Anda di halaman 1dari 17

Operational Research

https://doi.org/10.1007/s12351-018-0420-3

ORIGINAL PAPER

Optimal reactive power dispatch using water wave


optimization algorithm

Yongquan Zhou1,2 · Jinzhong Zhang1 · Xiao Yang1 · Ying Ling1

Received: 16 July 2017 / Revised: 11 April 2018 / Accepted: 16 August 2018


© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract
This paper presents water wave optimization (WWO) algorithm to solve the optimal
reactive power dispatch (ORPD) problem with the continuous and discrete control
variables in power system. The ORPD problem is defined as a complex, discrete,
constrained nonlinear combinatorial optimization problem. The WWO algorithm is
utilized to find the optimized values of control variables such as generator voltages,
tap positions of tap changing transformers and the amount of reactive compensation
devices to achieve minimized value of active power losses. The WWO algorithm
not only effectively avoids the shortcomings of local search and poor calculation
accuracy, but also accelerates the convergence rate to find the global optimal solu-
tion. The WWO algorithm is implemented on standard IEEE 30-bus power system
that is to verify the effectiveness and feasibility of the WWO algorithm to tackle
with the ORPD problem. Compared with other algorithms, the WWO algorithm can
find the set of the optimal solutions of control variables. The simulation experiment
indicates that the WWO algorithm has better overall performance to reduce the real
power losses.

Keywords  Water wave optimization algorithm · Optimal reactive power dispatch ·


Control variables · Active power losses · Simulation experiment

1 Introduction

With the development of the economy, the power load increases rapidly and the
ORPD problem is an important part for power system to operate safely, which has
aroused wide public attention. The power system achieves the optimal dispatch and

* Yongquan Zhou
yongquanzhou@126.com
1
College of Information Science and Engineering, Guangxi University for Nationalities,
Nanning 530006, China
2
Key Laboratory of Guangxi High Schools Complex System and Computational Intelligence,
Nanning 530006, China

13
Vol.:(0123456789)
Y. Zhou et al.

control of reactive power, which can improve the quality of voltage and reduce the
power transmission losses so as to reduce operating costs and enhance the level of
stable operation. The ORPD problem (Alsac and Stott 1974; Lee et al. 1985; Kena-
rangui and Seifi 1994; Lai 2005; Varadarajan and Swarupa 2008; Duman et al. 2012)
is a nonlinear combination optimization problem with discrete, complex, multi-con-
strained features. The traditional methods to solve the ORPD problem are interior
point method (Momoh et  al. 1994), linear programming (Deeb and Shahidehpur
1990), nonlinear programming method (Wu et al. 1994), Gradient method (Lee et al.
1985), quadratic programming method (Grudinin 1998), Newton method (Bjelogrlic
et al. 1990), which generally causes problems, such as large error, curse of dimen-
sionality, difficult to deal with discrete variable, so that it is difficult to obtain the
ideal result.
In recent years, the traditional methods have some shortcomings to solve ORPD
problem, the heuristic optimization algorithms have been come forward by some
scholars to solve the complicated problem. Such as bat algorithm (BA) (Yang and
He 2013), flower pollination algorithm (FPA) (Yang 2012), particle swarm optimi-
zation (PSO) (Kennedy and Eberhart 1995), sine cosine algorithm (SCA) (Mirjalili
2016), crow search algorithm (CSA) (Askarzadeh 2016), water wave optimization
(WWO) (Zheng 2015). The adapted genetic algorithm with adjusting population
size is proposed to solve the ORPD problem (Attia et  al. 2012). The quasi-oppo-
sitional teaching learning based optimization is applied to tackle with the ORPD
problem, which accelerates the convergence speed and improves calculation accu-
racy (Mandal and Roy 2013). The opposition-based gravitational search algorithm
has been put forward to solve the ORPD problem, which indicates that its robust-
ness and effectiveness for solving ORPD problem of power systems (Shaw et  al.
2014). The harmony search algorithm is used to solve the ORPD problem to find the
optimal control variables and the minimal active power loss (Khazali and Kalantar
2011). The gray wolf optimizer is posed to solve the ORPD problem and that is
able to achieve less power loss (Sulaiman et  al. 2015). Artificial bee colony algo-
rithm is presented to solve the ORPD problem with discrete and continuous control
variables (Mouassa and Bouktir 2016). Ant lion optimizer is proposed to solve the
ORPD problem in power system, which not only finds the set of optimal control var-
iables, but also gets real power loss (Mouassa et al. 2017). The colonial competitive
differential evolution has strong effectiveness and robustness to solve the optimal
economic load dispatch problem (Ghasemi et al. 2016). A novel teaching–learning-
based optimization algorithm is applied to solve optimal reactive power dispatch
problem, the result demonstrates that the optimization efficiency of the modified
algorithm is better than that of other algorithms (Ghasemi et al. 2015). The hybrid
algorithm combining modified teaching learning algorithm and double differential
evolution algorithm has faster convergence speed and better solutions in comparison
with other optimization algorithms (Ghasemi et al. 2014).
The WWO algorithm based on shallow wave theory mainly simulates propagation,
refraction and breaking operations to solve the optimization problem. The WWO algo-
rithm is applied to solve the ORPD problem to obtain the optimal control variables and
minimum power losses. The WWO algorithm not only speeds up the convergence rate
and calculation accuracy, but also strikes a balance between the global search and local

13
Optimal reactive power dispatch using water wave optimization…

search. The robustness and consistency of the WWO algorithm is beneficial to solve
the ORPD problem to find the global solution in power system. The WWO algorithm
has been tested in IEEE 30-bus in power system, the optimization result of the WWO
algorithm is better than that of other algorithms, which shows the WWO algorithm has
a strong global search ability.
The article has the following sections: Sect. 2 introduces the mathematical formula-
tion of the ORPD problem; Sect. 3 reviews WWO algorithm; the solution procedure of
the ORPD problem designed detailed in Sect. 4; the experimental results and analysis
presented in Sect. 5. Finally, Sect. 6 discusses the conclusion work.

2 Mathematical formulation

The ORPD problem is a constrained nonlinear combinatorial optimization problem in


power system. While satisfying all the specified constraints, we obtain the minimiza-
tion of the total transmission active power losses by setting the optimal values of con-
trol variables like reactive power output of generators, tap ratios of transformers and
reactive power output of shunt compensators. The total active power losses of the sys-
tem are minimized as the objective function, and the voltage quality has been improved
while reducing the active power losses of the system. The problem’s control decision
variables and their notations are given in Table 1.
The objective function of ORPD can be described as follows:
∑ ∑ ( )
min Pkloss = gk × v2i + v2j − 2 × vi × vj × cos 𝜃ij (1)
k∈NE k∈NE

Table 1  Control decision Nomenclature


variables and notations
WWO Water wave optimization
BA Bat algorithm
FPA Flower pollination algorithm
PSO Particle swarm optimization
SCA Sine cosine algorithm
CSA Crow search algorithm
Pkloss Active power loss of branch k
gk Conductance of branch k
NE A collection of all branches
𝜃ij Load angle difference between bus i and j
vi,vj Voltage amplitudes of the bus i and j
Pgi,Qgi Active and reactive power of generator bus i
Pdi,Qdi Active and reactive power of load bus i
gij,Bij Conductance and susceptance between bus i and j
Ng Number of generators
NB Number of bus in test system
NT Number of regulating transformer
NL Number of transmission lines

13
Y. Zhou et al.


where Pkloss is active power loss of branch k , k∈NE Pkloss represents total active
power losses in the transmission system, gk is the conductance of branch k , NE is a
collection of all branches, 𝜃ij is load angle difference between bus i and j , vi and vj
express the voltage amplitudes of the bus i and j , respectively.
The equality constraints can be generated as follow:
∑ ( )
Pgi − Pdi − vi vj gij cos 𝜃ij + Bij sin 𝜃ij = 0
(2)
j=Ni

∑ ( )
Qgi − Qdi − vi vj gij sin 𝜃ij − Bij cos 𝜃ij = 0
(3)
j=Ni

where Pgi and Qgi represent the active and reactive power of generator bus i  , Pdi and
Qdi represent the active and reactive power of load bus i  , respectively. The gij and Bij
represent conductance and susceptance between bus i and j.
The inequality constraints can be generated as follow:
Reactive power generation constraint for each generator bus:

min ≤ Q ≤ Qmax , i ∈ N
Qgi (4)
gi gi g

Voltage magnitude constraint for each bus:

vimin ≤ vi ≤ vmax
i
, i ∈ NB (5)

Transformer tap-setting constraint:

Tkmin ≤ Tk ≤ Tkmax , k ∈ NT (6)

The power limit constraint of transmission lines:


S ≤ Smax , l ∈ N
l l L (7)
In order to make the result satisfy the constraints, the minimum value of the active
power losses is taken as the objective function, the penalty function is used to tackle
with inequality constraints. So that we can get the augmented objective function as
follow:

∑ ∑
NG

NB

NL
FP = Pkloss + k1 × f (Qgi ) + k2 × f (Vi ) + k3 × f (Slm ) (8)
k∈NE i=1 i=1 m=1

⎧0 if xmin ≤ x ≤ xmax

f (x) = ⎨ (x − xmax )2 if x > xmax (9)
⎪ (xmin − x)2 if x < xmin

13
Optimal reactive power dispatch using water wave optimization…

Fig. 1  Different wave shapes in deep and shallow water

Table 2  Correspondence between problem space and population space


Problem space Population space

The search space of the problem Seabed area


Solve each solution of the problem A water wave with height h and wavelength 𝜆
The evaluation function value of each solution The closer sea level to the seabed, the higher the
fitness value. On the contrary, the lower the
fitness value

where k1 , k2 and k3 are penalty function coefficient, and the value is 10,000, xmin
and xmax are the value rang of dependent variables for each generator bus.

3 WWO algorithm

The WWO algorithm mainly simulates the motion of waves to solve the optimiza-
tion problem, and propagation, refraction and breaking can effectively balance the
global search and local search. For each wave, the fitness value of the water wave
is related to wave height and wavelength. As can be seen from Fig.  1, the water
wave has a higher fitness value, its wave height is higher and wavelength is longer
in the shallow water areas; the water wave has a lower fitness value, its wave height
is lower and wavelength is shorter in the deep water areas. The correspondence
between problem space and population space is given in Table 2.

13
Y. Zhou et al.

3.1 Propagation

When each wave is performing propagation, wave height and wavelength will
change accordingly because the seabed is uneven. The new water x′ is obtained by
propagation the original wave x . Assuming that the position of the original water
wave is x(d) and that of the new water wave is x� (d) , the location update formula
as follows:
x� (d) = x(d) + rand(−1, 1) ⋅ 𝜆L(d) (10)
where rand(−1, 1) is a uniformly distributed random number between − 1 and 1,
L(d) is the length of d th for search space, 𝜆 is wavelength of water wave x . If the
position of wave x′ is not within the search range, then we randomly give a new posi-
tion in the search space. Compared with the fitness value of original water wave x ,
if that of the new wave x′ is higher, wave x is replaced by x′ , wave height is defined
as hmax . On the contrary, wave x is retained and wave height is decreased by one,
which indicates that the wave energy is depleted.
The wavelength 𝜆 is updated as follows:
/
𝜆=𝜆⋅𝛼
−(f (x)−f
min +𝜀) (fmax −f
min +𝜀) (11)

where 𝛼 is wavelength attenuation coefficient, fmin and fmax are minimum and
maximum fitness values, 𝜀 is a positive integer and very small to avoid the denomi-
nator is zero.

3.2 Breaking

Increasing the wave energy, it makes the crests steeper and steeper. When the
speed of crest exceeds that of wave propagation, the water wave will break into
a series of solitary waves. In WWO algorithm, the optimal water wave x∗ is per-
formed by breaking operation to improve the diversity of the population. The
location update formula is shown below:
x� (d) = x(d) + N(0, 1) ⋅ 𝛽L(d) (12)
where 𝛽 is a broken wave coefficient, L(d) is the length of the search space for dth . If
the fitness value of all solitary waves obtained by breaking operation not better than
that of the water wave x∗ , x∗ is retained; on the contrary, the water wave x∗ will be
replaced by an optimal solitary wave in the population.

3.3 Refraction

In wave propagation, the water wave energy continues to decrease and eventually
becomes zero. When the water wave height is zero, refraction operation is used to
avoid the search stagnation, which speeds up the convergence of WWO algorithm

13
Optimal reactive power dispatch using water wave optimization…

and improves the accuracy of the solution. For water wave x , its refraction opera-
tion formula is shown below:
( )
(x ∗ (d) + x(d)) |x ∗ (d) − x(d)|
x� (d) = N , (13)
2 2

where x∗ is the optimal solution that found in the current position, N(𝜇, 𝜎) is used
to generate mean 𝜇 and standard deviation 𝜎 Gaussian random number. In fact, this
allows wave x to learn from the current optimal wave x∗ . At this time, the wave
height of wave x′ is hmax , the update formula of wavelength is as follow:
f (x)
𝜆� = 𝜆 (14)
f (x� )

The solution procedure of the WWO algorithm is given in Table 3.

4 The solution procedure of ORPD problem

The effectiveness and feasibility of the WWO algorithm is stable, which accelerates
convergence rate to find the global optimal solution. The WWO algorithm is pro-
posed to solve the ORPD problem that can obtain minimized value of power losses
in power system. The correspondence between the ORPD problem space and WWO
algorithm space is shown in Table 4.

Table 3  The solution procedure Algorithm WWO


of the WWO algorithm
1 Randomly initialize a water wave population P with n
waves (solutions), initialize wavelength
 reduction coefficient 𝛼 , breaking coefficient 𝛽 , and
wavelength reduction coefficient 𝜆.
2 while stop the termination criterion is not satisfied do
3    for each water wave x ∈ P do
4     Propagate x to a new x′ by using formula (10);
5     if f (x� ) < f (x) then
6        if f (x� ) < f (x∗ ) then
7           Break x′ by using Eq. (12);
8           Update x∗ with x′;
9        Replace x with x′;
10     else
11        Decrease x.h by one;
12        if x.h = 0 then
13          Refract x to a new x′ by using Eq. (13) and
(14);
14    Update the wavelengths by using Eq. (11);
15 return x∗.

13
Y. Zhou et al.

Table 4  Correspondence between ORPD and WWO


The ORPD problem space The WWO algorithm space

A collection contains all the optimization schemes A water wave population P with (n1 , n2 , … , nk )
(x1 , x2 , … , xk ) to solve ORPD problem waves
An optimal optimization scheme for solving ORPD An optimal water wave
problem
The objective evaluation function of the ORPD prob- The fitness function of the WWO algorithm
lem

In the coding of water waves, the code length is the total number of the con-
trol variables and the variables are arranged in a certain order. In this paper, the
sequence is the generator bus voltages, tap ratios of transformers and reactive
power output of shunt compensators, which constitutes a complete individual
structure of the water wave. The values of the control variables are within the
valid range. The number of generators is n1 , the number of reactive compensa-
tion point is n2 , and the number of transformer branches is n3 , so the code length
of a water wave is n = n1 + n2 + n3.
The encoding formats are different for different control variables. The genera-
tor bus voltages are continuous variables, which is a real number encoding, such
as formula (15). Tap ratios of transformers and reactive power output of shunt
compensators are discrete variables, and the randomly generated numbers are
rounded, such as formula (16).
x1 = xmin + rand(1, 1) ∗ (xmax − xmin ) (15)

x2 = xmin + round(rand(1, 1) ∗ K) ∗ r (16)


where xmax and xmin are maximum and minimum values of the control variable
range, K is total number of gears, r is unit change of discrete variables, rand(1, 1) is
random function, round() is a rounding function.
The flowchart of WWO algorithm to solve ORPD problem is depicted in
Fig.  2 and the solution procedure of the ORPD problem is given in Table  5,
VG1 , VG2 , VG5 , VG8 , VG11 , VG13 represent reactive power output of generators (gen-
erator bus voltages), T6−9 , T6−10 , T4−12 , T28−27 represent tap ratios of transformers,
and QC3 , QC10 , QC24 represent reactive power output of shunt compensators.

5 Experiment results and analysis

The section mainly contains two important contents: Content 5.1 simply intro-
duces the experimental setup; Content 5.2 minutely expounds the WWO algo-
rithm to solve the ORPD problem.

13
Optimal reactive power dispatch using water wave optimization…

Start

Read system data, bus data, line


data, and unit data

Initialization of population
set parameters

Map control variables from


water waves into laod flow data

Evaluation
Obtain power losses from power flow calculation
(MATPPOWER), then find the fittest water wave

Whether the condition are meet or not ?

Yes
No

Each water wave x performs propagation operation


by Eq.(10) to produce a new water wave x

No The wave height h of water wave x is reduced


f(x )<f(x) ?
by one, which represents energy loss.

Yes

x takes the place of x in the population xh=0 ? iter=iter+1

Yes

Water wave x performs refraction


f(x )<f(x*) ?
operation by Eq.(13) and (14)

Yes

Water wave x performs the breaking


Updata the wavelength of each wave by Eq.(11).
operation by Eq.(12), x replaces x*

Calculate the fitness of all water waves by


Eq.(8) to obtain value of power losses

No
Whether the condition are meet or not ?

Yes

Output the optimal control variables and


minimize the total active power losses

End

Fig. 2  Solution methodology flow chart of OPRD

13
Y. Zhou et al.

Table 5  The solution procedure of ORPD problem

1 Read system data, bus data, line data, and unit data. Define control variables
(VG1 , VG2 , VG5 , VG8 , VG11 , VG13 , T6−9 , T6−10 , T4−12 , T28−27 , QC3 , QC10 , QC24 ) within their permissible
range;
2 Randomly initialize a population P of n waves (solutions), and initialize wavelength reduction
 coefficient 𝛼 , breaking coefficient 𝛽 , and wavelength reduction coefficient 𝜆.
3 Map control variables from water waves into power flow data and calculate the power losses of
each water wave using Eq. (8);
4 while stop the termination criterion is not satisfied do
5   for each water wave x ∈ P do
6     Propagate x to a new x′ by using formula (10);
7     if f (x� ) < f (x) then
8       if f (x� ) < f (x∗ ) then
9         Break x′ by using Eq. (12);
10         Update x∗ with x′;
11       Replace x with x′;
12    else
13       Decrease x.h by one;
14       if x.h = 0 then
15       Refract x to a new x′ by using Eq. (13) and (14);
16     Update the wavelengths by using Eq. (11);
17     Calculate the fitness function of each water wave using Eq. (8);
18   return x∗;
19   Output minimized value of active power losses and optimized values of control variables.

5.1 Experimental setup

All of the algorithms were programmed in MATLAB R2012a, numerical experi-


ment was set up on AMD Athlont (tm) II*4640 processor and 4 GB memory.

5.2 The WWO algorithm to solve the ORPD problem

The IEEE30-bus, 41 branch test system is applied to verify the feasibility and effec-
tiveness of the WWO algorithm. The test system mainly contains thirteen control
variables: six generators placed at the buses 1, 2, 5, 8, 11 and 13, where bus 1 is
taken as slack bus, 2, 5, 8, 11 and 13 are taken as PV generator buses and the rest

Table 6  Control variables List of control variables Min. Max. Type


Limits
Generator bus 0.90 1.1 Continuous
Load bus 0.90 1.1 Continuous
Transformer-tap 0.95 1.05 Discrete
Shunt reactive compensator 1 20 Discrete

13
Optimal reactive power dispatch using water wave optimization…

are taken as PQ load buses; four tap ratios of transformers between buses 6–9, 6–10,
4–12 and 27–28; three shunt compensators install at bus 3, 10 and 24. The control
variables limits are shown in Table 6.
Compared with other optimization algorithms, such as BA (Yang and He 2013),
FPA (Yang 2012), PSO (Kennedy and Eberhart 1995), SCA (Mirjalili 2016) and
CSA (Askarzadeh 2016), the calculation accuracy of the WWO algorithm is better.
The important optimization parameters are given in Table 7.
For all algorithms, the size of population is 50, the maximum number of itera-
tions is 200, and the number of independent runs is 30. In order to more effectively
demonstrate the superiority of the WWO algorithm, Best, Worst, Mean and Std are
taken as the most important judgment criteria, where Best represents the optimal fit-
ness value, Worst represents worst fitness value, Mean represents mean fitness value
and Std represents standard deviation. The optimal value can reflect the convergence
accuracy, and the standard deviation can reflect the stability performance. The con-
trol variables and active power losses obtained by different algorithms are given in
Table  8. Ranking is based on standard deviation, which can effectively obtain the
effective degree of the algorithms to solve the ORPD problem.
For WWO algorithm, it mainly simulates the propagation, refraction and break-
ing operations to find the global optimal solution in the entire search space. The
process of propagation optimization can be seen as the process of moving from the
deep water area to the shallow water area, so that the wave with high fitness value
is exploited in a small scale, and the wave with low fitness value is explored in a
large scale. The refraction operation effectively avoids the search stagnation of the

Table 7  The important parameters of six algorithms


Algorithm Parameters Values

BA Pulse frequency range f [0, 2]


Echo loudness A 0.25
Decreasing coefficient 𝛾 0.5
FPA Switch probability 𝜌 0.8
PSO Constant inertia 𝜔 0.7298
First acceleration coefficient c1 1.4962
Second acceleration coefficient c2 1.4962
SCA Constant a 2
Random number r2 [0, 2𝜋]
Random number r4 [0,1]
CSA Random number r [0, 1]
Flight length f 2
Awareness probability A 0.1
WWO Wavelength 𝜆 0.5
Wave height hmax 6
Wavelength reduction coefficient 𝛼 1.0026
Breaking coefficient 𝛽 [0.01, 0.25]
Maximum number kmax of breaking directions min(12, D∕2)

13
Y. Zhou et al.

Table 8  Values of control variables after optimization


Variables BA FPA PSO SCA CSA WWO

VG1 1.0925 1.0835 1.0868 1.0885 1.0356 1.1000


VG2 1.0572 1.0805 1.0564 1.0723 0.9167 1.0861
VG5 1.0218 0.9534 1.0002 0.9894 1.0263 1.0550
VG8 0.9433 1.0491 1.0114 1.0328 0.9872 1.0627
VG11 1.0609 1.0083 1.0637 0.9899 0.9575 1.0829
VG13 0.9333 1.0985 1.0523 1.0903 0.9996 1.0992
T6−9 0.9806 0.9519 0.9960 1.0130 1.0196 1.0176
T6−10 1.0108 0.9770 0.9864 1.0252 1.0222 0.9643
T4−12 1.0377 1.0294 1.0204 1.0483 1.0240 0.9798
T28−27 1.0449 1.0170 0.9678 0.9801 1.0403 0.9683
QC3 19.8074 13.7981 14.7614 7.7134 6.0417 18.2857
QC10 14.6633 12.6445 11.7118 14.4309 13.2038 19.8140
QC24 11.3836 14.3583 11.8721 15.2393 12.7867 11.0903
Best 17.5576 17.0225 15.6142 16.5413 16.4704 15.9441
Worst 44.2782 20.8336 18.4493 18.7694 18.6908 15.9856
Mean 28.6135 18.6158 16.5854 17.5600 17.6214 15.9565
Std 6.4988 0.8939 0.6538 0.5533 0.5488 0.0091
Rank 6 5 4 3 2 1

algorithm, and removes energy-depleted waves to accelerate the convergence of


the algorithm. The breaking operation produces an optimal solitary wave, which
improves the calculation accuracy of the algorithm to some extent. These three oper-
ations can effectively balance the WWO algorithm’s global search and local search
to find the optimal solution, so that the WWO algorithm has good robustness and
feasibility when it is applied to solve the complex optimization problem.
The WWO algorithm is applied to solve the ORPD problem, which has a strong
global search ability to find the global optimal solution. Compared with other opti-
mization algorithms, such as, BA, FPA, PSO, SCA and CSA, the WWO algorithm
finds the optimal control variables and the optimal active power losses, which indi-
cates that the WWO algorithm can effectively solve this problem. The WWO algo-
rithm’s optimal fitness value, worst fitness value, mean fitness value and standard
deviation are the best in all algorithms. The optimal fitness value of the WWO algo-
rithm is superior to other algorithms, which indicates that the WWO algorithm has
higher calculation accuracy. The worst fitness value and mean fitness value of the
WWO algorithm are smaller than that of other algorithms, which indicates that the
WWO algorithm has a good overall optimization performance. The standard devia-
tion of the WWO algorithm is the smallest in all algorithms, which indicates that the
WWO algorithm has strong stability.
The WWO algorithm is an efficient search method in which the propaga-
tion, refraction, and breaking operations provide a good search mechanism for
the algorithm. The WWO algorithm can effectively avoid search stagnation to
find the global optimal solution in search space. As can be seen from Fig. 3, all

13
Optimal reactive power dispatch using water wave optimization…

18.5
BA
FPA
18 PSO

Total Active power losses (MW)


SCA
CSA
17.5 WWO

17

16.5

16

15.5
0 20 40 60 80 100 120 140 160 180 200
Iterations

Fig. 3  Evolution curves of fitness value

optimization algorithms are used to solve the ORPD problem, the WWO algo-
rithm has faster convergence speed and higher calculation accuracy, which indi-
cates that the WWO algorithm can effectively balance exploration and exploi-
tation. The results indicate that the WWO algorithm finds the optimal control
variables and the optimal active power losses. The standard deviation is one
of the important tools for detecting the stability of an algorithm. The standard

45

40

35

30

25

20

15
BA FPA PSO SCA CSA WWO

Fig. 4  ANOVA test of global minimum

13
Y. Zhou et al.

Table 9  p_value of the Algorithm Results


Wilcoxon rank-sum test results
p_value h

WWO versus BA 3.02E−11 1


WWO versus FAP 3.02E−11 1
WWO versus PSO 1.87E−05 1
WWO versus SCA 3.02E−11 1
WWO versus CSA 3.02E−11 1

-12
10

-10
10

-8
10
p-value result

-6
10

-4
10

-2
10

0
10
WWO vs BA WWO vs FPA WWO vs PSO WWO vs SCA WWO vs CSA

algorithm comparison

Fig. 5  Wilcoxon rank test histogram

deviation is small, and the stability of the algorithm is better. As can be seen from
Fig. 4, compared with other algorithms, such as, BA, FPA, PSO, SCA and CSA,
the standard deviation of the WWO algorithm is smaller, which indicates that the
WWO algorithm has a strong stability to find the global optimal solution, and
the WWO algorithm has certain effectiveness and feasibility in solving ORPD
problem.
The Wilcoxon rank test (Wilcoxon 1944; Garcia et  al. 2008) is an important
judgment indicator, which can effectively reflect the relationship between the
WWO algorithm and other algorithms. The p_value of Wilcoxon rank-sum test
results are given in Table 9, where 1 indicates significant difference, 0 indicates
no statistical difference. The histogram of the test is given in Fig. 5.

13
Optimal reactive power dispatch using water wave optimization…

Table 10  execution time of Algorithm BA FPA PSO SCA CSA WWO


different algorithms
Time execution (s) 87 141 103 140 145 90

All the algorithms are used to solve the ORPD problem, the number of the inde-
pendent runs is 30, and the average run time is given in Table 10, the average run
time of the WWO algorithm is bigger than that of the BA algorithm. Compared with
FPA, PSO, SCA and CSA, the average run time of the WWO algorithm is much
smaller. The histogram of execution time is given in Fig. 6.

6 Conclusions

The ORPD problem is a hybrid optimization problem with continuous variables and
discrete variables, the solution process is quite complex and the research hotspot is
to find an efficient solution. The WWO algorithm can effectively utilize propaga-
tion, refraction and breaking to find the global optimal solution in search space. The
propagation operation strikes a balance between exploration and exploitation; the
refraction operation can make the algorithm jump out of the local optimal solution to
accelerate the convergence speed; the breaking operation enhances the local search
to improve the calculation accuracy. In order to better verify the overall optimization

150
Time Execution (Seconds)

100

50

0
BA FPA PSO SCA CSA WWO
algorithm

Fig. 6  Execution time versus different algorithms

13
Y. Zhou et al.

performance of the WWO algorithm, WWO is applied to solve the optimal reac-
tive power dispatch problem. The objective of optimization is to find global optimal
control variables and minimum active power losses. Compared with BA, FPA, PSO,
SCA and CSA, the WWO algorithm has faster convergence speed and higher calcu-
lation accuracy. The standard deviation of the WWO algorithm is the smallest in all
algorithms and the independent run time is relatively small. The WWO algorithm
has strong stability and robustness to find the optimal solution of the problem. The
results indicate that the WWO algorithm is an effective and feasible method to solve
the optimal reactive power dispatch problem.
In the future research, the WWO algorithm is studied from two aspects. In the
first aspect, adding an effective strategy or combining with other optimization algo-
rithms to improve the overall performance of the WWO algorithm, and the modified
algorithm is applied to solve the optimal reactive power dispatch problem. In the
second aspect, the WWO algorithm will be applied to solve the multi-objective opti-
mal reactive power dispatch problem and large-scale electrical networks.

Acknowledgements This work was supported by the National Science Foundation of China under
Grant Nos. 61463007 and Project of the Guangxi Natural Science Foundation under Grant No.
2016GXNSFAA380264.

References
Alsac O, Stott B (1974) Optimal load flow with steady-state security. IEEE Trans Power Appar Syst PAS-
93 3:745–751
Askarzadeh A (2016) A novel metaheuristic method for solving constrained engineering optimization
problems: crow search algorithm. Comput Struct 169:1–12
Attia A-F, Al-Turki YA, Abusorrah AM (2012) Optimal power flow using adapted genetic algorithm with
adjusting population size. Electr Power Comp Syst 40(11):1285–1299
Bjelogrlic M et al (1990) Application of Newton’s optimal power flow in voltage/reactive power control.
IEEE Trans Power Syst 5(4):1447–1454
Deeb N, Shahidehpur SM (1990) Linear reactive power optimization in a large power network using the
decomposition approach. IEEE Trans Power Syst 5(2):428–438
Duman S et  al (2012) Optimal reactive power dispatch using a gravitational search algorithm. Gener
Transm Distrib IET6 6:563–576
Garcia S, Molina D et al (2008) A study on the use of nonparametric tests for analyzing the evolutionary
algorithms’ behaviour: a case study on the CEC’2005 special session on real parameter optimiza-
tion. J Heuristics 15:617
Ghasemi M, Ghanbarian MM, Ghavidel S et al (2014) Modified teaching learning algorithm and double
differential evolution algorithm for optimal reactive power dispatch problem: a comparative study.
Inf Sci 278(278):231–249
Ghasemi M, Taghizadeh M, Ghavidel S et  al (2015) Solving optimal reactive power dispatch problem
using a novel teaching–learning-based optimization algorithm. Eng Appl Artif Intell 39:100–108
Ghasemi M, Taghizadeh M, Ghavidel S et al (2016) Colonial competitive differential evolution: an exper-
imental study for optimal economic load dispatch. Appl Soft Comput 40:342–363
Grudinin N (1998) Reactive power optimization using successive quadratic programming method. IEEE
Trans Power Syst 13(4):1219–1225
Kenarangui R, Seifi A (1994) Fuzzy optimal reactive power control. Electr Power Syst Res 30(1):47–55
Kennedy J, Eberhart R (1995) Particle swarm optimization. In: Proceedings of the IEEE international
conference on neural networks, Perth, Australia, vol IV, pp 1942–1948
Khazali AH, Kalantar M (2011) Optimal reactive power dispatch based on harmony search algorithm. Int
J Electr Power Energy Syst 33(3):684–692

13
Optimal reactive power dispatch using water wave optimization…

Lai LL et al. (2005) Swarm intelligence for optimal reactive power dispatch. In: Transmission and distri-
bution conference and exhibition: Asia and Pacific, 2005 IEEE/PES IEEE pp 1–5
Lee KY, Park YM, Ortiz JL (1985a) A united approach to optimal real and reactive power dispatch. IEEE
Trans Power Appar Syst PAS-104 5:1147–1153
Lee KY, Park YM, Oritz JL (1985b) A united approach to optimal real and reactive power dispatch. IEEE
Trans Power Syst 104(5):1147–1153
Mandal B, Roy PK (2013) Optimal reactive power dispatch using quasi-oppositional teaching learning
based optimization. Int J Electr Power Energy Syst 53(1):123–134
Mirjalili S (2016) SCA: a sine cosine algorithm for solving optimization problems. Knowl-Based Syst
96:120–133
Momoh JA, Guo SX, Ogbuobiri EC, Adapa R (1994) The quadratic interior point method solving power
system optimization problems. IEEE Trans Power Syst 9(3):1327–1336
Mouassa S, Bouktir T (2016) Artificial bee colony algorithm for discrete optimal reactive power dispatch.
In: International conference on industrial engineering and systems management IEEE
Mouassa S, Bouktir T, Salhi A (2017) Ant lion optimizer for solving optimal reactive power dispatch
problem in power systems. Eng Sci Technol Int J 20:885–895
Shaw B, Mukherjee V, Ghoshal SP (2014) Solution of reactive power dispatch of power systems by an
opposition-based gravitational search algorithm. Int J Electr Power Energy Syst 55(1):29–40
Sulaiman MH et  al (2015) Using the gray wolf optimizer for solving optimal reactive power dispatch
problem. Appl Soft Comput 32(C):286–292
Varadarajan M, Swarupa KS (2008) Differential evolution approach for optimal reactive power dispatch.
Appl Soft Comput 8(4):1549–1561
Wilcoxon F (1944) Individual comparisons by ranking methods. Biom Bull Biom 1(6):112–114
Wu YC, Debs AS, Marsten RE (1994) A direct nonlinear predictor–corrector primal–dual interior point
algorithm for optimal power flows. IEEE Trans Power Syst 9(2):876–883
Yang XS (2012) Flower pollination algorithm for global optimization. In: Proceedings of the unconven-
tional computation and natural computation, lecture notes computer science, vol 7445, pp 240–249
Yang XS, He X (2013) Bat algorithm: literature review and applications. Int J Bio-Inspired Comput
5(3):141–149
Zheng YJ (2015) Water wave optimization: a new nature-inspired metaheuristic. Comput Oper Res
55:1–11

13