalent to 2.9 cents/kWh [22]. The terms associated with
ρSEL represents the total income from selling power to the LDC.
Minimize Energy Consumption:
" #
X X X X
i
J2 = Pi Si,t + PLI Lt + PLDC t
+ gi Si,t (2.2)
t∈T i∈A i∈{BPV,ESS} i∈{HT,WH}
i6∈{LI,ESS,BPV}

In (2.2), the customer seeks to minimize the energy consumption from household
appliances, lighting system, charging power of ESS and PV panel battery, and gas
consumption.
Maximize Comfort Level:
" #
X X
J3 = Θin set
i,t − Θi (2.3)
t∈T i∈{AC,HT,WH,FR}

The temperature deviations of certain appliances from their preferred set points represents
the customer comfort level. Therefore, maximizing the customer’s comfort level can be
attained by minimizing the temperature deviations from the pre-defined set points.

20
2.2.2 Operational Constraints of the Household Appliances

In this section, the constraints of the HEMS mathematical model are presented. The typical
household appliances / devices comprises: HVAC, dishwasher, fridge, lighting system, pool
pump, stove, washer and dryer, and water heater, in addition to an ESS and a PV panel.
Home Power Balance and Peak Load Constraints:
X X
H
Pi Si,t = PLDC,t + PDHi,t + PPV,t
H
, ∀t ∈ T (2.4)
i∈A i∈{B,ESS}

X X X
LDC
Pi Si,t + Pi,t − PDHi,t − PPHV,t ≤ Pbtmax (2.5)
i∈A i∈{B,ESS} i∈{B,ESS}

Constraint (2.4) ensures that the total power demand of the REH appliances is met by the
power drawn from the grid, discharged to the house from the ESS and PV panel battery,
and that generated by the PV system. It should be noted that while the PV panel batteries
represent stationary storage, the ESSs denote the mobile storage devices in the house, such
as PEVs. In (2.5), the peak load constraint ensures that the total load of the REH is within
a limit specified by LDC, Pbtmax .

Heating, Ventilation, and Air Conditioning System operation

The following constraints are used to represent the HVAC operation, considering the
household preferences such as minimum and maximum temperature inside the house.

Θmin max
in,t ≤ Θin,t ≤ Θin,t , ∀t ∈ T (2.6)

0 or 1 if t∈ T , i ∈ {AC,HT}
Si,t = (2.7)
0 if t∈
/ T , i ∈ {AC,HT}

1 if Θin Θmax
in,t , i = AC
Si = t=0 (2.8)
t=1 0 if Θin Θmin
in,t , i = AC
t=0

21
 
1 if Θin Θmin
in,t , i = HT
Si = t=0 (2.9)
t=1 0 if Θin Θmax
in,t , i = HT
t=0

SAC,t + SHT,t ≤ 1, ∀t ∈ T (2.10)

In (2.6), the AC and HT operations are based on a pre-set minimum (Θmin in ) and maximum
max
(Θin ) temperature inside the REH. Constraint (2.7) is used to allow the operation of
the HVAC at certain set of time intervals only. As a result, the AC system operates when
the inside temperature is above Θmaxin ; while, the HT system operates when the inside
min
temperature is less than Θin , as shown in (2.8) and (2.9), respectively. Constraint (2.10)
prevents simultaneous operation of the AC and HT.
The indoor temperature depends on many factors such as the activity level of the
customer and the ambient temperature, as given below:
h i
Θin,t = Θin,t−1 + τ βAC ALt − αAC Si,t + ρAC (Θout,t − Θin,t ) , ∀i = AC, t ∈ T (2.11)
h i
Θin,t = Θin,t−1 + τ βHT ALt + αHT Si,t − ρHT (Θout,t − Θin,t ) , ∀i = HT, t ∈ T (2.12)

Ui,t − Di,t = Si,t − Si,t−1 , ∀i ∈ {AC,HT}, t ∈ T (2.13)

Ui,t + Di,t ≤ 1, ∀i ∈ {AC,HT}, t ∈ T (2.14)

t+M
X U Ti
Si,k ≥ MUTi − M (1 − Ui,t ), ∀i ∈ {AC,HT}, t ∈ T (2.15)
k=t

Xi −1
t+MDT
Si,k ≤ M (1 − Di,t ), ∀i ∈ {AC,HT}, t ∈ T (2.16)
k=t

Equations (2.11) and (2.12) models the change in the indoor temperature of the REH as a
result of HVAC operation, the changes in the activity level of the customer, or the changes
in the ambient temperature. Equations (2.13) and (2.14) are the coordinating constraints
for the start-up and shut-down decisions of the AC and HT. In addition, minimum up-time
(MUTi ) and minimum down-time (MDTi ) of the AC and HT are considered in (2.15) and
(2.16).

22
Dishwasher

The dishwasher mathematical model is presented below:


0 or 1 if t∈ T , i ∈ {DW}
Si,t = (2.17)
0 if t∈
/ T , i ∈ {DW}

Ui,t − Di,t = Si,t − Si,t−1 , ∀i = DW, t ∈ T (2.18)

Ui,t + Di,t ≤ 1, ∀i = DW, t ∈ T (2.19)

X
Si,t = ROTi , ∀i = DW, t ∈ T (2.20)
t∈T

t+MST
Xi
Si,k ≤ MSTi + M (1 − Ui,t ), ∀i = DW, t ∈ T (2.21)
k=t

t
X
Ui,k ≤ Si,t , ∀i = DW, t ∈ {EOTi + MUTi + 1, LOTi } (2.22)
k=t−MUTi +1

t
X
Di,k ≤ 1 − Si,t , ∀i = DW, t ∈ {EOTi + MDTi + 1, LOTi } (2.23)
k=t−MDTi +1

The dishwasher operates within the pre-determined time intervals preferred by the
customer as modeled in (2.17). Minimum up-time, minimum down-time, and required
operation time in addition to the maximum successive operation time constraints are
expressed mathematically in (2.20) to (2.23).

Fridge

The following mathematical model pertaining to fridge operation is formulated considering

the technical aspects of the fridge as well as the customer preferences.

0 or 1 if t∈ T
Si=FR,t = (2.24)
0 if t∈
/T

23
 
1 if Θi = Θmax
i,t
Si=FR = t=0 (2.25)
t=1 0 if Θi = Θmin
i,t
t=0

Θmin
i ≤ Θi,t ≤ Θmax
i , ∀i=FR, t ∈ T (2.26)
h i
Θi,t = Θi,t−1 + τ βi ALi,t − αi Si,t + γi , ∀i=FR, t ∈ T (2.27)

The time intervals where the fridge can be in operation are modeled in (2.24). The
constraint (2.25) ensures that if the initial temperature of the fridge is above the pre-
determined maximum temperature, then it should be in operation; or else it should be in a
standby mode. The REH’s preferred minimum (Θmin max
FR ) and maximum (ΘFR ) temperature
inside the fridge is specified in (2.26). The effect of the activity level, ON and OFF states,
on the fridge operation are represented by βF R , αF R , and γF R respectively and presented
in (2.27).

Lighting System

The model of the lighting system is affected by two factors: daylighting (Lout
t ) and the
min
minimum illumination required at time t (Lt ).

Lt + Lout
t ≥ (1 + Kt )Lmin
t , ∀t ∈ T (2.28)

The constraint (2.28) ensures that the illumination level at time t is greater than the
minimum required illumination. In addition, the lighting system load of the REH is price
elastic, which is modeled using Kt , which is equal to zero during the peak prices and unity
during the off-peak prices.

Pool Pump

The installation of pool pump along with the swimming pool is required to allow the REH
to keep the water quality relatively high by treating the swimming pool water for a certain

24
period of time during the day. The mathematical model of its operational constraints are
as follows: 
0 or 1 if t∈ T ,
Si=PMP,t = (2.29)
0 if t∈
/T
X
Si,t = ROTi , ∀i = PMP, t ∈ T (2.30)
t∈T

Ui,t ≥ Si,t − Si,t−1 , ∀i = PMP, t ∈ T (2.31)

t
X
Ui,t ≤ Si,t , ∀i = PMP, t ∈ {EOTi + MUTi + 1, LOTi } (2.32)
k=t−MUTi +1

t
X
Ui,t ≤ 1 − Si,t−MDTi , ∀i = PMP, t ∈ {EOTi + MDTi + 1, LOTi } (2.33)
k=t−MDTi +1

t+MST
Xi
Si,k ≤ MSTi + M (1 − Ui,t ), ∀i = PMP, t ∈ T (2.34)
k=t

The above set of constraints ensure that the operation of pool pump is within the customer’s
preferred pool operation time.

Stove

The mathematical model of the stove is presented to help determine the optimal operation
time, based on the REH customer’s objective, within their preferred operation time period.
The mathematical optimization model of the stove is shown below.

0 or 1 if t∈ T , i = STV
Si,t = (2.35)
0 if t∈
/ T , i = STV

Ui,t ≥ Si,t − Si,t−1 , ∀i = STV, t ∈ T (2.36)

25
 X
Si,t = ROTi , ∀i = STV, t ∈ T (2.37)
t∈T

t
X
Ui,k ≤ Si,t , ∀i = STV, t ∈ {EOTi + MUTi + 1, LOTi } (2.38)
k=t−MUTi +1

t+MST
Xi
Si,k ≤ MSTi + M (1 − Ui,t ), ∀i = STV, t ∈ T (2.39)
k=t

Washer and Dryer

The mathematical model of the washer and dryer is presented below:


0 or 1 if t∈ T , i = {W, Dry}
Si,t = (2.40)
0 if t∈
/ T , i = {W, Dry}

Ui,t − Di,t = Si,t − Si,t−1 , ∀i = {W, Dry}, t ∈ T (2.41)

Ui,t + Di,t ≤ 1, ∀i = {W, Dry}, t ∈ T (2.42)

X
Si,t = ROTi , ∀i = {W, Dry}, t ∈ T (2.43)
t∈T

t+MST
Xi
Si,k ≤ MSTi + M (1 − Ui,t ), ∀i = {W, Dry}, t ∈ T (2.44)
k=t

t
X
Ui,t ≤ Si,t , ∀i = {W, Dry}, t ∈ {EOTi + MUTi + 1, LOTi } (2.45)
k=t−MUTi +1

t
X
Di,k ≤ 1 − Si,t , ∀i = {W, Dry}, t ∈ {EOTi + MDTi + 1, LOTi } (2.46)
k=t−MDTi +1

The constraints (2.40) to (2.46) are modeled to provide the REH customer with the optimal
operation schedule for both the washer and dryer based on the customer preferences. The
washer and dryer are required to operate in a chronological sequence, where the operation
of the dryer should follow the operation of the washer. The following constraints are

26
included to coordinate the sequential operation of the washer and dryer.

t−MUT
XW
UDRY,t ≤ UW,t−k
k=t−MW,DRY

∀i = {W, Dry}, t ∈ {EOTi + MW,DRY , LOTi − MUTW } (2.47)

SDRY,t + SW,t ≤ 1, ∀t ∈ T (2.48)

X X
UDRY,t = UW,t (2.49)
t∈TDRY t∈TW

Water Heater

Similar to the mathematical model of the fridge, the water heater model is presented below:

0 or 1 if t∈ T , i = WH
Si,t = (2.50)
0 if t∈
/ T , i = WH

1 if Θi,t=0 = Θmin
i,t , i = WH
Si,t=1 = (2.51)
0 if Θi,t=0 = Θmax
i,t , i = WH

Θmin max
i,t ≤ Θi,t ≤ Θi,t , ∀i = WH, t ∈ T (2.52)
h i
Θi,t = Θi,t−1 + τ αi Si,t − βi HWUt − γi , ∀i = WH, t ∈ T (2.53)

Constraints (2.50) to (2.53) represent the mathematical model of the water heater.
The allowable operating time is restricted by (2.50), while the initial operational decision
is determined based on constraint (2.51). The allowable water temperature within the
appliances is constrained by (2.52), and dynamically modeled as in (2.53).

27
Photovoltaic Panel with a Battery

REHs with PV panels are assumed to be equipped with a 3.6 kW PV panel; the modified
model with more detailed and realistic architecture is discussed below:
Solar Panel :
PP V,t = PPLDC BP V H
V,t + PP V,t + PP V,t , ∀t ∈ T (2.54)

As shown in (2.54), the PV panel generates power, and the HEMS optimizes its usage by
distributing it to the battery, household, and grid,
Battery:
hn o n o
BP V
EBP V,t = EBP V,t−1 + τ PLDC,t + PPBP V
V,t
LDC H
η1 − PBP V,t + PBP V,t /η2 ], ∀t ∈ T (2.55)

min max
EBP V ≤ EBP V,t ≤ EBP V , ∀t ∈ T (2.56)

Equation (2.55) presents the inter-temporal changes of energy level of the battery and
(2.56) defines the limits on its energy level. The charging and discharging relationships are
formulated as follows:
BPV BPV
PCBPV ,t = PLDC,t + PPV,t , ∀t ∈ T (2.57)
LDC H
PDBPV ,t = PBPV,t + PBPV,t , ∀t ∈ T (2.58)

In (2.57), the battery can be charged from the PV panel power, or grid power; (2.58)
denotes that the battery can discharge power to the house, or sell to the grid.

PCBPV ,t ≤ SCBPV ,t PCmax

BPV
, ∀t ∈ T (2.59)

PDBPV ,t ≤ SDBPV ,t PDmax

BPV
, ∀t ∈ T (2.60)

Constraints (2.59) and (2.60) represent the maximum charging and discharging power of
the battery; the charging and discharging processes do not occur simultaneously, as ensured
by the following:
SCBPV ,t + SDBPV ,t ≤ 1, ∀t ∈ T (2.61)

28
Energy Storage System

ESS LDC H
EESS,t = EESS,t−1 + τ [PLDC,t η1 − (PESS,t + PESS,t )/η2 ], ∀t ∈ T (2.62)
min max
EESS ≤ EESS,t ≤ EESS , ∀t ∈ T (2.63)
ESS
PLDC,t ≤ SCESS ,t PCmax
ESS
, ∀t ∈ T (2.64)
LDC H
PESS,t + PESS,t ≤ SDESS ,t PDmax
ESS
, ∀t ∈ T (2.65)

SCESS ,t + SDESS ,t ≤ 1, ∀t ∈ T (2.66)

Equation (2.62) relates the change in energy level of the ESS as affected by the power
drawn, and discharged by the ESSs to the grid, and to the house. Note that the self-
discharge phenomenon of BESS is neglected in this thesis and hence not included in the
mathematical models of the ESS. Constraint (2.63) ensures that the ESS energy level is
within the minimum and maximum limits; and limits on charging and discharging power
of the ESS are given by (2.64) and (2.65), respectively. Constraint (2.66) ensures that the
charging and discharging process does not occur simultaneously.
Table 2.1 shows the typical parameters and preferences for the REH.

2.3 Energy Storage Systems

2.3.1 Energy Storage System Characteristics

The main characteristics and terms associated with ESSs are discussed here pertain to
their integration with the power grid [67].

1. Storage capacity: it quantifies the available energy in the ESS after a full charging
process, and is defined by the total energy stored (kWh). The amount of available
energy that can be used, is governed by the Depth of Discharge (DoD). Furthermore,
fast charging and discharging processes cause the ESS’s efficiency to deteriorate,
which decreases the obtained energy from the storage capacity.

29
 Table 2.1: Parameters of the HEMS Model of a Typical REH [66]
Device Parameter Setting Value Parameter Setting Value Parameter Setting Value
Θf r0 3 αf r 2.75 Pf r 0.6
Fridge (FR) Θmin
fr 0 βf r 0.605
Θmax
fr 10 γf r 0.14
Θwh0 50 αwh 4.4 Pwh 0.6
Water Heater (WH) Θmin
wh 48 βwh 0.068 HR wh 0.297
Θmax
wh 58 γwh 0.05 CG 0.029
EOT 65 ROT 12 Pstv 1.5
Stove (STV)
LOT 88 MUT 4 MSOT 12
EOT 65 MUT 8 PDW 0.7
Dishwasher (DW)
LOT 92 MDT 4 MSOT 8
EOT 64 MUT 8 Pw 0.45
Washer (W)
LOT 92 MDT 4 MSOT 8
EOT 64 MUT 8 Pdry 1.1
Dryer (DRY)
LOT 92 MDT 4 GapW,Dry 8
Lighting EOT 1 LOT 96 Pli 0.15
EOT 29 MUT 1 Ppmp 0.75
Pool (PMP)
LOT 96 MDT 1 ROT 40
Emin 1 PCmax
ESS
0.45 EOT 1
Energy Storage System (ESS) max
Emax 3.6 PESS 0.45 LOT 96
Emin 1 PCmax
B
0.45 EOT 1
PV Panel Battery (BPV)
Emax 3.6 PDmax
B
0.45 LOT 96

2. SOC: it indicates the charging level of an ESS with respect to the maximum amount
of energy that can be stored (considering its degraded state), typically expressed in
percent.

3. Ramp rate: the rate of change of the ESS output power. It is usually expressed in
terms of kW/s.

4. Round-trip efficiency: the ESS efficiency can be evaluated based on the energy loss
during charging and discharging processes. The round-trip efficiency represents the
relationship between the fraction of energy stored in the ESS to that can be retrieved.

5. Cycle and cycle life: a cycle of ESS can be defined as a discharging process followed
by a complete recharging process. The number of charge-discharge cycles that can

30
 be performed by an ESS until it reaches the stage when it is no longer suitable for a
given application is called Cycle Life.

6. C-Rate: under ideal conditions, C-Rate = 1 means the total energy of the ESS is
depleted in one hour, C-Rate = 2 means the energy depletion rate is doubled, i.e.,
depletes in 30 minutes, and so on.

7. Derating factor: it is a multiplier applied to the ESS energy capacity or rated power
in order to take into account the SoH and degradation of the ESS.

The above characteristics vary in their values from one ESS technology to another.
Different ESS technologies are available to provide specified services to the power grid.
This thesis assumes BESS as the sole form of ESS which is present in a distribution system.
BESS can be classified as solid-state and flow batteries. The first includes Li-ion, lead-acid
(PbO2 ), nickel-cadmium, and sodium sulphur technologies (NaS), while the latter includes
vanadium redox (VRB), polysulphide bromide, and zinc-bromine. BESS operational life
and cycle life depends upon the number of charge-discharge cycles. In addition, two factors
play an essential role in determining the amount of energy that can be delivered and the
degradation of the BESS, namely, the SOC and the DoD.
BESSs are one of the most mature energy storage technologies. These are able to deliver
electrical energy using the chemical energy generated by electrochemical reactions [68].
The capacity of BESS varies from less than 10 kWh to several MWh with efficiency range
between 60 − 90% depending in two factors: operational cycle and electrochemistry type.
Li-ion batteries have found several applications in power systems, their main advantages
include high efficiency, fast response time, and significantly high life cycle [69].

2.4 Cycle Counting Using RCA

The RCA is a well-known methodology applied in material fatigue analysis, which can be
used for fatigue life prediction [70]. it captures and extracts the irregular cycles from a
loading history of an object.

31
 The RCA has recently been used [71] to count the number of operational cycles of a
BESS which is subject to degradation in its life due to charging and discharging stress.
The stress on the BESS is represented in the form of DoD. In other words, a high DoD
reduces the lifetime of the BESS, as shown in Figure 2.1 [72]. In addition, the operation
of the BESS usually has irregular cycles, which can be counted using the RCA and used
to study their impact on the asset life.

8000

7000

6000
Cycles to failure

5000

4000

3000

2000

1000

0
10 20 30 40 50 60 70 80
DOD(%)
Figure 2.1: Typical cycles to failure vs. DoD curve of Li-ion batteries [72].

The RCA counting method includes four main steps: hysteresis filtering, peak-valley
filtering, discretization, and a counting method (e.g. Four Point Method or Pagoda roof
method). Hysteresis filtering aims to eliminate the short cycles which do not significantly
contribute to the total damage of BESS. Peak-valley filtering comes after to identify the
peaks and the valleys reversal in the operation profile of ESS. A peak is formed in the
SOC profile when the change in SOC is negative, while a valley is formed when a change
in the SOC is positive. Figure 2.2 and 2.3 presents an example SOC profile, where the
alphabetical enumeration shows peak and valley points over time. Also, the green and red
lines demonstrate the charging and discharging reversals, respectively, where each reversal
represents a half cycle with a particular amplitude. In Pagoda roof method, the green
and red lines represent flow of water from a number of Pagoda roofs. A half cycle can be
counted at the end of a green or red line, which shows the termination of a water flow.

32
 B J
100
F
90
D H
80

70

60
SoC (%)

A G K
50
4
C 5
40
2 I
30
3
E
20

10
1

1 2 3 4 5 6 7 8 9 10

Time Interval (hr)

Figure 2.2: Peak filtering of an ESS operation profile.

The following cases show the possible termination scenarios of the water flow:

• The end of the flow can happen when a higher peak is reached, in peak filtering, or
lower valley, in valley filtering. In Figure 2.3, water flow 1 encounters a lower valley
C as compared to its starting valley A, and this counted as a half-cycle.

• A water flow merges with another water flow which started at an earlier peak or
valley. As in Figure 2.2, the flow-2 merges with flow-1.

• A water flow arrives at the end of the time horizon, as shown in Figure 2.3, flow-3.

33
 1 3
B J
100
F
90 2 4
D H
80

70 5

60
SoC (%)

A G K
50
C
40
I
30
E
20

10

1 2 3 4 5 6 7 8 9 10

Time Interval (hr)

Figure 2.3: Valley filtering of an ESS operation profile.

Table 2.2 shows the range of each reversal and the cycle count. The obtain table from
RCA then reduces by combining the cycles with the same range. Finally, the counted cycle
can be used to estimate the cumulative damage on an asset such as BESSs.

2.5 Artificial Neural Network

The NN can be defined as [73], “A neural network is an interconnected assembly of

simple processing elements, units or nodes, whose functionality is loosely based on the

34
 Table 2.2: Rainflow Counting Algorithm Result
From To Absolute Range Cycle Count
A B 50 0.5
B C 60 0.5
C D 40 0.5
D E 60 0.5
E F 70 0.5
F G 40 0.5
G H 30 0.5
H I 50 0.5
I J 70 0.5
J K 50 0.5

x1 xN
1
N1 w11
nin
1
P
N1

w1h nout
1
W1

β1
.. .. P
N1
XN X

wi1 Γ
nin nout
h
Wh
h
P
xi Nh
xN wih
i
Ni

βh

Input Layer Neurons Hidden Layer Neurons Output Layer

Figure 2.4: NN structure with a single hidden layer and output.

35
animal neuron. The processing ability of the network is stored in the inter-unit
connection strengths, or weights, obtained by a process of adaptation to, or learning from,
a set of training patterns”. An NN consists of different layers, namely, input layer, hidden
layers, and output layer, as in Figure 2.4. Each layer consists of a number of basic units
called neurons. The input of the NN goes through different mathematical
transformations, as follows:

Pre-processing input and post-processing output

Before training a NN, data preparation is required to normalize and standardize the input
and output variables to re-scale the data into an acceptable range for the NN, hence
improve its performance [74]. Different scaling techniques can be applied in pre-processing
the inputs such as Min-Max normalization, mean normalization, and Z-score normalization.
Min-Max normalization, for example, maps the range of features as follows:

2 (xi − x)
xN
i = −1 (2.67)
(x − x)

Where, xN
i is the normalized vector of xi . The equation maps the feature xi , given that
its minimum and maximum values x and x, respectively, falls between [-1,1].

Training

The input neurons pass the pre-processed data to the neurons of the subsequent layers. A
neuron i ∈ I of the input layer is linked with all neurons of the hidden layer h ∈ H by
weights w as shown below:
 
w11 w12 · · · w1h
 w21 w22 · · · w2h 
 
[w] = 
 .. .. . . . .. 
 (2.68)
 . . . 
wi1 wi2 ... wih

36
 The input to a hidden layer neuron can be represented as follows:

I
X
nin
h = wih xN
i + βh , ∀h (2.69)
i=1

The first component of (2.69) represents the weighted sum of the NN inputs, while the
second component is the bias (βh ) of the hidden layer neuron h. The input to a hidden
layer neuron, then, is processed using an activation function to calculate the hidden layer
neuron’s output. There are several types of activation functions such as Sigmoid function,
Hyperbolic Tangent Function, and Softmax Function. Using the commonly used Sigmoid
activation function, the hidden layer neuron output is:

2
nout
h = −1 (2.70)
1 + exp (−2 nin
h)

The calculated output nout

h of the hidden layer neuron using (2.70) is passed through
the activation function of the output layer neuron, which can be a linear transfer function,
as follows:
XNH
N
X = nout
h Wh + Γ (2.71)
h

Finally, the normalized output (X N ) of the output layer neuron represents the
normalized predicted output value, which can be post-processed to obtain the required
output value, as presented below:

(X N − X N )(X − X)
X= +X (2.72)
2

2.6 Cooperative Game Theory

Game theory can be defined as the field of mathematical modeling of the interaction
between a number of rational players [75], where each player’s decision making process is
impacted by other players. Such competitive behavior is classified into two categories: non-

37
cooperative and cooperative games. The first presents the strategic behavior of individual
players without an external entity that forces rules, so as to maximize their own benefits.
On the other hand, the second is governed by a set of rules to maximize the benefit of
a coalition of players. In this thesis, a coalitional game is considered, where a group of
BESSs acts together to maximize their payoffs and provides different services to the LDC.
The redistribution of the total system saving between BESS’s owners are achieved using
the Shapley value criterion.
The Shapley value criterion associates a unique payoff for every individual within a
coalition. The payoffs should satisfy the following axioms [76]:

• Additivity: A sum of the cost saving, which a player receives as a member of every
coalition formed with other players, represents the cost saving allocated to that player.

• Dummy player: If the total value of different possible coalitions remains the same
when a player joins the coalition, it is called a dummy player.

• Efficiency: The sum of the players’ payoffs represents the value of the grand coalition.
In other words, the payments given to the players in a coalition represents the total
payoffs that can be achieved from their cooperation.

• Symmetry: If two players are identical to each other with no change in the coalition’s
value, those players are symmetrical.

Therefore, the Shapley value criterion represents the marginal contribution of a player to
a coalition, which is mathematically given as follows [77]:

v(h) = v(S) − v(S − h) (2.73)

The marginal contribution of a player h to a coalition S is given by the change in the

utility value of the coalition after h’s joining the coalition. The sum of a player h’s marginal
contributions is divided by c(q), the number of all formed coalitions of size q that contains
the player h, and n is the total number of players. The Shapley value can be calculated as

38
follows: n
1 X 1 Xh i
φh = v(S) − v(S − h) (2.74)
n q=1 c(S) h∈S

where,
(n − 1)!
c(q) = (2.75)
(n − q)!(q − 1)!

2.7 Summary

In this chapter, some essential background topics required for this research were introduced.
The concepts of REH and HEMS with their mathematical model were discussed. A brief
discussion of the RCA and NN were presented. Finally, cooperative game theory definition
and the Shapley value were discussed.

39
Chapter 3

Flexibility of Residential Loads for

Demand Response Provisions in
1
Smart Grid

This chapter proposes a two-stage optimization framework wherein multiple HEMSs

simultaneously optimize their respective energy consumption patterns and determine
their flexibility provisions, which is communicated to the LDC. The LDC aggregates the
controllable demand profile and the flexibilities of each HEMS to optimize its operational
performance and hence the peak reduction signals which are sent to the HEMSs. Studies
are carried out considering a 33-bus distribution system coordinating with 1,295 houses,
each with varying customer preferences and objectives, to demonstrate the applicability
of the proposed scheme.
The mathematical model of the HEMS was discussed in Chapter 2; the LDC’s
operational model and the proposed coordination between the REHs and the LDC is
discussed here.

1
This chapter has been published as:
O. Alrumayh and K. Bhattacharya, “Flexibility of Residential Loads for Demand Response Provisions in
Smart Grid,” in IEEE Transactions on Smart Grid, vol. 10, no. 6, pp. 6284-6297, Nov. 2019.

40
 3.1 Nomenclature
Indices and Sets
hj Index for household located at bus j

i Index for household appliances, i ∈ A

j,k Index for bus (j = s, substation), (j, k) ∈ N

q Index for Energy Storage Device (ESS) and Battery (B)

t Index for time, t ∈ T

H Set of households

Parameters

γ FLEX Objective function weight assigned by LDC on scheduled flexibility

min , E max
Eq,h Minimum and maximum energy level, kWh
j ,t q,hj ,t

EBInitial , EBFinal Initial and final energy level of battery, kW h

AR
Eq,h Energy level at arrival, kWh
j ,t

Gj,k Conductance of feeder between buses, p.u

gi,hj Gas consumption rate, m3 /hr

PCmax
q,h
, PDmax
q,h
Maximum charging/discharging power, kW
j j

PD0hj Connected load in house, kW

Pgmin
j
, Pgmax
j
Minimum and maximum generation level, p.u

PLIhj Rated power of the lighting system, kW

Pi,hj Rated power of appliance, kW

d
Pj,t Active power drawn by residential loads, p.u
Imp
Pj,t Total power purchased by LDC from residential loads, p.u
CAP
Pj,t Peak load of the feeder, p.u

41
PhFLEX
j ,t
Customer flexibility available, kW
FLEX
Pj,t Aggregated bus-wise flexibility, kW

PP Vhj ,t PV panel generation in house, kW

Qdj,t Reactive power drawn by residential loads, p.u

tAR DEP
hj , thj Electric vehicle arrival/departure from/to house hj , min

Vjmin , Vjmax Minimum and maximum voltage level, p.u

Yj,k Magnitude of admittance matrix element

β Peak load factor

η1 , η2 Charging/discharging efficiency, %

ωhj Charging level factor at departure of household hj

ρTOU
t ,ρSEL , ρGAS Time of use price $/kWh, selling price to the LDC $/kWh, and Gas price $/m3

τ Length of the time interval

θj,k Angle of complex Y-bus matrix element, p.u

Θset
i,hj Preferred temperature setting of appliance, ◦ C

Variables

Eq,hj ,t Energy level, kWh

Lhj ,t Illumination level

PCB,hj ,t Total charging power drawn by Battery, kW

PCPB,h
V Generated power by PV solar panel used to charge the battery, kW
j ,t

PCLDC
q,h ,t
Charging power drawn by device, kW
j

PDB,hj ,t Total discharged power from Battery, kW

PDHq,h Discharging power from device to house, kW

j ,t

PDLDC
q,h ,t
Discharging power from device to LDC, kW
j

42
 g
Pj,t Active power drawn from substation at bus, p.u
FLEX
Pj,t Active power flexibility scheduled, p.u
H
PLDC Delivered power by LDC to meet house demand, kW
h j ,t

i
PLDC Delivered power by LDC to meet appliance i demand, kW
h j ,t

H
PPV,h PV generated power consumed by household appliances, kW
j ,t

LDC
PPV,h Generated power by PV solar panel sold to LDC in house, kW
j ,t

Phmax
j ,t
House hj maximum demand, kW

Pbhmax
j ,t
Updated house maximum demand, kW

Qgj,t Reactive power drawn from substation at bus, p.u

QFLEX
j,t Reactive power flexibility, p.u

SCq,hj ,t Binary charging status of device, ON/OFF

SDq,hj ,t Binary discharging status of device, ON/OFF

Si,hj ,t Binary status of device, ON/OFF

Vj,t Voltage at bus, p.u

δj,t Voltage angle at bus, p.u

Θin
i,hj ,t Temperature level of appliance,◦ C

αhj ,t flexibility index of customer

43
3.2 Proposed Coordination Framework and
Mathematical Models

In [22] and [65], a comprehensive representation of a generic REH was presented, and which
was equipped with an HEMS. The REH was equipped with the following appliances: fridge,
water heater, lighting system, air-conditioning, dishwasher, washer, dryer, stove, and pool,
an ESS and a rooftop PV panel (see Fig.3.1). The charging and discharging operations of
the ESS are controlled by the HEMS; the ESS is charged by the power drawn from the
grid, while at peak demand hours, it can discharge power to the grid and earn revenue.
The PV panel battery can be charged either by the power drawn from the grid or from
the PV panel directly. The stored energy in the PV panel battery can be used to supply
the REH or sold to the grid. The HEMS controls the PV panel operation also, i.e., how
much is allocated to meet the house load, to charge the PV panel battery, or sell to the
grid. Indeed, the ESS and the PV panel battery creates an additional power demand of the
REH when in charging mode. The PV generation depends on the weather conditions and
PV panel features. Note that while the PV panel batteries represent stationary storage,
the ESSs essentially account for the mobile storage devices present in the house, such as
PEV.
In this work, a two-stage framework is proposed to aggregate and coordinate the
contributions of a large number of HEMS within REHs to provide a flexibility service to
the LDC. The HEMS mathematical model is formulated by significantly improving upon
the model in [22], which optimizes the operations of typical home appliances and ESSs to
determine the optimal interactions with the LDC such as the power to be sold to or
purchased from the LDC, while considering different objectives of the REH. Thereafter, a
residential feeder is considered, where, at each bus, several such REHs, equipped with
HEMS are assumed being served by the LDC; each home receives a control signal from
the LDC pertaining to a cap on its hourly load, based on which each HEMS re-optimizes
their loads to meet the prescribed cap. This creates an aggregated flexibility capacity at
each bus, through load shifting, HEMS optimization, and LDC imposed load cap at the
feeder bus. This proposed decentralized approach maintains privacy of the individual
customers, not requiring them to provide their electricity usage profiles to the LDC; each

44
 Utility Grid

𝑳𝑫𝑪
𝑷𝑭𝑳𝑬𝑿
𝒉𝒋 ,𝒕 𝑷𝑳𝑫𝑪
𝑫𝒒,𝒉,𝒕 + 𝑷𝑷𝑽,𝒉,𝒕
𝑷𝒎𝒂𝒙
𝒉,𝒕
𝑅𝐸𝐻 𝑀𝑜𝑑𝑒𝑙
𝐿𝐷𝐶
𝑃𝐵,ℎ,𝑡 𝐿𝐷𝐶
𝑃𝑃𝑉,ℎ,𝑡
Circuit
Breaker
𝐻
𝑃𝑃𝑉,ℎ,𝑡
PV PANEL
𝐻
𝑃𝐿𝐷𝐶 ℎ ,𝑡

𝑃𝐷𝐻𝐸𝑆𝑆 ,ℎ,𝑡
𝑃𝐷𝐿𝐷𝐶
𝐸𝑆𝐷 ,ℎ,𝑡

𝑃𝐶𝑃𝑉
BATTERY WITH
𝐵,ℎ,𝑡 𝑃𝐶𝐿𝐷𝐶
𝐸𝑆𝑆 ,ℎ,𝑡
ESS
PV PANEL
(BPV) 𝑃𝐶𝐿𝐷𝐶
𝐵,ℎ ,𝑡
𝑃𝐷ℎ𝐵,ℎ ,𝑡

Figure 3.1: Schematic of a Residential Energy Hub.

HEMS achieves its own objective, while the overall framework converges to an
equilibrium, to the satisfaction of all parties. Such coordination between the HEMS and
the LDC can improve grid operations by flattening the system load profile, improving bus
voltages, reducing losses, and LDC’s operating costs. Indeed, the work considers a
realistic scenario where multiple REHs with different behaviors, preferences, and
locations, are coordinated to provide flexibility to the LDC taking advantage of their
inherent elasticity.
Figure 3.2 shows a detailed layout of the proposed coordination scheme. This layout
has three levels, namely, REH level, LDC level, and coordination level.

3.2.1 Residential Energy Hub and HEMS (Stage: Ia, Ib, Ic)

The following three stages are required to present different REH customer consumption
behaviors, to formulate the HEMS mathematical model, and to evaluate the REH

45
 STAGE Ia: Development of large data set for unique representation of each HEMS parameters
 Example of HEMS model parameters: early/late operation time of washer and dryer, minimum/maximum
preferred inside house temperature, required operation time of dishwasher.

STAGE Ib: Development of HEMS mathematical model of large set of REHs

REH LEVEL

 Household appliances: Fridge, Dishwasher, Washer, Dryer, Water Heater, Pool Pump, PV Panel, ESS, and
Lighting System.
 HEMS objective can be: Minimize Cost, minimize energy, or maximize comfort level.
 Output: the operation of each appliance and the total load profile of the house.

STAGE Ic: Flexibility evaluation

 The available flexibility at each REH can be obtained by comparing the customer load profile while
minimizing cost or maximizing comfort with that obtained from minimizing energy. The difference represents
the REH flexibility.
 Each REH flexibility is represented by a flexibility index. The total flexibility at a location in the system is
hence obtained.
LDC LEVEL

STAGE II: Development of OPF to represent LDC

 Objective: minimizing the cost of power drawn by LDC to meet the residential demand.
 Subject to: power balance equations, angle and voltage limits, locational demand response availability, power
drawn limit, and hourly peak demand.
 Output: Total energy cost, total system losses, and required demand response.
COORDINATION LEVEL

STAGE III: Development of a coordination scheme to provide flexibility services by utilizing individual
REH flexibility
 Customers with flexibility at a particular location in the distribution system can participate.
 If flexibility services required at a bus, the REH participates to meet the requirement based on their evaluated
flexibility index.

Figure 3.2: Required levels and stages of the proposed coordination scheme.

flexibility.
Stage-Ia: In order to represent each HEMS individually, unique objective functions,
activity levels, and customer preferences (e.g. the required operation time and minimum
up/down time of washer, dryer, stove) are generated randomly.
Stage-Ib: The HEMS mathematical model is formulated and solved considering the
mathematical model (2.4)-(2.66) discussed in Chapter-2 with some modifications to present
the two-way communication between residential loads and the LDC, discussed later.

46
Objective Functions

Three different objectives are used for the multiple HEMSs considered in this work, as
given below:
Minimize Cost: the objective of REH hj located at bus j is to minimize its total cost
of energy, net of revenue earned.
"
X n X X
J1hj = ρTOU
t Pi,hj Si,hj ,t + PLIhj Lhj ,t
t∈T i∈A z∈LI
i6∈{LI,ESS,BP V }
X o X
i
+ PLDC hj ,t
+ PD0hj + ρGAS gi,hj Si,hj ,t
i∈{BPV,ESS} i∈{HT,W H}
#
n X o
− ρSEL PDLDC
i,h ,t
LDC
+ PPV,hj ,t
, ∀ hj ∈ H (3.1)
j
i∈{BPV,ESS}

In (3.1), the terms within the first curly brackets represents the power drawn by
household appliances, the lighting system, the charging power drawn by the ESS and PV
panel battery, and the uncontrolled load; which all together, denote the total power
drawn from the grid. The term associated with ρGAS presents the total cost of gas
consumption. The discharged power from ESS and PV panel battery accounts for the
power sold to the grid.
Minimize Energy Consumption: the REH minimizes its total energy consumption
over the day.
"
X X X
J2hj = Pi,hj Si,hj ,t + PLIhj Lhj ,t +
t∈T i∈A z∈LI
i6∈{LI,ESS,BPV}
#
X X
i
PLDCh
+ PD0hj + gi,hj Si,hj ,t , ∀hj ∈ H (3.2)
j ,t
i∈{BPV,ESS} i∈{HT,WH}

Maximize Comfort Level: the REH maximizes its comfort level by minimizing the

47
temperature deviations of certain appliances from their preferred set points.
" !2 #
X X
J3hj = Θin
i,hj ,t − Θset
i,hj , ∀hj ∈ H (3.3)
t∈T i∈{AC,HT,WH,FR}

Note that (3.3) is linearized using the method reported in [78] to arrive at an MILP
model.
Appliance Constraints: the REH hj appliances operational model was discussed in
detail in Chapter-2 and it included the equations (2.4) to (2.66). In the present work, the
REH model is further improved by incorporating the following constraints:
Peak Load Constraint: Ensures that the total load of REH hj is within a limit
specified by the LDC, Pbhmax
j ,t
.
X X X
LDC
Pi,hj Si,hj ,t + Pi,h j ,t
− PDHi,h − PPHV,hj ,t ≤ Pbhmax
j ,t
, ∀hj ∈ H (3.4)
j ,t
i∈A i∈{BPV,ESS} i∈{BPV,ESS}

In (3.4), the REH load includes the appliances’ consumption; charging power of ESS
and PV panel battery, net of the power discharged from the ESS; the PV panel battery;
and that generated by the PV panel to supply some of the REH loads.

Pbhmax
j ,t
= Phmax
j ,t
FLEX
− αhj ,t Pj,t , ∀t ∈ T ; ∀hj ∈ H ; ∀j ∈ N (3.5)

Constraint (3.5) represents the two-way communication between the LDC and the
HEMS- the updated maximum demand of REH hj , Pbhmax j ,t
, is based on the previous
max
maximum allowable demand, Phj ,t , and a portion of the flexibility requested by the LDC
at bus j; αhj ,t represents the REH’s flexibility index, which is explained in Stage-III.
REH Power Balance: Ensures that the total power demand of the household
appliances is met by the power drawn from the grid and power discharged to the REH
from the ESS and PV panel battery, and that generated by the PV system, as given
below:
X X
H
Pi,hj Si,hj ,t = PLDC,h j ,t
+ PDHi,h H
+ PPV,hj ,t
, ∀t ∈ T ; ∀hj ∈ H (3.6)
j ,t
i∈A i∈{B,ESS}

48
 The optimal decisions of the HEMS includes the operating decisions of appliances, the
energy sold or purchased to/from the utility grid, the charging/discharging decisions of the
ESS and the PV panel battery, and the resulting aggregated load profile of the REH.
Stage-Ic: In this last stage of the house level, flexibility provisions of the REH is
determined as follows:
" # " #
X X
4 Phj ,t = Pi,hj Si,hj ,t − Pi,hj Si,hj ,t , ∀t ∈ T ; ∀hj ∈ H (3.7)
i∈A J1hj or J3hj i∈A J2hj

4Ph ,t
αhj ,t = P j , ∀t ∈ T ; ∀hj ∈ H (3.8)
4Phj ,t
h∈H

The flexibility of the REH, given by (3.7), is obtained from the difference between its
scheduled load at an hour using the energy cost minimizing model with that obtained from
the energy consumption minimizing model. For some REHs seeking to maximize comfort,
the flexibility is the difference between its scheduled load at an hour using the maximizing
comfort level model with that obtained from the energy consumption minimizing model.
The total flexibility at a bus is the sum of the flexibility from all REHs connected at
P
the bus, given by, 4Phj ,t . The REH flexibility index αhj ,t in (3.8), is obtained by
h∈H
normalizing an REH’s flexibility by the total flexibility at that bus. The optimal flexibility
is proportionally allocated to each REH based on the value of αhj ,t .

3.2.2 LDC Operations Model (Stage-II)

The objective of the LDC is to minimize total system losses and a weighted component of
the total flexibility, as given below:

N X
X h1 X N   XN i
2 2 FLEX FLEX
J= Gj,k Vj,t + Vk,t − 2Vj,t Vk,t cos(δj,t − δk,t ) + γ Pj,t (3.9)
t∈T
2 j=1 k=1 j=1

49
 No. of Houses Located at Bus i

35 2

31 3

31 19 42 4

31 20 5 31 23
21

31 21 6 147 24
21

31 22 7 147 25
70
8
Pj,tFLEX) sent by LDC to neighborhood
70
Flexibility request (
21 9
26 LDC
21
21 10
21 27

15 11
28
21 Send Aggregated neighborhood information to LDC
12
(Pj,tImp, Pj,td , ∑∆Ph ,t j)
21 29
42
21 13
70 30 h∈H

42 14
52 31
21 15
73 32
21 16
21 33
21 17

31 18

Figure 3.3: Schematic of coordination between HEMS and LDC.

and subjected to the following power flow equations,

N
X
g Imp FLEX d
Pj,t + Pj,t + Pj,t − Pj,t = Vj,t Vk,t Yj,k
k=1
 
cos θj,k,t + δk,t − δj,t , ∀t ∈ T ; ∀(j, k) ∈ N (3.10)

N
X
Qgj,t + QFLEX
j,t − Qdj,t = − Vj,t Vk,t Yj,k
k=1
 
sin θj,k,t + δk,t − δj,t , ∀t ∈ T ; ∀(j, k) ∈ N (3.11)

Where,
H  X
X 
Imp
Pj,t = PDLDC
i,hj ,t
+ LDC
PPV,h j ,t
, ∀t ∈ T ; ∀j ∈ N (3.12)
h i∈{B,ESS}

50
In (3.11), Qdj,t and QFLEX
j,t
d
are obtained from Pj,t FLEX
and Pj,t a using constant load power
Imp
factor of 0.9, Pj,t is the aggregate power import by the LDC from all REHs located at
bus j, as given in (3.12). And, given that,

H 
X X X
d
Pj,t = Pi,hj Si,hj ,t + PLIhj Lhj ,t
h i∈A z∈LI
i6∈{LI,ESS,BP V }
X 
i
+ PLDC h
+ PD0hj , ∀t ∈ T ; ∀j ∈ N (3.13)
j ,t
i∈{BPV,ESS}

d
In (3.13), Pj,t represents the aggregated load at bus j and time t from multiple REHs. The
LDC operations model also includes constraints on bus voltages and power drawn from the
substation, as given below:

Vjmin ≤ Vj,t ≤ Vjmax , ∀t ∈ T ; ∀j ∈ N (3.14)

g
Pgmin
j
≤ Pj,t ≤ Pgmax
j
, ∀t ∈ T ; ∀j = s (3.15)

3.2.3 Coordination of Multiple HEMS and LDC (Stage-III)

A novel coordination scheme between the multipleREHs and the LDC is presented herein.

• It is assumed that all REH are equipped with HEMS which receive input updates
such as customer preferences and weather forecast, as shown in Fig.3.4 under “Data
collection and analysis”.

• The REHs are clustered in three groups based on their objective function and their
possession of ESS and PV panel. The HEMSs then optimizes the operations, using
the optimization models OPT.1, OPT.3, and OPT.5, of their respective household
appliances considering one of: cost minimization or comfort maximization objective,
and appliance operational constraints, see Fig.3.4. Furthermore, the last process at
the house level is to evaluate the available household flexibility.

51
• The resulting optimal load profile and other optimal decisions, such as how much
power the house can sell to the distribution grid, and the available flexibility are
Imp d FLEX
communicated to the LDC, one day ahead, Pj,t , Pj,t , and Pj,t (see Fig.3.3 and
3.4).

• The LDC develops a bus-wise load profile by aggregating the customers’ optimized
load profiles, executes the LDC operations model, and hence evaluates its system
operating conditions considering its objective and constraints (3.9)-(3.15). In
addition to the mentioned constraints, the LDC can take advantage of the inherent
flexibility of the customers by imposing a peak demand cap. Such decision is
limited by the customers’ evaluated flexibility, see constraint (3.16). This cap
changes the operation schedules of the participating customers and benefits the
cap
LDC. The peak load of the feeder at bus j, hour t, is Pj,t ; when the LDC has to
FLEX
request for flexibility from customers, it sends a request signal Pj,t , as shown in
FLEX
Fig.3.3; with an appropriate choice of β, where Pj,t is limited by the aggregated
customers’ flexibility, at bus j, as given below:

cap FLEX FLEX FLEX

βPj,t ≤ Pj,t ≤ Pj,t if Pj,t ≥ 0, ∀ j ∈ N
cap FLEX FLEX FLEX
− βPj,t ≥ Pj,t ≥ Pj,t if Pj,t ≤ 0, ∀j∈N (3.16)

FLEX
In (3.16), Pj,t positive denotes that the customers are willing to reduce their
consumption at bus j and time t, from their optimal schedule to the minimum energy
FLEX
consumption operation. While, Pj,t negative indicates the customer is willing to
increase its consumption.

• From this analysis, the hour-wise optimal flexibility use required by the LDC is
determined considering the bus-wise flexibility obtained from (3.7)-(3.8).

• This bus-wise, hourly, optimal flexibility is then proportionally allocated to each

house based on the flexibility index αhj ,t of the customer, and communicated to
individual HEMS. The respective HEMS in turn incorporates these signals as a ‘peak
demand cap’ (3.5), and re-optimizes their respective appliance schedules.

52
 • Note that, a household will only submit to the LDC its aggregate load profile, how
much it would sell/buy to/from the grid, and how much flexibility provision is
available next day. There will be no exchange of private information of the
customer or its activity schedules.

• The HEMS and the LDC optimization models are interlinked by external signals, as
Imp d FLEX
shown in Fig.3.3, wherein the LDC receives the Pj,t , Pj,t , and Pj,t signals from
FLEX
HEMS while the HEMS receives Pj,t signal from the LDC, based on which the
optimal decisions of each, are determined. This is a two-stage sequential process
involving information exchange and effective communication between the houses and
the LDC.

The proposed coordination scheme enables each REH to carry out its desired operation
based on its own objective while preserving information privacy and security while at
the same time providing a flexibility service to the LDC to improve grid operations and
reliability.

3.3 Results and Discussions

3.3.1 Case Study

In this work, the LDC is responsible for managing a 33-bus distribution system and meeting
the demand of 1,295 houses connected at its various buses. The number of houses at
each bus is shown in Fig.3.3 (within the circles), which is obtained assuming that the
connected load at a house is 7 kW. In order to represent each HEMS individually, unique
objective functions, activity levels, and customer preferences (i.e, the required operation
time and minimum up/down time of washer, dryer, stove, etc.) are randomly generated.
Furthermore, the TOU rates applicable in Ontario, Canada, are used in this study [79].
The list of parameters and their values, associated with a single typical HEMS model is
provided in the Appendix.
Three customer clusters are considered to evaluate the flexibility of residential
customers in providing DR services to the LDC: Cluster-1: 80% of the customers own

53
 REH Optimization Models

LDC Data Collection and Coordination Between LDC and

Data Collection and Analysis Cluster-1 Optimization Model REH Loads

OPT.1: Min Energy Cost (3.1) Data Collection Data Collection

Data Collection S.t. (2.4)-(2.53)
Load Profile of REHs  LDC receives the following  LDC sends bus-wise hourly
 Time of Use (TOU) ∆𝑃ℎ𝑗 ,𝑡 Power Sold to LDC hourly data for next day: data:
 Weather Data Cluster-1 from REHs Power Sold to LDC Bus-wise Flexibility Request
 Photovoltaic (PV) Power Bus-wise Load Profiles Hourly Flexibility Index of an REH
forecast OPT.2: Min Energy Consumption
(3.2) Bus-wise Flexibility Profiles Coordination
S.t. (2.4)-(2.53)
Data Analysis Bus-wise flexibility distribution
Optimization Model obtained as follows:
Simulate households appliances  Distribute𝑷𝑭𝑳𝑬𝑿
operational preferences and LDC 𝒉𝒋,𝒕 based on
each individual REH flexibility
settings
Cluster-2 𝐼𝑚𝑝 index 𝜶𝒉𝒋,𝒕 , see eq.(3.7)-
Cluster-1 𝑃𝑗,𝑡 OPT.7: Min Power Losses (3.9) (3.8)
Preferences: S.t. (3.10)-(3.16)
OPT.3: Min Energy Cost (3.1)
 Early/Late/Required Operation S.t. (2.4)-(2.66) 𝑃𝑗,𝑡𝑑
time of: Household owns PV and BPV
Stove Washer hj belongs to cluster? Cluster-2 ∆𝑃ℎ𝑗 ,𝑡
Outcome

54
Dishwasher Dryer b.10% of REH Cluster-2
belongs to Cluster-2. OPT.4: Min Energy Consumption  LDC model outcome: Is Yes
𝛼ℎ𝑗 ,𝑡 = 0
Settings: (3.2)
Power Losses ?
Cluster-3 S.t. (2.4)-(2.66) 𝑃ℎ𝐹𝐿𝐸𝑋
Voltage Profiles 𝑗 ,𝑡 Confirm the
 Preferred/Max/Min Household owns PV and BPV
Temperature for: Operation Cost current REH
No schedule
Fridge Water Heater Optimal Flexibility Use
HVAC Revised REH Operation Schedule
Cluster-3

Others: OPT.5: Max Comfort (3.3) Update REH Model

S.t. (2.4)-(2.66) Is 𝒎𝒂𝒙
 Pool pump operation settings Household owns PV and BPV No  Update 𝑷𝒉𝒋,𝒕 in Eq.(2.5) by
𝑃ℎ𝐹𝐿𝐸𝑋
𝑗 ,𝑡
=0 from
such as min/max up and down calculating new 𝑷𝒎𝒂𝒙
𝒉𝒋,𝒕
time. ∆𝑃ℎ𝑗 ,𝑡 ? Eq.(3.5).
 ESD current state of charge and Cluster-3  Run OPT.1, OPT.3, and OPT.5
min/max charge discharge limits. Yes to obtain the final schedule
OPT.6: Min Energy Consumption
of each individual REH and
Note: Sample of HEMS parameters (3.2) ∆𝑃ℎ𝑗 ,𝑡 the load profile of LDC.
presented in Chapter-2 S.t. (2.4)-(2.66) Confirm the current
ℎ𝑗
Household owns PV and BPV REH load schedule

Note: Optimization Model No. (OPT)

Calculate Flexibility index of each REH in Cluster-1,2,3 , 𝜶𝒉𝒋,𝒕

see Eq.(3.8)
Figure 3.4: Coordination of multiple HEMS and LDC.
typical household appliances and seek to minimize their energy cost; Cluster-2: 10%
customers own typical household appliances, ESS, and PV panel, and seek to minimize
their energy cost; and Cluster-3: 10% customers own typical household appliances, ESS,
and PV panel, and seek to maximize their comfort level. Each cluster of customers are
distributed at all buses uniformly. It is important to mention that many households do
not change their initial consumption level either because of their limited flexibility in
consumption patterns or their limited affect on the LDC system operation. The
customers from Cluster-2 and -3 do not impact the system peak load since they own
RESs which reduce their energy consumption during peak hours. Consequently, the
results of Cluster-1 customers are only discussed in this paper. Different peak reduction
requests (PRR) values have been considered to evaluate the customers flexibility, as
follows:

• Case-1: Base case, no PRR requested by LDC, β = 1

• Case-2: PRR request by LDC is 5%, i.e., β = 0.95;

• Case-3: PRR request by LDC is 10%, i.e., β = 0.9;

• Case-4: PRR request by LDC is 20%, i.e., β = 0.8;

3.3.2 HEMS Level Outcomes

At the house level, Table-3.1 presents one typical household’s appliance-wise consumption
on a summer day (house #400 located at bus-14). Figure 3.5 (A) and (B) shows a
residential customer’s load profile while minimizing energy and energy cost, respectively.
In addition, it presents the customer flexibility in Fig.3.5 (C), and this information is
communicated to the LDC.
It can be noted that the energy cost rate of the household increases from Case-I (Base
Case) to the cases with flexibility provisions, as a result of the PRR imposed by the LDC,
which limits the power that can be drawn from the grid at low price hours. Fig.3.6 shows
the ON/OFF operation of the dishwasher in house #400, which depends on the required

55
 F LEX
Table 3.1: Energy Consumption and Cost of REH #400 per Day, for Different Pj,t
CASE 1 (Base) CASE 2 (5% PRR) CASE 3 (10% PRR) CASE 4 (20% PRR)
Device (kWh) ($) (kWh) ($) (kWh) ($) (kWh) ($)
fridge 1.425 0.501 1.463 0.520 1.425 0.522 1.425 0.517
Elec. 0.225 0.078 0.225 0.078 0.225 0.078 0.225 0.078
water heater
Gas 0.111 0.013 0.111 0.013 0.111 0.013 0.111 0.013
lighting 1.245 0.453 1.245 0.453 1.245 0.453 1.245 0.453
AC 5.225 1.855 5.363 2.038 5.363 2.001 5.225 1.982
Dishwasher 0.350 0.091 0.350 0.131 0.350 0.161 0.350 0.142
Washer 0.225 0.059 0.225 0.074 0.225 0.100 0.225 0.089
Dryer 0.555 0.144 0.555 0.181 0.555 0.262 0.555 0.225
Stove 1.125 0.293 1.125 0.393 1.125 0.483 1.125 0.497
Pool 1.875 0.538 1.875 0.641 1.875 0.627 1.875 0.614
Total 12.361 4.024 12.536 4.522 12.499 4.700 12.361 4.608
Cost of Energy ($/kWh) 0.3255 0.361 0.376 0.373

A Residential customer load A Residential customer flexibility

7
2 6 Evaluated Flexibility (C)
1.8 Min Energy (A) 5
1.6 4
Power (kW)

1.4 3
Power (kW)

1.2 2
1 1
0.8
0
0.6
-1
0.4
0.2 -2
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
0
time (15 min interval)
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
time (15 min interval)
7 Aggregated loads and flexibility of residential
Min Cost (B)
6 customers
12000
5
Total Load Flexibility (D)
Power (kW)

10000
4
8000
Power (kW)

3 6000
2 4000
1 2000

0 0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 -2000
time (15 min interval) -4000
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
time (15 min interval)

Figure 3.5: Load and flexibility profiles at residential and grid levels.

56
 13 14 15 16 17 18 19 20 21 22
Time (hours) 23 24

Case 1 Case 2 Case 3 Case 4

Figure 3.6: Optimal operation of selected appliance (dishwasher) at house #400.

operation time (i.e., eight time intervals), operation interval (i.e., from 2 PM to midnight),
and the minimum time gap between two consecutive operations (minimum of two time
intervals). In Case-1 (Base Case), the operation of the dishwasher is scheduled at two
periods, each of four time intervals. It can be noted that the optimal schedule changes
with different peak caps signals (β values) imposed by the LDC.
Fig.3.7 shows the temperature deviation inside house #400, refrigerator and the
operation of the water heater for different values of β (different peak caps) imposed by
the LDC. Note that the range specified by the customer for the minimum and maximum
temperature, governs the degree of customer flexibility that would be available.
Fig.3.8 illustrates the variation in the total household consumption between Case-1 and
Case-2; and Case-1 and Case-3, respectively. It is noted that the peak hours of power usage
of the house is shifted as a result of peak signals sent by the LDC.

3.3.3 LDC Level Outcomes

At the LDC side, Fig.3.9 presents the total load profile of the distribution system. It can
be noted from the figure that TOU tariffs, which are applicable in the present work, causes
peak rebound effects at hours 19-21, which means that a peak demand is created when

57
 21.5

21
o

20.5

20

19.5

19

18.5
0 3 6 9 12 15 18 21 24

Time (hr)

8
o

0
0 3 6 9 12 15 18 21 24

Time (hr)

59
o

57

55

53

51

49

47

45
0 3 6 9 12 15 18 21 24

Time (hr)
Case 1 Case 2 Case 3 Case 4 ϴmax ϴmin

Figure 3.7: Temperature profile inside house #400, refrigerator, and water heater.

58
 9

0
0 3 6 9 12 15 18 21 24

Case 1 Time (hr) Case 2

7
Demand (kW)

0
0 3 6 9 12 15 18 21 24

Case 1 Time (hr) Case 3

Figure 3.8: Optimal load profile for House hj at different PRR requests.

59
 10.5

9.5

Demand (MW) 8.5

7.5

6.5

5.5

4.5

3.5

2.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time(hr)
CASE1 CASE2 CASE3 CASE4

Figure 3.9: Total load profile of LDC.

the TOU price reduces to the off-peak price levels at hour 19. In order to circumvent this
situation, the LDC utilizes the available flexibility, inherent in the customers’ consumption
patterns (i.e, as shown in Fig.3.5), to improve the total load shape and reduce the peak
demand at those hours.
Table 3.2: LDC Load Profile Characteristic and Power Losses
Case Minimum Load Peak Load Average Load Power Losses
(kW) (kW) (kW) (Rate of Change%)
Case 1 (0% PRR) 2722.1 10515.78 5175.45 —–
Case 2 (5% PRR) 2735.5 (0.49%) 10150.97 (-3.47%) 5178.01 (0.05%) -1.341%
Case 3 (10% PRR) 2754.7 (1.2%) 9580.18 (-8.9%) 5187.2 (0.23%) -6.124%
Case 4 (20% PRR) 2765.7 (1.6%) 9164.81 (-12.85%) 5198.2 (0.44%) -17.830%

Table 3.2 illustrates the improvements in the total load profile of the LDC and system
losses, for the different cases considered. It is important to note that the proposed
coordination scheme not only shifts the peak load, but it also reduces it, and increases
both the minimum and the average load of the distribution system, thus flattening the
load profile. It is assumed that the LDC requested for a reduction in the peak load
between 6 to 10 PM. As noted from Table 3.2, there is a significant reduction in the peak
load while the shifted controllable loads increase the total average load of the system.

60
 1
0.995
0.99
0.985
Voltage (pu)

0.98
0.975
0.97
0.965
0.96
0.955
0.95
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Bus
CASE1 CASE2 CASE3 CASE4

Figure 3.10: Average voltage at each bus at different PRR requests.

Moreover, the inherent flexibility of the customers reduces the system losses as well.
Fig.3.10 shows an improvement in the average voltage profile at each bus over the day,
when the proposed DR scheme is applied, and justifies the usefulness of the proposed
coordination scheme toward efficient grid operations.

Table 3.3: Model Statistics

House Level Grid Level
OPT.1 OPT.3 OPT.5 OPT.2 OPT.4-6 OPT.7
Model Type MILP NLP
Solver Type CPLEX SNOPT
Single Variables 2,223 3,857 2,223 2,223 3,857 19,203
Single Equations 2,403 3,638 2,400 2,442 3,553 9,527
Discrete Variables 962 1,328 952 992 1,344
Resource Usage, s 0.2 0.254 0.134 0.188 0.232 8.6
Absolute Gap 7 3 7.5 6 9
Relative Gap 0.09 0.05 0.09 0.07 0.09
*Optimization Model (OPT)

61
3.3.4 Computational Aspects

The proposed REH model is an MILP problem which is solved using CPLEX solver; the
LDC operations model is a nonlinear programming (NLP) problem, solved using SNOPT
solver, in both cases on the GAMS platform [80]. The optimization models are executed
on a Dell PowerEdge R810 server, Windows 10 Education 64-bit operating system, with 4
Intel(R) core(TM) i7 processors and 16 GB of RAM. The individual HEMS optimization
programs are executed in parallel by individual customers, and each require less than 30
seconds to solve. Different optimization models are assigned for each individual HEMS,
see Fig.3.4. Table 3.3 presents an overview of the model statistics of the HEMS and LDC
model.
The proposed coordination scheme can be implemented in real-time or day ahead,
without any difficulty. It is important to note that the maximum amount of PRR that can
be requested by the LDC, can vary based on the customers’ willingness to change their
consumption patterns. In fact, some customers do not influence the grid operation, so the
LDC does not impose such a peak cap to limit the flexibility of their energy consumption.

3.4 Summary
This chapter proposed a novel, two-stage coordination scheme between multiple HEMS
and the LDC in order to enhance the grid operational efficiency, reduce peak demand,
minimize power losses; and reduce customers’ energy cost. At the residential customer level,
each individual HEMS optimizes the household appliances schedule based on pre-defined
requirements and goals by the household. In addition, each customer’s inherent flexibility
is evaluated using the proposed approach, which is aggregated with other neighborhood
households to provide flexibility services to the LDC. Bus-wise aggregated loads, power
sold to the LDC, and total available flexibility from households, are communicated to the
LDC system operator, who evaluates its system operation and determines the bus-wise
PRR. Each customer is sent a peak cap signal by the LDC.
The proposed two-stage coordination scheme captures the procurement of flexibility
provisions by the LDC in real-time. An advantage of this two-stage coordination scheme

62
is the disaggregation and distribution of the computational burden amongst each entity,
as compared to that in a centralized optimization model which would need to optimize
thousands of REHs as well as the LDC’s operations simultaneously.

63
Chapter 4

Inclusion of Battery SoH Estimation

in Smart Distribution Planning with
1
Energy Storage Systems

Chapter 3 proposed the coordination of a large number of HEMSs to evaluate their inherent
flexibility, and hence provide a DR service to the LDC. This chapter, on the other hand,
examines the impact of multiple self-optimized HEMSs operating in an uncoordinated
manner, on the distribution grid. Thereafter, a novel NN based SoH estimator for a
Li-ion battery based BESS is proposed, which is incorporated within a framework and
mathematical model for planning and integrating BESS in the distribution grid.

1
This chapter has been accepted for publication in:
O. Alrumayh, S. Wong, and K. Bhattacharya, “Inclusion of Battery SoH Estimation in Smart Distribution
Planning with Energy Storage Systems,” IEEE Transactions on Power Systems, (available in IEEE Xplore
Early Access).

64
 4.1 Nomenclature

LDC Operations and Planning Model with ESS

Indices and Sets

hj Index of houses, hj ∈ H

j, k Index of buses in distribution system, (j, k) ∈ N

n, y Index of year, (n, y) ∈ Y

t Index of time, t ∈ T

Parameters

CPF Fixed installation cost of ESS, $

CPV , CEV Variable installation cost of ESS associated with power ($/kW) and energy
($/kWh), respectively

C OMF Fixed operation and maintenance cost of ESS, $/kW-year

C OMV Variable operation and maintenance cost of ESS, $/kWh

C REP Replacement cost of ESS, $/kW

DODj Depth of discharge limit of ESS j, p.u

DY Number of seasonal representative days

EPR, EPR Limits of energy to power ratio

HY Number of hours per year, h

d
Pj,y,t Residential load demand, p.u

P min , P max Power limits of distribution substation, p.u

P STD , E STD Standard power/energy capacity of ESS available in market

R Discount rate, %

V min , V max Voltage limits, p.u

65
αDOD Maximum allowable change in SOC

η CH /η DCH Charging and discharging efficiency of ESS, %

Υy,t Electricity price, $/kWh

Variables

Ej,y,t Actual energy capacity of ESS, p.u

J1 , J2 , J3 ESS installation, operation, and replacement cost, respectively, $.

NUP , NUE Integer multiplier of standard size of ESS

OTj,y,t , OYj,y Number of hours/years ESS in service, respectively

SS , QSS
Pj,y,t Active/reactive power drawn from substation, p.u
j,y,t
d , Qd
Pj,y,t Active/reactive power demand from residential loads, p.u
j,y,t
INST , E INST
Pj,y Installed power/energy capacity of ESS, p.u
j,y
Rate , E Rate
Pj,y Rated power/energy capacity of ESS, p.u
j,y
CH , P DCH
Pj,y,t Active power to be charged/discharged to/from ESS, p.u
j,y,t

SoCj,y,t State of charge of ESS, p.u

SoH
d j,y,t Estimated SoH of ESS, %

Vj,y,t Voltage at bus, p.u

INST , Z REP
Zj,y Binary installation and replacement decisions of ESS, respectively.
j,y
PI , Z PR
Zj,y Binary presence indicator of ESS after installation / replacement.
j,y

∆SoCj,y,t Maximum allowable change in SoCj,y,t , p.u

δj,y,t Voltage angle at bus, p.u

Degradation Model

Indices and Sets

K Index of cycle, K ∈ K

66
Parameters

δK Depth of discharge of Kth cycle

σK State of charge of Kth cycle

TK Ambient temperature during Kth cycle

βsei , αsei Parameters of the early degradation model

Variables

ft , fc Calendar and cycle aging, respectively

fsei Degradation function of ESS during early cycles

fd Degradation function of ESS

SoHj,y,t Reference SoH of ESS, %

aK Cycle indicator, [0.5, 1]

Neural Network Model

Indices and Sets

i Index of input, i ∈ I

o Index of hidden layer neuron, o ∈ O

Parameters

βo , Γ Hidden/output layer neuron bias

nin out
o,j,y,t , no,j,y,t Input and output of a hidden layer neuron

wi,o , Wo Input weight and layer weight, respectively

xij,y,t , xi,N
j,y,t Actual and normalized input i
N
Xj,y,t , Xj,y,t Actual and normalized output of NN

67
4.2 Proposed Planning Framework and Mathematical
Models

The proposed ESS planning framework considering the inclusion of a Li-ion battery SoH
estimation model is presented in Fig. 4.1. The framework comprises the following steps:

• The REH Operations Model (discussed in Section II.A) and based on [81] is executed
to develop bus-wise load profiles. These load profiles are used to simulate an LDC
Operations Model (discussed in Section II.B) to obtain a set of SOC profiles of the
ESS.

• Using the RCA (discussed in Section II.C) on these SOC profiles, the Li-ion based
ESS cycle parameters are determined, which are input to a degradation model [51]
to obtain the reference SoH (discussed in Section II.D).

• The SOC and SoH profiles, so obtained, are used to train a NN, and the function
relationship of the SoH is extracted (discussed in Section II.E).

• This functional relationship is included in the LDC planning model to determine the
optimal plan decisions. The estimated SoH profiles obtained from the planning model
are send back to the NN-based estimator to re-train and improve the SoH estimation
function for revising the plan decisions. The framework arrives at the optimal plan
when the mismatch in the estimation of SoH with the reference, is minimal.

4.2.1 REH Operations Model

The HEMS are residential controllers that carry out scheduling of the REH including all
house appliances and power interchanges with the external grid, considering the customer’s
preferences and objectives, such as minimizing its daily energy cost. In this work, the REH
operations model is taken from [81], and simulated for daily load profiles over a 10 year
horizon (i.e. 365 x 10 days).

68
 Start
Set counter c = 0

Execute REH c = c+1

Operations SoCj,y,t
Model for hj Collect ESS
minimizing SoC profiles SoCj,y,t
cost [81] [wi,o , Wi , βi , Γ]c
(Section 4.2.1) Execute LDC
Planning Model
Execute RFA Train NN
to obtain final
(Section 4.2.3) (Section 4.2.5)
Create bus-wise plan of ESS
LDC load profiles [CA,CM,CN,CB,CE]c (Section 4.2.6)
No
Evaluate SoH of SoHj,y,t
ESS using [51]
Execute LDC (Section 4.2.4)
Operations
Model with initial
Is |∆SoHj,y,t | ≤  ?
placement of ESS SoCj,y,t SoHj,y,t SoH
d
(Section 4.2.2)

Yes
End

Figure 4.1: Proposed ESS planning framework considering smart REHs and battery SoH.

4.2.2 LDC Operations Model

After completing the simulation of all REHs, bus-wise load profiles are created over a period
of ten years. The LDC operations model including ESS is then simulated to analyze the
operational decisions of the ESS.

Objective Function

minimize LDC’s daily operation cost, comprising the cost of importing power from external
grid at electricity price Υy,t to meet the residential load demand.

Y X
X T
SS
J= Υy,t Pj=1,y,t (4.1)
y=1 t=1

subject to the following constraints:

69
Load Flow Equations

Linearized power flow equations, as in [82] are used:

J
X  
SS d DCH CH
Pj,y,t − Pj,y,t + (Pj,y,t − Pj,y,t ) = Kj,k,1 (δj,y,t − δk,y,t ) + Kj,k,2 (Vj,y,t − Vk,y,t ) (4.2)
k=1,k6=j

N
X
QSS
y,t − Qdj,y,t = Kj,k,2 (δj,y,t − δk,y,t ) + Kj,k,1 (Vj,y,t − Vk,y,t ) (4.3)
k=1,k6=j

where,
x2j,k rj,k xj,k
rj,k +x2j,k
2 2 +x2
rj,k j,k
Kj,k,1 = , Kj,k,2 = , ∀(j, k) ∈ J (4.4)
xj,k xj,k

These are subject to bus voltage constraints and limits on power drawn from the
substation, as given below:
Vjmin ≤ Vj,y,t ≤ Vjmax (4.5)

P min ≤ Pj=1,y,t
SS
≤ P max (4.6)

The following equations present the operational constraints of the ESS:

DCH
Pj,y,t
SoCj,y,t+1 = SoCj,y,t + η CH Pj,y,t
CH
− (4.7)
η DCH

− ∆SoCj,y,t ≤ SoCj,y,t − SoCj,y,t−1 ≤ ∆SoCj,y,t (4.8)

where,
∆SoCj,y,t = αDOD Ej,y,t (4.9)

The inter-temporal change in the SOC level is defined by (4.7), and the same is limited
by minimum and maximum allowable changes, given by (4.8). Equation (4.9) states that
the value of the maximum allowable change in SOC is a fraction, αDOD , of Ej,y,t .

(1 − DoD) Ej,y,t ≤ SoCj,y,t ≤ Ej,y,t (4.10)

70
 SoCj,y,t=1 = SoCj,y,t=24 = 0.5 Ej,y,t (4.11)
CH Rate
Pj,y,t ≤ Pj,y (4.12)
DCH Rate
Pj,y,t ≤ Pj,y (4.13)

In (4.10), the physical capacity of the ESS limits its SOC considering the maximum
allowable DoD. It is assumed that the SOC of the ESS at the beginning and end of
the daily operation are equal, equation (4.11), to prevent the optimization model from
choosing the maximum SOC at the beginning of the day and fully discharge at the end of
the day. Moreover, the rated power of the ESS constraints the power charging/discharging
capability from/to the ESS in (4.12) and (4.13).

4.2.3 RainFlow Counting Algorithm

The RCA is a well known approach in the analysis of fatigue data [70]. It has been used
in some recent works [83] and [84] for analysis of ESS operation, wherein the RCA counts
the number of irregular cycles within a given operation period. The algorithm requires the
SOC profile of the ESS as an input, as shown in Fig. 4.1, in order to obtain the following:

• Cycle amplitude (CA).

• Cycle mean value (CM).

• Cycle number (CN ).

• Cycle begin and end times (CB, CE).

The results obtained from applying the RCA are used to calculate the variables of the
degradation model, discussed next.

4.2.4 Degradation Model of ESS

In [51], a mathematical model was proposed to evaluate lithium-ion battery cell life
considering calendar and cycle aging. Calendar aging represents the battery’s inherent

71
degradation over time, as a function of the average temperature (T ) and the average SOC
of the battery (σ). Cycle aging represents the loss of life due to the charging and
discharging of the battery, represented as a function of the DoD, SOC, and temperature.
This is given as follows [51]:

K
X
fd (t, δ, σ, T ) = ft (t, σ, T ) + ak fc (δK , σK , TK ) (4.14)
K

The first term in (4.14) represents the effect of calendar aging (ft ) while the second term
represents the degradation as a result of cycle aging (fc ). In addition, ESS degradation
during its operation in the early cycles is given by [51],

fsei = βsei fd (4.15)

The SoHj,y,t of ESS can be represented as follows:

SoHj,y,t = 1 − αsei · e−fsei − (1 − αsei ) · e−fd (4.16)

where αsei denotes the portion of the ESS capacity lost during the early operation.
The third term in (4.16) represents the degradation due to calendar and cycle aging. In
order to apply the degradation model (4.14)-(4.16) to irregular cycle operation of ESS, the
outcomes of the RCA are used as inputs to the degradation model, as follows [51]:

• δK = 2 CAK .

• σK = CMK .

• σ = Average value of CM.

• TK = Mean temperature between start and end time of Kth cycle.

• T = Mean value of the temperature profile.

The following subsection presents a novel SoH estimation approach (i.e. SoH)
d for an
ESS in a distribution system that does not require RCA.

72
4.2.5 Proposed NN Based ESS SoH Estimator

This paper adopts the degradation model in [51] to propose a novel approach for estimating
the SoH of the ESS. For this purpose, an NN-based mathematical model of the SoH is
developed to be embedded within the planning model in order to take into account the
impact of calendar and cycle aging in determining the optimal ESS size, location, and
operation decisions.
In order to estimate the SoH of ESS (SoH),
d a supervised learning technique is applied
to train the NN with outcomes of the LDC operation of model, i.e., SOC profiles of 365
days for 10 years operation, as shown Fig. 4.1. The required output vector of the NN is
the target SoH (SoHj,y,t ), which is determined as follows: (a) Use RCA to find CA, CB,
CE, CN, and CM. (b) Calculate the input of the degradation model δK , σK , σ, TK , and
T , as discussed in Section 4.2.4 and shown in Fig. 4.1. (c) Evaluate the ESS SoH using
the degradation model adopted in Section 4.2.4, the output of which represents the target
vector of the NN. As a result, a [87, 600 × 2] order matrix training set and a [87, 600 × 1]
target vector are used for the NN training. Accordingly, the SoH d of an ESS located at bus
(j) at year (y) and time (t) can be expressed as a function of SOC and operation time, as
given below:
SoH
d j,y,t = f (SoCj,y,t , OTj,y,t ) (4.17)

The NN has one hidden layer with two hidden layer neurons, which is obtained by trial-
and-error; the NN is trained using Marquardt learning technique in MATLAB [85]. The
resulting structure of the NN-based SoH Estimator is shown in Fig. 4.2. The mathematical
representation of the NN function is developed as follows:

• Pre-processing input: The inputs of the NN xij,y,t = {SoCj,y,t , OTj,y,t } are normalized
to lie in the interval [−1, 1] using mapminmax function in MATLAB, as shown below:

2(xij,y,t − xij,y,t )
xi,N
j,y,t = + xi,N
j,y,t , (4.18)
xij,y,t − xij,y,t

The underline and overline notations denote the respective minimum and maximum
values of the inputs.

73
 β1

x1j,y,t w11
x1,N
j,y,t nout
1,j,y,t W1
w12
N
Xj,y,t Xj,y,t
x2j,y,t
x2,N w21 nout
j,y,t w22 2,j,y,t W2
Γ

β2

Figure 4.2: Structure of the proposed NN-based SoH estimator.

• Hidden layer activation function: The pre-processed inputs with appropriate weights,
wi,o [∀i ∈ 1, ..., H, o ∈ 1, ..., O], are summed up at every hidden layer neuron. In
addition, each hidden layer neuron has its own bias (βo ). The summed signals present
the input to a hidden layer neuron (nin o ), as given below:

I
X
nin
o,j,y,t = wi,o xi,N
j,y,t + βo (4.19)
i=1

This input value is passed through an activation function (i.e. tansig), which
transforms nin out
o,j,y,t into an output signal (no,j,y,t ), as follows:

2
nout
o,j,y,t = −1 (4.20)
1 + exp (−2 nin
o,j,y,t )

• Output layer function: The calculated nout o,j,y,t represents the input of the output
layer, which is also multiplied by a layer weight (Wo ). Ultimately, a linear function
N
is applied in the output layer, which results in the output of the NN, Xj,y,t , as shown
below:

74
 O
X
N
Xj,y,t = nout
o,j,y,t Wo + Γ (4.21)
o
N
The obtained output from the NN (Xj,y,t ) represents the normalized SoH of the ESS
(SoHN
j,y,t ).

• Post-process output: The obtained output (2.71) is post-processed to obtain the

estimated output (i.e. SoH), as follows (refer Fig. 4.2):

N − X N )(X
(Xj,y,t j,y,t j,y,t − Xj,y,t )
Xj,y,t = + Xj,y,t (4.22)
2

The mathematical model (4.22) of the SoH of the ESS at bus j at year y and time t is
incorporated in planning model, discussed in the following subsection.

4.2.6 LDC Planning Model

The LDC planning model seeks to optimally allocate, size, and replace a number of ESS
at different buses in the distribution grid, considering the following objective function.

Objective Function

J = J1 + J2 + J3 (4.23)

Equation (4.23) comprises the ESS installation cost (J1 ), operation cost (J2 ), and ESS
replacement cost (J3 ), as follows:
ESS Installation Cost: comprises power capacity cost ($/kW), energy capacity cost
($/kWh), and a fixed installation cost ($), as given below:

N
Y X
" #
X 1
J1 = (CPV Pj,y
INST
+ CEV Ej,y
INST
+ C F Zj,y
INST
) (4.24)
y=1 j
(1 + R)y

75
 Operation Cost: comprises fixed and variable operation and maintenance cost (O&M),
the cost associated with ESS charging and discharging, and the cost of buying power from
the external grid, as given below:

N
Y X
" T
X C OMF Rate 365 X 1
J2 = y
P j,y +
y=1 j
(1 + R) DY t (1 + R)y
!#
C OMV (Pj,y,t
CH DCH
+ Pj,y,t SS
) + Υy,t Pj=1,y,t ) (4.25)

Replacement Cost: included when the ESS reaches its end of life and has to be replaced,
as given below:

Y XN
X C REP
J3 = E INST Zj,y
y j,y
REP
(4.26)
y=1 j
(1 + R)

The objective function is subject to the following constraints:

Selection Decisions of ESS

The selection of the ESS size should be based on standard unit of power and energy capacity
ratings available in the market, modeled as follows:

INST
Ej,y = NUE E STD (4.27)
INST
Pj,y = NUP P STD (4.28)

NUE and NUP are integer variables that determine the rated energy and power capacities
respectively, based on the available standard sizes in the market, E STD (e.g. 50 kWh)
and P STD (e.g. 50 kW). Furthermore, the energy capacity of the ESS for a certain power
rating, is determined based on its energy to power ratio, as follows:

76
 INST INST INST
EPR Pj,y ≤ Ej,y ≤ Pj,y EPR (4.29)

INST
The installation year decision (Zj,n ) is activated only once during the planning
horizon, as indicated below:

y
X
INST
Zj,n ≤1 (4.30)
n=1

The variable ZPI

j,y indicates the presence of ESS after an installation decision. The
INST
following constraint coordinates the installation year decision (Zj,n ) in the presence of
ESS: y
X
PI INST
Zj,y = Zj,n (4.31)
n=1

The binary decision variable ZPR REP

j,y indicates the presence of ESS after replacement (Zj,y,t )
and is modeled as follows: y
X
PR REP
Zj,y = Zj,n (4.32)
n=1

The two binary indicators ZPI PR

j,y and Zj,y are used in counting the number of operation years
after installation of a new ESS.
The installed energy/power capacity determines the rated capacity of ESS after the
year of installation, as follows:
y y
X X
Rate INST Rate INST
Ej,y = Ej,n ; Pj,y = Pj,n (4.33)
n=1 n=1

It is important to account for ESS energy capacity degradation in order to achieve

an optimal plan involving sizing, year, and location of the ESS. Therefore, the de-rated
ESS energy capacity can be modeled considering its estimated SoH and rated capacity, as
follows:
d j,y,t E Rate
Ej,y,t = SoH (4.34)
j,y

77
The following constraint models the estimated SoH of the ESS at installation or
replacement, as follows:
d j,y,t=1 ≥ Z INST + Z REP
SoH (4.35)
j,y j,y

It is assumed that ESS installation or replacement is done at the beginning of the year, at
which time the battery SoH is 100%; else, SoH
d is calculated as per Section 4.2.5.

Inclusion of SoH Estimator

The first step to estimate the SoH of the ESS is the pre-processing of the required inputs (i.e.
SoCj,y,t and OTj,y,t ). The inputs to the extracted NN mathematical model are normalized
and scaled to the range of [−1, 1], as below:

PI
2(SoCj,y,t − SoCj Zj,y )
SoCN
j,y,t = −1 (4.36)
SoC − SoCj

In (4.36), SoCj is assumed to be 20% of the ESS installed capacity. SoCN

j,y,t = 1 when
N PI
SoCj,y,t = SoC, and SoCj,y,t = −1 when Zj,y = 0. Similarly, the following equation
normalizes the operation time of the ESS.

2 OTj,y,t
OTN
j,y,t = −1 (4.37)
T
where,
PI PI
OTj,y,t = HY(OYj,y − Zj,y ) + (t − 1)Zj,y (4.38)

In (4.38), the first term on the RHS denotes the total number of ESS operational hours
from installation/replacement until the previous year, while the second term denotes the
number of hours of operation in the current year. Furthermore, note that OYj,y is a
function of installation and replacement decisions, which represents the total number of
years of operation after ESS is installed/replaced, as follows:
y y
X X
PI REP
OYj,y ≤ Zj,n +M Zj,n (4.39)
n=1 n=1

78
 y y
X X
PI REP
OYj,y ≥ Zj,n −M Zj,n (4.40)
n=1 n=1
y y
X X
PR REP
OYj,y ≤ Zj,n + M (1 − Zj,n ) (4.41)
n=1 n=1
y y
X X
PR REP
OYj,y ≥ Zj,n − M (1 − Zj,n ) (4.42)
n=1 n=1

Equations (4.39) and (4.40) coordinate the counting of OYj,y,t for an ESS in service after
installation, while (4.41) and (4.42) counts the number of years an ESS is in service after
a replacement.
N
The normalized inputs (i.e. SoCj,y,t and OTN
j,y,t ) are passed to the hidden layer of the
NN, as follows:

2
nout
o,j,y,t = N +w OTN
−1 (4.43)
1 + e−2 (w1,o SoCj,y,t 2,o j,y,t +βo )

dN
Finally, SoHj,y,t , which is the normalized output of the NN, can be expressed as follows:

dN =
X
SoHj,y,t nout
o,j,y,t Wo + Γ (4.44)
o∈O

Budget Constraint

The net present value (NPV) of the installation cost should not exceed the NPV of the
allocated budget:
J1 ≤ Budget (4.45)

The constraints of the LDC operations model, discussed in Section 4.2.2 (i.e. (4.2) to
(4.13)), are further included in the planning model in addition to the above constraints.

79
 Off-Peak Mid-Peak On-Peak
6
TOU price TOU price TOU price

5
Power (kW)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (hr)

Figure 4.3: Residential load profile.

Table 4.1: Tuned Weights and Biases of the NN
Parameter Hidden Layer Output Layer
w11 −0.164E − 4 w12 0.2218 W1 −0.4328
Weight
w21 0.447E − 6 w22 −0.0661 W2 −16.5719
Bias β1 −0.5117E − 2 β1 0.3317E − 3 Γ 0.3260E − 2

4.3 Results and Discussions

In this work, a 33-bus test system is used to determine the optimal ESS size, location and
year of installation / replacement considering the impact of its calendar and cycle aging on
its SoH. The residential load demand is assumed to increase 3% annually. The planning
period is 10 years; each year is represented by one day, and each day’s operational aspects,
such as load and generation profiles and charge / discharge decisions, are modeled hourly.
The representative day is the one which has the highest peak demand of that year. This is
a conservative approach, using the worst-case scenario and planning accordingly, so that

80
 Table 4.2: ESS Optimal Plan (Considering Scenario-1 Load Profile).
Operation Cost Installation Cost
Case Study
Optimal Plan Decisions* (NPV-OC) (NPV-OC)
(2, 9, 400, 1500) (4, 5, 450, 1800)
Case-1 (8, 4, 150, 600) (17, 7, 400, 1600)
$64,000 $5,988,000
(Proposed Approach) (19, 5, 500, 2000) (23, 10, 500, 2000)
(30, 6, 250, 1000)
Total Rated Power/Energy: 2.65 MW/10.5 MWh Total: $6,052,000
(2, 7, 300, 1200) (4, 5, 500, 2000)
Case-2 (8, 5, 500, 2000) (15, 9, 250, 1000)
$68,000 $5,815,000
(Without Degradation) (17, 4, 200, 800) (19, 10, 300, 1150)
(23, 5, 150, 600) (30, 6, 250, 1000)
Total Rated Power/Energy: 2.5 MW/9.75 MWh Total: $5,881,000
(2, 5, 500, 2000) (4, 10, 400, 1600)
Case-3 (8, 7, 250, 1000) (15, 9, 400, 1600)
$77,000 $6,844,000
(Fixed Degradation Rate) (17, 6, 250, 1000) (19, 8, 250, 1000)
(23, 4, 400, 1600) (30, 5, 500, 2000)
Total Rated Power/Energy: 2.95 MW/11.8 MWh Total: $6,921,000
*(Bus number, Year of Installation, Rated Power (kW), Rated Energy (kWh))

the system has sufficient redundancy and margin. The choice of one representative day
per year reduces the computational burden of the planning model.
To implement the planning framework proposed in Fig. 4.1, 2, 600 individual REH
operations models [81] are executed to obtain their daily optimal load profiles, and hence
the LDC’s load profiles at each bus are created. Figure 4.3 shows the distribution of the
REH loads at each hour, indicating the minimum and maximum, 25th and 75th
percentiles, and the median consumption of the customers. Note that during the on-peak
TOU prices, the variation in customer load is small, which depicts harmonized operation
of REHs. The HEMS helps the household in shifting its demand from on-peak TOU price
to off-peak price hours.
The LDC operations model is executed for (365 x 10 days) considering the respective
REH load profiles to obtain the corresponding optimal SOC profiles of the ESS. This
data set is divided into a training set (60%), validation set (20%), and testing set (20%).
The Dividerand function of the MATLAB NN-Toolbox [85] is applied to divide the data
set using random indices. The training stage takes place to build the NN-based SoH
estimator. Fig. 4.4 presents the error distribution between SoHj,y,t and SoH d with the
number of incidences. The main outcomes of the NN, as stated in Fig. 4.1, are the input
weights, layer weights, and biases at hidden and output layer neurons. Table 4.1 shows

81
Table 4.3: ESS Optimal Plan Using the Proposed Approach (case-1) for Different Load
Profile Scenarios
Optimal Plan Decisions*
(2, 9, 400, 1500) (4, 5, 450, 1800)
Scenario-1 (8, 4, 150, 600) (17, 7, 400, 1600)
IEEE RTS Load (19, 5, 500, 2000) (23, 10, 500, 2000)
(30, 6, 250, 1000)
Total: 2.65 MW/10.5 MWh
Scenario-2 (2, 5, 600, 2000) (4, 3, 950, 1550)
50% IEEE RTS load (8, 10, 350, 1400) (19, 7, 150, 600)
50% REH penetration (23, 9, 250, 1000)
Total: 2.3 MW/6.55 MWh
Scenario-3 (4, 5, 500, 2000) (8, 4, 150, 600)
IEEE RTS Load (15, 5, 450, 1800) (17, 8, 500, 2000)
20% PV penetration (-ve load) (19, 10, 350, 1400) (23, 6, 350, 1400)
Total: 2.3 MW/9.2 MWh
Scenario-4 (2, 4, 550, 1900) (17, 10, 350, 1400)
50% IEEE RTS load (19, 5, 700, 2000) (23, 3, 250, 300)
50% REH penetration (30, 7, 300, 1200)
20% PV penetration (-ve load)
Total: 2.15 MW/6.8 MWh
*(Bus number, Year of Installation, Rated Power (kW), Rated Energy (kWh))

the resulting parameters of the NN, which are used to construct the SoH functional
relation given in (4.44).
Next, the proposed LDC planning model is executed to obtain the optimal ESS plan
decisions over the 10 year horizon including ESS energy and power capacity, location, and
installation years, considering the following case studies:

• Case-1: Estimated SoH-integrated LDC planning model.

• Case-2: Without degradation of ESS.

• Case-3: Fixed annual degradation rate of 5%.

Furthermore, each case has been studied considering four load profile scenarios:

82
Table 4.4: NPV of Operation and Installation Costs for Various Cases and Scenarios
($1000)
Scenario-1 Scenario-2 Scenario-3 Scenario-4
Case
NPV-OC NPV-IC NPV-OC NPV-IC NPV-OC NPV-IC NPV-OC NPV-IC
1 64 5,988 65 4,800 62 5,364 69 4,630
2 68 5,815 66 4,106 53 5,264 68 4,038
3 77 6,844 76 4,980 68 6,142 69 4,772

104
4
Training
3.5 Validation
Test

2.5
Intsances

1.5

0.5

0
0.13
0.58
1.03
1.48
1.93
2.38
2.83
3.28
3.73
4.18
-4.37
-3.92
-3.47
-3.02
-2.57
-2.12
-1.67
-1.22
-0.77
-0.32

10-4

Figure 4.4: Training, validation, testing histogram of NN-based SoH estimator.

• Scenario-1: System demand profile is modeled using the IEEE RTS load.

• Scenario-2: System demand profile is a mix of 50% penetration of REHs and rest
modeled using the IEEE RTS load.

• Scenario-3: System demand profile is a mix of 20% penetration of PVs and rest
modeled using the IEEE RTS load.

• Scenario-4: System demand profile is a mix of 50% penetration of REHs, 20%

penetration of PVs, and rest modeled using the IEEE RTS load.

83
 Two load profiles are used: IEEE RTS load and REHs. Assuming that the connected
load of an REH is 3 kW, the number of REHs connected at a bus is determined by dividing
the IEEE RTS peak load at a bus by 3 kW. Each REH model is then executed to build a
bus-wise load profile [81]. The total demands of Scenario-1 and Scenario-2 are shown in
Fig. 4.5. In Scenario-3 and Scenario-4, it is assumed that PV facilities are commissioned
at buses 3, 17, and 30 in years 3, 6, and 8, respectively, with their power capacity being
20% of the load at that bus. Note that, in the system load profile, the PVs appear as a
negative load.
The case studies, Table 4.2, examine the impact the proposed approach vs less accurate
SoH models in LDC planning framework while the (load profile) scenarios, Table 4.3,
examine the impact of load mix and PV penetration on ESS sizing, siting, and year of
installation / replacement. These tables also show the optimal ESS planning combinations
for each scenario and case; and Table 4.4 shows the associated operation and installation
cost NPVs.
It is seen from Table 4.2 that in Case-1, the LDC needs to invest in a total ESS capacity
of 2.65 MW/10.5 MWh across seven ESS units. The NPV of the ESS installation cost is
M$5.99. In Case-2, eight ESS units are invested in by the LDC although the total ESS
capacity is 2.5 MW, 9.75 MWh for a total cost that is 2.8% lower than Case-1. This is
because the SoH of the ESS units are considered to remain unchanged in Case-2. On the
other hand, Case-3 requires 2.95 MW, 11.8 MWh of ESS capacity when using a fixed-rate
degradation.
Table 4.3 shows the LDC optimal plan for the four scenarios. In the first scenario, the
total capacity of ESS units is higher in than Scenario-2 as a consequence of the former’s
longer duration of peak demand (Fig. 4.5). In Scenario-3, the integration of PV results
in an approx. 10% reduction in installed ESS capacity compared to Scenario-1. However,
the integration of PV units does not significantly influence the decisions of ESS in the
presence of REHs, as seen comparing Scenario-2 with Scenario-4, because the maximum
PV power output occurs during the mid-day and does not coincide with the peak demand
of the uncoordinated REHs which appear at the onset of off-peak TOU prices.
Scenario-1 has total LDC NPV costs around M$6.05, M$5.88, and M$6.92 for Case-1,

84
 5
Aggregated Demand
Demand / Power drawn from Substation (MW) 4.5 Imported Power

3.5

2.5

1.5

0.5

0
1 2 3 4 5 6 7 8 9 10
Representitive Day of Year
(a) Scenario-1

6
Aggregated Demand
Demand / Power drawn from Substation (MW)

Imported Power
5

0
1 2 3 4 5 6 7 8 9 10
Representitive Day of Year
(b) Scenario-2

Figure 4.5: Total distribution system demand and power drawn from substation for Case-1
(Proposed Approach) over the planning period.

85
Case-2, and Case-3, respectively (Table 4.4). On average, LDC cost in Case-2 is around
8% lower than in Case-1 since, for the former, ESS SoH is fixed at 100%, which results in
over-estimation of the available energy capacity of the Li-ion batteries. On the other hand,
in Case-3 the average LDC cost is around 10% higher than Case-1 because of the under-
estimation of ESS SoH when using the fixed-rate degradation, leading to higher operation
and installation costs.
Figure 4.6 shows the SoH and SoHd profiles of the ESS located at bus-8 over a 24-
hour operation considering the proposed degradation model of Case-1. It is noted that
the proposed NN-based mathematical model captures the hourly SoH profile (SoH)
d due to
calendar and cycle aging with a relatively small error, as compared to SoH.
Figure 4.7 shows the reference SoH, SoHj,y,t , the estimated SoH from Case-1, SoH,
d and
the SoH profile considering a fixed deterioration of 5%, as in Case-3, over the plan period.
SoHj,y,t is obtained using the off-line degradation model proposed in [51] and discussed in
Section 4.2.4. In contrast, the SoH
d profile is calculated using the proposed NN-based SoH
estimator model, which is shown precise compared to SoHj,y,t .
SoHj,y,t obtained from the different load mix scenarios are shown in Fig. 4.8; Scenario-1
has consistently the lowest SoH. It is noted that the load mix has some influence on the
battery degradation: the penetration of PV units and the presence of REHs improve the
overall SoH of Li-ion batteries by around 1% every year, as a consequence of the adoption
of RES or REH, which results in a reduction in the peak duration of the system load and
requires steep discharge from ESS.
Figure 4.5 shows the LDC’s aggregated system demand and the power it imports over
the substation in Scenarios-1 and -2 for the representative day in each planning year. The
ESS units supply the net difference between demand and imports. Figure 4.9 shows the
total demand and power imported from external grid over a day in the last year of the
planning period for Scenario-1 and -2. The SOC of ESS located at bus 23 is also included,
which shows high / low SOC during low / high RTS and REH demand, respectively.
Figure 4.10 shows the bus voltage profiles during peak demand hours: hour-18 in
Scenario-1 and hour-20 in Scenario-2 of the terminal year. Note that the voltages are
always within pre-specified limits (i.e. 0.9 to 1.0 p.u.). Scenario-2 shows a higher deviation

86
 100

99.99

99.98

99.97
SOH(%)

99.96

99.95

99.94

99.93
2 4 6 8 10 12 14 16 18 20 22 24
Time (hr)

Figure 4.6: Comparison of SoH profiles of ESS located at Bus-8 over a day.

100

95

90

85
SoH (%)

80

75

70

65
0 1 2 3 4 5 6 7 8 9 10
Year

Figure 4.7: Comparison of SoH profiles of ESS located at Bus-8 over the plan period.

87
 100
Scenario-1
Scenario-2
95 Scenario-3
Scenario-4

90

85
SoH (%)

80

75

70

65
1 2 3 4 5 6 7 8 9 10
Year

Figure 4.8: Estimated SoH profiles of optimal ESS at Bus-23 for various scenarios.

in voltage from the nominal value of 1.0 p.u. as compared to Scenario-1 as a consequence
of the higher demand from mixed loads in the former. The voltage profile in Case-1 is also
lower in Scenarios-1 and -2 than the voltage profiles in Case-2 and -3.

4.4 Summary

A novel NN-based SoH estimator was developed using a large cluster of smart loads,
simulated to represent the total load of the distribution grid; and a large data set of
ESS operations, simulated to mimic the LDC’s behavior in controlling the ESSs. The
hence extracted functional relationship of SoH was integrated within a planning model,
which included a large penetration of smart REHs. The NN model was re-trained using
the updated outcomes of the planning model to improve the plan decisions. The results
showed a relatively small error between the estimated and the reference SoH of ESS. The
case studies demonstrated the impact of neglecting calendar and cycle aging and under- /
over-estimation of the ESS capacity on the plan decisions of the LDC.

88
 6 120

Demand and Power Drawn from Substation (MW)

System Demand Imported Power State of Charge
110
5 100

90
4
80

SoC (%)
70
3
60

50
2
40

1 30

20

0 10
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (hr)
(a) Scenario-1

6 120
Demand and Power Drawn from Substation (MW)

System Demand Imported Power State of Charge

110
5 100

90
4
80
SoC (%)

70
3
60

50
2
40

1 30

20

0 10
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (hr)
(b) Scenario-2

Figure 4.9: Total demand, power imported from external grid, and ESS operation over
one-day of year-10.

89
 1
Case-1
Case-2
0.99 Case-3

0.98
Voltage Level (p.u.)

0.97

0.96

0.95

0.94

0.93
5 10 15 20 25 30
Bus Number
(a) Scenario-1

1
Case-1
Case-2
0.99 Case-3

0.98
Voltage Level (p.u.)

0.97

0.96

0.95

0.94

0.93
5 10 15 20 25 30
Bus Number
(b) Scenario-2

Figure 4.10: System voltage profiles for different case studies, for Scenario-1 and -2, during
peak demand in the terminal year.

90
Chapter 5

Cooperative Operation of Battery

Energy Storage Systems
Participating in Flexibility Services
1
Provisions

In the previous chapter, LDC-owned BESSs are optimally sized and sited in the distribution
grid to help meet distribution grid constraints and minimize total system costs. On the
other hand, this chapter investigates the investor-owned BESSs capability in providing
the LDC with flexibility services. It proposes a cooperative game theory-based approach
to optimally distribute the total system saving between several BESSs participating in
flexibility service provision in the distribution grid. This work assumes that the participants
in flexibility service provision are all Li-ion battery-based ESSs and, furthermore, that the
specific flexibility service attained from these batteries are based on the C-Rate under
1
An earlier version of this work has been published in:
O. Alrumayh, S. Wong, H. Alharbi and K. Bhattacharya, “Incentives for Demand Response and Flexibility
Services Procured from Energy Storage Systems,” Proc. IEEE Power & Energy Society General Meeting
2020.
1
The present work to be submitted for publication in:
O. Alrumayh, S. Wong and K. Bhattacharya, “Cooperative Operation of Battery Energy Storage Systems
Participating in Flexibility Services Provisions,” in IEEE Transactions on Power System.

91
 which it must operate to offer such service. The proposed approach ensures the maximum
benefit of LDC from the cooperative behavior and the flexibility of BESSs; furthermore,
it ensures fairness in distributing savings between the ESSs by taking into account the
marginal contribution of the participant to the coalition.

5.1 Nomenclature
Indices and Sets

j, k Index of buses in distribution system, (j, k) ∈ J .

j, h Index of a participant location in distribution grid, (j, h) ∈ J .

s Index of services provided by BESS, s ∈ S.

t Index of time, t ∈ T .

Parameters

as , bs , cs Linearized BESS degradation cost function.

CB Total cost of BESS, $.

CRs Maximum change rate in SOC.

C-Rates Change rate in SOC.

Ej , Ej Minimum and maximum of ESS, p.u.

n Total number of BESS investors.

q Size of a BESS coalition.

SoC, SoC Minimum and maximum of SOC, %.

σtHEP , σ Peak Hourly electricity price and peak price, $/kWh.

σtTOU Time-of-use price, $/kWh.

d
Pj,t Hourly distribution grid demand, p.u.

92
η CH , η DCH Charging and discharging efficiency of ESS, %.
γs , ωs Coefficients of fitting degradation function.
τ Time interval, 15 min.

LDC Operation Model:

Variables
SS
Pj,t Active power drawn from substation, p.u.
P Peak Peak power drawn from substation, p.u.
QSS
j,t Reactive power drawn from substation, p.u.
Vj,t Voltage level, p.u.
δj,t Voltage angle, p.u.

BESS Operation Model:

Variables
DoD Depth of discharge of BESS, p.u.
DGCt,s Degradation cost, $.
Ej,t Energy stored in BESS, p.u.
Ls Number of cycles of BESS, p.u.
CH , P DCH
Pj,t Active Power to be charged/discharged to/from ESS at time of use price, p.u.
j,t

SoCj,t SOC of BESS, p.u.

SoCL
j,t Linearization of the product term of SOC and binary variable, p.u.
∆SoCj,t,s Change in SOC of BESS, p.u.
+
∆SoCj,t,s Positive change in SOC of BESS, p.u.
−
∆SoCj,t,s Negative change in SOC of BESS, p.u.
−/+
Zj,t,s , Zj,t,s Binary variable.

93
5.2 Proposed Flexibility Services Framework

5.2.1 Flexibility Services: Definitions

The lifetime of Li-ion batteries is impacted by many factors such as DoD, C-Rate and
ambient temperature as well as operational decisions. Various operational states of a
Li-ion battery are modeled in this work based on its ability to provide different ranges of
flexibility services which are defined in terms of the SOC limits of the BESS. These are
−/+ −/+
low flexibility (LF) denoted by s1 , moderate flexibility (MF) denoted by s2 and high
−/+
flexibility (HF) denoted by s3 ; their respective ranges of operation are given in Table
5.1. The change in SOC of a service s is dependent on one of the main characteristics of
BESSs, which is C-Rate. As shown in Table 5.1, LF, MF, and HF have C-Rate values of
1, 2 and 4 respectively. The maximum rate of change (CRs ) of a service, limits the change
in SOC, which can be calculated as follows:

C-Rates
CRs = (SoC − SoC), ∀s ∈ S (5.1)
τ

In (5.1), SoC and SoC denote the minimum (i.e. 20%) and maximum SOC (i.e. 100%)
of BESS, τ is the number of time intervals per hour; τ = 4 is selected, denoting time
intervals of 15 minutes. Therefore, it can be easily seen from (5.1) that CRLF = 20%,
providing low flexibility, while CRHF = 80%, which provides high flexibility.

Table 5.1: Flexibility Service Definition

Service Notation |∆SoC| (%) C-Rate
−/+
Low Flexible (LF) s1 0 - 20 1
−/+
Moderate Flexible (MF) s2 20 - 40 2
−/+
High Flexible (HF) s3 40 - 80 4

Referring to Figure 5.1, say at time t − 2, the BESS SoCt−2 is at minimum level of 20%
(point-A). The battery can be charged over the ranges of s+ + +
1 , s2 , or s3 which corresponds
to changes in SOC from 0% - 20%, 20% - 40%, or 40% - 80%, respectively. The selected

94
 CRMF < |∆SoCt−1 | ≤ CRHF |∆SoCt | ≤ CRLF CRLF < |∆SoCt+1 | ≤ CRMF

100%
D
s+
2
90%
s+
80% B 1

C s+
1
70%
s+
3 s−
1
60%
SoC

s−
1
50%
s+
2 s−
2
40%
s−
2
30%
s+
1 s−
3 s−
3
20% A

t-2 t-1 t t+1

Time Interval (15 min)

Figure 5.1: The proposed flexibility services classifications based on change in SoC.

service at time t − 1 is shown to be s+

3 for ∆SoCt−1 = 60%, which takes the SOC to 80%
(point-B) and meets the following constraint:

CRMF < |∆SoCt−1 | ≤ CRHF , ∀t ∈ T (5.2)

Equation (5.2) shows that the absolute change in SOC is limited by CRMF and CRHF ,
which are 40% and 80%, respectively. From point B, the BESS can opt for services s+1,
− − − −
s1 , s2 and s3 based on their corresponding ranges. As an example, s1 is selected for
∆SoCt = 10% and the SOC profile at time t reaches point-C, where the change in SOC
belongs to LF service as shown below:

95
 0 ≤ |∆SoCt | ≤ CRLF , ∀t ∈ T (5.3)

Equation (5.3) defines the limits of LF service, where CRLF equals 20%. Finally, from
− − −
point-C, the services s+ +
2 , s1 , s1 , s2 and s3 are available based on their corresponding ranges
of SOC. The service s+ 2 is selected for ∆SoCt+1 = 30% and the SOC profile reaching point
D at time t + 1. The mathematical representation of the range of MF service is shown
below:

CRLF < |∆SoCt+1 | ≤ CRMF , ∀t ∈ T (5.4)

It should be noted that the selection of a service will impact the number of cycles of the
BESS. Also note that out of the six services defined in this proposed framework, namely,
− − −
s+ + +
1 , s1 , s2 , s2 , s3 and s3 , each time interval will have a subset of available services which
will depend on the SOC of the preceding interval.

5.2.2 BESS Degradation Cost Function

Figure 5.2 demonstrates the relationship between the number of cycles of BESS and the
DoD at different C-Rates. It shows an exponential drop in number of cycles with the
increase of DoD. For C-Rate = 1, the experimental data is represented by its curve fit
equation, given as follows [86]:

Ls = γs DoDωs , ∀s ∈ S (5.5)

Ls represents number of cycles of a BESS at DoD for a particular service s, γs and ωs

are the fitting function coefficients, where each of the flexibility services- LF, MF, and HF
have their own set of parameters, which will be presented in a latter section.
After obtaining γs and ωs from (5.5), the degradation cost model of the battery at time
t and service s can be represented as follows [87]:

96
 8000
C-rate= 1
C-rate= 2
7000
C-rate= 4

6000
Number of Cycles

5000

4000

3000

2000

1000

0 10 20 30 40 50 60 70 80 90
DoD (%)

Figure 5.2: Number of cycles of BESS versus DoD at different C-Rates.

CB
DGCt,s = | ((1 − SoCt )ωs − (1 − SoCt−1 )ωs ) |, ∀ t ∈ T ,s ∈ S (5.6)
γs

In order to take into account the impact of the change in SOC over a time interval,
different values SoCt and SoCt−1 are used in (5.6) to calculate DGCt,s for different services.
Using these data, the degradation cost of the BESS for different SoCt and SoCt−1 scenarios
and C-Rates are plotted as shown in Figure 5.3. It is noted that higher C-Rate values (e.g.
C-Rate = 4) results in a relatively high degradation cost as compared to lower C-Rate (e.g.
C-Rate = 1). For instance, the discharge of BESS from an SoCt−1 value of 100% to SoCt
of 20% will have a degradation cost of about 2 $/kWh. On the other hand, the maximum
change in SOC within the limits of LF service will result in maximum degradation cost of
1 $/kWh.
In order to incorporate (5.6) into an MILP problem, it can be linearized using multi-

97
 Figure 5.3: Degradation cost of BESS for different flexibility services.

linear regression as follows [87]:

DGCt,s = as SoCt,s + bs SoCt−1,s + cs ∆SoCt,s ∀ t ∈ T ,s ∈ S (5.7)

In (5.7), the linearized degradation cost DGCt,s is presented. Service LF, MF, and
HF have their individual degradation cost coefficients as , bs , and cs , which gives different
weights for their associated variables.

5.2.3 LDC Operations Considering BESS Flexibility Services

The distribution system accommodates several distributed BESSs which are contracted to
provide any of the three previously defined flexibility services – LF, MF, and HF. It is also
assumed that the LDC purchases energy from the wholesale energy market at the Hourly

98
Electricity Price (HEP), while its energy arbitrage with the distributed BESSs is carried
out at TOU prices. The LDC is responsible for meeting the system demand while seeking
to minimize its total operation cost, as given below:

T
X J X
X T  
Peak P eak
J =σ P + σtHEP Pj,t
SS
− CH
Pj,t DCH
− Pj,t σtTOU
t j t
J X
X T X
S  
+ as SoCj,t Zj,t,s + bs SoCj,t−1 Zj,t,s + cs ∆SoCj,t,s (5.8)
j t s

The objective function (5.8) comprises four components: the first component represents
the demand charge paid by the LDC depending on its measured peak load during a given
operating period; the second, the total cost of buying energy from the electricity market at
HEP; the third, the net cost of selling / buying energy to / from the distributed BESSs at
TOU prices; and the final, the degradation cost of BESS operation at different flexibility
services, as discussed in Section 5.2.1 and 5.2.2. The objective function (5.8) is subject
to the nodal demand-supply balance constraints for active and reactive power, as given
below:

J
X  
SS d DCH CH
Pj,t − Pj,t + Pj,t − Pj,t = Kj,k,1 (δj,t − δk,t ) + Kj,k,2 (Vj,t − Vk,t ) ,
k=1,k6=j

∀t ∈ T ; ∀(j, k) ∈ J (5.9)

N
X
QSS
t − Qdj,t = Kj,k,2 (δj,t − δk,t ) + Kj,k,1 (Vj,t − Vk,t ), ∀t ∈ T ; ∀(j, k) ∈ J (5.10)
k=1,k6=j

where,
x2j,k rj,k xj,k
2 +x2
rj,k 2 +x2
rj,k
j,k j,k
Kj,k,1 = , Kj,k,2 = , ∀(j, k) ∈ J (5.11)
xj,k xj,k
The linearized power flow equations (5.9) and (5.10) are given in [45] and adopted here to

99
represent the power balance equations. Additionally, the following constraints are imposed:

SS
0 ≤ Pj,t ≤ PtSS , ∀t ∈ T ; j = 1 (5.12)

d CH DCH
Pj,t + Pj,t − Pj,t ≤ P Peak , ∀t ∈ T ; j ∈ J (5.13)

0.9 ≤ Vj,t ≤ 1.05, ∀t ∈ T ; ∀(j, k) ∈ J (5.14)

Equation (5.12) denotes the limits on power imported by the LDC from the external
grid, which is limited by the capacity of the distribution transformer. Constraint (5.13)
calculates the system peak over the day. Constraint (5.14) ensures that the voltages at
each bus are within predefined ranges.
The BESSs operational constraints are given below:

dch
Pj,t
Ej,t = Ej,t−1 + τ η ch Pj,t
ch
−τ , ∀t ∈ T , j ∈ J (5.15)
η dch
Ej ≤ Ej,t ≤ Ej , ∀t ∈ T , j ∈ J (5.16)
ch
0 ≤ Pj,t ≤ Pj Zj,t,ch , ∀t ∈ T , j ∈ J (5.17)
dch
0 ≤ Pj,t ≤ Pj Zj,t,dch , ∀t ∈ T , j ∈ J (5.18)

Zj,t,ch + Zj,t,dch ≤ 1, ∀t ∈ T , j ∈ J (5.19)

Equation (5.15) represents the inter-temporal changes of energy level of the BESS taking
into account its charging and discharging efficiencies. Constraints (5.16) ensures that the
energy levels of the BESSs are within their rated capacities. In addition, the charge and
discharge decisions are limited by the rated power of the BESS, as given in (5.17) and
(5.18).
The SOC of the BESS is the level of charge of the facility relative to its rated capacity,
as given below:

Ej,t
SoCj,t = , ∀t ∈ T , j ∈ J (5.20)
Ej

100
  P
SoCj,t−1 +
 ∆SoCj,t,s , if t 6= {0, T }
SoCj,t = s∈{s1 ,s,2 ,s3 } , ∀j ∈ J (5.21)
0.5, otherwise


SoC ≤ SoCj,t ≤ SoC, ∀t ∈ T , j ∈ J (5.22)

The inter-temporal changes in SOC of the BESSs are functions of their change in SOC
as given in (5.21). Constraint (5.22) limits the SOC levels within their lower and upper
limits, which are assumed to be 20% and 100%, respectively, in the present work. The
change in SOC of BESSs are modeled as follows:

0 ≤ |∆SoCj,t,LF | ≤ CRLF Zj,t,LF , ∀t ∈ T , j ∈ J (5.23)

CRLF Zj,t,M F < |∆SoCj,t,M F | ≤ CRM F Zj,t,M F , ∀t ∈ T , j ∈ J (5.24)

CRM F Zj,t,HF < |∆SoCj,t,HF | ≤ CRHF Zj,t,HF , ∀t ∈ T , j ∈ J (5.25)

Zj,t,LF + Zj,t,M F + Zj,t,HF ≤ 1, ∀t ∈ T , j ∈ J (5.26)

The constraints (5.23) to (5.25) denotes the limits of the changes in SOC of BESSs for
service LF, MF, and HF, respectively. The ∆SoCj,t,s is limited by the maximum rate of
change CRs , as discussed earlier in (5.1). Constraint (5.26) ensures that only a single
service, LF, MF, or HF, can be selected at a time.
The above formulation contains some nonlinear terms, as in the objective function (5.8)
and the constraints (5.23) to (5.25), which can be easily reformulated as follows [88]:

SoCLj,t,s ≥ 0, ∀t ∈ T , j ∈ J (5.27)

SoCLj,t,s ≤ SoCj,t , ∀t ∈ T , j ∈ J (5.28)

SoCLj,t,s ≤ Zj,t,s M, ∀t ∈ T , j ∈ J (5.29)

SoCLj,t,s ≥ SoCj,t − M (1 − Zj,t,s ), ∀t ∈ T , j ∈ J (5.30)

101
 Constraints (5.27) to (5.30) represent the linearization of the product of the two
variables appearing in (5.8): the binary variable (Zj,t,s ) and the continuous variable
(SoCj,t ). The linearized term is introduced as a positive variable SoCLj,t,s . The constraint
(5.28) ensures that SoCLj,t,s is less than or equal to SoCj,t . In (5.29) and (5.30), SoCLj,t,s is
forced to be equal to SoCj,t if Zj,t,s = 1; otherwise, SoCLj,t,s equals to zero.
The term |∆SoCj,t,s | in (5.23) to (5.25) can be linearly formulated as below [88]:
+ −
∆SoCj,t,s ≥ 0, ∆SoCj,t,s ≤ 0, ∀t ∈ T , j ∈ J (5.31)
+ −
|∆SoCj,t,s | = ∆SoCj,t,s − ∆SoCj,t,s , ∀t ∈ T , j ∈ J (5.32)
+ + +
CRs−1 SoC Zj,t,s < ∆SoCj,t,s ≤ CRs SoC Zj,t,s , ∀t ∈ T , j ∈ J (5.33)
− − −
− CRs SoC Zj,t,s ≤ ∆SoCj,t,s < −CRs−1 SoC Zj,t,s , ∀t ∈ T , j ∈ J (5.34)
+ −
Zj,t,s + Zj,t,s ≤ Zj,t,s , ∀t ∈ T , j ∈ J (5.35)

Two auxiliary variables are introduced to reformulate the absolute change in SOC;
+ −
∆SoCj,t,s and ∆SoCj,t,s converts the range of the change in SOC to two components. The
boundaries of the positive and negative changes in SOC are modeled as in (5.33) and
+ −
(5.34), respectively. The binary variables Zj,t,s and Zj,t,s are used to ensure that either
+ −
∆SoCj,t,s or ∆SoCj,t,s should be zero. Constraint (5.35) coordinates the binary variables of
the linearized change in SOC and the binary variable associated with the selected service
to be provided.

5.2.4 Allocation of Cost Savings

First, an LDC operations model without any BESS units is executed and the total
system cost is determined. Next, the LDC operations model including distributed BESSs,
presented in Section 5.2.3, is executed to find the operations cost with the flexibility
service provisions from the BESS. The difference between the two system operation costs
is the total contribution of the distributed BESSs to system savings.
In this work, the Shapley value criterion [89], which is well known in the literature of

102
cooperative game theory, is used to allocate the total system savings among the BESSs
for their flexibility service provisions, on the basis of their marginal contributions to the
savings. This concept provides a unique allocation strategy among the participants in a
coalition game [89]. It is assumed that the LDC schedules and dispatches the distributed
BESS units. In return, the LDC distributes the savings accrued among the BESSs units.
The Shapley value of a cooperative game is given as follows:

n
1 X 1 X
φh = v(h) (5.36)
n q=1 c(q) h∈H

where,
v(h) = v(H) − v(H − h) (5.37)

(n − 1)!
c(q) = (5.38)
(n − q)!(q − 1)!

The second summation in (5.36) represents the sum of all marginal contributions of a
BESS h to all possible subsets of BESS units H, which is obtained from (5.37). The first
summation calculates the number of all formed coalitions of size q that contains the BESS
h, and n is the total number of BESSs, as in (5.38).
From the above discussions, the notion of contribution of the BESSs to system savings
can be utilized to develop a financial compensation scheme for their flexibility service
provisions. The payment for flexibility services by a BESS unit can be represented as
follows:

ρh = fh + φh (5.39)

Equation (5.39) includes two components: a fixed payment component fh , and a

variable payment component φh . The fixed payment can be payable to the BESS for
participating, collaborating and providing control access to their flexible asset to the
LDC. The BESSs would receive this component of payment even if the LDC keeps a
BESS unit in idle mode. The variable component of payment in (5.39) would be a

103
function of the contribution of BESS h to the system savings, as obtained using the
Shapley value criterion.

5.3 Results and Discussions

In this example, the LDC is responsible for managing a 33-bus distribution system, the
same as that used in the previous chapters [90], and meeting the demand of its customers.
It is assumed that three investor-owned BESS units are located at buses j = 2, 23, and 30,
which is in line with the findings of optimal BESS placement from the previous chapter;
they are sized 800 kW/200 kWh, 2400 kW/600 kWh, and 3200 kW/800 kWh, respectively.
The LDC aims to minimize its total operation cost, given in (5.8) which includes the cost
of flexible operation of the BESSs. The following scenarios have been analyzed:

• SC-1: Normal operation with a typical given HEP profile.

• SC-2: Sudden 100% increase in HEP in the interval 1-2 PM.

• SC-3: Sudden 30% increase in system demand during the interval 8-9 PM, which is
close to the peak-load hour of the system.

The total system load profile and the HEP profile over an operation day of August 1,
2018, are shown in Figure 5.4 and 5.5, respectively. The system peak demand is 0.4 p.u.,
which is equivalent to 4 MW. Shown in Figure 5.5, the HEP varies over the day between
1.2 and 2.2 $/kWh.

5.3.1 Calculation of Shapley Value of BESS Investors

LF Service

In this subsection only the LF service from BESSs is considered in the SC-1 (normal
operation) scenario, in order to demonstrate in detail the calculation of the Shapley values
from the system savings and the marginal contributions of the BESSs. First, the LDC

104
 0.45

0.4

0.35
System Demand (p.u.)

0.3

0.25

0.2

0.15

0.1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

Figure 5.4: Load profile of the LDC over a day.

2.4

2.2

2
HOEP ($/kWh)

1.8

1.6

1.4

1.2

1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

Figure 5.5: HEP profile over a day.

105
operations model (i.e. (5.8) to (5.35)) is executed considering all possible BESS coalitions
to calculate their contribution to the total system savings. As stated earlier, three BESS
units (B-2, B-23 and B-30) are participating in LF service provisions, and hence they can
form one of the seven different coalitions, as presented below:

H = {B-2} V {B-23} V {B-30} V

{B-2,B-23} V {B-23,B-30} V {B-30,B-2} V
{B-2,B-23,B-30} (5.40)

After executing the LDC operations model, the total system savings for each coalition
is reported in Table 5.2. The savings accrued from single-facility coalitions namely, B-2,
B-23, and B-30 operating independently, are $30, $117, and $153, respectively. The BESS
units can also form paired coalitions such as B-2, B-23, B-23, B-30, and B-30, B-2, which
accrues total system savings of $139, $244, and $176, respectively. Also, in the grand
coalition when all BESS units are operating simultaneously, the total system savings is
$266. It should be noted that the total system savings are calculated with reference to the
LDC system operation cost when there are no BESSs.

Table 5.2: Total Savings in a BESS Coalition for LF Service

Coalition (H) Total System Savings ($)
B-2 30
B-23 117
B-30 153
B-2, B-23 139
B-23, B-30 244
B-30, B-2 176
B-30, B2, B-23 266

From the knowledge of total system savings in each coalition, the marginal contribution
of a BESS unit can be determined. Table 5.3 presents the marginal contributions of B-2 to
each coalition in which it can be part of; these are obtained by subtracting the savings of

106
the coalition without B-2 from the savings obtained from a coalition in which B-2 is part
of. The corresponding Shapley value of B-2 for the LF service is obtained as $25. Similar
to the calculations shown in Table 5.3, Shapley values can be calculated for other BESS
units B-23 and B-30 as well, as shown in Table 5.4 and 5.5. It should be noted that the
sum of the Shapley values of all BESS units for a given service, should be equal to the
total system savings achieved in the grand coalition. For instance, the sum of the Shapley
values for the three BESS units in the present example, for the LF service provisions, is
$25 + $102 + $139 = $266, which is the savings accrued in the grand coalition, as noted
from 5.2.

Table 5.3: Marginal Contributions and Shapley Value Calculation for BESS B-2
Marginal Contribution of B-2 in coalition H Weight on H
Coalition (H) 1 1
v(h) = v(H) − v(H − h)
n c(H)
B-2 $30 - $0 = $30 1/3
B-2, B-23 $139 - $117 = $22 1/6
B-2, B-30 $176 - $153 = $23 1/6
B-2, B23, B-30 $266 - $244 = $22 1/3
Shapley Value ($) 30/3 + 22/6 + 23/6 + 22/3 ≈ $25

Table 5.4: Marginal Contributions and Shapley Value Calculation for BESS B-23
Marginal Contribution of B-23 in coalition H Weight on H
Coalition (H) 1 1
v(h) = v(H) − v(H − h)
n c(H)
B-23 $117 - $0 = $117 1/3
B-23, B-2 $139 - $30 = $109 1/6
B-23, B-30 $244- $153 = $91 1/6
B23, B-2, B-30 $266 - $176 = $90 1/3
Shapley Value ($) 117/3 + 109/6 + 91/6 + 90/3 ≈ $102

107
 Table 5.5: Marginal Contributions and Shapley Value Calculation for BESS B-30
Marginal Contribution of B-30 in coalition H Weight on H
Coalition (H) 1 1
v(h) = v(H) − v(H − h)
n c(H)
B-30 $153 - $0 = $153 1/3
B-30, B-2 $176 - $30 = $145 1/6
B-30, B-23 $244 - $117 = $127 1/6
B-30, B-2, B-23 $266 - $139 = $126 1/3
Shapley Value ($) 153/3 + 145/6 + 127/6 + 126/3 ≈ $139

MF and HF Services

This subsection discusses the provision of LF, MF, and HF services being available
individually, for scenario SC-1. Table 5.6 presents the savings accrued in the grand
coalition of the BESSs for each of these services, and a fourth case when all the flexibility
services are available simultaneously – but only one of them is selected. It is noted that
the grand coalition savings for service LF are very close to that when all services are
provided simultaneously. This implies that the LF service is the most selected service
while MF or HF only leads to a reduction in the total savings. Also note that when all
services are available simultaneously, the grand coalition savings are the highest. Similar
to the discussion presented in Section 5.3.1, Tables 5.7, 5.8, and 5.9 present the marginal
contributions and allocations of savings among B-2, B-23, and B-30, respectively.

Table 5.6: Grand Coalition Savings for a given Flexibility Service

Service Provided Grand Coalition Savings ($)
LF Only 266
MF Only 247
HF Only 184
LF & MF & HF Available Simultaneously 267

As seen from the tables, the size of the BESS plays a crucial role in its marginal
contribution. The BESS B-2, for example, represents 12.5% of the total installed BESS
capacity, and receives the lowest share of savings, and correspondingly the Shapley value

108
 Table 5.7: Allocation of Saving of BESS B-2 (800 kW/200 kWh)
Marginal Contribution of B-2
v(h) = v(H) − v(H − h) Weight on H
Coalition (H) 1 1
n c(H)
LF ($) MF ($) HF ($) ALL ($)
B-2 30 22 18 30 1/3
B-2, B-23 23 21 23 23 1/6
B-2, B-30 23 28 27 23 1/6
B-2, B-23, B-30 22 22 18 22 1/3
Shapley Value ($) 25 23 20 25

Table 5.8: Allocation of Saving of BESS B-23 (2400 kW/600 kWh)

Marginal Contribution of B-23
v(h) = v(H) − v(H − h) Weight on H
Coalition (H) 1 1
n c(H)
LF ($) MF ($) HF ($) ALL ($)
B-23 117 103 64 120 1/3
B-23, B-2 109 102 69 112 1/6
B-23, B-30 91 98 82 92 1/6
B-23, B-2, B-30 90 92 73 91 1/3
Shapley Value ($) 102 98 71 104

Table 5.9: Allocation of Saving of BESS B-30 (3200 kW/800 kWh)

Marginal Contribution of B-30
v(h) = v(H) − v(H − h) Weight on H
Coalition (H) 1 1
n c(H)
LF ($) MF ($) HF ($) ALL ($)
B-30 153 128 85 160 1/3
B-30, B-2 145 134 93 152 1/6
B-30, B-23 127 122 102 131 1/6
B-30, B-2, B-23 126 123 97 131 1/3
Shapley Value ($) 139 126 93 144

109
 Table 5.10: Share of Savings Among BESSs
Shapley Value Share in Total Average of the Share in
BESS ($) Savings (%) Total Savings
(%)
LF MF HF LF MF HF
B-2 25 23 20 9.4 9.3 10.9 9.8
B-23 102 98 71 38.4 39.7 38.6 38.9
B-30 139 126 93 52.2 51 50.5 51.2
* Capacity share of B-2: 12.5%, B-23: 37.5%, B-30: 50%

of BESS B-2 approximately equals its percentage share of installed capacity within the
distribution system.
Table 5.10 presents the distribution of savings among the BESS units using the
Shapley value criterion for individual flexibility services. Each of the BESS unit’s share in
total savings for LF service is obtained by dividing the corresponding Shapley values by
the savings in the grand coalition (i.e. $266), as reported in Table 5.6. Similarly, the
BESS unit’s share in total savings for MF and HF services are obtained by dividing the
corresponding Shapley values by the savings in the grand coalition (i.e. $247 and $184,
respectively), obtained from Table 5.6. The averages of the share in total savings
considering all the services, for B-2 is 9.8%, for B-23 is 38.9% and for B-30 is 51.2%. It is
noted that these average shares in savings are in line with their capacity shares of 12.5%,
37.5%, and 50% with reference to the total BESS capacity in the system. It may be
inferred that B-30 has the highest impact on total system flexibility since it has the
largest share in system savings.

5.3.2 Operational Performance of BESS Flexibility Services in

Different Scenarios

Figure 5.6 shows the operation of the BESSs for different flexibility services; the changes
in the SOC are presented for every 15 min operating interval. It is noted from Figure
5.6(a) that the increments/decrements in SOC with the LF service are smaller than other

110
 1.2
B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(a) LF
1.2
B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(b) MF
1.2
B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(c) HF
1.2
B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(d) All Services

Figure 5.6: SOC of BESSs providing different flexibility services in Scenario SC-1.

111
 1.2
High Electricity Price Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(a) LF
1.2
High Electricity Price Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(b) MF
1.2
High Electricity Price Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(c) HF
1.2
High Electricity Price Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(d) All Services

Figure 5.7: SOC of BESSs for different flexibility services in Scenario SC-2.

112
 1.2
High Demand Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(a) LF
1.2
High Demand Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(b) MF
1.2
High Demand Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(c) HF
1.2
High Demand Period B-2 B-23 B-30

0.8
SOC (p.u.)

0.6

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (t)

(d) All Services

Figure 5.8: SOC of BESSs for different services in Scenario SC-3.

113
services. For example, all the batteries require an hour to fully discharge between 9 to
10 AM when providing LF service, while B-23 and B-30 requires only 30 minutes to fully
discharge, when providing MF service. The fastest response is noted in HF service, when
B-23 and B-30 discharges from high to low SOC in 15 minutes (Fig. 5.6(c)). Each service
impacts the total operation cost of the BESS. The need for larger changes in the SOC in
a short time results in a high degradation cost of the BESS, as given in (5.8).
Figure 5.7 shows the operation of BESS B-30 in scenario SC-2 which assumes a 100%
increase in the HEP between 1-2 PM, from its price reported in Fig. 5.5. The operation
of the BESS units are exactly identical to scenario SC-1 until the price spike has occurred
at 1 PM, after which the flexibility services come into play. It is noted that all the BESS
units are discharged fully during the period of price spike in the LF service or when all
services are provided simultaneously. In the case of MF service B-23 and B-30 are fully
discharged, while B-2 is discharged to 30% SOC within 30 min, as noted in Fig. 5.7(b).
In the case of HF service all the BESS units are discharged to 30% SOC within 15 min, as
noted in Fig. 5.7(c). Finally, Figure 5.7(d) shows that when all the services are available
simultaneously, 80% of the BESS stored energy is depleted during the price spike interval,
and the LF service is primarily selected over the other services.
In should be noted that since the overall objective of the LDC is to minimize its system
operation cost, the BESS units discharge when the price spike occurs in order to reduce
the amount of power purchased from the external grid at the high price while supplying
the LDC demand internally, thereby maximizing the savings.
In scenario SC-3 the LDC system demand during hour 20-21 increases by 30%. It is
again assumed that the BESS operations until that point of time were identically following
the base-case operation of the day-ahead, as was determined from scenario SC-1 (Figure
5.6) and that all the BESS units were at low SOC levels at hour-20. When the demand
spike occurs, if the BESS units would start charging in order to discharge and meet the
demand spike, it would create an instantaneous further increase in demand which would
be detrimental for the system. Hence, the BESS units continue to stay idle at their low
SOC levels, and do not provide any flexibility services, as shown in Fig. 5.8.

114
5.4 Summary

In this chapter, a novel flexibility services framework was developed based on the
capabilities of BESSs in providing different C-Rate levels, namely, low flexibility (LF),
moderate flexibility (MF), and high flexibility (HF). A novel mathematical model was
proposed, from the perspective of the LDC, to optimally procure the LF, MF and HF
services in the operations horizon. This model included a degradation cost formulation of
the BESS units based on the new flexibility services definitions proposed. Thereafter, a
cooperative game-theory based approach was applied wherein the Shapley value criterion
was used to allocate the total system savings among the flexible BESS asset owners, and
hence a pricing scheme for flexibility services was developed. The proposed approach
ensured that the flexible resources were financially compensated based on their marginal
contribution to the system savings.

115
Chapter 6

Conclusions

6.1 Summary

This thesis focused on the flexibility provisions from REHs and BESSs in operation and
planning frameworks of distribution systems. Chapter-1 presented the motivation and
related literature review relevant to the work presented in the subsequent chapters. In
addition, Chapter-1 outlined the main research objectives and finally the overall layout of
the thesis.
In Chapter-2, brief overviews of the background topics related to the research objectives
were presented. The concept of REH and the mathematical model for the optimal operation
of the HEMSs were discussed. The characteristics and various technologies of ESS were
discussed along with their applications in smart grids. The RCA was introduced, and
followed by a brief introduction on NNs. Also, a brief discussion on cooperative game
theory was presented with particular focus on the concept of Shapley value criterion.
In Chapter-3, a novel two-stage scheme was proposed to coordinate the interaction
between the large number of REHs and the LDC. A flexibility evaluation approach was
developed to quantify the available flexibility from individual REHs. The evaluated
flexibility, along with the REHs load profiles were communicated to the LDC in order to
study their impact on the distribution grid operation. A new mathematical model was

116
proposed for the LDC system operator to determine the bus-wise PRRs, which were
communicated back to the REHs based on their individual flexibility indices.
The main advantage of the proposed two-stage coordination scheme was the
avoidance of a centralized optimization model which would need to optimize thousands of
REHs as well as the LDC’s operations, simultaneously. Therefore, the proposed scheme
disaggregated and distributed the computational burden amongst each entity in the
distribution grid.
In Chapter-4, a novel NN-based SoH estimator was developed considering the
presence of a large cluster of REHs, as introduced in Chapter-3, and a large data set of
BESS operations, simulated to mimic the LDC’s behavior in controlling these units.
Subsequently, the obtained structure of the NN was converted to a functional
relationship, which was incorporated within a BESS planning problem. The weights and
biases of the mathematical representation of the NN were updated iteratively to improve
the plan decisions. The proposed approach showed high accuracy in estimating the SoH
of the BESS. Different case studies were carried out to investigate the impact of
neglecting calendar and cycle aging and under / over-estimation of the BESS capacity on
the LDC’s plan decisions. In addition, several scenarios were simulated to study the
influence of the load mix and PV penetration on battery degradation.
In Chapter-5, new definitions for flexibility services from BESS units were developed,
namely, low flexibility (LF), moderate flexibility (MF), and high flexibility (HF) services.
The impact of each of these services on the BESS degradation cost were captured and
modeled. A novel LDC operation model was proposed to optimally procure the LF, MF and
HF services while taking into account the degradation cost model. Thereafter, the Shapley
value criterion was applied to allocate the total system savings among the BESS units for
their respective services. The proposed approach ensured that the flexible resources were
financially rewarded based on their marginal contribution to the system savings.
The following conclusions can be drawn from the thesis:

• The presence of a large number of uncoordinated REHs can reshape the total load
of the distribution system and might result in a high peak-to-average power ratio.
Studies revealed that the procurement of flexibility from REHs significantly

117
 enhanced the grid operational efficiency, reduced peak demand, and minimized the
power system losses.

• The integration of BESSs in the distribution grid can significantly contribute to the
LDC’s operational flexibility. However, it is important to consider their SoH while
making operational and planning decisions. The proposed novel BESS degradation
model integrated within the BESS planning problem presented a very accurate
estimate of the batteries’ SoH. This helped the LDC in avoiding under /
over-estimation of the BESSs available capacity during the operation phase, and
the required rated power and energy capacity in the planning phase.

• Scenario studies were carried out to analyze the impact of load mix on the plan
decisions and the degradation of BESSs. Load profiles with extended duration peaks
resulted in a larger installed capacity of BESSs. The studies also revealed that the
level of PV penetration significantly reduced the required rated capacity of batteries.
Although the considered system load profile had a peak during the daytime, because
of the presence of REHs, the system peak demand was shifted and did not coincide
with the peak power generation from the PV panels, hence requiring lesser installation
capacity of BESS.

• Studies revealed that BESSs can provide flexibility services to the LDC particularly
during demand and price spikes. The contribution of a BESS unit to flexibility was
noted to be proportional to its capacity share in total installed capacity.

6.2 Contributions
The main contributions of the research presented in this thesis can be summarized as
follows:

• The mathematical model of the REH, previously reported, was extensively improved
by taking into account a comprehensive set of individual appliance operation models
and their optimal operation, considering the customer’s preferences, behavior, and
objectives.

118
• A new REH model was presented considering different levels of interactions in
terms of power interchanges between several devices internally within the REH and
externally with the LDC.

• A novel concept of residential load flexibility was proposed and hence a flexibility
index was defined to quantify the available flexibility from an REH, aggregated from
that provided by various appliances, ESS, and PV panel.

• A novel two-stage coordination scheme and associated novel models of the REH and
LDC were proposed that took into account distinctly individual objectives of each
HEMS as well as that of the LDC, seeking to enhance the grid operational efficiency,
and create an aggregated flexibility provision for the system.

• A novel NN based degradation model was proposed to estimate the SoH of Li-ion
batteries of an BESS by considering a large data set of BESS operations for NN
training. This data set was obtained by simulating the LDC operations in controlling
the BESS in the presence of a large cluster of uncoordinated REH loads.

• The proposed SoH model of the BESS was incorporated into the planning model to
determine the optimal energy capacity, power rating, location and year of
installation / replacement of BESSs while internalizing battery capacity
degradation due to cycling and aging effects.

• A novel flexibility services framework was developed based on the capabilities of

BESSs in providing different C-Rate levels, namely, low flexibility (LF), moderate
flexibility (MF), and high flexibility (HF). A novel mathematical model was
proposed, from the perspective of the LDC, to optimally procure the LF, MF and
HF services in the operations horizon.

• A cooperative game theory based approach was applied to allocate the accrued
financial benefits among the BESS units for their flexibility services. The Shapley
value was applied to fairly distribute the total savings from the BESS flexibility
provision based on their individual marginal contributions, and hence a new pricing
scheme for flexibility services was proposed.

119
6.3 Future Work

Based on the work presented in this thesis, the following issues can be examined in the
future:

• With the increase in PEV penetration, there is a need to consider their smart
charging and how they impact the system operation. This involves optimizing the
PEV charging load in the REHs to minimize the customer’s cost. It is also worth
investigating the effect of smart charging PEVs on the flexibility provided by an
REH.

• The study of flexibility of smart loads and BESSs need be extended further to examine
their impact on the reliability of the distribution system.

• The proposed SoH estimation approach can be incorporated within the bids / offers
of BESS units and included within electricity market settlement models.

• The BESS flexibility service provision framework can be improved to take into
account the impact of uncertainty in demand and electricity prices on the BESS
operation by incorporating model predictive control approach.

• The proposed flexibility services framework can be extended to be included as

products for trading within a wholesale electricity market auction model.

120
References

[1] International Energy Agency, “World energy outlook 2020,” IEA, Tech. Rep., 2020.
[Online]. Available: www.iea.org/reports/world-energy-outlook-2020

[2] Independent Electrcity System Operator, “Options for the future of peaksaverplus®
residential demand response.” [Online]. Available: www.ieso.ca/-/media/files/ieso/
document-library/sac/2016/sac-20160210-peaksaverplus-presentation.pdf?la=en

[3] Independent Electricity System Operator, “Global adjustment (ga).” [Online].

Available: www.ieso.ca/power-data/price-overview/global-adjustment

[4] Federal Energy Regulatory Commission, “Assessment of demand response and

advanced metering.” [Online]. Available: www.ferc.gov/sites/default/files/2020-04/
09-07-demand-response.pdf

[5] E. Lannoye, D. Flynn, and M. O’Malley, “Evaluation of power system flexibility,”

IEEE Trans. Power Syst., vol. 27, no. 2, pp. 922–931, May 2012.

[6] N. Ruiz, I. Cobelo, and J. Oyarzabal, “A direct load control model for virtual power
plant management,” IEEE Trans. Power Syst., vol. 24, no. 2, pp. 959–966, May 2009.

[7] P. Favre-Perrod, “A vision of future energy networks,” in 2005 IEEE Power

Engineering Society Inaugural Conference and Exposition in Africa, July 2005, pp.
13–17.

121
 [8] M. Geidl, G. Koeppel, P. Favre-Perrod, B. Klockl, G. Andersson, and K. Frohlich,
“Energy hubs for the future,” IEEE Power and Energy Magazine, vol. 5, no. 1, pp.
24–30, Jan 2007.

[9] Ontario’s Independent Electricity System Operator. IESO York Region Non
Wires Alternatives Demonstration Project. Accessed: Feb. 15, 2021. [Online].
Available: https://www.ieso.ca/en/Sector-Participants/Engagement-Initiatives/
Engagements/IESO-York-Region-Non-Wires-Alternatives-Demonstration-Project

[10] Federal Energy Regulatory Commission. FERC Order No. 2222: A New Day
for Distributed Energy Resources. Accessed: Feb. 15, 2021. [Online]. Available:
https://www.ferc.gov/media/ferc-order-no-2222-fact-sheet

[11] Independent, Electricity Market Operator, “18 month outlook: An assessment of the
reliability and operability of the ontario electricity system,” 2018.

[12] Canadian Energy Research Institute, “Electricity storage systems: Applications

and business cases,” 2019. [Online]. Available: https://ceri.ca/assets/files/
Study 180 Full Report.pdf

[13] W. J. Cole and A. Frazier, “Cost projections for utility-scale battery storage,” National
Renewable Energy Lab.(NREL), Golden, CO (United States), Tech. Rep., 2019.

[14] K. Abdulla, J. de Hoog, V. Muenzel, F. Suits, K. Steer, A. Wirth, and S. Halgamuge,

“Optimal operation of energy storage systems considering forecasts and battery
degradation,” IEEE Transactions on Smart Grid, vol. 9, no. 3, pp. 2086–2096, 2018.

[15] A. H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load control with

price prediction in real-time electricity pricing environments,” IEEE Trans. Smart
Grid, vol. 1, no. 2, pp. 120–133, Sept 2010.

[16] M. Pedrasa, E. Spooner, and I. MacGill, “Robust scheduling of residential distributed

energy resources using a novel energy service decision-support tool,” in Innovative
Smart Grid Technologies (ISGT), 2011 IEEE PES, Jan 2011, pp. 1–8.

122
[17] J. H. Yoon, R. Baldick, and A. Novoselac, “Dynamic demand response controller based
on real-time retail price for residential buildings,” IEEE Trans. Smart Grid, vol. 5,
no. 1, pp. 121–129, Jan 2014.

[18] Y. Wang, X. Lin, and M. Pedram, “Adaptive control for energy storage systems in
households with photovoltaic modules,” IEEE Trans. Smart Grid, vol. 5, no. 2, pp.
992–1001, March 2014.

[19] Y. Ozturk, D. Senthilkumar, S. Kumar, and G. Lee, “An intelligent home energy
management system to improve demand response,” IEEE Trans. Smart Grid, vol. 4,
no. 2, pp. 694–701, June 2013.

[20] ——, “An intelligent home energy management system to improve demand response,”
IEEE Trans. Smart Grid, vol. 4, no. 2, pp. 694–701, June 2013.

[21] A. Anvari-Moghaddam, H. Monsef, and A. Rahimi-Kian, “Optimal smart home energy

management considering energy saving and a comfortable lifestyle,” IEEE Trans.
Smart Grid, vol. 6, no. 1, pp. 324–332, Jan 2015.

[22] M. Bozchalui, S. Hashmi, H. Hassen, C. Canizares, and K. Bhattacharya, “Optimal

operation of residential energy hubs in smart grids,” IEEE Trans. Smart Grid, vol. 3,
no. 4, pp. 1755–1766, Dec. 2012.

[23] A.-H. Mohsenian-Rad, V. Wong, J. Jatskevich, R. Schober, and A. Leon-Garcia,

“Autonomous demand-side management based on game-theoretic energy consumption
scheduling for the future smart grid,” IEEE Trans. Smart Grid, vol. 1, no. 3, pp. 320–
331, Dec 2010.

[24] T. H. Chang, M. Alizadeh, and A. Scaglione, “Real-time power balancing via

decentralized coordinated home energy scheduling,” IEEE Trans. Smart Grid, vol. 4,
no. 3, pp. 1490–1504, Sept 2013.

[25] A. C. Luna, N. L. Diaz, M. Graells, J. C. Vasquez, and J. M. Guerrero, “Cooperative

energy management for a cluster of households prosumers,” IEEE Trans. Consum.
Electron, vol. 62, no. 3, pp. 235–242, August 2016.

123
[26] T. Logenthiran, D. Srinivasan, and T. Z. Shun, “Demand side management in smart
grid using heuristic optimization,” IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1244–
1252, Sept 2012.

[27] D. Wang, S. Ge, H. Jia, C. Wang, Y. Zhou, N. Lu, and X. Kong, “A demand response
and battery storage coordination algorithm for providing microgrid tie-line smoothing
services,” IEEE Trans. Sustain. Energy, vol. 5, no. 2, pp. 476–486, April 2014.

[28] Y. Guo, M. Pan, Y. Fang, and P. Khargonekar, “Decentralized coordination of energy

utilization for residential households in the smart grid,” IEEE Trans. Smart Grid,
vol. 4, no. 3, pp. 1341–1350, Sept 2013.

[29] G. Brusco, A. Burgio, D. Menniti, A. Pinnarelli, and N. Sorrentino, “Energy

management system for an energy district with demand response availability,” IEEE
Trans. Smart Grid, vol. 5, no. 5, pp. 2385–2393, Sept 2014.

[30] Safdarian, A. and Fotuhi-Firuzabad, M. and Lehtonen, M., “A distributed algorithm

for managing residential demand response in smart grids,” IEEE Trans. Ind. Informat,
vol. 10, no. 4, pp. 2385–2393, Nov 2014.

[31] A. Safdarian, M. Fotuhi-Firuzabad, and M. Lehtonen, “Optimal residential load

management in smart grids: A decentralized framework,” IEEE Trans. Smart Grid,
vol. 7, no. 4, pp. 1836–1845, July 2016.

[32] N. G. Paterakis, O. Erdinç, I. N. Pappi, A. G. Bakirtzis, and J. P. S. Catalão,

“Coordinated operation of a neighborhood of smart households comprising electric
vehicles, energy storage and distributed generation,” IEEE Trans. Smart Grid, vol. 7,
no. 6, pp. 2736–2747, Nov 2016.

[33] B. Moradzadeh and K. Tomsovic, “Two-stage residential energy management

considering network operational constraints,” IEEE Trans. Smart Grid, vol. 4, no. 4,
pp. 2339–2346, Dec 2013.

[34] K. McKenna and A. Keane, “Residential load modeling of price-based demand

response for network impact studies,” IEEE Transactions on Smart Grid, vol. 7, no. 5,
pp. 2285–2294, Sept 2016.

124
[35] K. Hartwig and I. Kockar, “Impact of strategic behavior and ownership of energy
storage on provision of flexibility,” IEEE IEEE Trans. Sustainable Energy, vol. 7,
no. 2, pp. 744–754, April 2016.

[36] J. Ponocko and J. V. Milanovic, “Forecasting demand flexibility of aggregated

residential load using smart meter data,” IEEE Trans. Power Syst., 2018.

[37] A. Anvari-Moghaddam, H. Monsef, and A. Rahimi-Kian, “Optimal smart home energy

management considering energy saving and a comfortable lifestyle,” IEEE Trans.
Smart Grid, vol. 6, no. 1, pp. 324–332, Jan 2015.

[38] O. Kilkki, A. Alahäivälä, and I. Seilonen, “Optimized control of price-based demand

response with electric storage space heating,” IEEE Trans Ind. Informat, vol. 11,
no. 1, pp. 281–288, Feb 2015.

[39] V. C. Cunha, R. Torquato, T. R. Ricciardi, W. Freitas, and B. Venkatesh, “Assessing

energy storage potential to facilitate the increased penetration of photovoltaic
generators and electric vehicles in distribution networks,” in 2017 IEEE Power Energy
Society General Meeting, July 2017, pp. 1–5.

[40] R. Li, W. Wang, and M. Xia, “Cooperative planning of active distribution system
with renewable energy sources and energy storage systems,” IEEE Access, vol. 6, pp.
5916–5926, 2018.

[41] S. Wen, H. Lan, Q. Fu, D. C. Yu, and L. Zhang, “Economic allocation for energy
storage system considering wind power distribution,” IEEE Trans. Power Syst.,
vol. 30, no. 2, pp. 644–652, 2015.

[42] M. Sedghi, A. Ahmadian, and M. Aliakbar-Golkar, “Optimal storage planning

in active distribution network considering uncertainty of wind power distributed
generation,” IEEE Trans. Power Syst., vol. 31, no. 1, pp. 304–316, 2016.

[43] A. S. A. Awad, T. H. M. EL-Fouly, and M. M. A. Salama, “Optimal ess allocation for

load management application,” IEEE Trans. Power Syst., vol. 30, no. 1, pp. 327–336,
2015.

125
[44] H. Hamidpour, J. Aghaei, S. Pirouzi, S. Dehghan, and T. Niknam, “Flexible, reliable,
and renewable power system resource expansion planning considering energy storage
systems and demand response programs,” IET Renewable Power Generation, vol. 13,
no. 11, pp. 1862–1872, 2019.

[45] H. Akhavan-Hejazi and H. Mohsenian-Rad, “Energy storage planning in active

distribution grids: A chance-constrained optimization with non-parametric probability
functions,” IEEE Trans. Smart Grid, vol. 9, no. 3, pp. 1972–1985, May 2018.

[46] F. Luo, K. Meng, Z. Y. Dong, Y. Zheng, Y. Chen, and K. P. Wong, “Coordinated

operational planning for wind farm with battery energy storage system,” IEEE Trans.
Sustainable Energy, vol. 6, no. 1, pp. 253–262, 2015.

[47] T. Qiu, B. Xu, Y. Wang, Y. Dvorkin, and D. S. Kirschen, “Stochastic multistage

coplanning of transmission expansion and energy storage,” IEEE Trans. Power Syst.,
vol. 32, no. 1, pp. 643–651, Jan 2017.

[48] H. Alharbi and K. Bhattacharya, “Stochastic optimal planning of battery energy

storage systems for isolated microgrids,” IEEE Trans. Sustain. Energy, vol. 9, no. 1,
pp. 211–227, Jan 2018.

[49] I. Alsaidan, A. Khodaei, and W. Gao, “A comprehensive battery energy storage

optimal sizing model for microgrid applications,” IEEE Trans. Power Syst., vol. 33,
no. 4, pp. 3968–3980, 2018.

[50] T. Alharbi, K. Bhattacharya, and M. Kazerani, “Planning and operation of

isolated microgrids based on repurposed electric vehicle batteries,” IEEE Trans. Ind.
Informat., vol. 15, no. 7, pp. 4319–4331, July 2019.

[51] B. Xu, A. Oudalov, A. Ulbig, G. Andersson, and D. S. Kirschen, “Modeling of lithium-

ion battery degradation for cell life assessment,” IEEE Trans. Smart Grid, vol. 9, no. 2,
pp. 1131–1140, March 2018.

[52] Y. Jiang, J. Zhang, L. Xia, and Y. Liu, “State of health estimation for lithium-ion
battery using empirical degradation and error compensation models,” IEEE Access,
vol. 8, pp. 123 858–123 868, 2020.

126
[53] X. Tan, Y. Tan, D. Zhan, Z. Yu, Y. Fan, J. Qiu, and J. Li, “Real-time state-of-health
estimation of lithium-ion batteries based on the equivalent internal resistance,” IEEE
Access, vol. 8, pp. 56 811–56 822, 2020.

[54] G.-w. You, S. Park, and D. Oh, “Real-time state-of-health estimation for electric
vehicle batteries: A data-driven approach,” Applied Energy, vol. 176, pp. 92–103,
2016.

[55] C. O. Adika and L. Wang, “Non-cooperative decentralized charging of homogeneous

households’ batteries in a smart grid,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp.
1855–1863, 2014.

[56] H. M. Soliman and A. Leon-Garcia, “Game-theoretic demand-side management with

storage devices for the future smart grid,” IEEE Trans. Smart Grid, vol. 5, no. 3, pp.
1475–1485, 2014.

[57] W. Tushar, B. Chai, C. Yuen, S. Huang, D. B. Smith, H. V. Poor, and Z. Yang,

“Energy storage sharing in smart grid: A modified auction-based approach,” IEEE
Transactions on Smart Grid, vol. 7, no. 3, pp. 1462–1475, 2016.

[58] C. P. Mediwaththe, M. Shaw, S. Halgamuge, D. B. Smith, and P. Scott, “An incentive-

compatible energy trading framework for neighborhood area networks with shared
energy storage,” IEEE Transactions on Sustainable Energy, vol. 11, no. 1, pp. 467–
476, 2020.

[59] H. Wang and J. Huang, “Incentivizing energy trading for interconnected microgrids,”
IEEE Trans. Smart Grid, vol. 9, no. 4, pp. 2647–2657, 2018.

[60] L. Han, T. Morstyn, and M. McCulloch, “Incentivizing prosumer coalitions with

energy management using cooperative game theory,” IEEE Trans. Power Syst.,
vol. 34, no. 1, pp. 303–313, 2019.

[61] C. Feng, F. Wen, S. You, Z. Li, F. Shahnia, and M. Shahidehpour, “Coalitional game-
based transactive energy management in local energy communities,” IEEE Trans.
Power Syst., vol. 35, no. 3, pp. 1729–1740, 2020.

127
[62] P. Hu, R. Karki, and R. Billinton, “Reliability evaluation of generating systems
containing wind power and energy storage,” IET Generation, Transmission
Distribution, vol. 3, no. 8, pp. 783–791, August 2009.

[63] C. E. P. J. CA3 rdova, B. Asare-Bediako, G. M. A. Vanalme, and W. L. Kling,

“Overview and comparison of leading communication standard technologies for smart
home area networks enabling energy management systems,” in 2011 46th International
Universities’ Power Engineering Conference (UPEC), 2011, pp. 1–6.

[64] O. Alrumayh and K. Bhattacharya, “Model predictive control based home energy
management system in smart grid,” in 2015 IEEE Electrical Power and Energy
Conference (EPEC), Oct 2015, pp. 152–157.

[65] O. Alrumayh, “Optimal coordination of home energy management systems in a

distribution grid,” Master’s thesis, 2016.

[66] S. A. Hashmi, “Evaluation and improvement of the residential energy hub management
system,” Master’s thesis, University of Waterloo, Sept. 2010.

[67] “Ieee guide for the interoperability of energy storage systems integrated with the
electric power infrastructure,” IEEE Std 2030.2-2015, pp. 1–138, June 2015.

[68] R. A. Huggins, Energy Storage. Springer, 2010.

[69] M. Aneke and M. Wang, “Energy storage technologies and real life applications – a
state of the art review,” Applied Energy, vol. 179, pp. 350 – 377, 2016. [Online].
Available: www.sciencedirect.com/science/article/pii/S0306261916308728

[70] C. Amzallag, J. Gerey, J. Robert, and J. Bahuaud, “Standardization of

the rainflow counting method for fatigue analysis,” International Journal
of Fatigue, vol. 16, no. 4, pp. 287 – 293, 1994. [Online]. Available:
www.sciencedirect.com/science/article/pii/0142112394903433

[71] R. Dufo-Lopez, J. L. Bernal-Agustı́n, and J. Contreras, “Optimization of control

strategies for stand-alone renewable energy systems with hydrogen storage,”
Renewable energy, vol. 32, no. 7, pp. 1102–1126, 2007.

128
[72] E. Telaretti, M. Ippolito, and L. Dusonchet, “A simple operating strategy of small-
scale battery energy storages for energy arbitrage under dynamic pricing tariffs,”
Energies, vol. 9, no. 1, p. 12, 2015.

[73] K. Gurney, An introduction to neural networks. CRC press, 1997.

[74] C. M. Bishop et al., Neural networks for pattern recognition. Oxford university press,
1995.

[75] R. Myerson, “Game theory: Analysis of conflict harvard university,” Press,

Cambridge, 1991.

[76] D. Chattopadhyay, “An energy brokerage system with emission trading and allocation
of cost savings,” IEEE Trans. Power Syst., vol. 10, no. 4, pp. 1939–1945, 1995.

[77] S. Hart, “Shapley value,” in Game Theory. Springer, 1989, pp. 210–216.

[78] E. Castillo, A. J. Conejo, P. Pedregal, R. Garcia, and N. Alguacil, Building and solving
mathematical programming models in engineering and science. John Wiley & Sons,
2011, vol. 62.

[79] Ontario Energy Board. Electricity Rates. Accessed: Dec. 15, 2020. [Online]. Available:
www.oeb.ca/rates-and-your-bill/electricity-rates

[80] GAMS Development Corporation. General Algebraic Modeling System (GAMS).

Accessed: Dec. 15, 2020. [Online]. Available: www.gams.com

[81] O. Alrumayh and K. Bhattacharya, “Flexibility of residential loads for demand

response provisions in smart grid,” IEEE Transactions on Smart Grid, vol. 10, no. 6,
pp. 6284–6297, 2019.

[82] H. Yuan, F. Li, Y. Wei, and J. Zhu, “Novel linearized power flow and linearized opf
models for active distribution networks with application in distribution lmp,” IEEE
Trans. Smart Grid, vol. 9, no. 1, pp. 438–448, Jan 2018.

129
[83] Y. Niu and S. Santoso, “Sizing and coordinating fast- and slow-response energy storage
systems to mitigate hourly wind power variations,” IEEE Trans. Smart Grid, vol. 9,
no. 2, pp. 1107–1117, 2018.

[84] B. Vatandoust, A. Ahmadian, M. A. Golkar, A. Elkamel, A. Almansoori, and

M. Ghaljehei, “Risk-averse optimal bidding of electric vehicles and energy storage
aggregator in day-ahead frequency regulation market,” IEEE Trans. Power Syst.,
vol. 34, no. 3, pp. 2036–2047, 2019.

[85] M. H. Beale, M. T. Hagan, and H. B. Demuth, “Neural network toolbox™ user’s

guide,” The MathWorks, 2010.

[86] C. Zhou, K. Qian, M. Allan, and W. Zhou, “Modeling of the cost of ev battery wear
due to v2g application in power systems,” IEEE Transactions on Energy Conversion,
vol. 26, no. 4, pp. 1041–1050, 2011.

[87] N. Padmanabhan, M. Ahmed, and K. Bhattacharya, “Battery energy storage systems

in energy and reserve markets,” IEEE Trans. Power Syst., vol. 35, no. 1, pp. 215–226,
2020.

[88] T. S. B.A. McCarl. Linear Programming Modeling: Nonlinearities and Approximation.

Accessed: Dec. 15, 2020. [Online]. Available: http://agecon2.tamu.edu/people/
faculty/mccarl-bruce/mccspr/new09.pdf

[89] A. E. Roth, The Shapley value: essays in honor of Lloyd S. Shapley. Cambridge
University Press, 1988.

[90] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution systems for loss
reduction and load balancing,” IEEE Trans. Power Del., vol. 4, no. 2, pp. 1401–1407,
1989.

130