Particle Swarm Optimization Based Reactive Power Dispatch For Pow

University of Denver
Digital Commons @ DU
Electronic Theses and Dissertations Graduate Studies
1-1-2015
Particle Swarm Optimization Based Reactive Power Dispatch for

Power Networks with Distributed Generation
Xiao Kou
Follow this and additional works at: https://digitalcommons.du.edu/etd
Part of the Electrical and Computer Engineering Commons
Recommended Citation
Kou, Xiao, "Particle Swarm Optimization Based Reactive Power Dispatch for Power Networks with
Distributed Generation" (2015). Electronic Theses and Dissertations. 1035.
https://digitalcommons.du.edu/etd/1035
This Thesis is brought to you for free and open access by the Graduate Studies at Digital Commons @ DU. It has
been accepted for inclusion in Electronic Theses and Dissertations by an authorized administrator of Digital
Commons @ DU. For more information, please contact jennifer.cox@du.edu,dig-commons@du.edu.
PARTICLE SWARM OPTIMIZATION BASED REACTIVE POWER DISPATCH
FOR POWER NETWORKS WITH DISTRIBUTED GENERATION
__________
A Thesis
Presented to
The Faculty of Daniel Felix Ritchie School of Engineering and Computer Science
__________
In Partial Fulfillment
of the Requirements for the Degree
Master of Science
__________
by
Xiao Kou
August 2015
Advisor: Dr. David Wenzhong Gao

©Copyright by Xiao Kou 2015
All Rights Reserved

Author: Xiao Kou
Title: Particle Swarm Optimization Based Reactive Power Dispatch for Power Networks
with Distributed Generation
Advisor: Dr. David Wenzhong Gao
Degree Date: August 2015
Abstract
Reactive power is critical to the operation of the power networks on both safety
aspects and economic aspects. Unreasonable distribution of the reactive power would
severely affect the power quality of the power networks and increases the transmission
loss. Currently, the most economical and practical approach to minimizing the real power
loss remains using reactive power dispatch method.
Reactive power dispatch problem is nonlinear and has both equality constraints
and inequality constraints. In this thesis, PSO algorithm and MATPOWER 5.1 toolbox
are applied to solve the reactive power dispatch problem. PSO is a global optimization
technique that is equipped with excellent searching capability. The biggest advantage of
PSO is that the efficiency of PSO is less sensitive to the complexity of the objective
function. MATPOWER 5.1 is an open source MATLAB toolbox focusing on solving the
power flow problems. The benefit of MATPOWER is that its code can be easily used and
modified.
The proposed method in this thesis minimizes the real power loss in a practical
power system and determines the optimal placement of a new installed DG. IEEE 14 bus
system is used to evaluate the performance. Test results show the effectiveness of the
proposed method.
ii
Acknowledgements
I would like to express my sincere gratitude to my advisor Prof. David Wenzhong
Gao, for his selfless support during my two years study at the University of Denver. Dr.
Gao’s rigorous attitude towards scholarship and humility deeply impressed me and will
benefit me for life.
I am also grateful to the committee chair Prof. Caroline Li and committee
members, Prof. Jason Jun Zhang, and Prof. Amin Khodaei. Thank you for your time in
checking the thesis draft and giving your valuable suggestions.
My thanks also go to all my lab mates in the Renewable Energy and Power
Electronics Lab at the University of Denver. Your help and support are greatly
appreciated.
Last but not least, I would like to thank my family, for your unconditional love
and continuous support.
iii
Table of Contents
Chapter One: Introduction .................................................................................................. 1

1.1 Background and Motivation ....................................................................... 1
1.2 Reactive Power Compensation Techniques ................................................ 2
1.3.1 Single Power Factor Correction ......................................................... 3
1.3.2 Group Power Factor Correction ......................................................... 4
1.3.3 Bulk Power Factor Correction ........................................................... 5
1.3 Literature Review........................................................................................ 6
1.4 Expected Contribution ................................................................................ 9
Chapter Two: Global Optimization Methods ................................................................... 11

2.1 Simulated Annealing ................................................................................. 12
2.2 Genetic Algorithm .................................................................................... 15
2.2.1 Selection........................................................................................... 16
2.2.2 Crossover ......................................................................................... 16
2.2.3 Mutation ........................................................................................... 17
2.3 Particle Swarm Optimization .................................................................... 18
2.3.1 Difference between GA and PSO .................................................... 22
2.3.2 PSO Parameters Selection................................................................ 22
Chapter Three: Reactive Power Dispatch Problem Formulation ...................................... 23

3.1 Objective Function .................................................................................... 24
3.2 Constraints ................................................................................................ 24
3.2.1 Equality Constraints ......................................................................... 24
3.2.2 Inequality Constraints ...................................................................... 25
3.3 Exterior Penalty Function (EPF) Method ................................................. 26
3.4 MATPOWER............................................................................................ 27
3.4.1 The loadcase Function ..................................................................... 28
3.4.2 The savecase Function ..................................................................... 28
3.4.3 The runpf Function .......................................................................... 28
3.4.4 The get_losses Function................................................................... 29
3.4.5 Canonical Form of the Generator Information ................................ 29
3.4.6 Canonical Form of the Branch Information ..................................... 30
3.4.7 Canonical Form of the Bus Information .......................................... 32
3.5 Procedures of the PSO Based Reactive Power Dispatch .......................... 33
Chapter Four: Case Studies ............................................................................................... 36

4.1 IEEE 14 Bus System Data ........................................................................ 36
4.2 Reactive Power Dispatch without New DGs ............................................ 39
4.3 Reactive Power Dispatch with a New DG Operating at Rated Power ..... 42
4.4 Reactive Power Dispatch with a New DG Operating at Various Power .. 60
4.5 Discussion ................................................................................................. 68
Chapter Five: Conclusion and Future Work ..................................................................... 69

5.1 Conclusion ................................................................................................ 69
iv
5.2 Future Work .............................................................................................. 70
References ......................................................................................................................... 71
Appendix A MATLAB Code for Reactive Power Dispatch without New DGs ............ 76
Appendix B MATLAB Code for Reactive Power Dispatch with a New DG Installed on
PQ Bus .............................................................................................................................. 82
v
List of Tables
Table 2.1 SA Terminologies ..................................................................................... 12
Table 2.2 GA Terminologies .................................................................................... 15
Table 2.3 PSO Parameters Selection ........................................................................ 22
Table 3.1 Explanation of the Generator Name Columns .......................................... 30
Table 3.2 Explanation of the Branch Name Columns .............................................. 31
Table 3.3 Explanation of the Bus Name Columns .................................................... 33
Table 4.1 IEEE 14 Bus System Loads Parameters ................................................... 37
Table 4.2 IEEE 14 Bus System Generators Parameters ........................................... 37
Table 4.3 IEEE 14 Bus System Branches Parameters .............................................. 38
Table 4.4 IEEE 14 Bus System Reactive Power Injection Parameter ...................... 39
Table 4.5 Comparison of the Real Power Loss at Each Branch ............................... 41
Table 4.6 Optimization Results with a New DG Installed on Different Buses ........ 56
Table 4.7 Real Power Losses Comparisons .............................................................. 58
Table 4.8 Optimization Results when DG Deliver 25% Rated Power ..................... 62
Table 4.9 Optimization Results when DG Deliver 50% Rated Power ..................... 64
Table 4.10 Optimization Results when DG Deliver 75% Rated Power ................... 66
Table 4.11 Comparison of the Phase Angle Shift ..................................................... 67
vi
List of Figures
Figure 1.1 Five Bus System without Power Factor Correction .................................. 3
Figure 1.2 Single Power Factor Correction ................................................................ 4
Figure 1.3 Group Power Factor Correction ............................................................... 5
Figure 1.4 Bulk Power Factor Correction .................................................................. 5
Figure 2.1 Local Minimum Value versus Global Minimum Value .......................... 13
Figure 2.2 Flow Chart of the Simulated Annealing .................................................. 14
Figure 2.3 Selection Operation ................................................................................. 16
Figure 2.4 Crossover Operation ................................................................................ 16
Figure 2.5 Mutation Operation ................................................................................. 17
Figure 2.6 Flow Chart of the Genetic Algorithm ...................................................... 18
Figure 2.7 Flow Chart of the Particle Swarm Optimization ..................................... 21
Figure 3.1 Canonical Form of the IEEE 14 Bus System Generator Data ................. 29
Figure 3.2 Canonical Form of the IEEE 14 Bus System Branch Data ..................... 31
Figure 3.3 Canonical Form of the IEEE 14 Bus System Bus Data........................... 32
Figure 3.4 Flow Chart of the PSO Based Reactive Power Dispatch ........................ 35
Figure 4.1 IEEE 14 Bus System ............................................................................... 36
Figure 4.2 Coordinates of the Particle ...................................................................... 39
Figure 4.3 Loss Reduction Process .......................................................................... 40
Figure 4.4 Reactive Power Capacity of Enercon E82 (2010 FACTS-WT) .............. 43
Figure 4.5 Reactive Power Capacity of Enercon E66 (2002) ................................... 43
Figure 4.6 Generator Data of the Modified System when the DG is Installed on Bus
2......................................................................................................................... 44
3......................................................................................................................... 44
vii
6......................................................................................................................... 45
8......................................................................................................................... 45
Figure 4.10 Coordinates of the Particle when the New DG is connected to a PV Bus
........................................................................................................................... 45
Figure 4.11 Coordinates of the Particle when the New DG is on a PQ Bus ............. 46
Figure 4.12 Bus Data of the Modified System when the DG is Installed on Bus 4 . 46
4......................................................................................................................... 47
5......................................................................................................................... 48
7......................................................................................................................... 49
9......................................................................................................................... 50
Figure 4.20 Bus Data of the Modified System when the DG is Installed on Bus 10 50
10....................................................................................................................... 51
Figure 4.22 Bus Data of the Modified System when the DG is Installed on bus 11 51
11....................................................................................................................... 52
12....................................................................................................................... 53
viii
13....................................................................................................................... 54
14....................................................................................................................... 55
Figure 4.30 Comparison of the Loss Reduction ....................................................... 58
Figure 4.31 Loss Reduction Process when the new DG is Installed on Bus 3 ......... 60
Figure 4.32 Real Power Output Curve of Enercon E82............................................ 61
Figure 4.33 Generators Data when the New Installed DG Delivers 25% Rated Power
........................................................................................................................... 61
Figure 4.34 Loss Reduction Process when DG Delivers 25% Rated Power ............ 62
........................................................................................................................... 63
........................................................................................................................... 65
ix
Chapter One: Introduction
1.1 Background and Motivation
Reactive power is critical to the operation of the power networks on both safety
aspects and economic aspects. Rational reactive power dispatch scheme can improve the
power quality as well as reduce the real power loss. On the contrary, if the reactive power
is unreasonably allocated, then it will bring great economic losses and might even
threaten the security of the power grid.
It has been proved that the New York Blackout in 1977 and the Tokyo Blackout
in 1987 were both caused by the deficiency of reactive power during the peak load hours
[1]. These blackouts brought social disruptions and hundreds of millions of dollars in loss.
On August 14, 2003, the Northeast Blackout affected about 55 million people in the mid-
western part of the United States and Ontario province in Canada. One of the main
reasons that causing this blackout is also due to the reactive power shortage [2-3].
Reactive power also plays a prominent role in minimizing the real power loss of
the power networks. Reactive power dispatch approach can significantly reduce the
power factor angle of each bus, thus cutting the overall energy losses. Each year, a large
amount of electricity is wasted on the transmission or distribution lines around the world.
According to the estimations from the U.S. Energy Information Administration, the
annual transmission and distribution losses in the United States can reach as much as 6%
[4]. Moreover, most of this loss occurs at the distribution level. This real power loss not
1
only causes energy waste and produces extra carbon emission, but also increases the
generation cost.
Along with the development of the economic, the scale of the power grid also
keeps growing. In some areas, however, the construction and upgrading of the power grid
did not keep pace with the growth of the loads. Then a severe shortage of the reactive
power would appear. For the purpose of minimizing the real power loss, utility
companies can either change the structure of the power grid or replace the old wiring
with lower impedance lines. However, both of these methods requires investing large
amounts of money. The simplest and most economical way remains reactive power
dispatch method. In the early days, the starting point of reactive power dispatch is to
improve the power factor at each end user by installing reactive power compensators.
This approach, of course, can reduce the total power loss. But in order to get the
maximum profit, electricity grid designers have to take a more holistic view and calculate
the power flow.
1.2 Reactive Power Compensation Techniques
There are many reactive compensation techniques. Fig. 1.1 shows a simple five
bus system without using any reactive power compensators. This model is developed in
PowerWorld Simulator 17 [5]. The real and reactive power output of the generators are
304 MW and 129 MVar, respectively. Since there is no capacitor in this system, the
generator will take the whole burden of both real power loads and reactive power loads.
2
Running the simulation, a 4.27 MW real power loss and 8.55 MVar reactive power loss
can be achieved.
Figure 1.1 Five Bus System without Power Factor Correction
In order to reduce the power loss, reactive power dispatch techniques can be
employed. The easiest way to compensate the reactive power is to connect the capacitors
in parallel with the loads. This approach can be further divided into single power factor
correction, group power factor correction, and bulk power factor correction [6].
1.3.1 Single Power Factor Correction
In single power factor correction model, each load has a shunt capacitor. The
capacitor and the load it serves share the same switch, so no extra control devices are
needed in this scheme. When the capacitors inject reactive power, both transmission loss
and voltage drop will decrease. In this case, the real power loss is 3.84 MW and the
reactive power loss is 7.67 MVar. The disadvantage of this method is that the shunt
3
capacitors are not fully utilized all the time. Since the load and its shunt capacitor use the
same switch, the capacitor will not compensate reactive power if the load is turned off.
Figure 1.2 Single Power Factor Correction
1.3.2 Group Power Factor Correction
To overcome the defect of single power factor correction method, a more
effective way called group power factor correction was proposed. The PowerWorld
model for group power factor correction method is shown in Fig. 1.3.
In this method, instead of just compensating only one load, one capacitor could
handle a group of loads. Also, the reactive power injection are controlled by a
microprocessor based on the real-time reactive power demand. In the PowerWorld model,
the active power loss is 3.96 MW and the reactive power loss is 7.91 MVar, which are
slightly higher than those in single power factor correction method. However, the
efficiency of the shunt capacitors is greatly improved.
4
Figure 1.3 Group Power Factor Correction
1.3.3 Bulk Power Factor Correction
The third power factor correction method is called bulk power factor correction,
as depicted in Fig. 1.4. The shunt capacitor bank is in charge of the whole system, and it
is directly connected to the PV bus.
Figure 1.4 Bulk Power Factor Correction
5
Since most of the loads in the electric power systems are inductive loads, they
will consume large amounts of reactive power. The reactive power has to be obtained
from somewhere in the network. If all the reactive power is produced from one place,
then the real power loss will be enormous. This conclusion is demonstrated in Fig. 1.4,
both real power loss and reactive power loss is much greater than the previous two
schemes. Therefore, the principle of reactive power dispatch is compensating reactive
power at where the loads consume.
1.3 Literature Review
In this thesis, the approach of reactive power dispatch is to adjust the values of
control variables and find the optimal placement of new installed DG. Since both
objective function and equality constraints are nonlinear, the main emphasis is on
managing the nonlinear function problem with mixed discrete control variables.
The first reactive power dispatch method was suggested by N. M. Neagle and D.
R. Samson in 1956. In [7], Neagle and Samson analyzed two types of load models. In the
first model, the loads are equally distributed along the feeders. However, this type of
system is an idealized model, and it does not exist in real life. So in their second model,
they assume the magnitudes of the loads in a feeder are proportional to their distances
from the substation. The shortcoming of this method is that there is only one generator in
their models. Thus, this method is not suitable for those power grids with distributed
generation.
6
After that, interior-point method (or barrier method) is introduced to solve the
reactive power dispatch problems. The interior-point method belongs to linear
programming. The principle of linear programming is to expand the nonlinear functions
and constraints into Taylor’s series expansions. Only the first-order terms and the
constant terms need to be considered. The advantage of this method is linear
programming theory is mature and the computation time is short. Nonetheless, since the
objective function of reactive power dispatch problem is not convex, many local optima
exist. Therefore, the linear programming method is very likely to get trapped into one of
these local optima and cannot achieve a global optimal solution. Moreover, linear
programming ignores the higher-order terms, so the accuracy of the results can also be
affected.
The quadratic programming method is another method that can be used to solve
the reactive power dispatch problem [8]. Quadratic programming is more adaptable to the
nonlinear characteristic of reactive power dispatch problem than linear programming.
Furthermore, its nice convergence characteristic is also helpful. The disadvantage is that
quadratic programming does not work very well for the high dimensional problems. As
the dimension increase, the computation time would increase dramatically.
Since the 1990s, heuristic methods like Simulated Annealing (SA) and Genetic
Algorithm (GA) are getting more and more capable and grab plenty of attentions [9].
Simulated annealing imitates the heating and cooling process of the metal, and it
was suggested by Metropolis in 1953. SA use random search and iteration methods to
obtain the optimal solution. In metallurgy, the goal of annealing is to get the best metallic
7
crystal. When the cooling process is finished, the energy of the material becomes lowest.
This process is similar to optimizing the reactive power. The goal of reactive power
dispatch in this thesis is to get the minimum real power loss. The power loss will be
reduced as the iterations proceed. The disadvantage of SA is that the parameters of
simulated annealing need to be carefully chosen. Inappropriate parameters selection
would greatly increase the computation time. T. Sousa et.al put forward a modified
simulated annealing method to search for the optimal size and placement of compensators
in IEEE 14 buses system [10]. In [11], M. Gitizadeh proposes another fuzzy-based
reactive power dispatch method by using simulated annealing.
The genetic algorithm is based on mimicking the evolutionary process, and it was
presented by J. Howard in 1975. Genetic algorithm is totally different from the traditional
optimization methods. The feasible solutions in GA are compared to chromosomes.
Selection, crossover and mutation operation are repeatedly conducted to propagate better
individuals in the next generation. GA provides a framework for solving the nonlinear
multi-objective complex problems, and it has already been used in many areas, such as
control, signal processing, robotics, and economics. The advantages of GA are its
practicability, high-efficiency, and robustness. A genetic algorithm toolbox GAOT
(Genetic Algorithm Optimization Toolbox) was developed by North Carolina State
University and posted for free online. The disadvantage of GA is its prematurity and
diversity problems. In [12], GA was proposed to solve the reactive power devices
placement in IEEE 14 buses system. Another reactive power optimization by using GA is
performed in IEEE 34 buses system in [13].
8
The artificial neural network is an algorithm based on the neural networks of
animals [14]. The biggest advantage of artificial neural networks is its self-teaching
ability. For instance, given a certain amount of data, artificial neural networks algorithm
can gradually identify similar data through self-teaching. This benefit has important
significance on forecasting. Paper [15] presents a combined artificial neural networks and
fuzzy sets to solve the optimal reactive power control problems. In [16], the artificial
neural network algorithm is used on the online optimal shunt capacitors dispatch to
reduce the input variables and to get better computation speed.
To sum up, each algorithm has its advantages and disadvantages. Therefore, local
conditions need to be considered in choosing the appropriate algorithm for solving the
reactive power dispatch problems in different power networks.
1.4 Expected Contribution
This thesis summarizes the status of reactive power dispatch and compares
different global optimization methods. A modified MATPOWER code utilizing particle
swarm optimization algorithm is developed to solve the reactive power dispatch problem
in the power systems.
The expected contribution of this thesis mainly includes the following aspects:
1. Applying particle swarm optimization algorithm to adjust the values of control
variables (voltage magnitudes, tap positions, and shunt capacitance) in the power
networks to minimize the real power loss.
9
2. Identifying the optimal placement of a new installed distributed generator in an
existing power system.
3. Introducing MATPOWER 5.1 toolbox to calculate the power flow and manage
the equality constraints in the reactive power dispatch problems.
10
Chapter Two: Global Optimization Methods
In most cases, the objective functions in nonlinear optimization problems are not
convex. Traditional optimization methods (such as gradient-based approaches) can only
find local optimal values. Moreover, the results from traditional optimization methods
often have strong connections with the initial guess. To overcome these problems, global
optimization methods are suggested in this thesis.
Global optimization methods can only guarantee to achieve acceptable solutions.
Usually, finding the global optimal results will take plenty of time and resources.
Sometimes it is not profitable to do so. If the improvement is insignificant, then it is
probably a bad deal to take the time to find the global optimal solution. Therefore, if the
result is very close to the global optimal solution, it can be viewed as an acceptable
solution.
In global optimization methods, some concessions have to be made (for instance,
increasing their objective function values in some iterations) to allow potential solutions
to escape from the local optimum. Most of the time, there is no way to determine if a
global optimal value is already achieved or not, so global optimization methods usually
need to take plenty of iterations without bias. This requirement will in turn force the
scheme of the global optimization methods to be as simple as possible.
In the following sections of this chapter, three of the most representative heuristic
global optimization methods (simulated annealing, genetic algorithm, and particle swarm
11
optimization) are introduced. These three global optimization methods can be easily
programmed and are well suited for solving reactive power dispatch problems.
2.1 Simulated Annealing
Simulated annealing (SA) was proposed by Kirkpatrick in 1983. In order to
change the physical properties of the metal materials and increase the ductility, the metal
is first heated to its melting temperature and then cooled down [17]. The optimization
process of simulated annealing resembles the process of annealing in metallurgy. Table
2.1 lists some technical terms of simulated annealing and their corresponding annealing
words in metallurgy [18].
Table 2.1 SA Terminologies
Simulated Annealing Terms Annealing Terms in Metallurgy
Feasible solutions States of the system
Cost Energy
Control variables Temperature
Final Solution Frozen state
The initial design vectors of simulated annealing are randomly selected. Uphill
moves are occasionally allowed in order to escape from the local optimum value. For
example, searching for the minimum value in the curve of f(x) = x ∙ sin(x) within the
range of xϵ[−6, 6], as portrayed in Fig. 2.1. Suppose the initial guess position is at B.
Since traditional optimization methods only permit downhill moves, the next step may be
C, then D, and finally converges at A. While in simulated annealing, uphill moves are
12
allowed in some iterations, so the solution is possible to jump from C to F and eventually
find the another local optimal value at E. This sequence of event is repeated until SA
finds to the global optimum.
x sin(x)
F
2 B
1
D C
0
A
-1
-2
-3
-4
-5
E
-6 -4 -2 0 2 4 6
x
Figure 2.1 Local Minimum Value versus Global Minimum Value
The stopping criteria of SA vary for different problems. The optimization process
could be stopped if a given minimum value is obtained, a certain number of iterations are
conducted, or no obvious improvement is achieved after some iterations.
A flow chart of the simulated annealing is shown in Fig. 2.2. The selection of
parameter β is the crux of the simulated annealing algorithm. β determines the acceptance
rate of uphill steps and it is related to the Boltzmann probability distribution and the
13
temperature. If the value of β is too big, the potential solution may not have enough
energy to escape from the local optimum. Conversely, if β is too small, the solution will
wander all over the searching space.
Choose starting design X0
Calculate the fitness value f0
Randomly select a point on the surface of a unit

n-dimensions to get a search direction S
Use step size a,

Calculate f1=f(x0+aS)
Calculate △f=f1-f0
△f≤0 yes p=1
No
p=e^(-β×△f)
r=rand()
Design vector remain

r≤p No
unchanged
Yes
The step is accepted and the

design vector is updated
Figure 2.2 Flow Chart of the Simulated Annealing
14
2.2 Genetic Algorithm
The idea of genetic algorithm (GA) comes from imitating natural evolutionary
processes. Chromosomes are equivalent to the solutions, the initial population
corresponds to the initial design vectors of the first generation, and the fitness value
represents the evaluation of the solutions. Two types of genetic operators, crossover, and
mutation, define the methods of generating new populations. Immigrants are randomly
generated population to keep the diversity of the group [19].
Table 2.2 GA Terminologies
Biological Genetic Algorithm
Chromosome Possible solutions
Population A group of solutions
Allele Piece of the design vector
Locus The position of the allele in chromosome
Fitness Evaluation of possible solutions
Crossover Allele exchange between parents
Replacement of a random element in design

Mutation
vector with a random value
Three of the most crucial operations in the genetic algorithm are selection,
crossover, and mutation. The details of these three operations are introduced as below.
15
2.2.1 Selection
The goal of the selection is to select the “well-behaved” individuals from the
current population and to give them more opportunities to reproduce their children.
Figure 2.3 Selection Operation
The selection operation is based on the fitness value of the individuals with
specific standards. The standards vary from different problems. Those solutions that are
more accord with the standards will have a higher probability to be selected, but not
decisive. The theory of selection reflects the principle of Darwin's survival of the fittest.
2.2.2 Crossover
Crossover is another important genetic algorithm operator. Most of the new
individuals in the next generation are generated through crossover operation. The child
chromosome will inherit characteristics from both of its parents.
+
Figure 2.4 Crossover Operation
16
Every individual has a chance to randomly crossover with other individuals
within a population. For a single individual, the probability of exchanging part of its
chromosome with other individuals is called crossover rate. Papers [20-21] suggest that
using more than two parent chromosomes to participate in crossover operation would
achieve a better solution. Crossover operation embodies the idea of information exchange.
2.2.3 Mutation
Mutation operation randomly selected a group of individuals at the beginning of
the search. The selected individuals would have a certain probability of changing one or
more gene values in its chromosome. The mutation rate is tiny. However, it is crucial to
keeping the diversity of the population.
Figure 2.5 Mutation Operation
After these three operations, a new generation of chromosomes will be produced.
Since the “well-behaved” chromosomes have more chance to breed their children
chromosomes, the chromosomes in the new generations are expected to move towards
the best solutions [22]. Although the relatively “bad-behaved” chromosomes may be less
accord with the standards, they still contribute to the diversity of the groups. A flow chart
of the genetic algorithm is illustrated in Fig. 2.6.
17
Initial population
Chromosome coding
Calculate the fitness value of each

member in the current population
Select parents based on the fitness value
Produce children chromosomes through

mutation and crossover operations
No
Use the current children chromosomes to

produce the next population
Stopping criteria met
Yes
Decoding
Figure 2.6 Flow Chart of the Genetic Algorithm
2.3 Particle Swarm Optimization
Particle Swarm Optimization (PSO) was developed by J. Kennedy and R.
Eberhart in 1995 [23]. It was originally used for solving continuous nonlinear functions.
The idea of PSO comes from a simplified social system like bird flocking or fish
schooling.
18
Imagine a group of birds is searching for food in an n-dimension area (n equals
the number of control variables). None of these birds knows where the food is. However,
they know which bird is nearest to the food (assume the closest bird to the food is Bird A).
The best strategy for the rest of birds to find the food is following Bird A and searching
its neighboring area.
In PSO, each single solution (particle) can be viewed as a "bird". The position of
each particle can be expressed as 𝑥𝑥𝑖𝑖 = (𝑥𝑥𝑖𝑖1 , 𝑥𝑥𝑖𝑖2 , … , 𝑥𝑥𝑖𝑖𝑖𝑖 ). The initial solutions in PSO are
randomly selected and then PSO will continually search for optimal value by updating
the solutions in each iteration. The fitness value of the particle is related to the objective
function. And the velocity of the particles 𝑣𝑣𝑖𝑖 = (𝑣𝑣𝑖𝑖1 , 𝑣𝑣𝑖𝑖2 , … , 𝑣𝑣𝑖𝑖𝑖𝑖 ) is related to its pervious
velocity, global best known position, and local best known position. The velocity
indicates the directions of all the particles in the next iteration. The local best known
position is the best solution that achieved by each particle so far. The global best known
position is the best solution among all the achieved solutions. The inertia velocity part,
local best known position part, and global best known position part of the velocity reflect
the cooperation and competition mechanism in PSO.
Similar to GA, PSO also starts with a group of randomly generated solutions and
updates the solutions in each iteration. However, PSO uses historical data rather than
does crossover and mutation operations. The behavior of all the particles appears to be
managed by a control center. However, in reality, as formula 2.1 and formula 2.2
describe below, the principle of the PSO algorithm is quite straightforward.
v d +1 = k * ( w * v d + ϕ1 ⋅ rand () * ( p Best − x d ) + ϕ 2 ⋅ rand () * ( g Best − x d )) (2.1)
19
x d +1 = x d + v d +1 (2.2)
where w is the inertia weight factor,
ϕ1 and ϕ 2 are acceleration factors,
rand () is a random value between 0 and 1.
k is the constriction factor.
The acceleration factors handle the step sizes of the particles in the next iteration.
If the acceleration factors are too small, the particles may not have enough velocity to
reach the target regions. If the acceleration factors are too big, the particles may fly over
the optimal value. Appropriate selection of acceleration factors could avoid trapping into
local minimal and reduce the computation time.
Vmax limits the maximum velocity of each particle. If the velocity of a particle is
greater than maximum allowable velocity, then the velocity of that particle will be limited
to Vmax. Otherwise, the particle may also fly over the optimal solution. The maximum
velocity is specified by users depending on different problems.
The advantages of PSO is summarized in [24]:
1. PSO choose the directions of next step by cooperation and competition.
2. Fewer parameters need to be set compared to simulated annealing method and
genetic algorithm method.
3. The computation speed of PSO is less sensitive to the complexity of the
objective functions.
4. PSO algorithm applies to many fields.
The flow chart of the PSO algorithm is shown in fig. 2.7.

20
Randomly generate the location of each particle
Randomly generate the velocity of each particle
Evaluate the objective function value at each

particle
Set Gbest and Pibest
Update the velocity of each particle
Vi≤Vimax Yes Vi=Vimax
No
Xd+1=Xd+Vd
No

particle
Vali≤Val_Pibest Yes Val_Pibest=Vali, Pibest=Pi+1
No
Gbest=min(Pibest)
Yes
Results
Figure 2.7 Flow Chart of the Particle Swarm Optimization

21
2.3.1 Difference between GA and PSO
In genetic algorithm, the communication scheme is bidirectional, chromosomes
could share information with each other. However, the communication scheme is one-
way in PSO. Only the global best-known position could send its information to the other
particles. Another difference is that the concepts in PSO are very clear. PSO does not
need to do encoding and transform the original solution into binaries, and then do
decoding at the end of the search, thus making it much easier to understand.
2.3.2 PSO Parameters Selection
The selection of the PSO parameters for general problems is listed in Table 2.3.
Programmers may change some of these parameters based on different problems.
Table 2.3 PSO Parameters Selection
20-40 works well for most of the optimization problems.
Particle size However, as the dimension increase, the number of the
particle should also increase according.
Dimension of the particles Equals the number of control variables.
Domains of the particles Depends on the upper bound and lower bound constraints
Acceleration factor 2 ≤ ϕ1 = ϕ2 ≤ 4
Iteration number
Difference between the current best solution and the

Stopping criteria
previous best solution
No improvement after a certain number of iterations
22
Chapter Three: Reactive Power Dispatch Problem Formulation
The loads in the power system keep changing all the time. In order to maintain the
power system operating at the timely optimum state, the reactive power optimization
need be continuously conducted in theory. However, frequent switching operations are
not feasible in the practical application. These operations will not only bring extra
workload to the operator of the network, but also accelerate the aging of the equipment in
the power systems. Sometimes the frequent switching operations may even threaten the
safety operation of the network. Therefore, the number of switching operations and tap
positions changing operations are strictly limited.
Most of the existing models convert the dynamic model into the static model. [25-
26] suggest to divide a whole day into several intervals and then further divide each
interval into several periods. Within each of these periods, the discrete control variables
remain constant. Only the continuous control variables keep changing to reduce the
power loss. The minimum real power loss during a day is set as the optimization object.
The advantage of this method is that it can reduce the total power loss of the system
while significantly decrease the number of switching operations.
Relying on the load forecasting and wind speed prediction information, grid
operators can obtain the solutions of the reactive power dispatch at different wind
conditions in advance, and then match these solutions with the real situations to minimize
the real power loss.
23
3.1 Objective Function
Reactive power dispatch in power systems may have different goals. It can be
minimizing the real power loss, getting the best voltage quality, using minimum
capacitors or achieving maximum economic profit. In this thesis, the goal of the reactive
power dispatch is to get the minimum real power loss.
The real power loss of the system equals the sum of the real power loss on each
branch, and it can be described as:
( )
N
f : Ploss = ∑ g ij Vi 2 + V j2 − 2ViV j cos θ ij (3.1)
k =1
where N is the number of the branches,
𝑔𝑔𝑖𝑖𝑖𝑖 is the conductance of the branch between bus i and bus j,
𝑉𝑉𝑖𝑖 is the voltage magnitude of bus i,
𝑉𝑉𝑗𝑗 is the voltage magnitude of bus j,
𝜃𝜃𝑖𝑖𝑖𝑖 is the difference of phase angle between bus i and bus j.
3.2 Constraints
Reactive power dispatch problem has both equality constraints and inequality
constraints to process.
3.2.1 Equality Constraints
The equality constraints are the power balance equations, which can be described
by the equations below:
h1 : Pgi − Pdi − Vi ∑ V j (Gij cos θ ij + Bij sinθ ij ) = 0 (3.2)

h2 : Q gi − Qdi − Vi ∑ V j (Gij sin θ ij − Bij cos θ ij ) = 0 (3.3)
24
where 𝑃𝑃𝑔𝑔𝑔𝑔 is the real power generation at bus i,
𝑃𝑃𝑑𝑑𝑑𝑑 is the real power demand at bus i,
𝑄𝑄𝑔𝑔𝑔𝑔 is the reactive power generation at bus i,
𝑄𝑄𝑑𝑑𝑑𝑑 is the reactive power demand at bus i.
3.2.2 Inequality Constraints
The inequality functions are the ranges of the voltage magnitudes, tap positions of
the transformers, and reactive power injection. Some of the parameters are continuous, as
the voltage magnitudes. While some are discrete, like the tap positions of the
transformers and reactive power injection. The commonly used method to manage the
discrete values is viewing them as continuous values at the beginning of the optimization
and then mapping the continuous values back to the discrete values in the end. In this
thesis, the discrete variables are seen as continuous variables initially and then keep three
decimal places at the end of the search.
g1 : Vi min < Vi < Vi max (3.4)

g 2 : t min
j < t j < t max
j (3.5)
g 3 : Q gimin < Q gi < Q gimax
(3.6)
Based on the original parameters in IEEE 14 bus system, the range of voltage
magnitude in this thesis is set from 0.95 p.u. to 1.10 p.u. The range of the tap position is
set from 0.975 to 1.025. The reactive power injection of the compensators is set between
0 MVar to 20 MVar.
25
3.3 Exterior Penalty Function (EPF) Method
Reactive power dispatch problem is a constrained problem. In optimization, the
constrained problems are usually converted into unconstrained problems for convenience.
One of the commonly used methods to convert the constrained problem is adding exterior
penalty function terms to the objective function [27], which is also known as exterior
penalty function method, as represent in the formula 3.7.
Minimize: F : f + P X , rh , rg( ) (3.7)

xil ≤ xi ≤ xiu , i = 1,2,..., n
( )
where P rh , rg is the penalty function,
rh is the penalty multiplier for the equality constraint.
rg is the penalty multiplier for the inequality constraint.
F is called the augmented function. [28]
The equality constraint in this thesis will be automatically fulfilled by using
MATPOWER 5.1 toolbox, so only inequality constraints need to be concerned. Therefore,
the final objective function could be described as:
(
F = Ploss + ∑ rgi Vi − Vi lim ) + ∑ r (T
2
Ti i ) ( )
− Ti lim + ∑ rQi Qi − Q ilim ,
2
(3.9)
where

Vi ;Vi > Vi
max max
Vi =  min
lim
(3.10)
Vi ;Vi < Vi min

Ti max ;Ti > Ti max
Ti =  min
lim
(3.11)
Ti ;Ti < Ti min
Qimax ; Qi > Qimax
QGi =  min
lim
(3.12)
Qi ; Qi < Qimin
26
In EPF, if all the control variables are within the limits, the penalty function terms
would be zero. On the contrary, if the control variables exceed the limits, then the penalty
function terms would be added to the objective function to penalize the violation. The
penalty multipliers are always assigned big numbers in programming. When the penalty
multipliers keep increasing until approaching infinity, the constrained problem will
transform to the unconstrained problem. In reactive power dispatch, if the control
variables exceed the voltage limit, significant damages to the power systems would occur.
So the voltage magnitudes, tap positions, and reactive power injection have to be
carefully examined.
3.4 MATPOWER
In this thesis, MATPOWER 5.1 toolbox is introduced to calculate the power flow
and to fulfill the equality constraints.
MATPOWER is a pack of MATLAB M-files that is developed by Ray D.
Zimmerman, Carlos E. Murillo-Sánchez and Deqiang Gan in 1996 to meet the
computational requirements of the PowerWeb project [29]. In order to install the
MATPOWER 5.1 toolbox, MATLAB version 7 or later is suggested as a system
requirement. The biggest advantages of MATPOWER is its easiness to use and modify
the original code. Furthermore, MATPOWER is open source and posted for free, users
can download the toolbox at:
http://www.pserc.cornell.edu/matpower/
In this section, several useful input and output MATPOWER functions related to
this thesis are presented.
27
3.4.1 The loadcase Function
The loadcase function can load the case information from the struct, M-file or
MAT-file. The imported information is then saved in a struct. Users can change the
structure of the network by modifying the imported data when needed.
The standard format of using loadcase is: mpc=loadcase(casefile)
3.4.2 The savecase Function
The savecase function can save the information of the network to M-file or MAT-
file. These files can also be overwritten in case of need. In MATLAB 7.10 environment,
if the case file needs to be overwritten more than once in a single run, users need to
choose saving the case information in MAT-format. Otherwise, an error message would
appear, and the case information would remain unchanged.
The standard format of using savecase is: savecase(fname, mpc).
3.4.3 The runpf Function
The runpf function can calculate the power flow of the network. When calculating
the power flow, the runpf function has several different options. ‘NR’ refers to using
Newton’s method, ‘FDXB’ is the fast decoupled method, and ‘GS’ means using Gauss-
Seidel method. ‘AC’ is calculating the AC power flow of the system, and ‘DC’ is
calculating the DC power flow of the system. By default, runpf works at the AC power
flow mode and uses Newton Raphson’s method to compute the power flow.
The standard format of using runpf is: results = runpf(casedata).
28
3.4.4 The get_losses Function
The get_losses function can calculate the reactive power injection and power loss
in all branches, by using the following formulas.

𝑣𝑣𝑓𝑓
| 𝑗𝑗𝑗𝑗 −𝑣𝑣𝑡𝑡 |2
𝜏𝜏𝑒𝑒 𝑠𝑠ℎ𝑖𝑖𝑖𝑖𝑖𝑖
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖 = (3.13)
𝑟𝑟𝑠𝑠 −𝑗𝑗𝑥𝑥𝑠𝑠
𝑣𝑣𝑓𝑓 𝑏𝑏𝑐𝑐
𝑓𝑓𝑐𝑐ℎ𝑔𝑔 = | 𝜏𝜏𝑒𝑒 𝑗𝑗𝑗𝑗 𝑠𝑠ℎ𝑖𝑖𝑖𝑖𝑖𝑖 |2 (3.14)
2
𝑏𝑏𝑐𝑐
𝑡𝑡𝑐𝑐ℎ𝑔𝑔 = |𝑣𝑣𝑡𝑡 |2 (3.15)
2
The standard format of using get_losses function is loss = get_losses(results).
3.4.5 Canonical Form of the Generator Information
The canonical form of the generator information in MATPOWER 5.1 is:
[gen_bus, Pg, Qg, Qmax, Qmin, Vg, Mbase, status, Pmax, Pmin, pc1, pc2, qc1min,
qc1max, qc2min, qc2max, ramp_agc, ramp_10, ramp_30, ramp_q, apf]. The parameters
settings of the generator data in IEEE 14 bus system are presented in fig. 3.1, and some of
the most important generator name columns and their corresponding meanings are listed
in Table 3.1.
Figure 3.1 Canonical Form of the IEEE 14 Bus System Generator Data
29
Table 3.1 Explanation of the Generator Name Columns
Name Meaning
Gen_bus Bus number of the generator
Pg Real power generation
Qg Reactive power generation
Qmax Maximum reactive power output
Qmin Minimum reactive power output
Vg Voltage magnitude of the bus
Pmax Maximum real power output
Pmin Minimum real power output
3.4.6 Canonical Form of the Branch Information
The canonical form of the branch information in MATPOWER 5.1 is: [f_bus,
t_bus, br_r, br_x, br_b, rate_a, rate_b, rate_c, ratio, angle, angmin, angmax]. The
parameters settings of the branch data in the IEEE 14 bus system are presented in fig. 3.2.
Some of the most important branch name columns and their corresponding meanings are
listed in Table 3.2.
30
Figure 3.2 Canonical Form of the IEEE 14 Bus System Branch Data
Table 3.2 Explanation of the Branch Name Columns
Name Meaning
f_bus From bus number
t_bus To bus number
br_r Resistance of the branch
br_x Reactance of the branch
rate_a Long-term rating of the branch
rate_b Short-term rating of the branch
rate_c Emergency rating of the branch
31
Ratio Tap ratio of the transformer
Angle Phase shift angle of the transformer
Pf Real power injection at “from” bus side
Qf Reactive power injection at “from” bus side
Pt Real power injected at “to” bus side
Qt Reactive power injected at “to” bus side
3.4.7 Canonical Form of the Bus Information
The canonical form of the bus information in MATPOWER 5.1 is [bus_i,
bus_type, Pd, Qd, gs, bs, area, Vm, Va, base_kv, zone, Vmax, Vmin]. The parameters
settings of the bus data in the IEEE 14 bus system are presented in fig. 3.3. Some of the
most important bus name columns and their corresponding meanings are listed in Table
3.3.
Figure 3.3 Canonical Form of the IEEE 14 Bus System Bus Data
32
Table 3.3 Explanation of the Bus Name Columns
Name Meaning
Bus_i Bus number
Bus-type 1 means PQ bus, 2 means PV bus, and 3 means slack bus
Pd Real power load
Qd Reactive power load
Gs Shunt conductance
Bs Shunt susceptance
Area Bus area
Vm The magnitude of the voltage
Va The phase angle of the voltage
Zone Zone number
3.5 Procedures of the PSO Based Reactive Power Dispatch
To conclude, the main optimization steps of the PSO based reactive power
dispatch are as follows:
1 Load case information: in MATPOWER, IEEE 14 bus system data is saved in
case14.m file. Users can also create their personalized case by following the
format of the canonical forms of generators, buses, and branches.
2 Initialization: set the total iteration number, particle number, and initial velocity,
randomly assign the position of each particle in the design space. Then evaluate
33
the fitness of each particle and save the global best-known position, and the local
best-known position of each particle.
3 Update the positions and velocities: updating the position and velocity of each
particle by using formula 2.1 and formula 2.2. Then check whether the solution
violates the limit or not. If the solution exceeds the limits, use the EPF method to
penalize the violations.
4 Evaluate each particle: substitute the position of each particle into the objective
function to calculate the evaluation value.
5 Update local best-known position: if the current fitness value is smaller than the
historical best fitness value, update the local best-known position.
6 Update global best-known position.
7 Decide stopping criterion: determine if the iteration has reached the maximum
iteration number. If so, stop the optimization process and print the result;
otherwise, iter=iter+1, and go back to step 3.
The flow chart of the PSO based reactive power dispatch is illustrated in the Fig.
3.4.
34
Randomly generate the location of each particle
Randomly generate the velocity of each particle

particle
Set Gbest and Pibest
Update the velocity of each particle
Vi≤Vimax Yes Vi=Vimax
No
Wd+1=Wd+Vd
No
particle
Control variable
Penalize No
within limit
Yes
Vali≤Val_Pibest Yes Val_Pibest=Vali, Pibest=Pi+1
No
Gbest=min(Pibest)
Yes
Results
Figure 3.4 Flow Chart of the PSO Based Reactive Power Dispatch
35
Chapter Four: Case Studies
4.1 IEEE 14 Bus System Data
The performance of the proposed method is verified on IEEE 14 bus system. The
structure of the 14 buses network is shown in Fig. 4.1 [30].
Figure 4.1 IEEE 14 Bus System
There are two generators in the IEEE 14 bus system. One is at the slack bus; the
other one is at bus 2. Three synchronous condensers are located at bus 3, bus 6, and bus 8,
respectively. There are also three transformers and one shunt reactive power compensator
in this system. The total real power load is 259 MW and the total reactive power load is
73.5 MVar. Other detail information of this system is listed as below:
Table 4.1 IEEE 14 Bus System Loads Parameters
Load Bus number P (MW) Q (MVar)
2 2 21.7 12.7
3 3 94.2 19.0
4 4 47.8 -3.9
5 5 7.6 1.6
6 6 11.2 7.5
9 9 29.5 16.6
10 10 9.0 5.8
11 11 3.5 1.8
12 12 6.1 1.6
13 13 13.5 5.8
14 14 14.9 5.0
Table 4.2 IEEE 14 Bus System Generators Parameters
Bus Bus Voltage P Q Maximum Minimum

Generator
number type (p.u.) (MW) (MVar) Q Q
1 1 Slack 1.060 232.4 -16.9 0 0
2 2 PV 1.045 40 42.4 50 -40
3 3 PV 1.010 0 23.4 40 0
37
4 6 PV 1.070 0 12.2 24 -6
5 8 PV 1.090 0 17.4 24 -6
Table 4.3 IEEE 14 Bus System Branches Parameters
Tap
From Bus To Bus R X B
Position
1 2 0.01938 0.05917 0.0528
1 5 0.05403 0.22304 0.0492
2 3 0.04699 0.19797 0.0438
2 4 0.05811 0.17632 0.0340
2 5 0.05695 0.17388 0.0346
3 4 0.06701 0.17103 0.0128
4 5 0.01335 0.04211 0
4 7 0 0.20912 0 0.978
4 9 0 0.55618 0 0.969
5 6 0 0.25202 0 0.932
6 11 0.09498 0.19890 0
6 12 0.12291 0.25581 0
6 13 0.06615 0.13027 0
7 8 0 0.17615 0
7 9 0 0.11001 0
9 10 0.03181 0.08450 0
38
9 14 0.12711 0.27038 0
10 11 0.08205 0.19207 0
12 13 0.22092 0.19988 0
13 14 0.17093 0.34802 0
Table 4.4 IEEE 14 Bus System Reactive Power Injection Parameter
Bus number Reactive Power Injection (MVar)
9 19
4.2 Reactive Power Dispatch without New DGs
The MATLAB code for the reactive power dispatch without adding new DGs is
attached in Appendix A. Most of the time, there would be no obvious improvement on
the optimization result after conducting one hundred iterations. But in order to give the
particles enough opportunities to reach the global minimum, the stopping criteria of the
optimization process is set as the iteration number reaching two hundred. The size of the
swarm is fifty. The initial weight inertia is set as 0.9, and the final weight inertia is set as
0.4. As the iterations go on, the weight value will drop from 0.9 to 0.4. The position of
each particle is defined in a nine-dimensional space, as shown in Fig. 4.2:
V1 V2 V3 V6 V8 T1 T2 T3 S9
Figure 4.2 Coordinates of the Particle
In Fig. 4.2, V represents the voltage magnitudes at the slack bus or PV bus, T is
for the tap position of the transformer, and S9 is the reactive power injection at bus 9.
39
When the optimization process starts, the position of each particle will be continuously
updated until reaching the stopping criteria.
Fig. 4.3 shows the optimization process of reactive power dispatch without
installing new DG. At the beginning of the optimization process, the positions of the
particles are randomly selected. The global optimal real power loss is about 13.5 MW at
that time. As the particles continually update their positions towards the best solution, the
real power loss keeps decreasing. After 100 iterations, no obvious improvement can be
observed. Finally, the active power loss converges to 12.36 MW.
Minimum Real Power Loss

13.8
13.6
13.4
Real Power Loss (MW)
13.2
13
12.8
12.6
12.4
12.2
0 20 40 60 80 100 120 140 160 180 200
Iteration
Figure 4.3 Loss Reduction Process
40
Table 4.5 shows the real power loss on each branch before and after the particle
swarm optimization. Even though the active power loss in some branches are slightly
increased, (for instance, branch 6-12, 6-13, 9-14), the overall real power loss of the 14
buses system is significantly reduced.
Table 4.5 Comparison of the Real Power Loss at Each Branch
Branch number Before optimization (MW) After optimization (MW)
1-2 4.298 3.907
1-5 2.763 2.552
2-3 2.323 2.147
2-4 1.677 1.546
2-5 0.904 0.828
3-4 0.373 0.347
4-5 0.514 0.462
4-7 0 0
4-9 0 0
5-6 0 0
6-11 0.055 0.055
6-12 0.072 0.073
6-13 0.212 0.213
7-8 0 0
7-9 0 0
41
9-10 0.013 0.013
9-14 0.116 0.120
10-11 0.013 0.013
12-13 0.006 0.006
13-14 0.054 0.053
4.3 Reactive Power Dispatch with a New DG Operating at Rated Power
The second case study is about adding a new DG to the IEEE 14 bus system and
then optimizes the reactive power of the system by using PSO. Wind generator, solar
panels, and micro-turbine can all be chosen as an alternative of DG. In this thesis,
Enercon E82 wind turbine is selected as the new DG. Enercon E82 is a direct-drive
synchronous generator. Its rated power is 2000 kW. The wind generator is assumed to
operate at its rated power in the second case study.
In the next page, the shaded area in Fig. 4.4 describes reactive power capability of
Enercon E82 [31]. When the real power output is 0 MW, the wind generator can still
deliver as much as 1.2 MVar or absorb -1.0 MVar reactive power. Since 2007, some
commercial wind turbines have already been equipped with this kind of full reactive
power capability, which can produce full reactive power regardless of the wind
conditions. While the reactive power capacity of the early products is usually related to
their real-time real power outputs.
42
Figure 4.4 Reactive Power Capacity of Enercon E82 (2010 FACTS-WT)
Figure 4.5 Reactive Power Capacity of Enercon E66 (2002)

43
In order to apply the MATLAB code in the previous section into this case study,
several parameters need to be modified before executing the MATLAB code.
If the wind turbine is installed on a PV bus (e.g. bus 2), then both real and reactive
power capacity of the generator need to be changed. In Fig. 4.6, note that the active
power output of the generator at bus 2 is increased from 40 to 42.2, the range of reactive
power output is changed from [-40, 50] to [-41, 51.2].
Figure 4.6 Generator Data of the Modified System when the DG is Installed on Bus 2
Similar changes will also be performed in other cases, as shown in the graphs
from Fig. 4.7 to Fig. 4.9.
44
The position of each particle, in this case, will remain the same as in the previous
section. The MATLAB code in Appendix A can still be used to solve the optimal reactive
power dispatch problem in this case.
V1 V2 V3 V6 V8 T1 T2 T3 S9
Figure 4.10 Coordinates of the Particle when the New DG is connected to a PV Bus
If the wind turbine is installed on PQ bus (e.g. bus 4), in addition to modifying the
capacity of the real and reactive power to new parameters, the voltage magnitude of the
new installed DG bus should also be treated as a new control variable, as illustrated in the
fig. 4.11.
45
V1 V2 V3 V6 V8 Vr T1 T2 T3 S9
Figure 4.11 Coordinates of the Particle when the New DG is on a PQ Bus
where Vr represents the voltage magnitude of the new DG.
The MATLAB code in Appendix A will not be fit for this case. So the MATLAB
code in Appendix B is used for solving the reactive power dispatch problem. The
differences between the two codes are highlighted in the Appendix B. Other changes on
the bus data and generator data are presented in the graphs from fig. 4.12 to fig. 4.29.
Figure 4.12 Bus Data of the Modified System when the DG is Installed on Bus 4
In the generator data section of the modified 14 bus system, the parameters of the
new generator are added.
46
Similar changes are needed before executing the MATLAB code to calculate the
reactive power dispatch problems at the buses.
47
48
49
50
Figure 4.22 Bus Data of the Modified System when the DG is Installed on bus 11
51
52
53
54
After running the MATLAB code, the real power loss of the system, when DG is
installed on different buses, is shown in table 4.6 on the next page.
From the table, we can learn that the real power loss can be reduced by 7.89% by
simply applying PSO algorithm. After adding a new DG to the system and using PSO
algorithm to further adjusting the values of control variables, the real power loss can be
reduced by as much as 10.27%. A comparison of the loss reduction, when DG is installed
on different buses, is presented in Fig. 4.30.
55
Table 4.6 Optimization Results with a New DG Installed on Different Buses
With
DG DG DG DG DG DG DG DG DG DG DG DG DG
-out With
Control at at at at at at at at at at at at at
DG -out
Variables Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus
and DG
2 3 4 5 6 7 8 9 10 11 12 13 14
PSO
Voltage at 1.06 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10 1.10
Bus 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Voltage at 1.04 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08 1.08
Bus 2 5 3 7 6 7 7 7 7 6 7 7 4 4 2 4
Voltage at 1.01 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.04 1.05
Bus 3 0 2 7 6 8 8 7 7 6 8 8 3 3 7 2
Voltage at 1.06
56
Bus 4 7
Voltage at 1.07
Bus 5 4
Voltage at 1.07 1.06 1.08 1.07 1.06 1.07 1.08 1.79 1.07 1.10 1.07 1.07 1.05 1.07 1.05
Bus 6 0 4 5 6 4 2 7 0 4 0 7 2 4 6 1
Voltage at 1.08
Bus 7 6
Voltage at 1.09 1.06 1.10 1.10 1.04 1.10 1.10 1.01 1.10 1.04 1.10 1.09 1.10 1.08 1.04
Bus 8 0 4 0 0 5 0 0 7 0 6 0 8 0 6 6
Voltage at 1.09
Bus 9 5
Voltage at 1.06
Bus 10 7
Voltage at 1.06
Bus 11 6
Voltage at 1.04
Bus 12 5
Voltage at 1.06
Bus 13 4
Voltage at 1.02
Bus 14 5
0.97 1.01 0.98 1.00 0.97 0.99 0.97 0.99 0.99 0.98 1.01 0.98 0.99 0.98 0.99
Tap1
8 8 6 8 4 6 4 1 6 1 5 9 5 9 8
0.96 0.99 0.97 1.00 0.98 0.99 1.00 1.01 0.99 1.01 1.00 0.99 1.01 1.00 1.00
Tap2
9 4 3 0 9 4 2 6 4 9 5 0 1 4 9
0.93 1.00 1.02 1.01 1.01 0.99 1.03 1.00 1.03 1.00 1.01 1.00 0.99 0.98 1.00
Tap3
2 6 4 6 2 6 3 5 3 1 2 4 3 5 8
Shunt 9 19 16.9 15.2 20.0 14.4 10.1 19.4 17.9 16.3 10.3 15.4 8.9 10.9 9.3 19.0
57
13.3 12.3 12.1 12.0 12.0 12.0 12.0 12.0 12.0 12.0 12.0 12.0 12.0 12.1 12.0
Loss (MW)
93 36 53 17 84 97 81 62 64 33 49 96 80 12 88
Loss
10.2 10.1 10.0
Reduction 0 7.89 9.26 9.77 9.68 9.80 9.94 9.92 9.68 9.80 9.56 9.74
7 5 4
(%)
Loss reduction
12.00%
10.00%
8.00%
6.00%
4.00%
2.00%
0.00%
Figure 4.30 Comparison of the Loss Reduction
Table 4.7 shows the comparison of the real power loss of the original network,
optimization without new DG, and optimization with new DG. We can learn that the real
power loss on most of the branches is significantly reduced.
Table 4.7 Real Power Losses Comparisons
Branch # Before optimization Optimization without DG Optimization with DG
1-2 4.298 3.907 3.826
1-5 2.763 2.552 2.517
2-3 2.323 2.147 2.051
2-4 1.677 1.546 1.531
2-5 0.904 0.828 0.818
3-4 0.373 0.347 0.309
58
4-5 0.514 0.462 0.461
4-7 0 0 0
4-9 0 0 0
5-6 0 0 0
6-11 0.055 0.055 0.038
6-12 0.072 0.073 0.068
6-13 0.212 0.213 0.194
7-8 0 0 0
7-9 0 0 0
9-10 0.013 0.013 0.019
9-14 0.116 0.120 0.135
10-11 0.013 0.013 0.006
12-13 0.006 0.006 0.005
13-14 0.054 0.053 0.039
Fig. 4.31 shows the optimization process of the proposed method when the new
wind turbine is installed on bus 3. The initial real power loss of the system is at about
12.45 MW. The particles start to converge after conducting eighty iterations. Finally, the
total power loss of the system is 12.017 MW.
59
12.45
12.4
12.35
12.3
12.25
12.2
12.15
12.1
12.05
12
0 20 40 60 80 100 120 140 160 180 200
Iteration
Figure 4.31 Loss Reduction Process when the new DG is Installed on Bus 3
4.4 Reactive Power Dispatch with a New DG Operating at Various Power
In order to evaluate the performance of the proposed method at the various wind
speed conditions, the third case study compares the power loss of the modified 14 bus
system when the wind turbine delivers 25%, 50%, and 75% of its rated power,
respectively.
The real power output curve of Enercon E82 wind turbine is presented in Fig. 4.32
[32]. When the wind speed is at 7 m/s, the output of the wind turbine is 532 W, which is
about 25% of its rated power output. When the wind speed is about 9 m/s, the output of
wind generator is 1.18 kW, approximately 50% of its rated power output. Finally, when
60
the wind speed is about 10 m/s, the output of the wind generator is 1.58 kW, about 75%
of the rated power output.
Real Power Output Curve

2200
2000
1800
1600
Real Power Output (W)
1400
1200
1000
800
600
400
200
0
0 5 10 15 20 25
wind speed (m/s)
Figure 4.32 Real Power Output Curve of Enercon E82
When the wind speed is 7 m/s, the value of Pg at bus 9 is changed to 532 W, as
shown in fig. 4.33.
61
The process of reactive power optimization, when DG produce 25% of its rated
power output, is shown in fig. 4.34. Finally, the real power loss of the system converges
at 12.1845 MW.

12.8
12.7
12.6
12.5
12.4
12.3
12.2
0 20 40 60 80 100 120 140 160 180 200

Iteration
Figure 4.34 Loss Reduction Process when DG Delivers 25% Rated Power
Table 4.8 Optimization Results when DG Deliver 25% Rated Power
Control variable Values
Voltage at bus 1 1.100
62
Turn ratio 1 1.007
Turn ratio 2 0.991
Turn ratio 3 1.005
Shunt 9 10.118
shown in fig. 4.35.
power output, is shown in fig. 4.36. Finally, the real power loss of the system converges
on 12.1221 MW.
63
12.7
12.6
12.5
12.4
12.3
12.2
0 20 40 60 80 100 120 140 160 180 200

Iteration
64
Turn ratio 1 1.009
Turn ratio 2 0.993
Turn ratio 3 1.011
Shunt 9 10.478
shown in fig. 4.37.
power output, is shown in fig. 4.38. Finally, the real power loss of the system is 12.0793
MW.
65
12.7
12.6
12.5
12.4
12.3
12.2
12.1
12
0 20 40 60 80 100 120 140 160 180 200
Iteration
Turn ratio 1 1.0238
66
Turn ratio 2 0.9740
Turn ratio 3 0.9971
Shunt 9 11.0650
The phase angle shifts at the different real power output conditions are also
compared in Table 4.11. When the output increases, the phase angle shift declines.
Table 4.11 Comparison of the Phase Angle Shift
Bus
25% rated power 50% rated power 75% rated power Rated power
number
1 0 0 0 0
2 -4.571 -4.561 -4.551 -4.530
3 -11.705 -11.668 -11.663 -11.636
4 -9.52 -9.486 -9.471 -9.448
5 -8.182 -8.153 -8.128 -8.111
6 -13.432 -13.409 -13.325 -13.272
7 -12.529 -12.485 -12.441 -12.333
8 -12.529 -12.485 -12.441 -12.333
9 -14.062 -14.016 -13.931 -13.840
10 -14.222 -14.182 -14.095 -14.011
11 -13.955 -13.925 -13.838 -13.769
12 -14.231 -14.212 -14.123 -14.066
13 -14.31 -14.291 -14.2 -14.142
14 -15.107 -15.079 -14.985 -14.906
67
4.5 Discussion
By analyzing the results of the three case studies, the following conclusions can
be obtained:
1 Before the reactive power optimization, the reactive power in IEEE 14 bus system
is unreasonable distributed. Reactive power dispatch can significantly reduce the
real power loss of the system and improve the power quality.
2 Satisfying results can be achieved after about conducting 90 iterations, which
reflects the excellent searching ability of PSO algorithm for solving nonlinear
problems.
3 When a small capacity DG is added into the system, the real power loss would be
further reduced. As the output of the DG increases, the real power loss of the
system decreases.
68
Chapter Five: Conclusion and Future Work
5.1 Conclusion
Reactive power dispatch is a nonlinear optimization problem that contains both
continuous and discrete control variables. PSO is a heuristic global optimization
algorithm that possess of high efficiency and robustness. PSO is less sensitive to the
complexity of the objective functions. Therefore, it shows enormous potential for solving
reactive power dispatch problems.
This thesis uses the IEEE 14 bus system as the test system. Both PSO algorithm
and MATPOWER 5.1 toolbox are applied to reduce the real power loss in the power
networks. In order to avoid the control variables exceeding the limits, exterior penalty
function method is also employed. The main contribution of this thesis is as follows:
1 Reactive power dispatch approach can significantly reduce the power loss in
power systems, and this method is both cost-effective and can be easily employed
in real life.
2 PSO algorithm shows excellent searching ability in solving nonlinear
optimization problems. Applying PSO algorithm to address the reactive power
dispatch problems is technical feasible and can achieve considerable economic
benefits.
69
3 The mature MATPOWER 5.1 toolbox are introduced to calculate the power flow
and manage the equality constraints in PSO based reactive power dispatch. The
accuracy of the results and the robustness of the code get improved.
5.2 Future Work
This thesis solves the reactive power dispatch problem and determines the optimal
placement of newly installed DG in an existing power system. However, there still
appears to be some limitations and need to do further research.
1 PSO algorithm has excellent searching capability, but it is apt to plunge into local
minimum solutions. Further research needs to think about how to avoid premature
problems.
2 The running time of the code is five minutes on the laptop. Future work includes
improving the efficiency of the MATLAB code.
3 The modified test system only considers one DG. If more DGs are added to the
systems, the computation time would be dramatically increased. Further research
needs to simplify the power system model to reduce the computation time.
4 Due to the computation time limitation, this thesis only calculates the reactive
power dispatch problems when the wind generator operates at 7 m/s, 9 m/s, and
10 m/s. In the future research, the performance of the proposed method at other
wind speed conditions would also be calculated.
70
References
[1] A. M. Chebbo, M. R. Irving, M. J. H. Sterling, “Reactive power dispatch
incorporating voltage stability,” IET Proceedings on Generation, Transmission and
Distribution, 1992.
[2] R. Thomas, T. Mount, R. Schuler, W. Schulze, R. Zimmerman, D. Shawhan, and D.
Toomey, “Markets for reactive power and reliability: A white paper,” Eng. Econ.
Elect. Research Group, Cornell Univ., Ithaca, NY.
[3] J. Hanger, P. Adels, “Reactive Power and the Blackout,” Available:
http://www.energycentral.com/articles/article/529
[4] U.S. Energy Information Administration, “How much electricity is lost in
transmission and distribution in the United States?” Available:
http://www.eia.gov/tools/faqs/faq.cfm?id=105&t=3
[5] PowerWorld Simulator 17, Available: http://www.powerworld.com/download-
purchase/demo-software
[6] J. Kepka. “Reactive Power Compensation.” Wroclaw University of Technology.
[7] N. M. Neagle, D. R. Samson, “Loss Reduction from Capacitors Installed on Primary
Feeders,” Transaction of the American Institute of Electrical Engineers, Power
Apparatus and Systems, Part III, Vol. 75, Issue 3, 1956.
71
[8] J. A. Momoh, S. X, Guo, E. C. Ogbuobiri, and R. Adapa, “The quadratic interior
point method solving power system optimization problems,” IEEE Transaction on
Power System, vol. 9, no. 3, 1994.
[9] R. Dubey, S. Dixit, G. Agnihotri, “Optimal Placement of Shunt Facts Devices Using
Heuristic Optimization Techniques: An Overview,” Fourth International Conference
on Communication Systems and Network Technologies (CSNT), 2014.
[10] T. Sousa, J. Soares, Z.A. Vale, H. Morais and P. Faria,, “Simulated Annealing
metaheuristic to solve the optimal power flow,” IEEE Power and Energy Society
General Meeting, 2011.
[11] M. Gitizadeh, M. Kalanar, “Multi-objective fuzzy based reactive power and voltage
control in a distribution system using SA,” 11th International Conference on Hybrid
Intelligent Systems (HIS), 2011.
[12] Padhy, N.P.; Abdel-Moamen, M.A.; Praveen Kumar, B.J., “Optimal location and
initial parameter settings of multiple TCSCs for reactive power planning using
genetic algorithms,” IEEE Power Engineering Society General Meeting, 2004.
[13] A. Parizad, M. Kalantar, A. Khazali, “Application of HAS and GA in Optimal
placement of FACTS Devices Considering Voltage Stability and Losses,”
International Conference on Electric Power and Energy Conversion Systems, 2009.
72
[14] “Artificial neural network,” Available: https://en.wikipedia.org/wiki/Artificial_neura
l_network
[15] K.H. Abdul-Rahman, S.M. Shahidehpour, M. Daneshdoost, “AI approach to optimal
VAR control with fuzzy reactive loads,” IEEE Transactions on Power Systems,
Volume: 10, Issue: 1, 1995.
[16] Chao-Rong Chen, Hang-Sheng Lee, Wenta Tsai, “On-line Optimal Shunt Capacitors
Dispatch of Peak Power Systems,” IEEE/PES Transmission and Distribution
Conference and Exhibition: Asia and Pacific, 2005.
[17] Annealing (metallurgy), Available: https://en.wikipedia.org/wiki/Annealing_(metall
urgy)
[18] Simulated Annealing Terminology, Available: http://www.mathworks.com/help/gad
s/understanding-simulated-annealing-terminology.html
[19] Genetic Algorithm Terminology, Available: http://www.mathworks.com/help/gads/s
ome-genetic-algorithm-terminology.html
[20] Eiben, A. E. et al, "Genetic algorithms with multi-parent recombination".
Proceedings of the International Conference on Evolutionary Computation, 1994.
[21] Ting, Chuan-Kang, "On the Mean Convergence Time of Multi-parent Genetic
Algorithms without Selection," 2005.
[22] “Genetic Algorithm,” Available: https://en.wikipedia.org/wiki/Genetic_algorithm

73
[23] J. Kennedy, R. Eberhart, “Particle Swarm Optimization,” in IEEE International
Conference on Neural Networks, 1995.
[24] M.R. AlRashidi, M.E. El-Hawary, “A Survey of Particle Swarm Optimization
Applications in Electric Power Systems,” IEEE Transactions on Evolutionary
Computation, volume 13, issue 4, 2009.
[25] S.S. Sharif, J.H. Taylor, “Dynamic optimal reactive power flow,” Proceedings of the
American Control Conference, 1998.
[26] M.M.A. Salama, N. Manojlovic, V.H. Quintana, A.Y. Chikhani, “Real-time optimal
reactive power control for distribution networks,” Electrical Power and Energy
System, 1996.
[27] C.L. Hwang, F.A. Tillman, W. Kuo, “Reliability Optimization by Generalized
Lagrangian-Function and Reduced-Gradient Methods,” IEEE Transactions on
Reliability, Volume R-28, Issue 4, 1979.
[28] P. Venkataraman, “Applied Optimization with MATLAB Programming,” 2nd ed.
John Wiley & Sons, Inc., Hoboken, New Jersey, 2009.
[29] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas, "MATPOWER:
Steady-State Operations, Planning and Analysis Tools for Power Systems Research
and Education," Power Systems, IEEE Transactions on, vol. 26, no. 1, pp. 12-19, Feb.
2011.
74
[30] Iraj Dabbagchi and Rich Christie, 14 Bus Power Flow Test Case, Available:
https://www.ee.washington.edu/research/pstca/pf14/pg_tca14bus.htm
[31] T. Ackermann, “Wind Power in Power Systems,” 2nd ed. John Wiley & Sons, Inc.,
Hoboken, New Jersey, 2012.
[32] ENERCON Wind energy converters Product overview, Available:
http://www.enercon.de/p/downloads/EN_Productoverview_0710.pdf
75
Appendix A MATLAB Code for Reactive Power Dispatch without New DGs
%%%%%%%%%%% Reactive Power Dispatch %%%%%%%%%%%

%%%%%%%%%%%%%% Initialization %%%%%%%%%%%%%%
%Iteration=200, Swarm size=50
clear
clc
iter=0;
iteration=200;
particlenumber=50;
%Inertia Weight
w_max=0.9;
w_min=0.4;
w_temp=w_max;
w_step=(w_max-w_min)/iteration;
%Maximum and minimum limit

vol_min=0.95;
vol_max=1.10;
%Load IEEE 14 bus data

[baseMVA, bus, gen, branch]=loadcase(case14);
%Initialization of Swarm & velocity

%Control variables: vg1 (1.06), vg2 (1.045), vg3 (1.01), vg6 (1.07), vg8(1.09)
%Control variables: tp1(4-7 0.978), tp2(4-9 0.969), tp3(5-6 0.932)
%Control variables: shunt9(19)
%Random 50*9 matrix

Swarm=[unifrnd(0.95,1.10,particlenumber,5), ...
unifrnd(0.975,1.025,particlenumber,3),unifrnd(0,20,particlenumber,1)];
%Initial velocity is set to 0

Velocity=zeros(particlenumber,9);
for i=1:particlenumber
v1=Swarm(i,1); %v1
bus(1,8)=v1; %Vm, 8 is voltage magnitude (p.u.)
gen(1,6)=v1; %Vg, 6 is voltage magnitude setpoint (p.u.)
v2=Swarm(i,2); %v2
bus(2,8)=v2;
gen(2,6)=v2;
76
v3=Swarm(i,3); %v3
bus(3,8)=v3;
gen(3,6)=v3;
v6=Swarm(i,4); %v6
bus(6,8)=v6;
gen(4,6)=v6;
v8=Swarm(i,5); %v8
bus(8,8)=v8;
gen(5,6)=v8;
branch(8,9)=Swarm(i,6); %tp1 4-7, 9 is tap position

branch(9,9)=Swarm(i,7); %tp2 4-9
bus(9,6)=Swarm(i,9); %Shunt capacitor 9, 6 is BS
eval(['savecase (''case14_test' num2str(i) '.mat'', baseMVA, bus, gen, branch)']);

eval(['results',num2str(i),'=runpf(''case14_test', num2str(i) '.mat'')']);
eval(['losses',num2str(i),'=sum(real(get_losses(results',num2str(i),')))']);
%Penalty for bus voltage violation

bus_inf=bus(:,8);
for bus_num=1:14
if bus_inf(bus_num)>vol_max
penalty_bus(bus_num)=10000*(bus_inf(bus_num)-vol_max)^2;
elseif bus_inf(bus_num)<vol_min
penalty_bus(bus_num)=10000*(bus_inf(bus_num)-vol_min)^2;
else
penalty_bus(bus_num)=0;
end
end
penalty_bus_violation=sum(penalty_bus);
%Penalty for reactive generation violation

gen_inf=gen(:,3);
for gen_num=2:5
if gen_inf(gen_num)>gen(gen_num,4)
penalty_gen(gen_num)=1000*(gen_inf(gen_num)-gen(gen_num,4))^2;
elseif gen_inf(gen_num)<gen(gen_num,5)
else
penalty_gen(gen_num)=0;
end
77
end
penalty_gen_violation=sum(penalty_gen);
%Penalty for tap position violation

brch_inf=[branch(8,9); branch(9,9); branch(10,9)];
for brch_num=1:3
if brch_inf(brch_num)>1.025
penalty_brch(brch_num)=10000*(brch_inf(brch_num)-1.025)^2;
elseif brch_inf(brch_num)<0.975
else
penalty_brch(brch_num)=0;
end
end
penalty_brch_violation=sum(penalty_brch);
%Penalty function
losses(i)=eval(['losses',num2str(i)]);
Obj_fun_initial(i)=losses(i)+penalty_bus_violation+penalty_gen_violation+penalty_brch
_violation;
end
%%%%%%%%%%%%% Initialize Pbest and Gbest %%%%%%%%%%%%%

for j=1:particlenumber
Pbest(j,:)=Swarm(j,:);
Val_Pbest(j)=Obj_fun_initial(j);
end
[Val_Gbest,m]=min(Val_Pbest);
Gbest=Swarm(m,:);
Gbest_calc=repmat(Swarm(m,:),particlenumber,1);
%%%%%%%%%%%%%%%% PSO %%%%%%%%%%%%%%%%%%

losses_temp=zeros(1,particlenumber);
figure('NumberTitle', 'off', 'Name', 'Minimum Real Power Loss');
title('Minimum Real Power Loss');
ylabel('Real Power Loss (MW)');
xlabel('Iteration');
grid on;
hold on
for iter=1:iteration
78
R1=rand(particlenumber,9);
%2.05+2.05=4.1;
%2/abs(2-4.1-sqrt(4.1*4.1-4*4.1))=0.729
Velocity=0.729*(w_temp*Velocity+2.05*R1.*(Pbest-Swarm)+2.05*R2.*(Gbest_calc-
Swarm));
% Set maximum velocity

for v_iter=1:9
if v_iter==9
Outstep=Velocity(:,v_iter)>0.1;
Velocity(find(Outstep),v_iter)=0.1;
Outstep=Velocity(:,v_iter)<-0.1;
Velocity(find(Outstep),v_iter)=-0.1;
else
end
end
Swarm=Swarm+Velocity;
for k=1:particlenumber
v1=Swarm(k,1); %v1
v2=Swarm(k,2); %v2
bus(2,8)=v2;
gen(2,6)=v2;
v3=Swarm(k,3); %v3
bus(3,8)=v3;
gen(3,6)=v3;
v6=Swarm(k,4); %v6
bus(6,8)=v6;
gen(4,6)=v6;
v8=Swarm(k,5); %v8
bus(8,8)=v8;
gen(5,6)=v8;
branch(8,9)=Swarm(k,6); %tp1 4-7, 9 is tap position

branch(9,9)=Swarm(k,7); %tp2 4-9
79
bus(9,6)=Swarm(k,9); %Shunt capacitor 10, 6 is BS
eval(['savecase (''case14_test' num2str(k) '.mat'', baseMVA, bus, gen, branch)']);

eval(['results',num2str(k),'=runpf(''case14_test', num2str(k) '.mat'')']);
eval(['losses',num2str(k),'=sum(real(get_losses(results',num2str(k),')))']);

bus_inf=bus(:,8);
for bus_num=1:14
else
end
end

gen_inf=gen(:,3);
for gen_num=2:5
else
end
end
%Penalty for tap position violation

brch_inf=[branch(8,9); branch(9,9); branch(10,9)];
for brch_num=1:3
if brch_inf(brch_num)>1.025
elseif brch_inf(brch_num)<0.975
else
penalty_brch(brch_num)=0;
80
end
end
penalty_brch_violation=sum(penalty_brch);
%Penalty function
losses_temp(k)=eval(['losses',num2str(k)]);
Obj_fun_temp(k)=losses_temp(k)+penalty_bus_violation+penalty_gen_violation+penalt
y_brch_violation;
if Obj_fun_temp(k)<Val_Pbest(k)
losses(k)=losses_temp(k);
Val_Pbest(k)=Obj_fun_temp(k);
Pbest(k,:)=Swarm(k,:);
end
end
[Val_Gbest_temp,n]=min(Val_Pbest);
if Val_Gbest_temp<Val_Gbest
Val_Gbest=Val_Gbest_temp;
Gbest=Swarm(n,:);
Gbest_calc=repmat(Swarm(n,:),particlenumber,1);
end
w_temp=w_temp-w_step;
Val_Gbest_rec(iter)=Val_Gbest;
plot(Val_Gbest_rec);
drawnow;
end
81
Appendix B MATLAB Code for Reactive Power Dispatch with a New DG Installed
on PQ Bus
%%%%%%%%% Reactive Power Dispatch %%%%%%%%%

%%%%%%%%%% System initialization %%%%%%%%%%
%Iteration=200, Swarm size=50
clear
clc
iter=0;
iteration=200;
particlenumber=50;
%Inertia Weight
w_max=0.9;
w_min=0.4;
w_temp=w_max;
w_step=(w_max-w_min)/iteration;
%Maximum and minimum limit

vol_min=0.95;
vol_max=1.10;
%Load Modified IEEE 14 bus data

[baseMVA, bus, gen, branch]=loadcase(case14_bus4);
%Initialization of Swarm & velocity

%Control variables: vg1 (1.06), vg2 (1.045), vg3 (1.01), vg6 (1.07)
%vg8(1.09) vg4(1.019)
%Control variables: tp1(4-7 0.978), tp2(4-9 0.969), tp3(5-6 0.932)
%Control variables: shunt9(19)
%Random 50*10 matrix

Swarm=[unifrnd(0.95,1.10,particlenumber,6), ...
unifrnd(0.975,1.025,particlenumber,3),unifrnd(0,15,particlenumber,1)];
%Initial velocity is set to 0
Velocity=zeros(particlenumber,10);
for i=1:particlenumber
v1=Swarm(i,1); %v1
v2=Swarm(i,2); %v2
82
bus(2,8)=v2;
gen(2,6)=v2;
v3=Swarm(i,3); %v3
bus(3,8)=v3;
gen(3,6)=v3;
v6=Swarm(i,4); %v6
bus(6,8)=v6;
gen(4,6)=v6;
v8=Swarm(i,5); %v8
bus(8,8)=v8;
gen(5,6)=v8;
v4=Swarm(i,6); %v4
bus(4,8)=v4;
gen(6,6)=v4;
branch(8,9)=Swarm(i,7); %tp1 4-7, 9 is tap ratio

bus(9,6)=Swarm(i,10); %Shunt capacitor 9, 6 is BS
eval(['savecase (''case14_test' num2str(i) '.mat'', baseMVA, bus, gen, branch)']);

eval(['results',num2str(i),'=runpf(''case14_test', num2str(i) '.mat'')']);
eval(['losses',num2str(i),'=sum(real(get_losses(results',num2str(i),')))']);
bus_inf=bus(:,8);
for bus_num=1:14
else
end
end
gen_inf=gen(:,3);
for gen_num=2:6
83
else
end
end
%Penalty function
losses(i)=eval(['losses',num2str(i)]);
Obj_fun_initial(i)=losses(i)+penalty_bus_violation+penalty_gen_violation;
end
%%%%%%%%%%% Initialize Pbest and Gbest %%%%%%%%%%%

for j=1:particlenumber
Pbest(j,:)=Swarm(j,:);
Val_Pbest(j)=Obj_fun_initial(j);
end
[Val_Gbest,m]=min(Val_Pbest);
Gbest=Swarm(m,:);
Gbest_calc=repmat(Swarm(m,:),particlenumber,1);
%%%%%%%%%%%%%% PSO loop %%%%%%%%%%%%%%

losses_temp=zeros(1,particlenumber);
figure('NumberTitle', 'off', 'Name', 'Minimum Real Power Loss');
title('Minimum Real Power Loss');
ylabel('Real Power Loss (MW)');
xlabel('Iteration');
grid on;
hold on
for iter=1:iteration
%2.05+2.05=4.1;
%2/abs(2-4.1-sqrt(4.1*4.1-4*4.1))=0.729
Velocity=0.729*(w_temp*Velocity+2.05*R1.*(Pbest-Swarm)+2.05*R2.*(Gbest_calc-
Swarm));
% Set maximum velocity

for v_iter=1:10
if v_iter==10
84
else
end
end
Swarm=Swarm+Velocity;
for k=1:particlenumber
v1=Swarm(k,1); %v1
v2=Swarm(k,2); %v2
bus(2,8)=v2;
gen(2,6)=v2;
v3=Swarm(k,3); %v3
bus(3,8)=v3;
gen(3,6)=v3;
v6=Swarm(k,4); %v6
bus(6,8)=v6;
gen(4,6)=v6;
v8=Swarm(k,5); %v8
bus(8,8)=v8;
gen(5,6)=v8;
v4=Swarm(k,6); %v4
bus(4,8)=v4;
gen(6,6)=v4;
branch(8,9)=Swarm(k,7); %tp1 4-7, 9 is tap ratio

bus(9,6)=Swarm(k,10); %Shunt capacitor 10, 6 is BS
eval(['savecase (''case14_test' num2str(k) '.mat'', baseMVA, bus, gen, branch)']);

eval(['results',num2str(k),'=runpf(''case14_test', num2str(k) '.mat'')']);
eval(['losses',num2str(k),'=sum(real(get_losses(results',num2str(k),')))']);
85
bus_inf=bus(:,8);
for bus_num=1:14
else
end
end

gen_inf=gen(:,3);
for gen_num=2:6
else
end
end
%Penalty function
losses_temp(k)=eval(['losses',num2str(k)]);
Obj_fun_temp(k)=losses_temp(k)+penalty_bus_violation+penalty_gen_violation;
if Obj_fun_temp(k)<Val_Pbest(k)
losses(k)=losses_temp(k);
Val_Pbest(k)=Obj_fun_temp(k);
Pbest(k,:)=Swarm(k,:);
end
end
[Val_Gbest_temp,n]=min(Val_Pbest);
if Val_Gbest_temp<Val_Gbest
Val_Gbest=Val_Gbest_temp;
Gbest=Swarm(n,:);
Gbest_calc=repmat(Swarm(n,:),particlenumber,1);
end
w_temp=w_temp-w_step;
86
Val_Gbest_rec(iter)=Val_Gbest;
plot(Val_Gbest_rec);
drawnow;
end
87

Particle Swarm Optimization Based Reactive Power Dispatch For Pow

Uploaded by

Particle Swarm Optimization Based Reactive Power Dispatch For Pow

Uploaded by

University of Denver

Electronic Theses and Dissertations Graduate Studies

Particle Swarm Optimization Based Reactive Power Dispatch for

Follow this and additional works at: https://digitalcommons.du.edu/etd

Part of the Electrical and Computer Engineering Commons

of the Requirements for the Degree

Advisor: Dr. David Wenzhong Gao

All Rights Reserved

loss remains using reactive power dispatch method.

I would like to express my sincere gratitude to my advisor Prof. David Wenzhong

benefit me for life.

I am also grateful to the committee chair Prof. Caroline Li and committee

checking the thesis draft and giving your valuable suggestions.

and continuous support.

Chapter One: Introduction .................................................................................................. 1

Chapter Two: Global Optimization Methods ................................................................... 11

Chapter Three: Reactive Power Dispatch Problem Formulation ...................................... 23

Chapter Four: Case Studies ............................................................................................... 36

Chapter Five: Conclusion and Future Work ..................................................................... 69

Table 2.1 SA Terminologies ..................................................................................... 12

Table 2.2 GA Terminologies .................................................................................... 15

Table 2.3 PSO Parameters Selection ........................................................................ 22

Table 3.1 Explanation of the Generator Name Columns .......................................... 30

Table 3.2 Explanation of the Branch Name Columns .............................................. 31

Table 3.3 Explanation of the Bus Name Columns .................................................... 33

Table 4.1 IEEE 14 Bus System Loads Parameters ................................................... 37

Table 4.2 IEEE 14 Bus System Generators Parameters ........................................... 37

Table 4.3 IEEE 14 Bus System Branches Parameters .............................................. 38

Table 4.7 Real Power Losses Comparisons .............................................................. 58

Table 4.11 Comparison of the Phase Angle Shift ..................................................... 67

Figure 1.2 Single Power Factor Correction ................................................................ 4

Figure 1.3 Group Power Factor Correction ............................................................... 5

Figure 1.4 Bulk Power Factor Correction .................................................................. 5

Figure 2.2 Flow Chart of the Simulated Annealing .................................................. 14

Figure 2.3 Selection Operation ................................................................................. 16

Figure 2.4 Crossover Operation ................................................................................ 16

Figure 2.5 Mutation Operation ................................................................................. 17

Figure 2.6 Flow Chart of the Genetic Algorithm ...................................................... 18

Figure 2.7 Flow Chart of the Particle Swarm Optimization ..................................... 21

Figure 4.1 IEEE 14 Bus System ............................................................................... 36

Figure 4.2 Coordinates of the Particle ...................................................................... 39

Figure 4.3 Loss Reduction Process .......................................................................... 40

Figure 4.5 Reactive Power Capacity of Enercon E66 (2002) ................................... 43

Figure 4.30 Comparison of the Loss Reduction ....................................................... 58

Figure 4.32 Real Power Output Curve of Enercon E82............................................ 61

1.1 Background and Motivation

threaten the security of the power grid.

the power flow.

1.2 Reactive Power Compensation Techniques

Figure 1.1 Five Bus System without Power Factor Correction

1.3.1 Single Power Factor Correction

Figure 1.2 Single Power Factor Correction

1.3.2 Group Power Factor Correction

To overcome the defect of single power factor correction method, a more

efficiency of the shunt capacitors is greatly improved.

1.3.3 Bulk Power Factor Correction

is directly connected to the PV bus.

Figure 1.4 Bulk Power Factor Correction

schemes. Therefore, the principle of reactive power dispatch is compensating reactive

power at where the loads consume.

1.3 Literature Review

reactive power dispatch problems. The interior-point method belongs to linear

programming. The principle of linear programming is to expand the nonlinear functions

constant terms need to be considered. The advantage of this method is linear

nonlinear characteristic of reactive power dispatch problem than linear programming.

the dimension increase, the computation time would increase dramatically.

reduced as the iterations proceed. The disadvantage of SA is that the parameters of

simulated annealing need to be carefully chosen. Inappropriate parameters selection

in IEEE 14 buses system [10]. In [11], M. Gitizadeh proposes another fuzzy-based

reactive power dispatch method by using simulated annealing.

optimization methods. The feasible solutions in GA are compared to chromosomes.

practicability, high-efficiency, and robustness. A genetic algorithm toolbox GAOT

(Genetic Algorithm Optimization Toolbox) was developed by North Carolina State

placement in IEEE 14 buses system. Another reactive power optimization by using GA is

performed in IEEE 34 buses system in [13].

reduce the input variables and to get better computation speed.