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

net/publication/279963028

A Hybrid Particle Swarm Optimization Approach for Load-Flow Computation

Article  in  International journal of innovative computing, information & control: IJICIC · November 2013

CITATIONS READS

6 6,445

5 authors, including:

Camila Paes Salomon Germano Lambert-Torres

Universidade Federal de Itajubá (UNIFEI) PS Solutions, Itajuba, Brazil
52 PUBLICATIONS   157 CITATIONS    735 PUBLICATIONS   2,856 CITATIONS   

SEE PROFILE SEE PROFILE

Maurilio Pereira Coutinho C.H.V. de Moraes

Universidade Federal de Itajubá (UNIFEI) 51 PUBLICATIONS   40 CITATIONS   
43 PUBLICATIONS   144 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

asdafdhfhgdfghgd View project

Aprimoramento, Fabricação e Aplicação Piloto de Cabeças de Série de Dispositivo Inteligente para Reduzir a Demanda de Ponta em Equipamento de Linha Branca View
project

All content following this page was uploaded by Germano Lambert-Torres on 10 July 2015.

The user has requested enhancement of the downloaded file.

International Journal of Innovative
Computing, Information and Control ICIC International 2013
c ISSN 1349-4198
Volume 9, Number 11, November 2013 pp. 4359–4372

A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH

FOR LOAD-FLOW COMPUTATION

Camila Paes Salomon, Germano Lambert-Torres

Luiz Eduardo Borges da Silva, Maurilio Pereira Coutinho
and Carlos Henrique Valerio de Moraes
Department of System Engineering
Itajuba Federal University
Itajuba, MG 37500-903, Brazil
{ camila; germano; leborges; coutinho; valerio }@unifei.edu.br

Received November 2012; revised March 2013

Abstract. This paper presents a hybrid particle swarm based methodology for solving
load flow in electrical power systems. Load flow is an electrical engineering well-known
problem which provides the system status in the steady-state and is required by several
functions performed in power system control centers. The proposed approach for load flow
computation is based on the minimization of the power mismatches in the system buses
and applies a hybrid particle swarm optimization with mutation operation to achieve
this. The presented model searches for a greater convergence and a wider application in
comparison with traditional methods. As the proposed method is not a tangent method, it
is able to solve even non convex problems, unlike traditional methods. Numerical results
of the proposed methodology are presented for two different power system case studies.
Keywords: Particle swarm optimization, Load flow, Artificial intelligence, Electrical
power system, Computational intelligence

1. Introduction. Load flow studies are required by most functions performed in power
system control centers [1]. Load flow is an electrical engineering well-known problem
which provides the power system operation point in the steady-state [2,3]. This problem
is modeled by a set of non-linear equations, which is commonly solved by the application of
numerical methods [4,5]. Newton-Raphson approach – and its variants – is highlighted as a
numerical method to solve load flow, because of its good and quick convergence. However,
this method includes some difficulties or limitations because of the complex Jacobian
matrix calculation and inversion and also the dependence on good initial estimated values
to guarantee the convergence. Moreover, some current changes in the power system
characteristics, such as an occurrence of a higher resistance-to-reactance ratio (R/X),
may complicate the load flow convergence [6,7]. Finally, Newton-Raphson is a tangent
method, so it cannot solve non convex problems. For example, if part of the power
system is lost, because of a contingency, Newton-Raphson method cannot find the load
flow solution properly.
In [8] it is pointed out that the Newton-Raphson based approaches to solve the load flow
problem do not work well when the power system becomes highly stressed and when there
are nonlinear elements in the power network, which may occur because of the employ-
ment of flexible AC transmission systems (FACTS) devices [9]. [10] also highlights that the
Newton-Raphson method is highly sensitive to the initial settings of the variables. Other
problems in convergence are related in [11], which remarks that the Newton-Raphson

4359
4360 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

based methods do not behave well when the system is heavy loaded and when the Jaco-
bian matrix is ill-conditioned. Moreover, [12] observes that conventional methods have
difficulty to converge if the R/X ratio is high, and they indeed can fail to converge due
to an inappropriate choice of starting values.
Thus, researchers have been attempting new methods to solve the load flow equations,
looking for an easier and a more efficient implementation, as well as overcoming the
aforementioned limitations and convergence problems. Many of these new methods have
applied artificial intelligence techniques.
In the area of artificial intelligence, computational intelligence based algorithms have
been applied successfully to electrical engineering problems, and Particle Swarm Opti-
mization (PSO) is pointed out among these techniques. PSO algorithms are applied in
function optimization and they are based on the behavior of birds’ flocks searching for
food [13]. PSO applications have provided good convergence properties, ease of imple-
mentation and low computational time [14]. The PSO based methodologies have potential
to overcome limitations of the conventional methods for solving load flow because of the
technique nature and the algorithm structure, and they are powerful alternatives when
conventional methods fail to find the load flow solution [4]. An important characteristic
of PSO is that it is not a tangent method, it does not depend on so good initial estimative
values to converge to a best solution and so it is able to solve even non convex problems.
Ref. [15] proposed an adaptive PSO based method for normal and low voltage multiple
load flow solutions, which is important to the purpose of voltage stability assessment,
highlighting the potential of PSO to do this job, while standard methods are not suitable to
do this. In [14,16,17], the authors propose the PSO application to the optimal power flow
problem. In [18], the authors propose a PSO methodology for power system restoration. In
[19], it is presented PSO applied to voltage and reactive power control. Several researches
have also been implementing hybrid models, putting PSO together with other techniques,
for instance the Genetic Algorithm (GA) operators. In [20], it is proposed a Hybrid
PSO with Mutation applied to loss power minimization. In [21], it presents the particle
swarm optimization algorithm for solving the optimal distribution system reconfiguration
problem for power loss minimization, and in [6] it is presented a chaotic PSO algorithm
with local search to the load flow calculation. Finally, in [22], it is proposed a hybrid
algorithm based on combining fuzzy adaptive PSO and Differential Evolution for non-
convex economic dispatch.
This paper proposes the application of a hybrid particle swarm optimization approach
for load flow computation, combining PSO with a mutation operation. This methodology
is based on the minimization of the apparent power mismatches in the system buses. The
variables are continuous and must stay within the specified boundaries defined in the
system input data. Numerical results were obtained for 6-bus and 9-bus power system
case studies [23,24].

2. Load Flow Analysis in Electrical Power Systems. Load flow – or power flow –
provides the system status in the steady-state; it means the determination of the power
system operation point based on the previous knowledge of system parameters and some
variables of the system buses. The purpose of this study is to obtain the system buses
voltages in order to determine later the power adjustments in the generation buses and the
power flow in the system branches. Therefore, it is possible to obtain the amount of power
generation necessary to supply the power demand plus the power losses in the system
branches. Besides, the voltage levels must comply with the predetermined boundaries
and overloaded operations added to those in the stability limit must be prevented [4].
 A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4361

Each power system bus is associated with four variables, where two of them can be
controlled and the other two are results of the system conditions. These variables are:
real power (P ), reactive power (Q), voltage magnitude (|V |) and voltage phase angle (δ).
The power system buses are classified according to these mentioned variables. Type 1
Bus or PQ Bus: Pi and Qi are specified, and |Vi | and δi are calculated; Type 2 Bus or
PV Bus: Pi and |Vi | are specified, and Qi and δi are calculated; Type 3 Bus or Vδ Bus
(“Slack Bus”): |Vi | and δi are specified, and Pi and Qi are calculated.
The equation general form that represents the system power flow computation is given
by (1).
Pi − jQi − yi1 V1 Vi∗ − yi2 V2 Vi∗ − . . . − yin Vn Vi∗ = 0 (1)
where i = 1, . . . , n, bus number; Pi = real power generated or injected in the bus i; Qi
= reactive power generated or injected in the bus i; |Vi | = voltage magnitude of the bus
i; δi = voltage phase angle of the bus i; Vi = |Vi |ejδi , i.e., the voltage in the polar form;
Vi∗ = |Vi |e−jδi , i.e., the conjugate voltage; yik = element of the nodal admittance matrix
Ybus .
The nodal admittance matrix can be computed as follows: if i = k, yik is the sum of the
admittances that come out from the bus i; else yik is the admittance between the buses i
and k, multiplied by −1.
Equation (1) represents a complex and non-linear equations system, and its solution is
usually obtained through approximations adopting numerical methods.

3. Overview on Particle Swarm Optimization and Mutation Operation.

3.1. Particle swarm optimization. Swarm Intelligence is a kind of Artificial Intelli-
gence based on social behavior. An intelligent swarm relates to a population of interacting
individuals able to optimize a function or goal by collectively adapting to the environment
in which they are inserted. PSO has roots in artificial life and evolutionary computation;
it consists of an optimization procedure developed through the simulation of simplified
social models as bird flocks flying randomly in search for food [14,25].
PSO is applied to function optimization by using a population of individuals, i.e.,
a set of particles, where each one is a candidate to the approached problem solution.
These particles are distributed in the search space, each one having position and velocity
parameters at each time instant. Moreover, such particles have cognition about their own
performances and also about their neighbors’ performances. The best individual position
of a particle is called personal best, and the best position of all the particles is called global
best. The particles are assessed through a specific rule function at each time instant. The
rule function performs the interaction between the particles and the environment in which
they are inserted, and it is related to the problem modeling.
The PSO algorithm analyzes, at each time step, the particles displacement in search for
the global best and updating their parameters through defined equations. This process is
iterative and it proceeds until all the particles converge to the achieved global best, which
is adopted as the problem solution.
3.2. Mutation operation in genetic algorithms. Genetic Algorithm (GA) is a search
heuristic iterative procedure inspired by natural evolution and useful to optimization and
search problems [26]. GA uses a population of individuals, in which each one represents
a possible solution to the treated problem [27,28]. The GA accumulated information is
applied to decrease the search space and create new good solutions to the problem domain
[29].
Mutation is a genetic operator applied in order to cover better the problem search space,
maintaining genetic diversity and also preventing the convergence to a local best. The
4362 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

mutation operation is performed by altering one or more gene values in a chromosome,

randomly and with an associated probability to occur [20].

4. A Hybrid Particle Swarm Optimization Approach for Load Flow Computa-

tion. The proposed algorithm for solving load flow computation is named Hybrid Particle
Swarm Optimization with Biased Mutation (HPSOBM) [30]. This methodology is based
on the minimization of the power mismatches in the system buses. The particles posi-
tions are defined as the buses voltages, – the voltage phase angles and magnitudes, – as
depicted in (2), so they assume continuous values within the boundaries specified in the
input data.
xi = {δ2 , δ3 , . . . , δk , . . . , δn , V2 , V3 , . . . , Vk , . . . , Vn } (2)
where i = particle index; xi = particle i position; k = bus index; n = total number of
buses; δk = voltage phase angle at bus k; Vk = voltage magnitude at bus k. Note that
bus 1 is defined as the slack bus, so the particle position does not comprise it.
The rule function parameters are defined as scores and they must be minimized in
the HPSOBM algorithm. The scores are computed as the arithmetic mean of the buses
apparent power mismatch, ∆S. So the arithmetic mean of ∆S’s is the parameter to be
minimized by the algorithm rule function. Each particle has a personal score, i.e., the
value obtained by its personal best. The global score is the score associated to the global
best. The current score is the score obtained by a particle at the current iteration of the
process.
4.1. Algorithm startup. The algorithm begins generating the initial estimate value to
the particles positions, velocities, personal best values and global best values. The initial
estimation procedure is explained as follows. The voltage phase angles, which are one
kind of parameters of the particles positions, begin as random values within the specified
boundary. The voltage magnitudes, which are the other kind of parameters of the particles
positions, require an analysis on the bus type before the initial estimation. In the case of
a PQ bus, the voltage magnitude begins as a random value within the specified boundary;
in the case of a PV bus, the voltage magnitude is the rated value specified in the input
data. The initial velocities are null. The personal best parameters start as the associated
particle position values and the global best parameter starts as an arbitrary particle value.
The scores begin with large values in order to be minimized later.
4.2. Rule function. The iterations are initialized and then the rule function is succes-
sively applied. The procedure explained as follows is executed for each particle of the
population. Firstly, the buses voltages receive the particle position values. Thus, the
unknown power values of PV buses and V δ bus are computed using (1). The power flow
in the system branches is calculated using (3).
Sij = Pij + jQij = Vi (Vi∗ − Vj∗ )Yij∗ + Vi Vi∗ Ysh,i (3)
where Sij = complex apparent power between the buses i and j; Pij = real power between
the buses i and j; Qij = reactive power between the buses i and j; Vi = bus i voltage;
Vj = bus j voltage; Vi∗ = |Vi |e−jδi , i.e., the conjugate voltage; Vj∗ = |Vj |e−jδj , i.e., the
conjugate voltage; Yij = admittance between the buses i and j; Ysh,i = shunt admittance
of the bus i.
The real and reactive power mismatches of each bus are calculated as the sum of the
injected power in the approached bus. The apparent power mismatches arithmetic mean
is obtained and associated to the particle current score. It is also obtained the particle
position which has the biggest score until that instant time. This particle index is kept
and it is used in the mutation operation. The personal best updating verification is made,
 A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4363

and after all the particles pass through the described routine, the global best updating
verification is accomplished.
4.3. Particle parameters updating. The velocities as well as the position of the par-
ticles are updated according to (4), (5) and (6) [13,21,22].
v(t + 1) = w.v(t) + c1 .r1 .(p(t) − x(t)) + c2 .r2 .(g(t) − x(t)) (4)
x(t + 1) = x(t) + v(t + 1) (5)
(wmax − wmin ).t
w = wmax − (6)
ni
where i = particle index; t = iterations counter; ni = total number of iterations; v(t) =
particle i velocity at iteration t; x(t) = particle i position at iteration t; r1 , r2 = random
numbers between 0 and 1; c1 , c2 = acceleration coefficients, both set to a value of 2.0;
p(t) = particle i personal best found at iteration t; g(t) = global best found at iteration
t; w = velocity equation’s inertia weight; wmax = inertia weight maximum value, set to a
value of 0.7; wmin = inertia weight minimum value, set to a value of 0.2.
Then the mutation operation is applied to the worst particle of the current iteration,
i.e., the particle which has the biggest score, and because of this principle it is named as
Biased Mutation [30]. The procedure consists in adding a random value to the particle
position, according to (7) [22].
mx(k) = x(k) + 0.1.[(xmax − xmin ).r + xmin ] (7)
where k = mutated particle index; x(k) = particle position before the mutation operation;
mx(k) = particle position after the mutation operation; r = random number between 0
and 1; xmax = maximum value of the position, related to the specified boundary in the
input data; xmin = minimum value of the position, related to the specified boundary in
the input data.
4.4. Problem solution. The stop criterion is defined by iterations groups and by the
required accuracy for the global score. This tolerance was adopted as 10−5 (pu). The
simulation runs a group of a fixed number of iterations. In the end of the iterations
group, if the global score is bigger than the tolerance, so the algorithm runs another
group of iterations. The process continues until a maximum number of iterations groups
it can run. So the final global best is adopted as the load flow solution.
Note that the proposed methodology can provide several acceptable results for the same
load flow study, depending on the simulation. The reason is because each particle has a
random initial estimate value and the HPSOBM equations also employ random values, so
several solutions can be achieved for the same initial estimative [30]. However, numerical
traditional methods, starting with the same initial estimative values, achieve the same
final results, regardless of the program simulation.
Finally, in order to illustrate the proposed methodology and to compare with the con-
ventional one, Figures 1 and 2 present flowcharts of the Newton-Raphson method and the
HPSOBM method, respectively.
Regarding the flowchart presented in Figure 1, it is important to point out that the
mismatches equationing is a matrix equation which relates the voltage mismatches ∆θ’s
and ∆V’s with the power mismatches ∆P’s and ∆Q’s.
Analyzing the flowcharts depicted in Figures 1 and 2, and comparing the techniques
nature, it can be noted the main differences between them. HPSOBM is a stochastic
method based on a population of individuals searching for the global best. Because of
this stochastic nature, HPSOBM does not require gradient information, it utilizes the rule
function information and works in a random oriented way. HPSOBM is not a tangent
4364 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

Figure 1. Flowchart of the Newton-Raphson based load flow

method, so it can find the solution for non convex problems. On the other hand, Newton-
Raphson is a deterministic method, based on derivative calculation and provides good
results when the functions are continuous, convex and unimodal. Newton-Raphson also
depends on good initial estimative values to guarantee the convergence to an optimal
solution. In the HPSOBM algorithms, the individuals have the ability to adapt to the
environment in which they are inserted and to learn individually and with their neighbors,
through the rule function assessment and Equations (4)-(7). Moreover, the mutation
operation aids the particles to search along the search space, preventing a convergence
to a local best. These characteristics are not present in conventional numerical methods
and help HPSOBM method to obtain a better convergence and work even when these
conventional methods can fail. Therefore, the aforementioned HPSOBM characteristics
indicate its inherent potential to overcome some limitations found in the conventional
deterministic methods, including Newton-Raphson based algorithms to solve load flow.
As mentioned previously, a practical application for HPSOBM to solve load flow when
traditional methods have difficulty to converge could be when part of the power system is
lost, because of a contingency. In such situation, traditional tangent methods, as Newton-
Raphson and its variants, cannot converge to an optimal solution properly, because the
initial estimative values, normally 1 (pu) for voltage magnitudes and 0 (rad) for voltage
phase angles, are far away from the system current status. On the other hand, because
of the aforementioned characteristics, HPSOBM can find the load flow solution even in
this case.
A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4365

Figure 2. Flowchart of the HPSOBM load flow

4366 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

5. Numerical Results. In this section, the proposed methodology for load flow com-
putation is evaluated on two case studies: the 6-bus power system case study [23] and
the 9-bus power system case study [24]. A computational procedure has been developed
based on the HPSOBM methodology for solving the load flow. The software simulations
have been run on a 1.66-GHz Intel(R) T1600 PC. Five runs have been performed for each
test system study case. The selected results are the best solution over these five runs. The
obtained results of the proposed methodology for each case study are compared with those
obtained using a Newton-Raphson based computational method, which is a traditional
method applied to find the load flow solution, as mentioned in the last sections.

5.1. 6-bus power system case study. The first test system is a 6-bus power system
proposed in [23], as shown in Figure 3. This test system comprises six buses, including
three generation buses and three load buses, and 11 branches, composed by transmission
lines.
The simulations for this system held a population of 15 particles. The maximum number
of iterations was 4000, arranged into 40 groups of 100 iterations per group. Figure 4 shows
the global score decreasing along the iterations. The global score starts with 0.541057
(pu) at the first iteration and reaches 9.750141E-06 (pu) at the end of the iterations.
Tables 1 and 2 present the load flow results for the six bus power system obtained
through the application of a Newton-Raphson based method. Tables 3 and 4 present the
results obtained with the proposed methodology.

5.2. 9-bus power system case study. The second test system is a 9-bus power system
proposed in [24], as shown in Figure 5. This test system comprises nine buses, including

Figure 3. 6-bus power system, with all buses and branches

 A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4367

Figure 4. Simulation with 6-bus power system, global score vs. iterations
Table 1. 6-bus simulation applying a Newton-Raphson based method:
buses parameters results

k Vk δk Pk Qk ∆Pk ∆Qk ∆Sk

1 1.050000 0.000000 1.078688 0.159554 8.881784E-16 2.775558E-16 9.305365E-16
2 1.050000 −0.064068 0.500000 0.743507 1.403889E-05 2.428613E-15 1.403889E-05
3 1.070000 −0.074576 0.600000 0.896237 5.474126E-06 −3.330669E-16 5.474126E-06
4 0.989375 −0.073227 −0.700000 −0.700000 1.174903E-05 2.788370E-05 3.025790E-05
5 0.985449 −0.092085 −0.700000 −0.700000 2.679193E-05 3.770603E-05 4.625530E-05
6 1.004427 −0.103796 −0.700000 −0.700000 1.916657E-06 9.151917E-06 9.350463E-06
k = bus index, Vk = bus voltage module k, δk = bus voltage angle k, Pk = real power generated
at bus k, Qk = reactive power generated at bus k, ∆Pk = real power mismatch at bus k, ∆Qk =
reactive power mismatch at bus k, ∆Sk = apparent power mismatch at bus k.

Table 2. 6-bus simulation applying a Newton-Raphson based method:

load flow in the system branches

i j Pij Qij Pji Qji

1 2 0.286872 −0.154176 −0.277824 0.128172
1 4 0.435828 0.201193 −0.424953 −0.199322
1 5 0.355989 0.112537 −0.345254 −0.134492
2 3 0.029298 −0.122687 −0.028895 0.057280
2 4 0.330913 0.460513 −0.315863 −0.451227
2 5 0.155142 0.153519 −0.150163 −0.180054
2 6 0.262484 0.123989 −0.256652 −0.160109
3 5 0.191167 0.231729 −0.180232 −0.260937
3 6 0.437734 0.607227 −0.427701 −0.578597
4 5 0.040828 −0.049423 −0.040466 −0.027852
5 6 0.016142 −0.096627 −0.015646 0.038715
Pij(ji) = real power in the branch composed by the buses i − j (j − i),
Qij(ji) = reactive power in the branch composed by the buses i − j (j − i).
4368 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

Table 3. 6-bus simulation applying the proposed methodology: buses pa-

rameters results
k Vk δk Pk Qk ∆Pk ∆Qk ∆Sk
1 1.050000 0.000000 1.084409 0.231199 −4.440892E-16 −3.989864E-16 5.969970E-16
2 1.050000 −0.065011 0.500000 0.868622 −7.189685E-08 −4.822531E-15 7.189685E-08
3 1.070000 −0.075641 0.600000 0.988291 −5.389207E-07 3.122502E-16 5.389207E-07
4 0.986424 −0.072934 −0.700000 −0.700000 −2.561847E-07 −1.599371E-08 2.566835E-07
5 0.979661 −0.091177 −0.700000 −0.700000 −6.881101E-07 −1.618395E-07 7.068858E-07
6 1.001443 −0.104203 −0.700000 −0.700000 −5.672482E-05 −4.787121E-06 5.692646E-05

Table 4. 6-bus simulation applying the proposed methodology: load flow

in the system branches

i j Pij Qij Pji Qji

1 2 0.291155 −0.133932 −0.281839 0.152565
1 4 0.436949 0.238303 −0.425715 −0.193367
1 5 0.356305 0.165415 −0.345108 −0.123425
2 3 0.029829 −0.089712 −0.029424 0.091739
2 4 0.332800 0.501470 −0.316373 −0.468615
2 5 0.154944 0.195713 −0.149292 −0.178758
2 6 0.264266 0.166467 −0.258072 −0.148772
3 5 0.193275 0.283064 −0.180962 −0.256385
3 6 0.436149 0.650698 −0.425430 −0.597101
4 5 0.042088 −0.003962 −0.041720 0.004697
5 6 0.017082 −0.076549 −0.016442 0.078471

Table 5. 9-bus simulation applying a Newton-Raphson based method:

buses parameters results

k Vk δk Pk Qk ∆Pk ∆Qk ∆Sk

1 1.000000 0.000000 1.736411 −0.464677 0.000000E+00 −3.330669E-16 3.330669E-16
2 1.035000 −0.008727 1.500000 0.189550 −1.961729E-03 −5.551115E-17 1.961729E-03
3 1.029000 −0.090234 0.000000 0.600000 −5.090027E-02 2.199547E-02 5.544942E-02
4 1.027000 −0.117461 0.000000 0.000000 1.907532E-03 −1.109605E-02 1.125882E-02
5 1.012000 −0.155509 −0.550000 −0.270000 5.276012E-04 −2.922534E-03 2.969776E-03
6 1.022000 −0.157254 −0.370000 −0.180000 1.885152E-03 1.150218E-02 1.165564E-02
7 1.007000 −0.186052 −0.680000 −0.450000 −6.588536E-04 4.953215E-03 4.996842E-03
8 1.019000 −0.173835 −0.900000 −0.350000 −2.206755E-03 −4.251948E-03 4.790493E-03
9 1.003000 −0.228289 −0.750000 −0.280000 3.624526E-04 −3.718384E-03 3.736008E-03

two generation buses and five load buses, and 10 branches, composed by eight transmission
lines and two power transformers.
The simulations for this system held a population of 15 particles. The maximum number
of iterations was 8000, arranged into 40 groups of 200 iterations per group. Figure 6 shows
the decrease of the global score along the procedure iterations. The global score starts
with 1.405300 (pu) at the first iteration and achieves 8.653121E-04 (pu) at the end of the
iterations.
Tables 5 and 6 present the load flow results for the 9-bus power system obtained through
the application of a Newton-Raphson based method. Tables 7 and 8 present the results
obtained applying the proposed methodology.
 A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4369

Figure 5. 9-bus power system, with all buses and branches

Table 6. 9-bus simulation applying a Newton-Raphson based method:

load flow in the system branches

i j Pij Qij Pji Qji

1 3 1.736411 −0.464677 −1.736411 0.637215
2 4 1.501962 0.189550 −1.501962 −0.025241
3 5 0.752231 0.075248 −0.743055 −0.025590
3 8 1.035080 0.042368 −1.024945 0.043783
4 6 0.581620 0.014196 −0.578893 0.008910
4 7 0.918434 0.186151 −0.910940 −0.120376
5 7 0.192528 −0.003886 −0.191369 0.009715
6 8 0.207007 0.011096 −0.206555 −0.007640
7 9 0.422968 −0.001037 −0.420869 0.018821
8 9 0.333706 0.028261 −0.329494 −0.009899

It is possible to verify the effectiveness of the proposed methodology analyzing how small
the obtained power mismatches are. A remark must be made at this point. The 9-bus test
system voltage magnitudes obtained through the Newton-Raphson based method have
accuracy until the third decimal place, smaller than the proposed methodology accuracy,
so it obscures a direct comparison between the power mismatches obtained through these
two techniques. However, one can note that in both of the case studies the proposed
hybrid particle swarm optimization approach provides voltage modules and angles very
similar to those obtained through Newton-Raphson based methods. Moreover, the power
mismatches are properly minimized, most of them are smaller than the tolerance usually
accepted, which is about 10−4 (pu). So the aforementioned facts prove the effectiveness
4370 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

Figure 6. Simulation with 9-bus power system, global score vs. iterations

Table 7. 9-bus simulation applying the proposed methodology: buses pa-

rameters results
k Vk δk Pk Qk ∆Pk ∆Qk ∆Sk
1 1.000000 0.000000 1.790318 −0.466095 0.000000E+00 −1.720846E-15 1.720846E-15
2 1.035000 −0.011652 1.500000 0.193644 2.402975E-08 −2.997602E-15 2.402975E-08
3 1.029339 −0.093012 0.000000 0.600000 5.968895E-09 3.948885E-08 3.993742E-08
4 1.026682 −0.120277 0.000000 0.000000 1.056733E-06 1.195302E-06 1.595441E-06
5 1.011912 −0.158256 −0.550000 −0.270000 7.443516E-06 1.767402E-05 1.917751E-05
6 1.022208 −0.160093 −0.370000 −0.180000 4.100075E-06 1.469834E-05 1.525948E-05
7 1.006626 −0.188952 −0.680000 −0.450000 1.492413E-05 2.406256E-05 2.831495E-05
8 1.018853 −0.176731 −0.900000 −0.350000 −2.056931E-06 1.553391E-05 1.566951E-05
9 1.001943 −0.231275 −0.750000 −0.280000 2.681026E-03 7.226422E-03 7.707728E-03

Table 8. 9-bus simulation applying the proposed methodology: load flow

in the system branches

i j Pij Qij Pji Qji

1 3 1.790318 −0.466095 −1.790318 0.648855
2 4 1.500000 0.193644 −1.500000 −0.029645
3 5 0.752925 0.079939 −0.743727 −0.030161
3 8 1.037393 0.048148 −1.027214 0.038373
4 6 0.581006 0.006756 −0.578284 0.016305
4 7 0.918993 0.186797 −0.911484 −0.120885
5 7 0.193720 −0.002297 −0.192547 0.008198
6 8 0.208280 0.015274 −0.207821 −0.011768
7 9 0.424016 0.005664 −0.421904 0.012225
8 9 0.335036 0.033414 −0.330777 −0.014848

of HPSOBM methodology to find optimal solutions for the load flow problem minimizing
the power mismatches on the system buses.

6. Conclusions. This paper proposes a hybrid particle swarm optimization approach

for load flow computation. The methodology was evaluated on 6-bus and 9-bus power
 A HYBRID PARTICLE SWARM OPTIMIZATION APPROACH 4371

system case studies through the developed computational program. In both of the case
studies, the simulations results indicates that the proposed methodology presents accept-
able solutions for the buses power mismatches and the results are also better or as good
as those obtained through Newton-Raphson based methods.
The main advantages of this presented methodology are the ease and flexibility of
implementation, and its better convergence. The proposed HPSOBM method is not a
tangent method, so it is able to solve even non convex problems, unlike traditional methods
as Newton-Raphson and its variants. It is suggested for future works an improvement in
the procedure in order to achieve lower mismatches in the cases which they are larger
than 10−4 and also to evaluate its applicability in cases where the traditional methods
have limitations or fail to solve load flow.

Acknowledgments. The authors would like to thank CNPq, CAPES and FAPEMIG –
Brazilian research funding agencies, for the research scholarships, which supported this
work.

REFERENCES

[1] V. L. Paucar and M. J. Rider, On the use of artificial neural networks for enhanced convergence of
the load flow problem in power systems, Proc. of Intelligent Systems Applications to Power Systems,
pp.153-158, 2001.
[2] B. Stott, Review of load flow calculation methods, IEEE Proc., vol.62, pp.916-929, 1974.
[3] G. Lambert-Torres, L. E. B. da Silva, B. Valiquette, H. Greiss and D. Mukhedkar, A fuzzy knowledge-
based system for bus load forecasting, Fuzzy Logic Technology and Applications, pp.221-228, 1994.
[4] O. I. Elgerd, Electric Energy Systems Theory: An Introduction, McGraw-Hill Publishing Co. Ltd,
1975.
[5] G. Lambert-Torres, A. P. A. da Silva, V. H. Quintana and L. E. B. da Silva, Classification of power
system operation point using rough sets techniques, Proc. of IEEE International Conference on
Systems, Man and Cybernetics, Beijing, China, vol.3/4, pp.1898-1903, 1996.
[6] P. Acharjee and S. K. Goswami, Chaotic particle swarm optimization based reliable algorithm to
overcome the limitations of conventional power flow methods, Proc. of Power Systems Conference
and Exposition, 2009.
[7] A. P. A. da Silva, C. Ferreira, A. C. Zambroni and G. Lambert-Torres, A new constructive ANN and
its applications to electric load representation, IEEE Trans. on Power Syst., vol.12, no.4, pp.1569-
1575, 1997.
[8] T. O. Ting, K. P. Wong and C. Y. Chung, Hybrid constrained genetic algorithm/particle swarm
optimization load flow algorithm, IET Gener. Transm. Distrib., vol.2, pp.800-812, 2008.
[9] L. E. B. da Silva, L. E. L. de Oliveira, G. Lambert-Torres, V. F. da Silva, J. O. P. Pinto and B.
K. Bose, Improving the dynamic response of active power filters based on the synchronous reference
frame method, Proc. of Applied Power Electronics Conference, Dallas, USA, vol.2, pp.742-748, 2002.
[10] A. G. Bakirtzis, P. N. Biskas, C. E. Zoumas and V. Petridis, Optimal power flow by enhanced genetic
algorithm, IEEE Trans. on Power Syst., vol.17, pp.229-236, 2002.
[11] A. A. El-Dib, H. K. M. Youssef, M. M. El-Metwally and Z. Osman, Load flow solution using hybrid
particle swarm optimization, Proc. of the 2004 International Conference on Electrical, Electronic
and Computing Engineering, pp.742-746, 2004.
[12] M. Syai’in, K. L. Lian, N. C. Yang and T. H. Chen, A distribution power flow using particle swarm
optimization, IEEE Power and Energy Society General Meeting, pp.1-7, 2012.
[13] J. Kennedy and R. Eberhart, Swarm Intelligence, Morgan Kaufmann Publishers, 2001.
[14] K. S. Swarup, Swarm intelligence approach to the solution of optimal power flow, J. Indian Inst.
Sci., vol.86, pp.439-455, 2006.
[15] P. Acharjee and S. K. Goswami, An adaptive particle swarm optimization based method for nor-
mal and low voltage multiple load flow solutions, Joint International Conference on Power System
Technology and IEEE Power India Conference, POWERCON 2008, pp.1-6, 2008.
[16] I. Oumarou, D. Jiang and Y. Cao, Particle swarm optimization applied to optimal power flow
solution, Proc. of the 5th Conference on Natural Computation, pp.284-288, 2009.
 4372 C. P. SALOMON, G. LAMBERT-TORRES, L. E. B. DA SILVA ET AL.

[17] C. Kumar and P. Raju, Constrained optimal power flow using particle swarm optimization, Inter-
national Journal of Emerging Technology and Advanced Engineering, vol.2, no.2, pp.235-241, 2012.
[18] G. Lambert-Torres, H. G. Martins, M. P. Coutinho, C. P. Salomon and L. S. Filgueiras, Particle
swarm optimization applied to restoration of electrical energy distribution systems, Lecture Notes in
Computer Science, vol.5370, pp.228-238, 2008.
[19] H. Yoshida, K. Kawata, Y. Fukuyama, S. Takayama and Y. Nakanishi, A particle swarm optimiza-
tion for reactive power and voltage control in electric power systems considering voltage security
assessment, IEEE Trans. on Power Syst., vol.15, pp.1232-1239, 2001.
[20] A. A. A. Esmin, G. Lambert-Torres and A. C. Z. de Souza, A hybrid particle swarm optimization
applied to loss power minimization, IEEE Trans. on Power Syst., vol.20, pp.859-866, 2005.
[21] A. A. A. Esmin and G. Lambert-Torres, Application of particle swarm optimization to optimal power
systems, International Journal of Innovative Computing, Information and Control, vol.8, no.3(A),
pp.1705-1716, 2012.
[22] T. Niknam, H. D. Mojarrad and M. Nayeripour, A new hybrid fuzzy adaptive particle swarm opti-
mization for non-convex economic dispatch, International Journal of Innovative Computing, Infor-
mation and Control, vol.7, no.1, pp.189-202, 2011.
[23] A. J. Wood and B. F. Wollenberg, Power Generation, Operation and Control, 2nd Edition, John
Wiley & Sons Ltd., 1996.
[24] W. F. Alves, Proposition of Test Systems for Power Systems Computational Analysis, Master Thesis,
Fluminense Federal University, Niteroi, Brazil, 2007.
[25] J. Kennedy and R. Eberhart, The particle swarm: Social adaptation of knowledge, Proc. of IEEE
International Evolutionary Computation, pp.303-308, 1997.
[26] L. Davies, Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York, 1991.
[27] D. Beasley, D. Bull and R. Martin, An overview of genetic algorithms: Part 1, fundamentals, Tech-
nical Report, Inter-University Committee on Computing, 1993.
[28] D. E. Goldberg and J. H. Holland, Genetic algorithms and machine learning: Introduction to the
special issue on genetic algorithms, Machine Learning, vol.3, 1988.
[29] G. Lambert-Torres, H. G. Martins, M. P. Coutinho, L. E. B. da Silva, F. M. Matsunaga, R. A.
Carminati and J. C. Neto, Genetic algorithm to system restoration, Proc. of the 2009 World Congress
on Electronics and Electrical Engineering, 2009.
[30] C. P. Salomon, M. P. Coutinho, G. Lambert-Torres and C. Ferreira, Hybrid particle swarm opti-
mization with biased mutation applied to load flow computation in electrical power systems, Lecture
Notes in Computer Science, vol.6728, pp.595-605, 2011.

View publication stats