See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251543462

Optimal Placement of Distributed Generation Using Particle Swarm

Optimization

Article · January 2006

CITATIONS READS

66 4,744

2 authors, including:

Wichit Krueasuk
Sripatum University
6 PUBLICATIONS 82 CITATIONS

SEE PROFILE

All content following this page was uploaded by Wichit Krueasuk on 24 April 2014.

The user has requested enhancement of the downloaded file.

 Optimal Placement of Distributed Generation Using
Particle Swarm Optimization
Wichit Krueasuk Weerakorn Ongsakul
Department of Electrical Engineering School of Environment, Resources and
Faculty of Engineering Sripatum Development, Asian Institute of Technology
University, Bangkok, Thailand Pathumthani, Thailand
E-mail: [email protected] E-mail: [email protected]

ABSTRACT Optimal placement of DG (OPDG) in distribution

network is an optimization problem with continuous and
This paper proposes a particle swarm optimization discrete variables. Many researchers have used
(PSO) algorithm for optimal placement of distributed evolutionary methods for finding the optimal DG
generation (DG) in a primary distribution system to placement [6]-[9].
minimize the total real power loss. The PSO provides a
population-based search procedure in which individuals In fact, three types of DG are considered as follows:
called particles change their positions with time. During Type 1: DG is capable of supplying only real power;
flight, each particle adjusts its position according to its Type 2: DG is capable of supplying only reactive power;
own experience, and the experience of neighboring Type 3: DG is capable of supplying real power but
particles, making use of the best position encountered by consuming proportionately reactive power.
itself and its neighbors. Initially, the algorithm randomly
generates the particle positions representing the size and In [10], a Newton-Raphson algorithm based load flow
location of DG. Each particle will move from its current program is used to solve the load flow problem. The
position using the velocity and the distance from current methodology for optimal placement of only one DG type
best local and global solution reached. The velocity 1 is proposed. Moreover, the heuristic search requires
consists of inertia of the particle, memory, and exhaustive search for all possible locations which may
cooperation between particles. The proposed PSO not be applicable to more than one DG. Therefore, in
algorithm is used to determine optimal sizes and this paper, PSO method is proposed to determine the
locations of multi-DGs. Three types of DG are optimal location and sizes of multi-DGs to minimize the
considered and the distribution load flow is used to total real power loss of the distribution systems.
calculate the exact loss. Test results indicate that PSO
The organization of this paper is as follows. Section 2
method can obtain better results than the simple
addresses the problem formulation. The DG type and
heuristic search method on the 33-bus and 69-bus radial
heuristic search method are explained in Section 3. The
distribution systems. The PSO can obtain maximum loss
PSO algorithm is represented in Section 4. A PSO
reductions for each of three types of optimally placed
computation procedure on for the OPDG problem is
multi-DGs. Moreover, voltage profile improvement and
given in Section 5. Simulation result on the test systems
branch current reduction are obtained.
are illustrated in Section 6. Then, the conclusion is given
in Section 7.
Keywords- Distributed Generation, DG types, Optimal
DG size, Particle Swarm Optimization
2. PROBLEM FORMULATION
1. INTRODUCTION
The real power loss reduction in a distribution system is
required for efficient power system operation. The loss
Distributed Generation (DG) is a small generator spotted
in the system can be calculated by equation (1) [11],
throughout a power system network, providing the
given the system operating condition,
electricity locally to load customers [1]. DG can be an
alternative for industrial, commercial and residential n n
applications. DG makes use of the latest modern PL = ∑∑ Aij ( Pi Pj + Qi Q j ) + Bij (Qi Pj − Pi Q j ) (1)
technology which is efficient, reliable, and simple i =1 j =1

enough so that it can compete with traditional large

generators in some areas [2] [3]. where,
Rij cos (δ i −δ j )
Placement of DGs is an interesting research area due to Aij =
VV
i j
economical reason. Distributed generation systems (such
as fuel cells, combustion engines, microturbines, etc) can Rij sin (δ i −δ j )
Bij =
reduce the system loss and defer investment on VV
i j
transmission and distribution expansion. Appropriate
size and optimal locations are the keys to achieve it [4] where, Pi and Qi are net real and reactive power injection
[5]. in bus ‘i’ respectively, Rij is the line resistance between
bus ‘i’ and ‘j’, Vi and δi are the voltage and angle at bus wind turbines, induction generator is used to produce
‘i’ respectively real power and the reactive power will be consumed in
the process [12]. The amount of reactive power they
The objective of the placement technique is to minimize require is an ever increasing function of the active power
the total real power loss. Mathematically, the objective output. The reactive power consumed by the DG (wind
function can be written as: generation) in simple form can be given as in equation
N SC (9) as in the case of [13]
Minimize PL = ∑ Loss k
(2)
QDG = − ( 0.5 + PDG )
k =1 2
(9)
Subject to power balance constraints
N N The loss equation will be modified. After following the
∑ PDGi = ∑ PDi + PL
i =1 i =1
(3) similar methodology of the first two types, optimal DG
size can be found by solving equation (10).

≤ Vi ≤ Vi
min max
voltage constraints: Vi (4) 0.0032 Aii PDGi
3
+ PDGi [1.004 Aii + 0.08 Aii QDi − 0.08Yi ] + (10)
max
( X i − Aii PDi ) = 0
current limits: I ij ≤ I ij (5)
Equation (10) gives the amount of real power that a DG
where: Lossk is distribution loss at section k, NSC is should produce when located at but ‘i’, so as to obtain
total number of sections, PL is the real power loss in the the minimum system loss whereas the amount of
system, PDGi is the real power generation DG at bus i, reactive power that it consumes can be calculated from
PDi is the power demand at bus i. equation (9).

3. DG TYPE AND HEURISTIC METHODLOGY 4. PARTICLE SWARM OPTIMIZATION

Particle swarm optimization (PSO) is a population-based
3.1. DG TYPE 1 optimization method first proposed by Kennedy and
Eberhart in 1995, inspired by social behavior of bird
Certain type of DGs like photovoltaic will produce real
flocking or fish schooling [16]. The PSO as an
power only. To find the optimal DG size at but ‘i’, when
optimization tool provides a population-based search
it supplies only real power, the necessary condition for
procedure in which individuals called particles change
minimum loss is
their position (state) with time. In a PSO system,
n particles fly around in a multidimensional search space.
∑( A P − B Q )
1 (6)
Pi = PDGi − PDi = − ij j ij j
During flight, each particle adjusts its position according
Aii j =1 to its own experience (This value is called Pbest), and
j ≠i
according to the experience of a neighboring particle
(This value is called Gbest),made use of the best position
From equation (6), we obtain the following relationship:
encountered by itself and its neighbor (Figure 1).
n

∑( A P − B Q )
1 (7)
PDGi = PDi − ij j ij j
Aii j =1
j ≠i

Equation (7) gives the optimal DG size for each bus so

as to minimize the total real power loss.

3.2. DG TYPE 2

For synchronous condenser DG, it provides only reactive Figure 1: Concept of a searching point by PSO
power to improve voltage profile. To determine the
optimal DG placement, we again differentiate the loss This modification can be represented by the concept of
equation on either side with respect to Qi. The optimal velocity. Velocity of each agent can be modified by the
DG size for every bus in the system is given by equation following equation:
(8)
vidk +1 = ωvidk + c1rand ×( pbestid − sidk ) + c2rand × ( gbestd − sidk ) (11)
N

∑(A Q + Bij Pj )
1 (8)
QDGi = QDi − ij j
Aii j =1 Using the above equation, a certain velocity, which
j ≠i
gradually gets close to pbest and gbest can be calculated.
3.3. DG TYPE 3 The current position (searching point in the solution
space) can be modified by the following equation:
Here, we consider that the DG will supply real power
and in turn will absorb reactive power. In case of the
 s idk + 1 = s idk + v idk + 1 , i = 1, 2 , ..., n , (12) Start

d = 1, 2 , ..., m Input system data

(1)

where s k is current searching point, s k +1 is modified Calculate the original loss

using Bw-Fw Sweep (2)

searching point, v k is current velocity, v k +1 is modified

Initialize particle

velocity of agent i, v pbest is velocity based on pbest, population (3)

vgbest is velocity based on gbest, n is number of Calculate the total loss (4)

particles in a group, m is number of members in a Record Pbest (5),Gbest(6)

particle, pbesti is pbest of agent i, gbesti is gbest of the

group, ωi is weight function for velocity of agent i, ci
Update particle
position and velocity (7)

is weight coefficients for each term.

No Check the stopping
criterion (8)
The following weight function is used:
Particle Swarm
Yes
ωmax − ωmin Optimization
ωi = ωmax −
Print out location and
.k (13) size of DG (9)
kmax
End

where, ωmin and ωmax are the minimum and maximum

weights respectively. k and kmax are the current and Figure 2: PSO-OPDG computational procedure
maximum iteration. Appropriate value ranges for C1 and
C2 are 1 to 2, but 2 is the most appropriate in many
cases. Appropriate values for ωmin and ωmax are 0.4 and 6. SIMULATION RESULTS
0.9 [17] respectively.
The distribution test systems are the 33 bus [18] and 69
bus [19] systems. The 33 bus system has 32 sections
5. PSO PROCEDURE with the total load 3.72 MW and 2.3 MVar shown in
Figure 3. The original total real power loss and reactive
The PSO-based approach for solving the OPDG problem power loss in the system are 221.4346 kW and 150.1784
to minimize the loss takes the following steps: kVar, respectively. The 69 bus system has 68 Sections
with the total load of 3.80 MW and 2.69 MVar, shown in
Step 1: Input line and bus data, and bus voltage limits. Figure 4. The original total real and reactive power
Step 2: Calculate the loss using distribution load flow losses of the system are 230.0372 kW and 104.3791
based on backward-forward sweep. kvar, respectively. For PSO parameters, population
Step 3: Randomly generates an initial population (array) size=200, Maximum generation (kmax) = 100. The
of particles with random positions and velocities maximum number of DG is 3 for each type.
on dimensions in the solution space. Set the
iteration counter k = 0.
Step 4: For each particle if the bus voltage is within the
limits, calculate the total loss in equation (1).
Otherwise, that particle is infeasible.
Step 5: For each particle, compare its objective value
with the individual best. If the objective value is
lower than Pbest, set this value as the current
Pbest, and record the corresponding particle
position. Figure 3: The 33 bus radial distribution system.
Step 6: Choose the particle associated with the minimum
individual best Pbest of all particles, and set the
value of this Pbest as the current overall best
Gbest.
Step 7: Update the velocity and position of particle using
(11) and (12) respectively.
Step 8: If the iteration number reaches the maximum
limit, go to Step 9. Otherwise, set iteration index k
= k + 1, and go back to Step 4.
Step 9: Print out the optimal solution to the target
problem. The best position includes the optimal
locations and size of DG or multi-DGs, and the
corresponding fitness value representing the
minimum total real power loss.

Figure 4: The 69 bus radial distribution system.

 Table 1: PSO result of the 69 bus test system
Total real power loss Min Avg. Max

(kW)
80.1933 95.4714 203.2326
Average Time (sec.) 5.6341

For DG type 1, the convergence characteristic of the best

solution of PSO is shown in Figure 5. Figure 6 shows the
total real power loss from 100 trials of PSO-OPDG. The
average CPU time is 5.6341 second.

Figure 6: The total loss distribution from 100 trials of a

69 bus test system

The improvement in the voltage profile after optimally

placing the DGs is shown in Figure 7. Without DG, the
bus no. 64 has the lowest voltage of 0.8891 p.u. and the
bus voltage has improved to 0.9453 p.u. after installing
DG.

Figure 5: Convergence characteristic of the 69 bus test

system.

For the 33 and 69 bus systems, in Tables 2-4, the PSO

can obtain the same optimal size and location as the
heuristic search [10] for one types 1-3 DG. For the 33
bus system, one type-1 DG can reduce the total real and
reactive power loss by 47.49% & 43.12% compared to
28.29% & 27.55% and 26.01% & 23.05% for DG types
2 and 3, respectively. For three type-1 DGs, they can
further reduce the real and reactive power loss by
65.60% & 64.73% compared to 34.56% & 34.24% and
24.66% & 22.03% for DG types 2 and 3, respectively. In
the 69 bus system, three type-1 DGs can reduce the real Figure 7: Bus voltage before and after DG Installation
and reactive power loss by 69.19% & 65.81% compared for DG type-1.
to 35.53% & 33.78% and 29.71% & 29.13% for DG
types 2 and 3, respectively.

Table 2: Optimal DG placement for DG type 1

Table 3: Optimal DG placement for DG type 2

Table 4: Optimal DG placement for DG type 3

7. CONCLUSION generation: definitions, benefits and issues,”

Energy Policy 33, pp787-798, 2005.
In this paper, a particle swarm optimization for optimal [4] William Rosehart and Ed Nowicki, “Optimal
placement of multi-DGs is efficiently minimizing the Placement of Distributed Generation,” 14th
total real power loss satisfying transmission line limits PSCC, Sevilla, 24-28 June 2002.
and constraints. The methodology is fast and accurate
[5] Caisheng Wang and M. Hashem Nehrir,
in determining the sizes and locations. DG regulating
“Analytical Approaches for Optimal
bus voltage will be considered in future research work.
Placement of Distributed Generation Sources
in Power Systems,” IEEE Transactions on
8. ACKNOWLEDGEMENT Power Systems, vol.19, no.4, pp2068-2076,
2004.
The authors would like to thank Dr. Keerati
Chayakulkheeree, Head of Department of Electrical [6] T.Niknam, A.M. Ranjbar, A.R. Sirani,
Engineering of Sripatum University for his guidance in BMozafari and A. Ostadi, “Optimal Operation
PSO programming. of Distribution System with Regard to
Distributed Generation : A Comparison of
Evolutionary Methods,” IEEE Conference
REFERENCES IAS 2005, pp2690-2696, 2005.
[1] Thomas Ackermann, GÖran Andersson and [7] N.Mithulananthan, Than Oo and Le Van Phu,
Lennart SÖder, “Distributed generation: a “Distributed Generator Placement in Power
definition,” Electric Power Systems Research Distribution System Using Genetic Algorithm
57, pp195-204, 2001. to Reduce Losses,” Thammasat
Int.J.Sc.Tech.,vol.9,no.3, July-September,
[2] W.El-Khattam , M.M.A.Salama, “Distributed
pp55-62, 2004.
generation technologies, definitions and
benefits,” Electric Power Systems Research [8] Andrew Keane and Mark O’Malley, “Optimal
71, pp119-128, 2004. Allocation of Embedded Generation on
Distribution Networks,” IEEE Transactions
[3] G.Pepermans, J.Driesen, D.Haeseldonckx,
on Power System, vol.20, no.3, August,
R.Belmans, and W.D’haeseleer, “Distributed
pp1640-1646, 2005.
 [9] Kyu-Ho Kim, Yu-Jeong Lee,Sang-Bong Mr. Wichit Krueasuk received his
Rhee, Sang-Kuen Lee and Seok-Ku You, B.Eng from Sripatum University,
“Dispersed Generator Placement using Fuzzy- Thailand, degrees in electrical
GA in Distribution System,” IEEE engineering, in 2000.
Conference, pp1148-1152, 2002. Currently, he is a faculty member of
[10] Naresh Acharya, Pukar Mahat, and N. Sripatum University and also he is pursuing his
Mithulanathan, “An analytical approach for M.Eng. at AIT, Thailand. His research interest includes
DG allocation in primary distribution the optimization and Artificial Intelligence (AI)
network,” International Journal of Electrical application to distribution system, and energy
Power & Energy System, to be published. management systems.
[11] I.O. Elgerd, Electric Energy System Theory:
an Introduction. McGraw Hill., 1971. Dr. Weerakorn Ongsakul received his
[12] M. Ermis, H. B. Eratn, M. Demirekler, B. M. B.Eng. from Chulalongkorn University,
Saribatir, Y. Uctung, M. E. Sezer etal, Thailand, M.S. and Ph.D. from Texas
“Various Induction Generator Scheme for A&M University, USA, all degrees in
Wind Power Electricity Generation,” Electric electrical engineering, in 1988, 1991
Power Systems Research, vol.23, pp71-83, and 1994 respectively.
1992. Currently he is an Associate Professor at AIT,
[13] DTI, “Network Performance Benefits of Bangkok Thailand. His research interest includes AI
Energy Storage for a Large Wind Farm,” applications to power systems, parallel processing
http://www.dti.gov.uk/renewables/publication applications, power system operation & control, and
s/pdfs/kel002460000.pdf/, 2004. power system restructuring.
[14] W.D. Stevenson, Elements of Power System
Analysis. Tokyo: 3rd ed., McGraw Hill
Kogakush, Ltd., 1975.
[15] Hadi Saadat, Power System Analysis.
Singapore: McGraw Hill, 1999.
[16] Kennedy J and Eberhart R, “Particle Swarm
Optimizer,” IEEE International Conference
on Neural Networks (Perth, Australia), IEEE
Service Center Piscataway, NJ, IV, pp1942-
1948, 1995.
[17] Eberhart, R.C. and Shi, Y, “Comparing
inertial weights and Constriction factor in
particle Swarm optimization,” proceeding of
the 2000 International Congress on
Evaluationing Computation, San Diego,
California, IEEE Service Center, Piscataway,
NJ, pp84-88, 2000.
[18] M.A. Kashem, V. Ganapathy, G.B. Jasmon
and M.I. Buhari, “A Novel Method for Loss
Minimization in Distribution Networks,”
proceeding of International Conference on
Electric Utility Deregulation and
Restructuring and Power Technologies,
pp251-255, 2000.
[19] M.E. Baran and F.F. Wu, “Optimal Sizing of
Capacitor Placed on Radial Distribution
Systems,” IEEE Trans,Vol. PWRD-2, pp735-
743, 1989.

View publication stats