You are on page 1of 5

Load Flow Solution Using Hybrid Particle Swarm Optimization

Amgad A. EL-Dib, Hosam K.M. Youssef, Member, IEEE, M.M.EL-Metwally, Member, IEEE, Z. Osrnan, Member, IEEE
Electrical Power and Machines Department, Faculty of Engineering, Cairo University, Giza, Egypt

Abstract-Load flow (LF) is an important tool in the


planning and operation of power systems. It is
usually solved using conventional numerical
techniques like Newton-Raphson (NR) or GaussSeidel (GS) methods. This paper presents an
application of particle swarm optimization (PSO) in
solving the load flow problem as an optimization
problem. Examples on test systems are given to
demonstrate the validity and applicability of the
proposed method.

I. Introduction
Load flow studies are the backbone of power system
analysis and design. They are necessary for planning,
operation, economic scheduling, and exchange of
power between utilities. In addition, load flow analysis
is required for many other analyses such as transient
stability, voltage stability and contingency studies. One
of the most important aspects in power systems
nowadays is the voltage stability analysis. Most of the
methods used in determining the voltage stability or the
maximum loading of the system depend on the solution
of the Load flow.
It is well known that the problem of load flow is solved
using numerical techniques. Most of these techniques
depend on getting the inverse of the Jacobian matrix of
the system [ 13. Such techniques fail to solve the load
flow in some conditions, which may be summarized as
follows:
1. Heavy loaded system (near the maximum
loading point of the system).
2. Ill-conditioned Jacobian matrix.
Due to these conditions, the voltage stability analysis
that mainly depends on the solution of load flow
problem is conducted using different modified
techniques [ 2 ] .
Therefore, this paper proposes a different solution
technique for the problem of Load flow. In this paper,
the powerful particle swarm optimization technique is
used to solve the LF problem. PSO is an optimization
technique formulated by Kennedy and Eberhart [3]
inspired by the natural behavior of a population of birds
or insects. It simulates their performance during the
search for food. There have been another trials in
solving the LF problem by using Evolutionary
techniques. Genetic algorithm (GA) based algorithm
was used to solve the Load Flow problem [4]. The PSO
was used in some applications in the power system
field such as the solution of OPF or the Reactive Power
Planning problem [5].
This paper is organized as follows: Section I1 describes
the formulation of the LF problem as an optimization
problem. In section 111, a background on the PSO
0-7803-8575-6/04/$20.00 02004 IEEE

technique is introduced. Section IV describes the


application of the proposed method on the 6- & IEEE
14-bus test systems. In section V, the results are
analyzed.

11. Formulation of Load Flow as an Optimization


Problem
The load flow equations are simply the power balance
equations at each bus, both active and reactive powers.
The power balance equation expresses the fact that
there are no power losses in any bus, which means that
the input power to a bus equals the output power from
that bus. Therefore, the objective of the LF is to find
the voltage magnitudes and angles of the different
system buses that minimize the difference between the
input power and the output power from the bus. So the
LF problem can be formulated as an optimization
problem. The following is the mathematical derivation
of the LF problem as an optimization problem.
Let Vi: voltage magnitude of bus i
Vj: voltage magnitude of bus j
Si: voltage angle of bus i.
Sj: voltage angle of bus j.
Yij: magnitude of Yij element in bus admittance
matrix.
0,: angle of Yij element in bus admittance matrix.
Pgi:generated power at bus i.
Pli:load power at bus i.
Pti:net power injected to the system at bus i.
Qgi:generated reactive power at bus i.
Qe: load reactive power at bus i.
Qti:net reactive power injected to the system at bus i.
N: number of buses in the system
The load flow equations, at any given bus (i) in the
system, are as follows:
Active power: Pti= P, - Pli
Pti -Pgi+ P,i =o
Pti = zj vi vj Y, cos (0, + si - Sj)
cj vi vj Y, cos (0, + si - Sj) - P, + Pli =o

(1)
(2)
(3)

Reactive power: Qti= Qsi - Qli


Qti - Qgi + Qli =O
Qti= Zj Vi Vj Yij sin (0, + Si - Sj)
Cj Vi Vj Yij sin (0, + Si - Sj) - Qgi+ Qe =O

(5)
(6)

Where j=l:N

(4)

(7)
(8)

N= no. of buses

Let Fpi& Fqibe two functions defined as follows:


Fpi= Zj Vi Vj Yij COS (0, + Si - Sj) - Pgi+ Pli
Fqi= zj Vi Vj Yijsin (0, + Si - Sj) - Qgi+ Qe
742

(9)

(10)

Now the optimization problem can be formulatedas


follows:
Minimize

f(

v, 6 )

(11)

Subject to
Vslack= scheduled value of the slack bus voltage
V, = scheduled value of the PV-bus voltage
P, = scheduled value of PV-bus generated power
Npv: set of PV-buses
P E NPV
Where

f = C (Fpi)

(Fq,)

(12)

So the objective function of the load flow is to


minimize these functions to get the voltages and angles
of the buses, which satisfy the power balance equations.
111. Hybrid Particle Swarm Optimization (HPSO)

The particle swarm optimization algorithm was


originally introduced by Kennedy and Eberhart in 1995
as an alternative to the standard Genetic algorithm
(GA). The PSO was inspired by insect swarms and has
since proven to be a competitor to the GA whenit
comes to function optimization. The PSO model
consists of a number of particles moving around in the
search space, each representing a possible solution to a
numerical problem. Each particle has a position vector
(xi) and a velocity vector (y i), the position (Mi)
is the
best position encountered by the particle (i) during its
search and the position (gbest) is that of the best
particle in the swarm group.
In each iteration the velocity of each particle is updated
according to its best-encountered position and the best
position encountered among the group, using the
following equation:

where

: known as the constriction coefficient


: inertia weight
31 , a 2 : are random values different for each
particle and for each dimension between [0,2]
The position of each particle is then updated in each
iteration by adding the velocity vector to the position
vector.

Equation (13) consists of three terms: the first one


the second
best previous and current position. Finally, the effect of
the swarm group best experience on the velocity of
each individual in the group. This effect is considered
in equation (13
experience (the position of the best particle in the
swarm group) and the i-th
on.
Equation (14) simulates the flying of the particle

toward a new position. The role of the inertia weight w


is considered very important in PSO convergence
behavior [6]. The inertia weight is employed to control
the impact of the previous history of velocities on the
current velocity. In this way, the parameter w regulates
the trade-off between the global and local exploration
abilities of the swarm. A large inertia weight facilitates
global exploration (searching new areas), while a small
one tends to facilitate local exploration, i.e. fine-tuning
the current search area. A suitable value for the inertia
weight w usually provides balance between global and
local exploration abilities and consequently a reduction
on the number of iterations required to locate the
optimum solution. There has been a lot of research in
how to improve the performance of the PSO in means
of faster convergence and to make sure that the PSO
will not get stuck in a local minima [7]-[9]. The
improvements in the PSO are done by trying to have
some of the properties as in the GAbeside the PSO
own properties. One of the most powerful properties of
the GA is the ability to breed and produce better
individuals (children) than the old ones (parents). This
technique is used in the algorithm proposed by this
work. It is used to accelerate the solution of the
problem.
A hybrid model of the standard GA and thePSOis
introduced in [lo]. This model incorporates one major
aspect of the standard GA into the PSO, which is the
reproduction or breeding. Breeding is one of the core
elements that make the standard GA a powerful
algorithm. Therefore, a hybrid PSO with the breeding
property has the potential to reach a better optimum
than the standard PSO. The model for the breeding
process is as mentioned in [lo]:
For the positions of the children:
childl(xi)= pi * parentl(xi) + (1- pi) * parentz(xi) (15)
child2(xi)= pi * parent2(xi)+ (1- pi) * parentl(xi) (16)
For the velocity vectors of the children:
childl@) = (parentl@) + parent2@)) * Iparentl(y)l /
I parentl@) + parentz(y)l
(17)
child2(v) = (parentl(v) + parent2(v)) * Iparentz(v)l/
I parentl&) + parentz(y)l
(18)
where
pi is a uniformly distributed random number
between [0,1]
parentl(xi) : position vector of a randomly chosen
particle to take part in the breeding process.
parent2(xi) : position vector of a randomly chosen
particle to be the other parent in the breeding process.
childl(xi) : position vector of the first offspring
childz(xi): position vector of the first offspring
parentl&) : velocity vector of the first parent
parent2@) : velocity vector of the second parent

IV. Numerical Examples


In this section, the Ward-Hale 6-bus and the IEEE 14bus systems are used to show the applicability of the
proposed algorithm.
It is well known, as shown in the P-V curve fig. (l),
that at any given load on the system there are two
743

solutions for the load flow [ 111. One is a stable solution


(at point 1) and the other one is unstable (at point 2)
from the voltage stability point of view. This means
that as an optimization problem the LF has two global
minima. To avoid getting to the unstable global
minima, the initial positions of the particles were
chosen to be in the neighborhood of the stable solution.
This was accomplished by initializing the voltages
randomly in the range of 0.95 1.05 pu.
This approach gave good results and the PSO reached
the global minimum that is stable.

3
4
5
6

-0.55
-0.0
-0.3
-0.5

-0.13

-0.0
-0.18
-0.05

0.0
0.0
0.0
0.0
0.0

Node 1: is the slack bus.


SC: stands for shunt capacitors installed at this bus.

Bus

Table (2) Branch data


Impedance
Halfofline

System loading

Fig. (1)

P-V curve

a) Ward-Hale 6-bus system:


The Ward-Hale 6-bus system consists of two
generators, four load buses, and seven branches of
which two branches (2-3,4-5) are under load tap setting
transformers. The system is shown in figure (2). The
loads are given in table (1). The branch parameters are
given in table (2). Generator no. 1 is set as the slack bus
with voltage magnitude = 1.05 pu. The second
generator is a PV-bus with scheduled voltage
magnitude = 1.I pu and scheduled generated power =
0.5 pu. The setting of the LTC (2-3) is 0.909, while the
setting of the one (4-5) is 0.975. The parameters used in
the PSO model were as follows: number of particles =
100, w =1.075. The constriction factor was decreasing
with the number of iterations as follows:
= 0.95
= 0.94
k>400
= 0.935
k%OO
= 0.93
k> 1400

x
x
x
x

where k is the iteration counter.


The stopping criterion used was some specified
tolerance for the maximum Fpior Fqi.
This tolerance was 0.0001. The algorithm solved the
problem after (851) iterations with a tolerance of
(8.1644*10-5). To prove the effectiveness and
robustness of the method the loading of the system was
changed. The algorithm was also able to reach the
specified tolerance. Other parameters in the network
such the voltage magnitude of the slack bus, the voltage
or power generated at a PV-bus or the settings of the
were also changed and the proposed method
successfully solved the problem.

TI

9 1

Fig. (2)

Ward-Hale 6-bus system

b) IEEE 14-bus system:


The network shown in figure (3), consists of two
generators, 12 load buses. Three of these load buses are
P-V buses beside one of the generators. The other
generator is taken as the slack bus. The data ofthe
network is found in [9]. The operating conditions of the
system are shown in table (2). The algorithm was also
able to solve this system using the same parameters as
before. The algorithm needed about (4994) iterations to
reach a tolerance of (9.3366*10-5).

V. Analysis of the Results


The proposed algorithm was capable of solving the load
flow problem with the required tolerance. It was
successful in many cases where the load distribution
among the buses was changed both active and reactive
powers. It also solved the problem when the settings of
the slack bus, the PV-buses were changed. As a
comparison between the solution obtained by PSO
against the solution by Newton-Raphson technique,
744

table (4) shows both solutions. Another case was solved


where the system was near to its maximum loading
point of the system. This maximum loading was
obtained by the continuation power flow [2]. Therefore,
the PSO was able to solve the problem even near the
maximum loading point where the NR technique may
sometimes fail to solve the problem due to the
singularity of the Jacobian matrix.

4-4

VLConclusions

Fig. (3)

A new application for particle swarm optimization has


been developed. The PSO algorithm has been
strengthened using breeding technique similar to that
applied in Genetic algorithm (GA). The new suggested
algorithm has been applied to two test systems (HaleWard system and the IEEE 14-bus system). The results
proved the applicability and validity of the new
algorithm as a new tool for load flow solution that
could be helpful in other studies when problems are
encountered due to Jacobian singularity in the classical
techniques.

IEEE 14-bus system

Ta

References

9
10
11
12
13
14

-0.295
-0.090
-0.035
-0.061
-0.135
-0.149

-0.166
-0.058
-0.018
-0.016
-0.058
-0.05

0.19
0.0
0.0
0.0
0.0
0.0

Node 1: is the slack bus.


SC: stands for shunt capacitors installed at this bus.

1. Glem W. Stagg, Ahmed H. Alabiad


methods in Power system analysis Macgraw-Hill,
1968.
2. V. Ajjarapu
flow: A tool for steady state voltage stability
7,
No. 1, February 1992
3. J. Kennedy and R. Eberhart
In Proc. IEEE Int. Conf. Neural
Optimization
Networks, page pp. 19421948,1995
4. Kit Po Wong, An Li
he Load flow

Computation 1995, IEEE International Conference


on, Volume 1, 29 Nov.- 1 Dec. 1995
5. Y.
Fukuyama, S. Takayama, H.Yoshida,
KKawata, and Y. Nakanishi
optimization for reactive power and voltage control
Trans. on Power Systems, pages 1232-1239,2000
6 . Y. Shi and R. Eberhart
Proc. P ~ n n .
Conf. Evolutionary Program., Mar. 1998, pp. 591600
745

7. G. Ciuprina, D. Ioan and I. Munteanu


Intelligent-Particle Swarm Optimization in
Electromagnetics
Magn., vol. 38,
pp. 1037-1040,2002.
8. Vladimiro Miranda and Nuno Fonseca
Evolutionary Particle Swarm Algorithm (EPSO)
Applied to VoltageNar
141h PSCC,
Sevilla, 24-28 June 2002, Session 21, Paper 5.
9.
Estimation Using Hybrid Particle Swarm
Optimization Proc. Of IEEE Power Engineering
Society Winter Meeting, Columbus, 2001.
10. M. Lovbjerg, T. Kiel Rasmussen and T. Krink
Hybrid Particle Swarm Optimizer with Breeding
and Subpopulation
11. Kenji Iba, Hiroshi Suzuki, Masanao Egawa and
Condition with Nose Curve Using Homotopy
Systems, V01.6, No.2, May 1991 p.584-593

746

You might also like