Rochester Institute of Technology

RIT Scholar Works

Theses Thesis/Dissertation Collections

7-12-2018

Convex Optimization Approach to the Optimal

Power Flow Problem in DC-Microgrids with
Energy Storage
Abdullah Mohammed Alzahrani
[email protected]

Follow this and additional works at: https://scholarworks.rit.edu/theses

Recommended Citation
Alzahrani, Abdullah Mohammed, "Convex Optimization Approach to the Optimal Power Flow Problem in DC-Microgrids with
Energy Storage" (2018). Thesis. Rochester Institute of Technology. Accessed from

This Thesis is brought to you for free and open access by the Thesis/Dissertation Collections at RIT Scholar Works. It has been accepted for inclusion
in Theses by an authorized administrator of RIT Scholar Works. For more information, please contact [email protected].
Convex Optimization Approach to the
Optimal Power Flow Problem in
DC-Microgrids with Energy Storage
By
Abdullah Mohammed Alzahrani
A Thesis Submitted in Partial Fulfillment of the Requirements for the
Degree of
Master of Science in Electrical Engineering
Department of Electrical and Microelectronic Engineering
Kate Gleason College of Engineering
Rochester Institute of Technology, Rochester, New York
Supervised By
Dr. Juan Carlos Cockburn
Associate Professor, Department of Computer Engineering
Approved By:
Dr. Juan Carlos Cockburn
Primary Adviser, Department of Computer Engineering

Dr. Luis Carlos Herrera

Secondary Adviser, Department of Electrical Engineering

Dr. Santosh Kurinec

Committee Member, Department of Electrical and Microelectronic
Engineering

Dr. Panos P. Markopoulos

Committee Member, Department of Electrical Engineering

July 12, 2018

Copyright Warning

All rights reserved. No part of this publication may be reproduced, distributed, or trans-
mitted in any form or by any means, including photocopying, recording, or other elec-
tronic or mechanical methods, without the prior written permission of the publisher.

Use of this thesis is for the purpose of private study or scholarly research only. Users must
comply with the Copyright Ordinance. Anyone who consults this thesis is understood to
recognize that its copyright rests with its author and that no part of it may be reproduced
without the author’s prior written consent.

1
ACKNOWLEDGMENT

I would like to give my most sincere gratitude to Prof. Juan Carlos Cockburn for his ex-
pertise, understanding, patience, and conscientious and kindly guidance during my MSc
studies. Thank you very much for offering me the chance to study at Rochester Institute
of Technology (RIT), encouraging my research interests and helping me to develop. I ap-
preciate your vast knowledge and skill in many areas, and for taking time to support my
thesis work. Your advice and assistance has been invaluable and of great help in getting
everything done in a very timely manner.

A very special thanks goes out to Prof. Luis Carlos Herrera for giving me precious advice
in my research. Your guidance has been extremely valuable for me and will still be helpful
for my future career and life. I sincerely appreciate your efforts sharing your knowledge
and experience during the research of my thesis.

I must also acknowledge Prof. Santosh Kurinec and Prof. Panos P. Markopoulos. Thanks
for your kind help and support during my MSc studies at RIT.

I would also like to thank my family for the support they provided me through my entire
life and during my thesis studies in particular. I would like to take this opportunity to
thank my friends for their kind help during my MSc studies. Thanks all of you for bringing
me the gorgeous studying atmosphere and assisting me through those extremely tough
days.

I would like to thank all the professors and staff members of the Department of Electrical

2
Engineering, RIT for their extended cooperation and guidance. I also take this opportunity
to give thanks to all others who have given me support for the thesis or in other aspects of
my study at RIT.

3
 Abstract

Humanity is currently facing a global energy crisis. This is due to the shortage in the con-
ventional energy resources while the demand for energy is rising. In response to this cri-
sis, research in designing more energy efficient systems has gained significant importance.
The Microgrids (MGs) are one of the main key elements in giving significant momentum
to efficient decentralized energy generation.

From the perspective of MGs power management, economical scheduling for generators,
energy storage, and demand loads are critical. Performance optimization processes are
needed to minimize the operating costs while considering operational constraints.

In this thesis, the optimal power flow problem for managing energy sources with storage
devices is presented for dc microgrids. The power management model has been exam-
ined in various scenarios. One of them is based on a network of a six-bus power system,
including an energy storage device coupling at a certain bus. The other scenario is based
on the same model but including more energy storage devices.

After analyzing the results of these scenarios, several conclusions have been made such
as when the energy storage should charge/discharge to minimize costs. The study shows
the feasibility of optimal power flow operation in DC microgrids.
Contents

1 Introduction 10

1.1 Energy Storage Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.2 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 AC Optimal Power Flow Problem 16

2.1 Formulation of the Optimal Power Flow Problem . . . . . . . . . . . . . . . . 17

2.1.1 General Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1.2 Objective Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.1.3 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.1.4 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.1.5 Standard Optimal Power Flow (OPF) Problem . . . . . . . . . . . . . 20

2.2 Hardness of Optimal Power Flow (OPF) problem . . . . . . . . . . . . . . . . 22

2.2.1 Active Constraints: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.2.2 Non-Convexity: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.2.3 Switching Behavior: . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.3 Techniques to solve Optimal Power Flow (OPF) problem . . . . . . . . . . . 23

1
 2.3.1 Optimization Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.3.2 Traditional Optimization Methods . . . . . . . . . . . . . . . . . . . . 24

2.3.3 Convex Relaxation for Optimal Power Flow (OPF) problem . . . . . . 27

3 Optimal Power Flow without Energy Storage 33

3.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.2 Mathematical Model and Problem Formulation . . . . . . . . . . . . . . . . . 34

3.2.1 Analysis of Optimal Power Flow without Energy Storage . . . . . . 34

3.3 Demonstration and Evaluation of Proposed Optimal Power Flow (OPF) Method 37

3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.3.2 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.3.3 Solution to Nonlinear OPF . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.3.4 Solution to Convex Relaxation of OPF . . . . . . . . . . . . . . . . . . 40

3.3.5 Result Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4 Optimal Power Flow with Energy Storage 44

4.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.2 The Optimal Power Flow with Energy Storage . . . . . . . . . . . . . . . . . 45

4.2.1 Optimal Power Flow with Line Constraints . . . . . . . . . . . . . . . 48

4.2.2 Modeling the Energy Storage . . . . . . . . . . . . . . . . . . . . . . . 50

5 Demonstration and Evaluation of Proposed OPF Method with Energy Storage 52

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.2 CASE STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2
 5.2.1 Test Case of DC Microgrid System . . . . . . . . . . . . . . . . . . . . 52

5.2.2 Scenario-I, Parameters and Results . . . . . . . . . . . . . . . . . . . . 53

5.2.3 Demonstration of Scenario-I . . . . . . . . . . . . . . . . . . . . . . . . 54

5.2.4 System Load Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5.2.5 System Generation Data . . . . . . . . . . . . . . . . . . . . . . . . . . 58

5.2.6 State-of-Charge (SOC) . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.2.7 Parameters and Results of Scenario-II . . . . . . . . . . . . . . . . . . 61

5.2.8 Demonstration of Scenario-II . . . . . . . . . . . . . . . . . . . . . . . 62

5.2.9 System Load Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.2.10 System Generation Data . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5.2.11 State-of-Charge (SOC) . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.2.12 Analysis of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

6 Conclusion and Future Work 71

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

7 APPENDIX 73

7.1 Appendix-I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

7.2 Appendix-II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3
List of Figures

1.1 (a) Energy losses in traditional power grid . . . . . . . . . . . . . . . . . . . . 11

1.2 (b) Concept of new power generation systems . . . . . . . . . . . . . . . . . . 11

3.1 Summary of notation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.2 3-bus of Power system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.3 6-bus of Power System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Scenario I: A power system with energy storage at bus six. . . . . . . . . . . 54

5.2 The actual power loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5.3 The generator schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

5.4 The battery schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.5 The State-of-charge schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.6 Scenario II: A power system with two energy storages. . . . . . . . . . . . . . 63

5.7 The actual power loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.8 The generator schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5.9 The battery schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.10 The State-of-charge schedule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4
5.11 The flowchart of the convexification process . . . . . . . . . . . . . . . . . . . 70

5
List of Tables

3.1 DC Microgrid Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 Conductances Between Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.3 Matlab-Nonlinear programming solver Result . . . . . . . . . . . . . . . . . 40

3.4 CVX Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

5.1 The bound constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.2 The power capacity constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.3 The result of Scenario-I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5.4 The Bound Constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.5 The power capacity constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.6 The result of Scenario-II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

7.1 Power Load for Scenario-I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

7.2 Power Generation for Scenario-I. . . . . . . . . . . . . . . . . . . . . . . . . . 73

7.3 Power Load and rate of charge for Scenario-I. . . . . . . . . . . . . . . . . . . 74

7.4 Power Load for Scenario-II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

7.5 Power Generation for Scenario-II. . . . . . . . . . . . . . . . . . . . . . . . . . 75

6
7.6 Power Load and rate of charge for Scenario-II. . . . . . . . . . . . . . . . . . 76

7
Nomenclature

N The set of buses

G The set of generation buses.
E The set of transmission lines.
n∼ j Buses n and j connected by a line.
H Denotes hermitian transpose.
G The graph contains a set N of nodes and a set E of edges.
C The set of Complex numbers.
δ Voltage phase angles.
|V | Voltage magnitudes.
Pn Active power at bus n.
Qn Reactive power at bus n.
Vn Complex voltage at bus n.
yn j Admittance of line joining buses n and j.
Sn The net power injection at bus n.
In The current at bus n.
x The vector of state variables.
u The vector of control variables.
0 Positive Semidefinite
SDP Semidefinite Program.

8
SOCP Second-Order Cone Program.
LP Linear programming.
QP Quadratic programming.
NLP Nonlinear programming.
IPM Interior point method.
SLP Sequential Linear Programming.
SM Simplex Method.
LIM Lambda-iterative Method.
SQP Successive Quadratic Programming.
GT Gradient Techniques.
NT Newton’s Techniques.
KKT Karush–Kuhn–Tucker.

9
1 | Introduction

n recent decades, electrical power systems have been significantly changing whereas
I fixed power generation has been moving towards more distributed generation. The
increase in power demand and environmental concerns about conventional power gen-
eration are the main reasons for this transformation. An additional critical stimulus is
the substantial amount of energy losses in conventional methods. When power is pro-
duced from fossil fuels such as coal, hydrocarbons, fuel oil or natural gas, 40%–70% of the
energy in the current resources is wasted as heat. Additionally, 2% and 4% is lost in trans-
mission lines and distribution, respectively. Generally, as shown in Fig. 1.2.a, only 33% of
the energy input to generation makes it to the user as electricity. Nowadays, distributed
generation systems can be placed beside the load to utilize lower capacity of power gen-
eration resources, such as renewable energy sources, as seen in Fig. 1.2.b. Solar and wind
energy are the most popular types of renewable energy sources that can be integrated into
modern power systems and microgrids (MGs) [1, 2], and [3]. Microgrids consist of a col-
lection of such distributed generation systems coordinated with each other in a way that
increases the capacity of the system and improves the power quality [4, 5, 6, 7], and [8].

10
 Figure 1.1: (a) Energy losses in traditional power grid

Figure 1.2: (b) Concept of new power generation systems

11
There are three main advantages associated with microgrids, which can be classified as:
technical, economical, and environmental. From a technical perspective, distributed gen-
eration systems can support the energy of remote places and provide very high efficiency
while lacking the fixed source vulnerability of large-scale networks, which is a factor in
the failure of power systems [9], and [8]. From the economic point of view and according
to market research studies, distributed generation minimizes line losses and interruption
costs for consumers while reducing fuel cost [10]. The environmental advantages of mi-
crogrid systems are investigated in [11] and include reducing the emissions of pollutants
and greenhouse gases, smaller physical footprint, and easier integration of clean energy
sources.

The general configuration of a DC microgrid is shown in Fig. 1.2). Indeed, DC micro-

grids have several advantages over their AC counterparts such as improving the energy
efficiency and the performance of power controllers and energy management schemes for
regulating power flow [12, 13], and [14]. In addition, some devices such as wind turbines,
electric vehicles, and electronic appliances are more easily integrated into DC than AC
networks since these devices are already either DC in nature or have a different natural
frequency than the primary grid. DC microgrid systems are very durable and efficient in
handling voltage sags and frequency variations in the primary grid as compared with AC
systems, since DC microgrids do not need frequency synchronization.

In microgrids, energy sources may be controlled by a power management system to op-

timize the power flow within the network. The objective of power management depends
on the mode of operation such as grid connected or island modes. The main goals of
the power management in grid connected mode is to stabilize the system and reduce the
electricity price in the main grid. In island mode, the main objective is to optimize the
microgrid operation by minimizing the cost of local generation [15], and [16].

Increasing the efficiency of power and the maximum utilization of renewable energy is

12
fundamental in modern power systems. Energy storage can thus be an excellent technol-
ogy for this purpose [17], and [18]. When microgrids are connected to the main grid with
renewable resources, the energy storage guarantees none of the energy generated is lost.
For instance, if a renewable energy system such as wind is integrated to a DC microgrid
without energy storage devices, the surplus power produced will be wasted if the load is
not high enough. Only through the assistance of a dynamic energy storage system can util-
ities or system owners ensure they can store that energy for future use. Thus, microgrids
with energy storage are cost effective solutions to lower overall energy costs and improve
the grid’s efficiency.

1.1 Energy Storage Systems

Studying energy storage systems from a power management perspective becomes signif-
icantly important for several reasons. Some of these main aspects are sizing of the energy
storage, choice of the type of energy storage (e.g. lithium ion, flywheels, etc.) and optimal
charging/discharging strategies.

One of the earliest types of energy storages is hydro pumps, where the power is stored in
the form of water in a reservoir, whereas batteries store electrical energy in their chemistry.
Moreover, energy storage is a significant microgrid component which helps the main grid
in several ways such as ancillary services including load following, operational reserve,
renewable integration and relieving congestion and constraints. Indeed, energy storage
units must have the capability to respond sufficiently rapidly to transient power changes
in grid-connected mode and to keep themselves fully charged. There are also many other
applications that can utilize energy storage devices in microgrids, including consumers
energy management (minimize cost) and integration of renewable energy sources.

The control of such energy storage is crucial and implementing a three-level hierarchical

13
control system can be used for optimum performance. For microgrids and energy storage
device systems, hierarchical control can consists of three levels beginning with “primary”
(for imitating physical behaviors to ensure system stability), “secondary” (for synchroniz-
ing control loops to smoothly connect or disconnect the microgrid from the distributed
generation system), and “tertiary” (for controlling the power flow between the microgrid
and the main grid) control levels. In this study, we focus on the tertiary control level,
which relates to the control of power/energy of the microgrid resources [19].

In particular, tertiary control can be achieved solving an optimal power flow (OPF) prob-
lem: a type of optimization process with the goal of minimizing an objective function
(e.g. power generation cost, losses, etc.), subject to network and physical constraints [20],
and [21]. However, there are significant challenges in solving OPF problems due to its
nonlinearity, non-convexity, and large scales [22]. Several numerical solution techniques
are available for solving this nonlinear optimization problem. Although these techniques
have been found to have capability and reliability limitations [23, 24, 25, 26, 27], and [28].
In the past decade, convex and conic relaxations have been proposed to overcome these
limitations in particular for AC microgrids [29, 30], and [31].

Recently, semidefinite programming (SDP) techniques have been employed to solve the
OPF problem, as described [32]. This approach is advantageous compared to the other
techniques as it is convex, and easily available solvers exist for problems of moderate/large
size. In some cases, the solution is globally optimal [33, 34]. The work of [35] has shown
proof of concept using convex programming techniques, although the work only covers a
few simple, limited systems [36].

14
1.2 Thesis Outline

Following the introduction in Chapter 1, Chapter 2 introduces the OPF problem. We be-
gin by discussing the formulation and defining the general structure of an OPF problem,
objective functions, state and control variables, and constraints. Thereafter, we present the
difficulty of the OPF problem and the techniques to address it that are of common inter-
est in towards solving a global optimization problem. In Chapter 3, we formulate the OPF
without energy storage. In Chapter 4, we include energy storage into the OPF formulation.
In Chapter 5, we test the apply the semidefinite (SDP) relaxation technique on standard
test networks for the problem using a strategy based on a convex semi-definite program
for the OPF problem with energy storage. Then we conclude the thesis and explore a
future work in Chapter 6.

15
2 | AC Optimal Power Flow Problem

AC Power flow (PF) refers to the power network solution that contains complex voltages,
active and reactive power, and complex currents at every bus in the specific power system
[37]. In the AC PF problem, the relations between voltage/current and active/reactive
power is nonlinear. Hence, obtaining a feasible set of solutions can be challenging due to
this nonlinearity.

Moreover, the Optimal Power Flow (OPF) problem is an optimization problem with the
main goal of minimizing the total cost of generation while at the same time meeting the
power flow relationships at every bus. Hence, this problem is nonlinear and non-convex
which complicates its solution. Improvement and redesign of power networks has in-
creased the network complexity. For example, the onset of independent power and the
development and production renewable generation in smart grids further extended the
scope of OPF. Several contributions have been made to reformulate and improve the OPF
formulation to be more efficient to handle the non-convexity problem under various hy-
potheses [38].

OPF problems play an essential role in power network operations and design. OPF can
minimize power generation cost and losses, subject to network and physical constraints
[20], and [21]. This optimization is indispensable in power network operations lasting
minutes, hours, or even up to one day. The formulation of the OPF provides the optimal
electrical response of the power transmission system to a specific set of power injections

16
and loads.

2.1 Formulation of the Optimal Power Flow Problem

As mentioned previously, the Optimal Power Flow (OPF) problem involves minimizing
the total cost of generation while ensuring that the power flow equations are met at every
bus. In addition, generation, voltage, and other constraints can be imposed.

In this chapter, a detailed overview of the OPF problem will be presented. This formula-
tion will be used and expanded in the next subsections to describe algorithms that will be
used for solving the OPF problem.

2.1.1 General Structure

The standard optimal power flow problem can be formulated as follows [38, 39], and [40]:

m
Õ
min fn (x, u) (2.1)
x,u
n=1
subj. to (2.2)

g(x, u) = 0 (2.3)

h(x, u)− ≤ h(x, u) ≤ h(x, u)+ (2.4)

Where, fn (x, u) are cost functions, x is a vector of state variables, and u a vector of control
variables, which are usually the independent variables in an OPF. The equality constraints
(2.3) typically include the power flow equations at every bus. The inequality constraints
(2.4) physical bounds on the state and control variables.

17
2.1.2 Objective Function

The most common use of the OPF objective function is to reduce the total cost of the gener-
ated power, occasionally taking an account active transmission losses in the whole or part
of the power network as well. For the convergence, the generation cost functions are fre-
quently approximated by either linear or quadratic functions [41]. In some situations, the
cost is a function of both the generation active and reactive power in the network. Other
types of costs which can be included in the objective function include power transfer abil-
ity, investment cost, voltage profile, load shedding, environment impact, etc [38].

2.1.3 Variables

Analysis of the OPF equations reveals two types of variables: state variables and control
variables, also known as dependent and decision variables, respectively. In general, the
vector states x in the OPF problems describes the electrical state of the network. These
can be represented as voltage phase angles, voltage magnitudes, active power output of
the slack bus only, and reactive power of all generations at each bus, including the power
loads and line flows. These state variables can be represented as continuous in nature. On
the other hand, the control variables u in the OPF problems normally contain a subset of
the state variables and also variables describing control appliance adjustments, such as the
active power at the generator buses tap changes in transformer, position of the phase shift-
ing taps, and the status of the switched capacitors and reactors (switchable VAR sources).
These control variables can be represented as continuous or discrete. The selections of
the state variables x are imposed by the structure of PF equations utilized, whereas con-
trol variables vary among OPF formulations depending on the nature of the particular
problem [42], and [43].

18
2.1.4 Constraints

Constraints are normally considered an integral part of a specific optimization problem

playing a crucial role in its categorization, for example, convex vs non-convex. In the OPF
problem, the network variables have to be within allowable bounds or else damage the
electrical power network devices or cause a malfunction. Typically, the constraints are
classified as equality and inequality constraints [42], and [43].

In OPF problems, equality constraints are normally represented by the power flow system
equations [44]. Inequality constraints identify upper and lower bounds on the control
and state variables of the devices of the electrical power network and also the bounds
required to ensure network security and stability [40]. The inequality constraints on the
state variables and controls can be partitioned as follows:

• State Variables Bounds on

– Voltage Magnitude.

– Voltage Angle.

– Power Line.

• Control Variables Bounds on

– Active Power Generation.

– Reactive Power Generation.

– Volt Ampere Reactive Power.

– Voltage Generation.

– Transformer Tap Position.

Moreover, the inequality constraints can also define forbidden areas of branch flow lim-
its, interface limits, active/reactive power reserve limits, spinning reserve requirements,

19
transient security, transient stability, transient contingencies, environment constraints, ro-
tor angle stability, etc. [45].

2.1.5 Standard Optimal Power Flow (OPF) Problem

In this subsection, we present two mathematical models of the power system networks
and show their equivalence. Let the set of buses be N:={1, 2, 3, .., n}, the set of transmission
lines E ⊆ {N × N} of edges. Then n ∼ j denotes a line between the buses n and j.

2.1.5.1 Bus Injection Model (BIM):

The general bus injection model (BIM) is defined by the following power flow equations
that represent Kirchhoff’s laws at each bus

Õ
Sn = Vn (VnH − VjH )ynHj , ∀n ∈ N (2.5)
j:n∼ j

the superscript H denotes hermitian transpose. Let the set the solutions of the power flow
V for each S to be:
V := {V ∈ C(n+1) | V satisfies (2.5)} (2.6)

The buses in the power networks can be divided into three classes: slack, generation, and
load buses. The next quantities are generally identified at each bus: voltage phase angles
δ, voltage magnitudes |V |, active power output of the slack bus only Ps , and reactive power
of all generations Q n . Usually the voltage magnitudes and the angles are described by one
complex variable V rather than real and imaginary variables. Let us now consider the gen-
eral power network modeled by the subset of buses N:={1, 2, 3, .., n}, the set of transmission
lines E ⊆ {N × N} of edges, and the set of generation buses G ⊂ N.

• Pnd + iQ nd : Complex power of the load connected to bus n, ∀n ∈ N.

20
 g g
• Pn + iQ n : Complex power output of the generator connected to bus n, ∀n ∈ N

• Vn : Complex voltage at bus n, ∀n ∈ N.

• yn j : Admittance of the transmission line (n, j) ∈ E, yn j =gn j + ibn j .

Define V, P g , Q g , P d and Q d as the vectors formed stacking, in lexicographical order the

g g
values at all the buses of the corresponding variables Vn , Pn , Q n , Pnd , and Q nd ∀n ∈ N,
respectively. Assume fn (·) is a convex function representing the cost associated with bus
n ∈ N. The objective is to minimize the total cost of the system. This problem can then be
formulated as follows:

Õ
g
min
g
f = fn (Pn ) (2.7)
Pn ,Vn
n∈N
subj. to (2.8)

|Vn− | ≤ Vn ≤ |Vn+ |, ∀n ∈ N (2.9)

g− g g+
Pn ≤ Pn ≤ Pn , ∀n ∈ N (2.10)

Q− +
n ≤ Qn ≤ Qn , ∀n ∈ N (2.11)
n∈N
Õ  
g g
Pn + Q n = Pn + Q n + Vn
d d
Vn − Vj yn j , ∀n ∈ N
H
(2.12)
j:n∼ j

|Vn − Vj | ≤ ∆Vn+j , ∀(n, j) ∈ E (2.13)

g g
In this formulation Pn and Q n are controllable variables, and Pnd and Q nd are given external
g g
disturbances. The solution will give values of Pn and Q n , at every bus, that minimize the
cost and satisfy the physical and operational constraints of the network. In the remainder
g
of this work we specialize to DC power systems where Q n = Q nd = 0 and the constraints are
Pn− , Pn+ , Pnd , Vn− .

21
2.2 Hardness of Optimal Power Flow (OPF) problem

There are three main challenges that must be faced in the numerical solution of OPF prob-
lems: 1) the presence of constraints, 2) the nonlinear nature of constraints and 3) switching
elements and scale of the system.

2.2.1 Active Constraints:

Given a feasible point, the inequality requirements satisfied at this point with strict im-
balances are called latent or inactive, while those limitations satisfied with equality are
alluded to as binding or active constraints. Finding the dynamic inequality limitations
has a combinatorial viewpoint, and it is a troublesome piece of tackling the OPF prob-
lem. Indirect approaches exist for solving OPF without utilizing a middle optimization
approach to obtain the active sets at an optimal solution [46].

2.2.2 Non-Convexity:

Despite the fact that some OPF problems can be formulated with linear objective func-
tion, the problem is nonlinear due to the power flow equations being nonlinear as shown
in (2.12). Subsequently, the non-convex feasible sets may even be unconnected. Therefore,
the Karush–Kuhn–Tucker (KKT) conditions are not usually adequate for a global opti-
mum. Specifically, for AC power systems, the OPF problem is naturally non-convex and
allows for many local optimal solutions. Correspondingly, existing solution approaches
utilized extensively in practice depend on repeated optimization approaches, which can
only return local optimal solutions with the achievement of a global optimal solution. In
summary, the non-convexity of the OPF problem broadly prohibits obtaining a solution
in polynomial time, which makes OPF NP-hard to achieve [12].

22
2.2.3 Switching Behavior:

Tackling dual and integer variables that arise when representing node levels switch con-
trols in practical power systems introduces a significant level of difficulty due to the fact
the OPF becomes an integer programming problem. In addition increasing power net-
work size increases the computational complexity significantly [34].

2.3 Techniques to solve Optimal Power Flow (OPF) problem

2.3.1 Optimization Techniques

Many different approaches have been proposed to solve OPF problem, including linear
programming (LP), quadratic programming (QP), nonlinear programming (NLP), and in-
terior point methods (IPM). However, these approaches have not been widely used in in-
dustry according to [47]. As is to be expected there is no single approach that can be used
for all OPF problems, even though a few approaches might be more successful than others
in certain OPF problems. Some of the figures of merit used to compare algorithms uti-
lized in OPF problems, reported in [48], and [49] include accuracy, computational speed,
robustness, flexibility, scalability, solution quality, and rate of convergence. It is very chal-
lenging for a single algorithm to hold all these characteristics. However, solution quality,
robustness, rate of convergence, reliability, and scalability are of primary importance in se-
lecting an OPF optimization approach [49]. In this subsection, some of these approaches
are described in detail.

23
2.3.2 Traditional Optimization Methods

Examples of traditional optimization algorithms for solving the OPF problem include:
Newton-like method, sequential linear programming (SLP), non-linear programming (NLP),
simplex method (SM), lambda-iterative methods (LIM), successive quadratic program-
ming (SQP), and gradient-based methods [47], and [50]. The simplex approach is appro-
priate for LP-based OPF problems and can be directly applied to the DC-OPF formulation
[27], and [51]. SLP is an extension of LP described by [52] that allows for optimizing prob-
lems with nonlinear characteristics by a sequence of linear approximations. In certain
cases, a NLP formulation can be reduced to an LP by utilizing linear approximations of
the objective function and constraints around an initial estimate of the optimal solution
[23, 53], and [54]. SQP is a solution method for NLP problems, and similar to SLP, it solves
the original problem by solving a sequence of QP problems, of which solutions converge
to a global optimal solution of the original problem [55]. In most SQP implementations
for an OPF problem, conventional power flow equations are linearized at each iteration,
which can increment the computational effectiveness at the cost of an increase in the num-
ber iterations. Many studies have investigated the implementation of SQP for solving OPF
problems [56, 57], and [58]. The following technique centers on directly solving the NLP
problem rather than solving a sequence of LP or QP approximation problems.

2.3.2.1 Gradient Techniques (GT):

This approche was among the first efforts in the late 1960s to find solution to OPF prob-
lems. Gradient techniques can be split into three essential paths of exploration: reduced
gradient techniques (RG), conjugate gradient techniques (CG), and generalized reduced
gradient techniques (GRG). Gradient technique generally utilize the first-order derivative
vector O f (x k ) of the objective function at the first iterate x k to decide an enhancing heading.
Gradient technique are not difficult to ensure convergence of correct functions. However,

24
gradient technique is slow, i.e., requests additional repetition contrasted to higher-order
technique. Furthermore, since it does not use the second-order derivative, it may converge
to a local optimal point. Global optimality must be confirmed for nonconvex problems,
which eliminate most OPF formulations [20], and [12]. The RG technique has been used
to find a solution to the OPF problem by [59]. The CG technique is essentially an enhance-
ment and a change of the RG technique and is a standout amongst the most understood
iterative techniques for taking care of NLP problems with sparse systems of linear equa-
tions. Rather than utilizing the negative reduced gradient as the heading of descent, the
CG technique selects the descent trends such that it is conjugate to the past search trends
by including the present negative gradient vector to a scaled, linear combination of past
search trends. There are a few focal points of applying CG technique for resolving an OPF
problem, especially the enhancement and change of the characteristics over the RG tech-
nique [60]. The GRG technique is an augmentation and an expansion of the RG technique,
which empowers coordinate treatment of inequality and nonlinear constraints. Instead of
utilizing penalty functions, the GRG technique modifies the iterations by inserting slack
variables with all inequality limitations and all constraints are linearized about the present
working point. In this manner, the first problem is moved into a progression of subprob-
lems with linear constraints that can be determined by RG or CG techniques [61]. Be that as
it may, since linearization presents a hitch in the constraints, an extra advance is required
to adjust the variables toward the finish of every repetition with emphasis to recuperate
feasibility. In the OPF, this feasibility recuperation is performed by explaining a regular
power flow [62].

2.3.2.2 Newton’s Techniques (NT):

Newton’s technique (NT) is essentially a second-order method for unconstrained opti-

mization based on Taylor series expansion. The descent trend at a point x k is set to dk =
−H(x k )1 O f (x k ) , where H(x k ) indicates the Hessian matrix of f at point x k . Then the tech-

25
nique figures out a step size αk > 0 in trend dk satisfying confirmed step-size choosing
principles such as inexact line seeking with fall back. However, the technique does not
ensure to converge to a local optimum solution as the Hessian matrix H may not be pos-
itive semidefinite in an adequately large neighborhood of the lower or minimum point.
The NT requires the application of Lagrangian function when utilized to OPF problems.
Inequality constraints produced from the power system physical edges must either be ne-
glected or handled as equality, relying on whether they are bounded at either the optimal
solution or not. However, recognizing active inequality constraints is the main challenge
for Newton-based solutions for OPF problem, as mentioned in Section 2.2. In one study,
[63] presented a Newton-based OPF technique and another investigation by [64] discussed
a more effective approach applying the Lagrangian function. Thus researchers have made
momentous contributions to improve performance of power active inequality constraints
[65].

2.3.2.3 Interior Point Methods (IPM):

The challenge of implementing inequality constraints was a motivating factor for employ-
ing interior point methods (IPMs) to solve OPF problems. IPMs are a family of projec-
tive scaling approaches for solving linear and nonlinear optimization problems. IPMs
are used to define and follow a central path through the feasible district to the set of op-
timal solutions. The main point of feasibility enforcement is achieved either by utiliz-
ing partition terms in the augmented objective function or by manipulating the required
KKT conditions [66]. When utilized to solve OPF problems, IPMs have been developed
over the past decades. The most well-known techniques are the primal-dual interior-
point method (PDIPM) [67], Mehrotra’s predictor-corrector techniques [68], Gondzio’s
multiple-centrality corrections [69], and trust district techniques [70]. Among various
IPMs, PDIPMs are probably the most popular deterministic approaches studied in OPF
research. Granville [71] was the first to employ PDIPMs to the OPF problem, extending

26
first PDIPMs for LP and QP to the NLP case of reactive power dispatch. The main features
of Newton’s technique contain no requirement to define the active constraint set or have
an initial feasible solution. Several types of research on IPMs have focused on enhancing
PDIPMs by taking advantage of the framework of power flow constraints. For instance,
[72] suggested a changed PDIPM which utilizes a feature function to promote the conver-
gence properties for OPF. A simplified OPF formulation utilizing rectangular coordinates
and present mismatches described by [73] leads to a facilitation of Hessian matrix and can
reduce computational effort.

2.3.3 Convex Relaxation for Optimal Power Flow (OPF) problem

The traditional techniques for solving the non-linear and non-convex OPF problems have
been described in subsection 2.3.2. Some of these techniques rely on KKT conditions,
which can only assure a local optimal solution because of the non-convexity of the OPF
problem. Furthermore, the characteristics of the nonlinearity and the non-convexity of the
problem make the OPF problem NP-hard to solve. Taking into account that the OPF for-
mulation has been specified in subsection 2.1.5, the non-convexity of the problem dictates
power flow equation constraints (2.12). For each connection line (n, j) ∈ E, let yn j = an j +
ibn j indicate its admittance, and Y ∈ CN×N indicate the admittance matrix of the power
network, i.e.,


re f
−yn j ,

i f n , j and (n, j) ∈ E






(2.14)


Yn j := Íl∈N(n) ynl, i f n = j




Otherwise

 0,




where N(n) is known as set of buses at certain bus. Let In indicates the current of bus n,
and I := [I1 I2 ...IN ]T indicate the current vector of all buses, which can be written as YV,

27
and Sn = Pn + iQ n indicate the net power injection of bus n, i.e.,

Sn = Pn + iQ n = (PG n − PDn ) + i(QG n − Q Dn ) (2.15)

The state of each bus can be written as:

• Current balance

Õ
In = yn j (VN − Vj ), ∀n ∈ N (2.16)
j∈N(n)

• Power balance:

Sn = Vn In∗, ∀n ∈ N (2.17)

Define e1 , e2 ,..., e N as the standard basis vectors in R N such that (en ) j = 0 if n , j, (en ) j = 1
if n = j. Therefore,

Sn = Pn +iQ n = (PG n −PDn )+i(QG n −Q Dn ) = Vn In∗ = (en∗V)(en∗YV)∗ = en∗VV ∗Y ∗ en = T r(VV ∗Y ∗ en en∗ )

(2.18)

1. Semidefinite Relaxation (SDP):

As mentioned earlier, the nonlinearity and non-convexity of the OPF problem make
it NP-hard. Thus, let the non-linear term VV ∗ in equality constraint (2.3) be changed
by a new matrix variable M ∈ H N×N to define this constraint as a linear term and
to be convex, where H N×N indicates the set of N × N of Hermitian matrices. At the
same time, in order to make sure that the map from V to M is invertible, M must
be constrained to be both positive semidefinite and rank-one. The feasible set of
main OPF problem in subsection 2.1.5 can be equivalently formulated as follows (
for clarity, the transmission line constraints have been emitted but will be considered

28
later):

g− g g+
Pn ≤ Pn ≤ Pn , ∀n ∈ N (2.19)
g− g g+
Q n ≤ Q n ≤ Q n , ∀n ∈ N (2.20)

(Vn− )2 ≤ Mnn ≤ (Vn+ )2, ∀n ∈ N (2.21)

T R(MY ∗ en en∗ ) = (PG n − PDn ) + i(QG n − Q Dn ), ∀n ∈ N (2.22)

M = M∗ (2.23)

Rank(M) = 1 (2.24)

However, the rank constraint (2.24) makes OPF formulation non-convex. Eliminat-
ing this rank constraint from the optimization problem creates the SDP relaxation,
which is defined as a convex problem unless the objective function is a non-convex
function. SDP relaxation can be solved effectively and easily in polynomial time.
However, the main challenge is qualifying the solution for the relaxation to match
the optimal solution of the original non-convex problem. In some scenarios the re-
laxation is strict. In [34], it has been proved that the dual of the original OPF problem
is the same as the dual of the SDP relaxation; under certain circumstances, robust
duality holds between the SDP relaxation and the dual of the original OPF problem
with quadratic generation cost. Generally, the dual optimal value is only a lower
limit on the optimal objective value of the OPF problem and the lower limit may not
be strict in the presence of a nonzero duality gap. Moreover, it has been shown that
the global optimal solution to the OPF problem can be restored from the optimal so-
lution to the convex dual problem whenever the duality gap between the dual and
the original OPF problem is 0. In this scenario, the SDP relaxation is strict, the OPF
problem can be obtained either by utilizing SDP or dual relaxation. Reference, [34]
proposed solving the dual of the OPF problem instead of the primal SDP relaxation
as the number of variables in the primal SDP relaxation increases quadratically with

29
 the number of nodes in the network, while the number of variables in the dual prob-
lem grows linearly with the number of nodes. This uniqueness is especially valuable
when utilizing primal and dual interior-point algorithms for solving the primal SDP
and dual relaxations, respectively [34]. In their approach, a necessary and adequate
situation is provided to assure the existence of zero duality gap for the OPF prob-
lem, which is created for the standard IEEE benchmark systems with 14, 30, 57, 118,
and 300 buses including various randomly generated networks. Taking into con-
sideration the hardness of demonstrating this situation, another approach has been
studied to assure the existence of zero duality gap for the IEEE model. In this situ-
ation, a small perturbation is inserted into the admittance matrix that holds well in
practice. In addition, the duality gap is predicted to be zero for a large bus of power
systems due to the passivity of transmission lines and transformers [34]. More re-
cent approaches have been proposed with other sources of non-convexity in OPF,
such as variable transformer ratios, variable shunt elements and contingency con-
straints [74]. In addition to quadratic cost functions, the previous work on OPF with
arbitrary convex cost functions have been developed by [28]. In [28], they found that
because of the physics of the power network, every OPF problem can be effectively
solved in polynomial time after applying the following approximations:

• Write power balance equations as inequalities, and

• Place all equalities and the unknown parameters of the system.

2. Second-Order Cone Relaxation (SOCP):

SOCP relaxation a linear function is minimized over the intersection of an affine

set and the product of second-Order cones. SOCP relaxation is nonlinear convex
problem that involve linear and quadratic program as special cases. However, SOCP
relaxations have a much lower computational complexity than the SDP relaxations.
The main challenge of solving the OPF problem comes from the nonlinear equality

30
constraints in (2.12). To avoid this problem, one can approximate the feasible set of
OPF problem with a convex set via change of variables: ∀n ∈ N, define

Un j = VnVj , n ∼ j (2.25)

For each line n ∼ j where n < j, let H 2×2 as follow

 
 vn Un j 
Mn j =  ∈ H 2×2 (2.26)

U
 jn v j 
 

By introducing the new (slack) variable Un j , the feasible set of the OPF problem can
be expressed as

n∈N
Õ
g
Pn − Pnd = vn − Un j yn j , ∀n ∈ N (2.27)


j:n∼ j
g− g g+
Pn ≤ Pn ≤ Pn , ∀ n ∈ N (2.28)

vn− ≤ vn (κ) ≤ vn+, ∀ n ∈ N (2.29)

 
 vn U n j

Mn j =    0, ∀n ∼ j (2.30)
 
U
 jn v j 
 
Rank(M n j ) = 1 ∀n ∼ j (2.31)

In this transformation, the rank constraint is a non-convex set. To recover convexity

the feasible set can be relaxed by removing the rank constraint (2.31). If in addition
the cost function to be minimized is convex the problem becomes an SOCP, because
the set of positive semidefinite matrices defines a conic constraint. This will be re-
ferred as the SOCP relaxation of the OPF [12].

While contrasting the SDP relaxation and SOCP relaxation for OPF problem, we no-
N×(N+1)
tice that the SDP relaxations have a matrix variable M with 2 unknown entries;
as a consequence the number of decision variables in the SDP is N 2 . Subsequently,

31
it can be challenging to obtain effectively while increasing N. On the other hand,
the SOCP relaxations are more memory efficient than SDP relaxations because it en-
forces constraints only on the sub-matrices of M. Furthermore, the SOCP relaxations
have a lower computational complexity.

The SOCP relaxation is said to be exact when the solution for the original OPF prob-
lem satisfies the rank one constraint. When the relaxation is exact the solution of the
SOCP is also a global optimal solution of the original OPF problem [22].

A number of researchers have proposed different SOCP relaxations that guarantee

exactness that exploit particular characteristics of the power system [20]. For in-
stance, in [75] it is shown that their SOCP relaxation can be exact for radial networks.

Other approaches include [20] and [75]. They suggested some adjustments to the
OPF formulation with imposing slightly different voltage magnitude lower and up-
per bounds in (2.9). This makes it possible to prove that the SOCP relaxation for the
adjusted OPF problem is exact. There is no guarantee that the adjustment OPF prob-
lem will have a feasible set that is close to that of the original OPF problem. Hence,
the solution obtained with this approach may not be close to the global solution of
the original OPF problem. On the other hand, [20] and [75] reported favorable nu-
merical results for IEEE test networks demonstrating that the SOCP relaxations may
work well in practice but with no guarantees.

32
3 | Optimal Power Flow without Energy
Storage

3.1 Notation

In this section, the notation of the DC system is presented which will be used for the rest of
the paper. The DC model contains buses and lines connecting all buses with each others.
These lines can be linked in tree, mesh, and radial topologies, Fig. (3.1) shows.

Figure 3.1: Summary of notation.

33
The graph G :={N, E} contains a set N of nodes and a set E of edges. Each line connects
an ordered pair (n, j) of buses. If two buses n and j are connected by a tie line directly, we
denote (n, j) ∈ E by n ∼ j. For each bus n ∈ N, Vn represents its voltage, In represents the
current injection, and Pn represents the power injection. At each line, let yn j represents
the admittance, and In j represents the current flow between buses n ∼ j from n to j, also
g
Zn j := 1
ynj it is impedance. For each bus n ∈ E, let Pn (t) denotes the generation at time t
and Pnd (t) denote its constant load demand. Notice that all the above parameters are real
numbers since we are working in DC network.

3.2 Mathematical Model and Problem Formulation

3.2.1 Analysis of Optimal Power Flow without Energy Storage

In this section, we present a derivation of optimal power flow problem. This formulation
does not include energy storage elements and a fully connected 3 bus system as shown in
Fig.(3.2). The Power flows in network are controlled by the following physical rules:

In j = (Vn − Vj )yn j , ∀{n, j} ∈ E (Ohm’s Law)

Õ
In = (In j ) , ∀{n, j} ∈ N (Current balance)
j:n∼ j

Pn = Vn In , ∀{n, j} ∈ N (Power Balance)

Therefore, the power flow at bus n satisfies:

Õ
g
Pn − Pnd = Vn (Vn − Vj )yn j , ∀{n, j} ∈ N (3.1)
j:n∼ j

34
 Figure 3.2: 3-bus of Power system.

Then, the optimal power flow problem for this particular example is:

Õ
g
min
g
fn (Pn ) (3.2)
Pn ,Vn
n∈N
subj. to (3.3)
3
Õ
g
Pn − Pnd = Vn (Vn − Vj )yn j n = 1, . . . , 3 (3.4)
j:n∼ j
g− g g+
Pn ≤ Pn ≤ Pn , n = 1, . . . , 3 (3.5)

Vn− ≤ Vn ≤ Vn+, n = 1, . . . , 3 (3.6)

The cost (3.2) is a function of generation in each node The equality constraint (3.4) is non-
linear in Vn and nonconvex. It can be replaced by a linear constraint on VV T with a new
matrix variable U:=VV T and Vn = [V1, V2, V3 ]T . However, to make the map from V to U
invertible, U must be constrained to be positive semidefinite (SDP) with rank equal to
one. Relaxing the rank constraint (removing it) leads to a second-order cone programming

35
(SOCP) problem. The resulting OPF is:

Õ
g
g
min fn (Pn ), ∀n ∈ N (3.7)
Pn ,Vn,Mn j
n∈N
subj. to (3.8)
n∈N
Õ  
g
Pn − Pnd = vn − Un j Yn j , ∀n ∈ N (3.9)
j:n∼ j
g− g g+
Pn ≤ Pn ≤ Pn , ∀n ∈ N (3.10)

vn− ≤ vn ≤ vn+, ∀n ∈ N (3.11)

 
 vn Un j

Mn j =  0 ∀n ∼ j (3.12)
 
U
 jn v j 
 

As mentioned above, if the solution fo the SOCP problem satisfies the rank one constraint
it is also a global solution to the original non-convex OPF. Eliminating the rank one con-
straint results in a convex problem by enlarging the feasible set; however, the new feasible
set is larger than that of the original OPF problem. That is why checking the rank condition
a posteriory is necessary.

36
3.3 Demonstration and Evaluation of Proposed Optimal Power

Flow (OPF) Method

3.3.1 Introduction

This section demonstrates the OPF algorithm by testing a six-bus system described in [76].
We will begin by implementing the OPF algorithm using both a Matlab nonlinear pro-
gramming solver and using the special solver CVX which is a Matlab software for solving
a complex convex optimization problems; files required for this case study can be found
in [77] and [78]. The computer specification for running this algorithm is a Lenovo Intel
Core i5 DDR3 SDRAM (2410M Cache, 2.30 GHz) with 8GB memory.

3.3.2 System Description

The Six-bus system is illustrated in Fig.5.3 [76]. The unit data for this system are given by
Table. (3.1)

Figure 3.3: 6-bus of Power System.

37
 g+
Vn+
g−
BUS Cost Func- Vn− Pn Pn Pnd
tion
1 5 350 450 0 10,000 0
2 7 350 450 0 10,000 0
3 0 350 450 0 10,000 10,000
4 0 350 450 0 10,000 6,000
5 0 350 450 0 10,000 0
6 0 350 450 0 10,000 0

Table 3.1: DC Microgrid Parameters.

The conductances between each bus are shown in table.3.2.

From To BUS Conductances

BUS
1 5 0.5
5 3 0.02
3 4 0.02
4 6 0.02
6 2 0.5
6 5 0.02

Table 3.2: Conductances Between Lines

In this test system, there are two power loads (Pnd ) and two generators dispatched. The
power loads (P3d and P4d ) at bus 3 and 4 are set to be 10 KWh and 6 KWh, respectively. The
g g
cost function values are also set to P1 = 5 $/MWh and P2 = 7 $/MWh. At each bus, the
power variables are bounded to between 0 and 10 KWh and the voltage constraints are set
between 350 V and 450 V.

38
3.3.3 Solution to Nonlinear OPF

Based on the system described in the previous subsection, the general OPF problem is as
follows:

g g
min f = [5P1 + 7P2 ] (3.13)
X

h i
X T := P g P g V1 V2 V3 V4 V5 V6
1 2

subj. to (3.14)
g
P1 = P1 − P1d = V1 (V1 − V5 )y15 (3.15)
g
P2 = P2 − P2d = V2 (V2 − V6 )y26 (3.16)
g
P3 = P3 − P3d = V3 (V3 − V4 )y34 + V3 (V3 − V5 )y35 (3.17)
g
P4 = P4 − P4d = V4 (V4 − V3 )y43 + V4 (V4 − V6 )y46 (3.18)
g
P5 = P5 − P5d = V5 (V5 − V1 )y51 + V5 (V5 − V3 )y53 + V5 (V5 − V6 )y56 (3.19)
g
P6 = P6 − P6d = V6 (V6 − V2 )y62 + V6 (V6 − V4 )y64 + V6 (V6 − V5 )y65 (3.20)


 g
0 ≤ P1 ≤ 10K







 g
0 ≤ P2 ≤ 10K








 350 ≤ V[1,..,6] ≤ 450


g
[Pn , Pnd, Vn ] =

re f
V1r e f = 380








P3d = 10K








 P4d = 6K





39
Where V[1,..,6] represents the voltages at each bus.

3.3.3.1 Solution Report and Analysis

In this example, the optimal cost is 9, 5615$/hour.The final result is shown by

g
Bus Pn Vn
1 10,000 380
2 6,5164 375.48
3 0 366.36
4 0 366.42
5 0 366.84
6 0 366.80

Table 3.3: Matlab-Nonlinear programming solver Result

The solver takes 10 iterations to reach the minimum cost of the system with 1.123837 sec-
ond of total CPU time.

3.3.4 Solution to Convex Relaxation of OPF

Using the convex relaxation technique, we can obtain the following problem:

g g
min f = [5P1 + 7P2 ] (3.21)
X

h i
X T := P g P g v1 v2 v3 v4 v5 v6 Un j , ∀n ∼ j
1 2

40
 subj. to (3.22)
g
P1 = P1 − P1d = (v1 − U15 )y15 (3.23)
g
P2 = P2 − P2d = (v2 − U26 )y26 (3.24)
g
P3 = P3 − P3d = (v3 − U34 )y34 + (v3 − U35 )y35 (3.25)
g
P4 = P4 − P4d = (v4 − U43 )y43 + (v4 − U46 )y46 (3.26)
g
P5 = P5 − P5d = (v5 − U53 )y53 + (v5 − U51 )y51 + (v5 − U56 )y56 (3.27)
g
P6 = P6 − P6d = (v6 − U64 )y64 + (v6 − U5 )y65 + (v6 − U62 )y62 (3.28)


 g
0 ≤ P1 ≤ 10K







 g
0 ≤ P2 ≤ 10K








 350 ≤ V[1,..,6] ≤ 450


g
[Pn , Pnd, Vn ] =

re f
V1r e f = 380








P3d = 10K








 P4d = 6K





Where v[1,..,6] represents the voltage square at each bus.

Therefore, the Mn j in OPF can be formulated as follows: In the relaxation step, the rank
constraint Rank(Mn j ) is eliminated, but a positive semidefinite constraint Mn j  0 needs
to be kept. We refer to this relaxation as SDP hereafter.

 
 Mn j 
 

R :=  . ..

0 ∀i ∼ j (3.29)

 
Mn j 
 

 

41
3.3.4.1 Solution Report and Analysis

In this example, the optimal cost is 95244.5$/hour.The final result is shown by

g √ q
No. Pn vn Vn = vn Mn j = Vn2j
1 10,000 14440 380 13940
2 6,5164 14980 374.48 13773
3 0 13422 366.42 13424
4 0 13426 366.42 13439
5 0 13457 366.84 13440
6 0 13454 366.80 13456

Table 3.4: CVX Result

By (2.30), (2.31), and (3.29) the matrix of R can be obtained as follows:

  
  v1 U15  
 
  
 U
  51 v5  

  
   
  v2 U26  
  
   


U
 62 v6  

   
   
  v3 U34  
  
   


U
 43 v4  

R :=    0
  
  v4 U46  
  
   


U
 64 v6  

   
   
  v5 U53  
  
   


U
 35 v3  

   
  
  v6 U65  
 
  


U
 56 v5  
  
(3.30)

The solver takes 22 iterations to reach the minimum cost of the system with 2.51 second

42
of total CPU time, 0.11 second per iteration.

3.3.5 Result Summary

This chapter presented the OPF solution of six-bus system via the linear power system
model. The results showed that the CVX solver is more effective and accurate than non-
linear programming for large-scale system. Moreover, the minimum solution satisfied
constraints. Nevertheless, the final cost using the Matlab nonlinear programming solver
was 9, 5615$/hour, which was very close to the final cost using the CVX solver.

43
4 | Optimal Power Flow with Energy Storage

4.1 Notation

In this chapter, we first review the notation. A Graph G:={N, E} contains a set N of nodes
and a set E of edges. Each line connects an ordered pair (n, j) of buses. If two buses n and j
are connected by a tie line directly, we denote (n, j) ∈ E by n ∼ j. For each bus n ∈ N, Vn (κ)
represents its voltage. At each line let yn j represents the conductance between bus n ∼ j
g
from n to j. For each bus n ∈ N, let Pn (κ) denotes the generation, Pnd (κ) its load demand at
time κ and bn (κ) the value of the energy storage at time κ. The energy storage elements a
modeled by the difference equation (4.1). It can therefore be considered as the following
collection of discrete linear time-varying models (one for each τ = {1, ..., (M)}), whose state
at time κ (going backward in time) is κ ∈ τ. It also can be assumed that Ts represents the
time step of the time series. Moreover, rn (κ) represents the value of the power at time (κ)
where it charges when rn (κ) has a negative value; rn− < 0. Otherwise it discharges when
rn (κ) has a positive value; rn+ ≥ 0. In addition, Bn− and Bn+ represent the minimum and
maximum value of the energy storage, respectively.

44
4.2 The Optimal Power Flow with Energy Storage

An OPF problem optimizes both variables V and P over the feasible solutions of the bus
injection model in (3.4). Additionally, all voltage magnitudes must satisfy (3.11) where
the lower and upper bounds are represented by Vn− and Vn+ . The lower bound of voltage
magnitude is assumed to be greater than zero to avoid triviality in the system. The power
injection is also bounded as shown in (3.10), similar to the case of the voltage magnitudes,
where Pn− and Pn+ are given bounds on the injections at bus n. We now formulate an
OPF with energy storage and time-varying generation costs and demands. As mentioned
previously, bn (κ) represents the value of the energy storage at time κ at bus n. The amount
of the energy storage is modeled to follow the first order difference equation.

bn (κ + 1) = bn (κ) + αnrn (κ), ∀n ∈ N, κ∈τ (4.1)

where αn denotes the time interval [κ, κ +1]. Assuming the initial energy stored is bn (0) ≥ 0
at each bus n ∈ N
0 ≤ Bn− ≤ bn (κ) ≤ Bn+, ∀n ∈ N, κ∈τ (4.2)

where Bn− and Bn+ represent the minimum and maximum value of the energy storage at
bus n, respectively. Moreover, the charge rate of the battery are bounded by:

rn− ≤ rn (κ) ≤ rn+, ∀n ∈ N, κ∈τ (4.3)

where κ=1, ..., (τ + 1). The power flow constraint at each bus n ∈ N and time κ ∈ τ are thus:

n∈N
Õ  
g
Pn (κ) − Pnd (κ) − rn (κ) = vn (κ) − Un j (κ) Yn j , ∀n ∈ N, κ ∈ τ (4.4)
j:n∼ j

g
where Pn (κ) represents the generation at bus n at time κ and Pnd (κ) represents its load.

45
Thus, OPF problem with energy storage in DC microgrid can be formulated as follows:

M
Õ Õ 
hnκ (bn (τ))
g
g
min fn (Pn (κ)) + , ∀n ∈ N, κ∈τ (4.5)
Pn (κ),Vn (κ),bn (κ),rn (κ)
n∈N κ=1
subj. to (4.6)

Vn− ≤ Vn (κ) ≤ Vn+, ∀n ∈ N, κ∈τ (4.7)

g− g g+
Pn ≤ Pn (κ) ≤ Pn , ∀n ∈ N, κ∈τ (4.8)

rn− ≤ rn (κ) ≤ rn+, ∀n ∈ N, κ∈τ (4.9)

b− +
n ≤ bn (κ) ≤ bn , ∀n ∈ N, κ∈τ (4.10)

bn (κ + 1) = bn (κ) + αnrn (κ), ∀n ∈ N, κ ∈ τ (4.11)

n∈N
Õ  
g
Pn (κ) − Pnd (κ) − rn (κ) = Vn (κ) − Vn (κ)Vj (κ) yn j , ∀n ∈ N, κ ∈ τ
2
(4.12)
j:n∼ j

• The Initial Condition (I.C) of bn at κ = 0

bn (0) = µc, ∀n ∈ N (4.13)

• The Voltage Reference at bus nre f

re f
Vnr e f (κ) = V0 (4.14)

46
Now, we present the convex formulation of the OPF problem with energy storage:

M
Õ Õ 
hnκ (bn (τ))
g
g
min fn (Pn (κ)) + , ∀n ∈ N, κ∈τ (4.15)
Pn (κ),vn (κ),bn (κ),rn (κ),Un (κ)
n∈N κ=1
subj. to (4.16)

vn− ≤ vn (κ) ≤ vn+, ∀n ∈ N, κ∈τ (4.17)

g− g g+
Pn ≤ Pn (κ) ≤ Pn , ∀n ∈ N, κ∈τ (4.18)

rn− ≤ rn (κ) ≤ rn+, ∀n ∈ N, κ∈τ (4.19)

Bn− ≤ bn (κ) ≤ Bn+, ∀n ∈ N, κ∈τ (4.20)

bn (κ + 1) = bn (κ) + αnrn (κ), ∀n ∈ N, κ∈τ (4.21)

n∈N
Õ  
g
Pn (κ) − Pnd (κ) − rn (κ) = vn (κ) − Un j (κ) Yn j , ∀n ∈ N, κ ∈ τ (4.22)
j:n∼ j
 
 vn Un j 
Mn j =   0, ∀n ∼ j (4.23)

U
 jn v j 
 

Rank(Mn j ) = 1 (4.24)

where vn :=Vn2 and Un j :=VnVj ∀n ∼ j.

Since (4.24) is a non-convex constraint, it must be relaxed to arrive at the convex formula-
tion. Also, fn has to be a convex function.

• The Initial Condition (I.C) of bn at κ = 0

bn (0) = µc, ∀n ∈ N (4.25)

• The Voltage Reference at bus nre f

re f
Vnr e f (κ) = V0 (4.26)

47
 bn (κ)
Equation (4.1) is often written in term of the State-of-Charge, SOCn (κ) := Etot , where Etot
is the total energy. Hence it is equivalent to

SOCn (κ + 1) = SOCn (κ) + αnrn (κ), ∀n ∈ N, κ∈τ (4.27)

where, αn is Etot .
Ts
With respect to SOC inequalities (4.20) became

SOCn− ≤ SOCn (κ) ≤ SOCn+, ∀n ∈ N, κ∈τ (4.28)

where

g
• fn (Pn ) is a convex cost function, which represents generation.

• hnκ (bn (τ)) represents the terminal cost.

4.2.1 Optimal Power Flow with Line Constraints

It is also possible to limit the current going through any line n ∼ j. Let In j be a threshold
of the current flow through n ∼ j, then the line constraint can be formulated as:

(vn − Un j − U jn + v j )yn2 j ≤ In2j (4.29)

which in terms of original voltage variables is:

(Vn − Vj )2 yn2 j ≤ In2j (4.30)

48
Adding (4.29) to the constraints of the OPF problem leads to :

M Õ 
hnκ (bn (τ))
Õ
g
g
min fn (Pn (κ)) + , ∀n ∈ N, κ∈τ (4.31)
Pn (κ),vn (κ),SOCn (κ),rn (κ),Un (κ)
κ=1 n∈N
subj. to (4.32)

vn− ≤ vn (κ) ≤ vn+, ∀n ∈ N, κ∈τ (4.33)

g− g g+
Pn ≤ Pn (κ) ≤ Pn , ∀n ∈ N, κ∈τ (4.34)

rn− ≤ rn (κ) ≤ rn+, ∀n ∈ N, κ∈τ (4.35)

Bn− ≤ SOCn (κ) ≤ Bn+, ∀n ∈ N, κ∈τ (4.36)

SOCn (κ + 1) = SOCn (κ) + αnrn (κ), ∀n ∈ N, κ ∈ τ (4.37)

n∈N
Õ  
g
Pn (κ) − Pnd (κ) − rn (κ) = vn (κ) − Un j (κ) Yn j , ∀n ∈ N, κ ∈ τ (4.38)
j:n∼ j
 
vn (κ) − Un j (κ) Yn j ≤ Pn j (4.39)
 
v j (κ) − U jn (κ) Yjn ≤ P jn (4.40)
 
 vn Un j 
Mn j =   0, ∀n ∼ j (4.41)

U
 jn v j 
 

One way to keep some of the theoretical ensures is to force the line constraints in terms of
power flows alternatively. In specific, |In j | ≤ Iˆn j is equivalent to |Pn j | ≤ Vn Iˆn j . Assuming
that Vn is near to its nominal value, then |Pn j | ≤ Vn Iˆn j can be proximated by |Pn j | ≤ P̂n j for
some P̂n j ∈ R. Hence, the final formulation of the line constraints can be written as (4.39)
and (4.41).

(For detailed proofs, see T heor em − 7 in [20]).

49
4.2.2 Modeling the Energy Storage

Assuming that bus n can have at most one battery for energy storage, we denote SOCn
the charge of the battery at bus n. For simplicity, batteries were modeled as discrete-time
integrators:
SOCn (κ + 1) = SOCn (κ) + αnrn (κ), SOCn (0) given (4.42)

where αn [h/KW] is the event sampling and rn (κ) the rate of charge. In the rest of this
section, we will show how to express the dynamic equations imposed by the battery as
algebraic linear equations suitable for inclusion in a non dynamic optimization problem.

Consider a causal linear, time-invariant, discrete-time system with the state-space equa-
tions

x(κ + 1) = Ax(κ) + Bu(κ), x(0) = xo (4.43)

y(κ) = C x(κ) + Du(κ)

where x is the state vector and u the input vector.

The general expression for the solution of the state-space equations is

κ−1
κ
Õ
x(κ) = A xo + A(τ−1)−m Bu(m), κ≥0 (4.44)
m=0

In the context of OPF problems, τ is the index of the planning epoch. Suppose that the
problem involves (M) epochs, starting at M = 1. Then (4.44) can be written in matrix form

50
as:

 
 u(0) 
 
 
h i  u(1) 
x(M) − AM xo = AM−1 B · · · AB B 
 
.. 

 . 

 
 
u(M − 1)
 

Collecting all the states at each epoch in a vector, we obtain the following linear algebraic
equation:

 x(1)   A   B 0 ··· ··· 0  u(0) 

       

...
       
 x(2)   A2 
       
   
 AB
 B 0   u(1) 
 
 .   .  .. .. .. . .
 ..  −  ..  xo =  . ..  ..
  
     . . 



 ..   ..  .. ..
       
 .   .  . B 0 .
 M−2
A B
  
 
       
     M−1   
 x(M)  AM  A B AM−2 B · · · AB B u(M − 1)
      

For the particular model of the battery in (4.42) the above equations reduce to

 SOCn (1)  1 αn 0 · · · · · · 0   rn (0) 

       
 .. . . . .
       
. . .
      
 SOCn (2)  1 0   rn (1) 
       
..  .
 −  ..  SOCn (0) =
.
 .. .. .. ..  ..
(4.45)
  

 .     . . . 
 . 

.
..   ..   .. .. .
..
       
 . . . 0
   
  
       
αn · · · · · · · · · αn 
       
SOCn (M) 1 rn (M − 1)
       

51
5 | Demonstration and Evaluation of Pro-
posed OPF Method with Energy Storage

5.1 Introduction

This section demonstrates the DC-OPF problem with energy storage by testing a six-bus
system [76]. This problem is a nonconvex optimization problem whose objective function
has discontinuous first order derivatives that we solved through a special CVX solver and
ran on a Lenovo Intel Core i5 DDR3 SDRAM (2410M Cache, 2.30 GHz) with 8GB memory.

5.2 CASE STUDIES

5.2.1 Test Case of DC Microgrid System

The first test case considers DC microgrid system, with two separate power sources — the
DC network and a battery device — directly connected to the microgrid through a bus.
The battery device is added to the grid to provide sufficient power to the loads, in case of
the local power is not enough or if a generator fails.

52
5.2.2 Scenario-I, Parameters and Results

The microgrid, is a system of six buses shown in Fig. 5.1. The network consists of two
power generators at buses 1 and 2, two loads at buses 3 and 4, four link buses (i.e., without
generation nor load) and eight transmission lines. An energy storage unit has added to
bus 6. Since the full life of a battery device is directly related to the number of charging and
discharging cycles, there is a cost associated with it a per charge cycle [79]. The parameters
and the boundary conditions associated with the power generators, the SOC, and the bus
reference are given in Table. 5.1 and the power loads at different operational times are
given in Appendix − I Table. 7.1. The line constraints are given in Table. 5.2. The costs of
g g
power purchased from the generators are set to P1 = 5$/MWh and P2 = 7$/MWh. The
voltage magnitude bounds for all buses are set between 350 V and 450 V, and the voltage
re f
magnitude at the first bus defined as a reference bus, V1 , is 380 V.

g+
Vn+ SOCn+ rn+
g−
Vn− SOCn− rn−
re f
BUS Pn Pn Vnr e f
1 - - 0 10K 380 - - - -
2 350 450 0 10K - - - - -
3 350 450 - - - - - - -
4 350 450 - - - - - - -
5 350 450 - - - - - - -
6 350 450 - - - 0.2 0.8 -7K 7K

Table 5.1: The bound constraints.

BUS ∆Vn j (κ)− ∆Vn j (κ)+

3 To 4 -2K 2K
4 To 3 -2K 2K
5 To 6 -2K 2K
6 To 5 -2K 2K

Table 5.2: The power capacity constraints.

In this scenario, the transmission lines between buses 5 and 6 and between buses 3 and 4
are bounded. The storage battery is bounded by a capacity SOCn (κ)− and SOCn (κ)+ with a

53
charge or discharge rate (discussed earlier) as rn (κ)− and rn (κ)+ . The initial condition of the
storage battery was chosen to be 0.7 KWh. A total of 20 KWh of energy was proposed as
well as the sampling time of one hour, and the simulations were performed over one day.
The simulation results illustrate the advantages of including an energy storage under low
and high demand circumstances. During low demand, the energy is stored in the storage
unit and then released when the load is high, smoothing the total power injected into the
grid.

5.2.3 Demonstration of Scenario-I

The system in Fig. 5.1 includes a storage device at bus 6.

Figure 5.1: Scenario I: A power system with energy storage at bus six.

54
The following description as mentioning earlier: power generations are represented by
g
Pn , storage power is rn (κ), and the event sampling is αn . The battery is modeled by its
minimum and the maximum value of the energy storage rn (κ)− and rn (κ)+ , and the battery
State of Charge SOCn , in the range 0.2 and 0.8. This represents energy in this problem.

24 Õ
Õ
g g
g
min fn (5P1 (κ) + 7P2 (κ)) (5.1)
Pn (κ),vn (κ),SOCn (κ),rn (κ),Un (κ)
κ=1 n∈6
subj. to (5.2)

380 ≤ vn (κ) ≤ 450, ∀n ∈ 6, κ∈τ (5.3)

g
0 ≤ P1 (κ) ≤ 10K, κ ∈ τ (5.4)
g
0 ≤ P2 (κ) ≤ 10K, κ ∈ τ (5.5)

−7K ≤ r6 (κ) ≤ 7K, κ ∈ τ (5.6)

0.2 ≤ SOC6 (κ) ≤ 0.8, κ ∈ τ (5.7)

SOC6 (κ + 1) = SOC6 (κ) + α6r6 (κ), κ ∈ τ (5.8)

n∈6 
Õ 
g d (κ) − r (κ) =
P[1,2] (κ) − P[3,4] 6 vn (κ) − Un j (κ) Yn j , ∀n ∈ 6, κ ∈ τ (5.9)
j:n∼ j
 
−2K ≤ vn (κ) − Un j (κ) Yn j ≤ 2K (5.10)
 
−2K ≤ v j (κ) − U jn (κ) Yjn ≤ 2K (5.11)
 
M n j 
 

R := 
 . ..

  0, ∀n ∼ j (5.12)

 
M jn 
 

 

• The Initial Condition (I.C) of SOC6 at κ = 0

SOC6 (0) = 0.7 (5.13)

55
 • The Voltage Reference at bus nre f

re f
Vnr e f (κ) = 380 (5.14)

The problem was solved numerically using CV X. The results are summarized in Table. 5.3.

Optimal value 505927

Computation Time (s) 0.12

Table 5.3: The result of Scenario-I.

It was verified, a posteriory, that the rank one conditions on Mn j , ∀ n ∼ j were satisfied.
Hence the optimal value of the cost reported in the table is also global optimum for the
original OPF. The numerical values of power injections and nodal voltages are listed in
Appendix-I, Tables. 7.2 and 7.3.

56
5.2.4 System Load Data

The total power load of the DC microgrid for each hour is shown in Fig. 5.2. This includes
the power loads at buses 3 and 4. The power load at bus 3 is constant during the opera-
tional time. However, the power load at bus 4 is unsteady with the average hourly power
load of around 7.5 KWh.

Figure 5.2: The actual power loads

57
5.2.5 System Generation Data

The power generator schedule for the 24 hours is given in Fig. 5.3, which illustrates that all
generators during the operational time are within the bounds and satisfied the constraints.

Figure 5.3: The generator schedule

In Fig. 5.4, it can been seen that the battery schedule satisfied the constraints and bounds
during the operations. Moreover, battery charging and discharging happen during the
production times as can be seen by noting the graphs. The energy stored in the battery
is charging during 1 to 6, 11 to 16, and between 23 to 24. It is also discharging during 7
to 10 and between 17 to 22. The stored power is utilized to feed nearby loads and avoid
losses and the high cost on the system if faraway the DC generations are used to feed these
loads. Since the battery is not needed to hold any further power at the end of 25 hours, it

58
is economically beneficial to discharge all the battery energy so the net battery energy is
0. The total optimal cost for supplying the load is given in Table. 5.3.

Figure 5.4: The battery schedule

A list of the standby power battery values are given in Appendix-I, Table. 7.3. These values
are required at each hour to support the system by providing energy necessary for the
loads.

5.2.6 State-of-Charge (SOC)

A dynamic OPF problem has been solved due to the dynamics of the state-of-charge (SOC)
of the battery. As shown in Fig. 5.5, it can be seen experimentally when the battery is ei-
ther charged fully SOC = 100% or discharged to SOC = 0%.

59
Figure 5.5: The State-of-charge schedule

60
5.2.7 Parameters and Results of Scenario-II

In this scenario, the OPF formulation with energy storage dynamics are applied to ex-
amine the reliability and stability of the system. The DC microgrid is, shown in Fig. 5.6.
Presently the network consists of two power generators at buses 1 and 2, two loads at
buses 3 and 4, and four link buses (i.e., without including generation nor load) and eight
transmission lines. Furthermore, two energy storage units have been applied in the net-
work at buses 5 and 6. The main aim of including a battery device in this system is to
increase system security by acting as a standby when required. The parameters and the
constraints in the system associated with the power generators, the SOC, and the bus refer-
ence are shown in Table. 5.4, and the power loads at different operational time are given in
Appendix-II Table. 7.4. Moreover, the line constraints have been considered in this case as
shown in Table. 5.5. The costs of power purchased to the main grid throughout the oper-
g g
ation are also set to P1 = 5$/MWh and P2 = 7$/MWh. The voltage magnitude Vn bounds
for all buses are set between 350 and 450 per units (p.u.) and the voltage magnitude at
re f
first bus is defined as a reference bus, Vnr e f =380V.

g+
Vn+ SOCn+ rn+
g−
Vn− SOCn− rn−
re f
BUS Pn Pn Vnr e f
1 - - 0 10K 380 - - - -
2 350 450 0 10K - - - - -
3 350 450 - - - - - - -
4 350 450 - - - - - - -
5 350 450 - - - 0.2 0.8 -7K 7K
6 350 450 - - - 0.2 0.8 -7K 7K

Table 5.4: The Bound Constraints.

BUS ∆Vn j (κ)− ∆Vn j (κ)+

3 To 4 -2K 2K
4 To 3 -2K 2K
5 To 6 -2K 2K
6 To 5 -2K 2K

Table 5.5: The power capacity constraints

61
In this scenario, the transmission lines between buses 5 and 6 and between buses 3 and 4
are bounded. The storage battery is bounded by a capacity SOCn (κ)− and SOCn (κ)+ with
a charge or discharge rate (discussed earlier) as rn (κ)− and rn (κ)+ . The initial condition of
the storage battery was chosen to be 0.7 KWh. A total of 20 KWh of energy was proposed
as well as the sampling time of one hour, and the simulations were performed over one
day. The simulation results illustrate the advantages of including energy storage devices
under low and high demand circumstances.

5.2.8 Demonstration of Scenario-II

The system is illustrated in Fig. 5.6 and includes two storage devices at buses 5 and 6. Dur-
ing the power production, the energy is stored in the battery and released when the rn (κ),
value of the power at time (κ), is rn (κ)+ ≥ 0 and rn (κ)− ≤ 0, respectively.

62
 Figure 5.6: Scenario II: A power system with two energy storages.

g
The following illustration as given earlier: power generations are described by Pn , stor-
age power is rn (κ), and the event sampling is αn . The battery is modeled by its minimum
and the maximum values of the energy storage rn (κ)− and rn (κ)+ , and the battery state-of-
charge SOCn , in the range 0.2 and 0.8, this describes energy in this problem.

63
The optimization problem for this case is:

24
Õ
g g
5P1 (κ) + 7P2 (κ) (5.15)


g
min
Pn (κ),vn (κ),SOCn (κ),rn (κ),Un (κ)
κ=1
subj. to (5.16)

380 ≤ vn (κ) ≤ 450, ∀n ∈ 6, κ∈τ (5.17)

g
0 ≤ P1 (κ) ≤ 10K, κ ∈ τ (5.18)
g
0 ≤ P2 (κ) ≤ 10K, κ ∈ τ (5.19)

−7K ≤ r[5,6] (κ) ≤ 7K, κ ∈ τ (5.20)

0.2 ≤ SOC[5,6] (κ) ≤ 0.8, κ ∈ τ (5.21)

SOC[5,6] (κ + 1) = SOC[5,6] (κ) + α[5,6]r[5,6] (κ), κ ∈ τ(5.22)

n∈6 
Õ 
g
[5,6] (κ) = vn (κ) − Un j (κ) Yn j , ∀n ∈ 6, κ ∈ τ (5.23)
d (κ) − r
P[1,2] (κ) − P[3,4]
j:n∼ j
 
−2K ≤ vn (κ) − Un j (κ) Yn j ≤ 2K (5.24)
 
−2K ≤ v j (κ) − U jn (κ) Yjn ≤ 2K (5.25)
 
M n j 
 

R := 
 . ..

  0, ∀n ∼ j (5.26)

 
M jn 
 

 

• The Initial Condition (I.C) of SOC[5,6] at κ = 0

SOC[5,6] (0) = 0.7 (5.27)

• The Voltage Reference at bus nre f

re f
V1r e f (κ) = 380 (5.28)

64
The optimal value of the cost and the running time are shown in Table. 5.6.

Optimal value 1.62046e+06

Computation Time (s) 2.88

Table 5.6: The result of Scenario-II.

The numerical values of power injections and nodal voltages are listed in Appendix-II,
Tables 7.5 and 7.6.

65
5.2.9 System Load Data

Similarly to the previous scenario, the total power load of the DC micro-grids for each
hour is shown in Fig. 5.11. This includes the power loads at buses 3 and 4. The power
load at bus 3 is constant during the operational time. However, the power load at bus 4 is
unsteady with the average hourly power load of around 7.5 kWh.

Figure 5.7: The actual power loads

66
5.2.10 System Generation Data

The power generators schedule for 24 hours is given in Fig. 5.8. In this figure, it can been
seen that all generators during the operational time are within the bounds and satisfied
the constraints.

Figure 5.8: The generator schedule

In Fig. 5.9, it can be seen that the battery schedule satisfied the constraints and bounds
during the operations. Furthermore, battery charging and discharging happen during the
production times as can be seen by noting the graphs. The energy stored in the battery at
bus 5 is charging during 1 to 6, 8 to 10, 17 to 18, and between 22 to 23. It is also discharging
during at 7, 11 to 16, 18 to 21, and at 24. On the other hand, when the energy at bus
5 is stored in the battery, the battery at bus 6 is discharged from energy. In bus 6 the
battery starts to store at the beginning of the operation until hour 7, then it discharges

67
from hours 8 to 12, then back to restore the battery between 13 and 17, and next the battery
discharges during 18 to 22. The stored powers are utilized to feed nearby the loads and
avoid losses and the high cost on the system if faraway DC generations feed these loads.
Since the battery is not needed to hold any further power at the end of 25 hours, it is
economically beneficial to discharge all the battery energy so the net battery energy is 0.
The total optimal cost for supplying the load is given in Table. 5.6.

Figure 5.9: The battery schedule

A list of the standby of power battery values are given in Table. 7.6.

68
5.2.11 State-of-Charge (SOC)

A dynamic OPF problem has been solved due to the dynamics of the state-of-charge (SOC)
of the battery. As shown in Fig. 5.10, it can be seen experimentally when the battery is
either charged fully SOC = 100% or discharged to SOC = 0%.

Figure 5.10: The State-of-charge schedule

69
5.2.12 Analysis of Results

As it can be seen from above scenarios the SOCP is is exact, i.e. Rank(M n j ) = 1 holds,
∀n ∼ j and hence the solution found is also a global solution of the original OPF. The scale
of the problems solved in this work is small. In practice multi-bus DC systems may inte-
grate a large number of buses. The results of computation time suggests that the proposed
method may not be suitable to very large scale systems.

Therefore, following flowchart summarizes the SOPC relaxation approach used to solve
the OPF problem.

Figure 5.11: The flowchart of the convexification process

The solution of the SOCP includes a set of 2 × 2 positive semidefinite matrices Mn j , ∀ n j.

Hence it is necessary to check the rank constraint a posteriory. If it is satisfied then the
optimal solution is the global solution of the original OPF problem. If it is not satisfied
then, in general, nothing can be said [12].

70
6 | Conclusion and Future Work

6.1 Conclusions

In this work, we formulated an optimal power flow (OPF) problem with dynamic energy
storage as a static optimization problem. This results in a nonlinear programming prob-
lem. This problem was transformed to a problem with constraints over the cone of positive
semidefinite matrices and required the introduction of a rank suitable constraint. The rank
constraint was ignored resulting in am SOPC problem which can be solved numerically
as using SDP solvers. This approach was tested in various scenarios. One with a single
energy storage and another with more energy storage devices.

Analysis of the numerical result analysis of these scenarios shows that the optimal solution
is can be obtained, i.e. Rank(M n j ) = 1 holds, ∀n ∼ j.

Form the figures we can see that, as expected, the energy stored was scheduled to balance
the power of loads and generation sources. It can be seen that energy storage provides
additional flexibility to the system for handling load variations and power limits, with the
potential of providing improved power quality, stability, load following, peak reduction,
and reliability.

71
6.2 Future Work

There are several future research directions that have been revealed as a result of this work.
We summarize the most promising below.

• The most important next step is to integrate recent algorithms that explicitly enforce
the rank one constraint instead of ignoring it and checking it after the fact. There
are a few approaches that can be considered such as the well known nuclear norm
minimization [80] and also low-rank inducing norms recently proposed in [81].

• Another research direction to make this practical is necessary to reformulate the

problem as a distributed dynamic optimization problem. This will be necessary to
solve large-scale power flow problems but will also enable the implementation us-
ing low cost embedded system making it more affordable for less affluent and rural
communities.

• Finally, it is natural to consider the problem as a polynomial optimization problem

given that the nonlinearities are polynomial. This is already happening and will
bring powerful alternatives to find global solutions to optimal power flow problems.

72
7 | APPENDIX

7.1 Appendix-I

g g
Time (hr) P3d P4d Time (hours) Pg
Í
P1 P2
1 4K 2236.89 1 6214.7 142.84 6357.6
2 4K 1454.8 2 5942.3 143.11 6085.4
3 4K 1248.7 3 5870.6 143.18 6013.7
4 4K 1276.1 4 5880.1 143.17 6023.3
5 4K 1845.2 5 6078.2 142.96 6221.2
6 4K 3523.9 6 6664.4 10222 7686.4
7 4K 7148.9 7 7939.4 3481.2 11421
8 4K 9380.8 8 6781.5 5536.1 12318
9 4K 9794.5 9 6347.5 5553.8 11901
10 4K 9480.1 10 6677.2 5540.3 12218
11 4K 8951.7 11 7233.0 5517.7 12751
12 4K 8178.8 12 8050.1 5484.4 13535
13 4K 7615.9 13 8104.6 5482.0 13587
14 4K 7543.5 14 8079 5483 13562
15 4K 8071.9 15 8163.4 5479.8 13643
16 4K 9013.8 16 7167.6 6170 13338
17 4K 11005 17 6221.4 10K 16221
18 4K 14109 18 8492.4 10K 18492
19 4K 14249 19 8790.1 10K 18790
20 4K 13005 20 7940.7 10K 17941
21 4K 11393 21 7932.2 10K 17932
22 4K 9145.9 22 7028.5 6446.3 13475
23 4K 6354.0 23 7658.7 2938.1 10597
24 4K 3973.6 24 6821.9 1324.6 8146.5

Table 7.1: Power Load for Scenario-I. Table 7.2: Power Generation for Scenario-I.

73
 Time (hr) P3Loads P4Loads PLoads
Í
r6
1 4K 2236.89 6236.9 -16.810
2 4K 1454.8 5454.8 505.15
3 4K 1248.7 5248.7 642.67
4 4K 1276.1 5276.1 624.39
5 4K 1845.2 5845.2 244.60
6 4K 3523.9 7523.9 1.5456e-04
7 4K 7148.9 11149 -2.6321e-05
8 4K 9380.8 13381 -1343.2
9 4K 9794.5 13795 -2154.7
10 4K 9480.1 13480 -1538.0
11 4K 8951.7 12952 -501.96
12 4K 8178.8 12179 1013.3
13 4K 7615.9 11616 1626.3
14 4K 7543.5 11544 1675.6
15 4K 8071.9 12072 1222.8
16 4K 9013.8 13014 3.5765e-04
17 4K 11005 15005 -5.6655e-05
18 4K 14109 18109 -4473.9
19 4K 14249 18249 -4614.8
20 4K 13005 17005 -2801.1
21 4K 11393 15393 -110.20
22 4K 9145.9 13146 6.7061e-05
23 4K 6354.0 10354 7.3379e-05
24 4K 3973.6 7973.6 -1.4010e-04

Table 7.3: Power Load and rate of charge for Scenario-I.

74
7.2 Appendix-II

g g
Time (hr) P3d P4d Time (hr) Pg
Í
P1 P2
1 4K 2236.89 1 6414.5 143.05 6557.5
2 4K 1454.8 2 6415.2 143.05 6558.2
3 4K 1248.7 3 6415.4 143.05 6558.4
4 4K 1276.1 4 6415.4 143.06 6558.4
5 4K 1845.2 5 6415 143.07 6558
6 4K 3523.9 6 6664.4 1022 7686.4
7 4K 7148.9 7 7004.9 3480.9 10486
8 4K 9380.8 8 6912.4 5515.5 12428
9 4K 9794.5 9 6911 5516.2 12427
10 4K 9480.1 10 6912.2 5515.7 12428
11 4K 8951.7 11 7118.8 5506.7 12625
12 4K 8178.8 12 7121 5505.4 12626
13 4K 7615.9 13 7122.7 5505 12628
14 4K 7543.5 14 7122.8 5505 12628
15 4K 8071.9 15 7121.3 5505.2 12627
16 4K 9013.8 16 7082.7 6169.9 13253
17 4K 11005 17 6261.6 10K 16262
18 4K 14109 18 6222.8 10K 16223
19 4K 14249 19 62208 10K 16221
20 4K 13005 20 6237.8 10K 16238
21 4K 11393 21 6245.9 10K 16246
22 4K 9145.9 22 7054.6 6446.3 13501
23 4K 6354.0 23 7216.7 2938 10155
24 4K 3973.6 24 6821.8 1324.6 8146.5

Table 7.4: Power Load for Scenario-II. Table 7.5: Power Generation for Scenario-II.

75
Time (hr) P3Loads P4Loads P LLoads r5 + r6
Í
r5 r6
1 4K 2236.89 6236.9 191.00 -16.598 174.40
2 4K 1454.8 5454.8 452.67 505.08 957.75
3 4K 1248.7 5248.7 521.61 642.54 1164.1
4 4K 1276.1 5276.1 512.48 624.27 1136.8
5 4K 1845.2 5845.2 322.21 244.71 566.91
6 4K 3523.9 7523.9 8.3177e-03 3.2036e-03 1.1521e-02
7 4K 7148.9 11149 -886.16 -4.1974e-04 -886.16
8 4K 9380.8 13381 124.65 -1363.2 -1238.5
9 4K 9794.5 13795 537.59 -2191.4 -1653.8
10 4K 9480.1 13480 223.91 -1562 -1338.1
11 4K 8951.7 12952 -108.55 -512.41 -620.95
12 4K 8178.8 12179 -880.24 1034.2 153.97
13 4K 7615.9 11616 -930.12 1649.2 719.05
14 4K 7543.5 11544 -905.76 1697.5 791.74
15 4K 8071.9 12072 -986.95 1248.0 261.05
16 4K 9013.8 13014 -80.645 6.8607e-03 -80.639
17 4K 11005 15005 1129.4 -363.74 765.62
18 4K 14109 18109 -3720.4 -3946.0 -7666.4
19 4K 14249 18249 -3998.7 -4087.3 -8086.0
20 4K 13005 17005 -1517.5 -2831.0 -4348.4
21 4K 11393 15393 393.90 -772.0 -378.10
22 4K 9145.9 13146 24.843 1.2644e-03 24.844
23 4K 6354.0 10354 -419.24 1.4531e-03 -419.24
24 4K 3973.6 7973.6 -1.3429e-02 -2.7123e-03 -1.6141e-02

Table 7.6: Power Load and rate of charge for Scenario-II.

76
Bibliography

[1] T. Markvart, Solar electricity, vol. 6. John Wiley & Sons, 2000.

[2] B. Parida, S. Iniyan, and R. Goic, “A review of solar photovoltaic technologies,” Re-
newable and sustainable energy reviews, vol. 15, no. 3, pp. 1625–1636, 2011.

[3] G. K. Singh, “Solar power generation by PV (photovoltaic) technology: A review,”

Energy, vol. 53, pp. 1–13, 2013.

[4] T. Ackermann, G. Andersson, and L. Söder, “Distributed generation: a definition,”

Electric power systems research, vol. 57, no. 3, pp. 195–204, 2001.

[5] M. Barnes, J. Kondoh, H. Asano, J. Oyarzabal, G. Ventakaramanan, R. Lasseter,

N. Hatziargyriou, and T. Green, “Real-world microgrids-an overview,” in System of
Systems Engineering, 2007. SoSE’07. IEEE International Conference on, pp. 1–8, IEEE,
2007.

[6] N. Pogaku, M. Prodanovic, and T. C. Green, “Modeling, analysis and testing of au-
tonomous operation of an inverter-based microgrid,” IEEE Transactions on power elec-
tronics, vol. 22, no. 2, pp. 613–625, 2007.

[7] O. Palizban and K. Kauhaniemi, “Energy storage systems in modern grids—matrix

of technologies and applications,” Journal of Energy Storage, vol. 6, pp. 248–259, 2016.

77
 [8] H. Chen, T. N. Cong, W. Yang, C. Tan, Y. Li, and Y. Ding, “Progress in electrical energy
storage system: A critical review,” Progress in Natural Science, vol. 19, no. 3, pp. 291–
312, 2009.

[9] S. Abu-Sharkh, R. Arnold, J. Kohler, R. Li, T. Markvart, J. Ross, K. Steemers, P. Wil-

son, and R. Yao, “Can microgrids make a major contribution to UK energy supply?,”
Renewable and Sustainable Energy Reviews, vol. 10, no. 2, pp. 78–127, 2006.

[10] M. Basu, “Economic environmental dispatch using multi-objective differential evo-

lution,” Applied soft computing, vol. 11, no. 2, pp. 2845–2853, 2011.

[11] N. Kok, N. Miller, and P. Morris, “The economics of green retrofits,” Journal of Sus-
tainable Real Estate, vol. 4, no. 1, pp. 4–22, 2012.

[12] S. H. Low, “Convex relaxation of optimal power flow—part i: Formulations and

equivalence,” IEEE Transactions on Control of Network Systems, vol. 1, no. 1, pp. 15–
27, 2014.

[13] D. Salomonsson, L. Soder, and A. Sannino, “An adaptive control system for a DC
microgrid for data centers,” in Industry Applications Conference, 2007. 42nd IAS Annual
Meeting. Conference Record of the 2007 IEEE, pp. 2414–2421, IEEE, 2007.

[14] H. Kakigano, Y. Miura, and T. Ise, “Low-voltage bipolar-type DC microgrid for su-
per high quality distribution,” IEEE transactions on power electronics, vol. 25, no. 12,
pp. 3066–3075, 2010.

[15] C.-C. Lin, H. Peng, J. W. Grizzle, and J.-M. Kang, “Power management strategy for a
parallel hybrid electric truck,” IEEE transactions on control systems technology, vol. 11,
no. 6, pp. 839–849, 2003.

[16] K. Divya and J. Østergaard, “Battery energy storage technology for power sys-
tems—an overview,” Electric Power Systems Research, vol. 79, no. 4, pp. 511–520, 2009.

78
[17] C. W. Gellings, The smart grid: enabling energy efficiency and demand response. The Fair-
mont Press, Inc., 2009.

[18] J. M. Guerrero, L. Hang, and J. Uceda, “Control of distributed uninterruptible power

supply systems,” IEEE Transactions on Industrial Electronics, vol. 55, no. 8, pp. 2845–
2859, 2008.

[19] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. De Vicuña, and M. Castilla, “Hierarchical

control of droop-controlled AC and DC microgrids — a general approach toward
standardization,” IEEE Transactions on industrial electronics, vol. 58, no. 1, pp. 158–172,
2011.

[20] J. Li, F. Liu, Z. Wang, S. H. Low, and S. Mei, “Optimal power flow in stand-alone DC
microgrids,” arXiv preprint arXiv:1708.05140, 2017.

[21] J. Carpentier, “Contribution to the economic dispatch problem,” Bulletin de la Societe

Francoise des Electriciens, vol. 3, no. 8, pp. 431–447, 1962.

[22] L. Gan and S. H. Low, “Optimal power flow in direct current networks,” IEEE Trans-
actions on Power Systems, vol. 29, no. 6, pp. 2892–2904, 2014.

[23] O. Alsac, J. Bright, M. Prais, and B. Stott, “Further developments in LP-based optimal
power flow,” IEEE Transactions on Power Systems, vol. 5, no. 3, pp. 697–711, 1990.

[24] J. A. Momoh, R. Adapa, and M. El-Hawary, “A review of selected optimal power

flow literature to 1993. i. nonlinear and quadratic programming approaches,” IEEE
transactions on power systems, vol. 14, no. 1, pp. 96–104, 1999.

[25] F. Capitanescu and L. Wehenkel, “A new iterative approach to the corrective security-
constrained optimal power flow problem,” IEEE transactions on power systems, vol. 23,
no. 4, pp. 1533–1541, 2008.

79
[26] P. E. Onate and J. M. Ramirez, “Optimal power flow including transient stability
constraints,” in Transmission and Distribution Conference and Exposition, 2008. T&D.
IEEE/PES, pp. 1–9, IEEE, 2008.

[27] B. Stott and E. Hobson, “Power system security control calculations using linear pro-
gramming, part i,” IEEE Transactions on Power Apparatus and Systems, no. 5, pp. 1713–
1720, 1978.

[28] S. Sojoudi and J. Lavaei, “Physics of power networks makes hard optimization prob-
lems easy to solve,” in Power and Energy Society General Meeting, 2012 IEEE, pp. 1–8,
IEEE, 2012.

[29] R. Madani, S. Sojoudi, and J. Lavaei, “Convex relaxation for optimal power flow prob-
lem: Mesh networks,” IEEE Transactions on Power Systems, vol. 30, no. 1, pp. 199–211,
2015.

[30] J. A. Momoh, M. El-Hawary, and R. Adapa, “A review of selected optimal power flow
literature to 1993. ii. newton, linear programming and interior point methods,” IEEE
Transactions on Power Systems, vol. 14, no. 1, pp. 105–111, 1999.

[31] J. T. Betts, Practical methods for optimal control and estimation using nonlinear program-
ming. SIAM, 2010.

[32] X. Bai, H. Wei, K. Fujisawa, and Y. Wang, “Semidefinite programming for opti-
mal power flow problems,” International Journal of Electrical Power & Energy Systems,
vol. 30, no. 6, pp. 383–392, 2008.

[33] X. Bai and H. Wei, “Semi-definite programming-based method for security-

constrained unit commitment with operational and optimal power flow constraints,”
IET Generation, Transmission & Distribution, vol. 3, no. 2, pp. 182–197, 2009.

80
[34] J. Lavaei and S. H. Low, “Zero duality gap in optimal power flow problem,” IEEE
Transactions on Power Systems, vol. 27, no. 1, pp. 92–107, 2012.

[35] Z. Wang, F. Liu, Y. Chen, S. H. Low, and S. Mei, “Unified distributed control of stand-
alone DC microgrids,” IEEE Transactions on Smart Grid, 2017.

[36] S. Kim and M. Kojima, “Exact solutions of some nonconvex quadratic optimization
problems via sdp and socp relaxations,” Computational Optimization and Applications,
vol. 26, no. 2, pp. 143–154, 2003.

[37] G. Andersson, “Power system analysis,” Eidgenössische Technische Hochschule Zürich

(ETH), 2011.

[38] S. Frank, I. Steponavice, and S. Rebennack, “Optimal power flow: a bibliographic

survey i,” Energy Systems, vol. 3, no. 3, pp. 221–258, 2012.

[39] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and control, vol. 7.
McGraw-hill New York, 1994.

[40] W. Zhang, F. Li, and L. M. Tolbert, “Review of reactive power planning: objec-
tives, constraints, and algorithms,” IEEE transactions on power systems, vol. 22, no. 4,
pp. 2177–2186, 2007.

[41] Q. Jiang, H.-D. Chiang, C. Guo, and Y. Cao, “Power–current hybrid rectangular for-
mulation for interior-point optimal power flow,” IET generation, transmission & distri-
bution, vol. 3, no. 8, pp. 748–756, 2009.

[42] M. Abido, “Optimal power flow using particle swarm optimization,” International
Journal of Electrical Power & Energy Systems, vol. 24, no. 7, pp. 563–571, 2002.

[43] M. Modiri-Delshad, S. Koohi-Kamali, E. Taslimi, S. H. A. Kaboli, and N. Rahim, “Eco-

nomic dispatch in a microgrid through an iterated-based algorithm,” in Clean Energy
and Technology (CEAT), 2013 IEEE Conference on, pp. 82–87, IEEE, 2013.

81
[44] H. Bouchekara, M. Abido, and M. Boucherma, “Optimal power flow using teaching-
learning-based optimization technique,” Electric Power Systems Research, vol. 114,
pp. 49–59, 2014.

[45] I. Farhat and M. El-Hawary, “Optimization methods applied for solving the short-
term hydrothermal coordination problem,” Electric Power Systems Research, vol. 79,
no. 9, pp. 1308–1320, 2009.

[46] P. Gill, Electrical power equipment maintenance and testing. CRC press, 2008.

[47] K. Pandya and S. Joshi, “A survey of optimal power flow methods.,” Journal of Theo-
retical & Applied Information Technology, vol. 4, no. 5, 2008.

[48] J. Momoh, R. Koessler, M. Bond, B. Stott, D. Sun, A. Papalexopoulos, and P. Ris-

tanovic, “Challenges to optimal power flow,” IEEE Transactions on Power Systems,
vol. 12, no. 1, pp. 444–455, 1997.

[49] H. Wang and R. J. Thomas, “Towards reliable computation of large-scale market-

based optimal power flow,” in System Sciences, 2007. HICSS 2007. 40th Annual Hawaii
International Conference on, pp. 117–117, IEEE, 2007.

[50] J. M. Carrasco, L. G. Franquelo, J. T. Bialasiewicz, E. Galván, R. C. PortilloGuisado,

M. M. Prats, J. I. León, and N. Moreno-Alfonso, “Power-electronic systems for the grid
integration of renewable energy sources: A survey,” IEEE Transactions on industrial
electronics, vol. 53, no. 4, pp. 1002–1016, 2006.

[51] B. Stott and E. Hobson, “Power system security control calculations using linear pro-
gramming, part ii,” IEEE Transactions on Power Apparatus and Systems, no. 5, pp. 1721–
1731, 1978.

82
[52] R. E. Griffith and R. Stewart, “A nonlinear programming technique for the optimiza-
tion of continuous processing systems,” Management science, vol. 7, no. 4, pp. 379–392,
1961.

[53] X.-P. Zhang, C. Rehtanz, and B. Pal, Flexible AC transmission systems: modelling and
control. Springer Science & Business Media, 2012.

[54] R. Mota-Palomino and V. Quintana, “Sparse reactive power scheduling by a penalty

function-linear programming technique,” IEEE Transactions on Power Systems, vol. 1,
no. 3, pp. 31–39, 1986.

[55] O. Mangasarian, “Nonlinear programming: Sequential unconstrained minimization

techniques (Anthony V. Fiacco and Garth P. McCormick),” SIAM Review, vol. 12, no. 4,
p. 593, 1970.

[56] R. Burchett, H. Happ, and K. Wirgau, “Large scale optimal power flow,” IEEE Trans-
actions on Power Apparatus and Systems, no. 10, pp. 3722–3732, 1982.

[57] X. Lin, A. David, and C. Yu, “Reactive power optimisation with voltage stability con-
sideration in power market systems,” IEE Proceedings-Generation, Transmission and
Distribution, vol. 150, no. 3, pp. 305–310, 2003.

[58] G. Granelli and M. Montagna, “Security-constrained economic dispatch using dual

quadratic programming,” Electric Power Systems Research, vol. 56, no. 1, pp. 71–80,
2000.

[59] H. W. Dommel and W. F. Tinney, “Optimal power flow solutions,” IEEE Transactions
on power apparatus and systems, no. 10, pp. 1866–1876, 1968.

[60] R. Burchett, H. H. Happ, D. Vierath, and K. Wirgau, “Developments in optimal power

flow,” IEEE Transactions on Power Apparatus and Systems, no. 2, pp. 406–414, 1982.

83
[61] E. P. De Carvalho, A. dos Santos Junior, and T. F. Ma, “Reduced gradient method com-
bined with augmented lagrangian and barrier for the optimal power flow problem,”
Applied Mathematics and Computation, vol. 200, no. 2, pp. 529–536, 2008.

[62] J. Peschon, D. W. Bree, and L. P. Hajdu, “Optimal power-flow solutions for power
system planning,” Proceedings of the IEEE, vol. 60, no. 1, pp. 64–70, 1972.

[63] A. Sasson, F. Viloria, and F. Aboytes, “Optimal load flow solution using the hessian
matrix,” IEEE Transactions on Power Apparatus and Systems, no. 1, pp. 31–41, 1973.

[64] H. Happ, “Optimal power dispatch a comprehensive survey,” IEEE Transactions on

Power Apparatus and Systems, vol. 96, no. 3, pp. 841–854, 1977.

[65] G. Da Costa, C. Costa, and A. De Souza, “Comparative studies of optimization meth-

ods for the optimal power flow problem,” Electric Power Systems Research, vol. 56, no. 3,
pp. 249–254, 2000.

[66] J. Nocedal and S. Wright, Numerical Optimization. Springer, 2006.

[67] R. A. Jabr, “A primal-dual interior-point method to solve the optimal power flow dis-
patching problem,” Optimization and Engineering, vol. 4, no. 4, pp. 309–336, 2003.

[68] G. L. Torres and V. H. Quintana, “An interior-point method for nonlinear optimal
power flow using voltage rectangular coordinates,” IEEE transactions on Power Sys-
tems, vol. 13, no. 4, pp. 1211–1218, 1998.

[69] G. L. Torres and V. H. Quintana, “On a nonlinear multiple-centrality-corrections

interior-point method for optimal power flow,” IEEE Transactions on Power Systems,
vol. 16, no. 2, pp. 222–228, 2001.

[70] K. Karoui, L. Platbrood, H. Crisciu, and R. A. Waltz, “New large-scale security con-
strained optimal power flow program using a new interior point algorithm,” in Elec-

84
 tricity Market, 2008. EEM 2008. 5th International Conference on European, pp. 1–6, IEEE,
2008.

[71] S. Granville, J. Mello, and A. Melo, “Application of interior point methods to power
flow unsolvability,” IEEE Transactions on Power Systems, vol. 11, no. 2, pp. 1096–1103,
1996.

[72] A. El-Bakry, R. A. Tapia, T. Tsuchiya, and Y. Zhang, “On the formulation and theory of
the newton interior-point method for nonlinear programming,” Journal of Optimiza-
tion Theory and Applications, vol. 89, no. 3, pp. 507–541, 1996.

[73] X. Zhang, S. Petoussis, and K. Godfrey, “Nonlinear interior-point optimal power flow
method based on a current mismatch formulation,” IEE Proceedings-Generation, Trans-
mission and Distribution, vol. 152, no. 6, pp. 795–805, 2005.

[74] J. Lavaei, “Zero duality gap for classical opf problem convexifies fundamental non-
linear power problems,” in American Control Conference (ACC), 2011, pp. 4566–4573,
IEEE, 2011.

[75] J. Zhang, “On distributed optimization methods for solving optimal power flow prob-
lem over electricity grids,” 2016.

[76] X. Yao, “Study on DC arc faults in ring-bus DC microgrids with constant power
loads,” in Energy Conversion Congress and Exposition (ECCE), 2016 IEEE, pp. 1–5, IEEE,
2016.

[77] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex programming,
version 2.1.” http://cvxr.com/cvx, Mar. 2014.

[78] M. Grant and S. Boyd, “Graph implementations for nonsmooth convex programs,”
in Recent Advances in Learning and Control (V. Blondel, S. Boyd, and H. Kimura, eds.),

85
 Lecture Notes in Control and Information Sciences, pp. 95–110, Springer-Verlag Lim-
ited, 2008. http://stanford.edu/~boyd/papers/graph_dcp.html.

[79] K. M. Chandy, S. H. Low, U. Topcu, and H. Xu, “A simple optimal power flow model
with energy storage,” in Decision and Control (CDC), 2010 49th IEEE Conference on,
pp. 1051–1057, IEEE, 2010.

[80] B. Recht, M. Fazel, and P. A. Parrilo, “Guaranteed minimum-rank solutions of lin-

ear matrix equations via nuclear norm minimization,” SIAM review, vol. 52, no. 3,
pp. 471–501, 2010.

[81] C. Grussler and P. Giselsson, “Low-rank inducing norms with optimality interpreta-
tions,” arXiv preprint arXiv:1612.03186, 2016.

86