You are on page 1of 13

Ain Shams Engineering Journal xxx (xxxx) xxx

Contents lists available at ScienceDirect

Ain Shams Engineering Journal


journal homepage: www.sciencedirect.com

Electrical Engineering

Optimized controller design for islanded microgrid using non-dominated


sorting whale optimization algorithm (NSWOA)
Quazi Nafees Ul Islam ⇑, Ashik Ahmed, Saad Mohammad Abdullah
Department of Electrical and Electronic Engineering (EEE), Islamic University of Technology (IUT), Board Bazar, Gazipur 1704, Bangladesh

a r t i c l e i n f o a b s t r a c t

Article history: Swarm-based nature-inspired optimization techniques are developing rapidly due to their universal
Received 26 August 2020 acceptance and capability in solving various real-life challenges efficiently. Hybridization of swarm intel-
Revised 29 December 2020 ligence based optimization algorithms with multi-objective based solution techniques is creating a wide
Accepted 8 January 2021
door in the field of optimization. The main focus of this work is, developing a hybrid Non-dominated
Available online xxxx
Sorting Whale Optimization Algorithm (NSWOA), where swarm-based Whale Optimization Algorithm
(WOA) is hybridized with multi-objective based, non-dominated sorting technique. This is done to
Keywords:
develop an algorithm with efficient optima searching ability and faster computational speed. The appli-
Multi-objective
Dynamic load
cation of NSWOA in optimizing the controller parameters of an islanded microgrid consisting of both sta-
Static load tic and dynamic load has been also described. SPSS software has been used to compare the performance
Non-dominated sorting of proposed NSWOA with non-dominated sorting genetic algorithm-II (NSGA-II) and Strength Pareto
SPSS Evolutionary Algorithm (SPEA) technique in optimizing the controller parameters of islanded microgrid
model with multi-objective problem. It is obtained that, NSWOA requires an average of 4 iterations to
reach the best optimum solution, which is less than other existing algorithms. Moreover, the computa-
tional time required by NSWOA is 2.9201 s, which proves that, it converges at a much faster rate com-
pared to existing NSGA-II and SPEA algorithms.
Ó 2021 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).

1. Introduction and implemented in various important fields of study, for instance,


in [4], WOA based proportional integral derivative (PID) controller
Metaheuristic optimization algorithms have gained much was introduced for automatic generation control of interconnected
acceptance over the years in solving various optimization prob- system consisting of renewable energy sources. Similarly, refer-
lems. These algorithms perform the optimization process in a ence [5] shows designing of proportional-integral (PI) controller
stochastic manner, which implies that the operation of these algo- utilizing WOA in order to improve the performance of photovoltaic
rithms has been based on random operators. Evolutionary and power systems; [6] shows the application of WOA in estimation of
swarm intelligence-based algorithms such as genetic algorithm parameters of single- and multiple-diode photovoltaic model. In
(GA), simulated annealing (SA), particle swarm optimization [7]WOA was used in proton exchange membrane (PEM) fuel cell
(PSO), ant colony optimization (ACO), firefly algorithm (FA), etc. (FC) model in order to achieve their unidentified parameters.
are categorized under the class of metaheuristic algorithms [1,2]. Moreover, in [8] shows implementation of WOA in designing the
These algorithms have been inspired by some biological or natural controller parameters of Sugeno Fuzzy Logic for improving the
phenomena. Mirjalali and Lewis first introduced the whale opti- fault ride through capability of wind power plants. In reference
mization algorithm (WOA) in 2016, which was inspired by the [9], a hybrid whale Particle Swarm Optimization algorithm
hunting behavior of whales [3]. In the WOA technique, a specific (HWPSO) has been developed combining WOA with Particle
breed of whale known as humpback whales has been considered Swarm Optimization (PSO) to address the limitation of the PSO
where these whales use their unique bubble net feeding method algorithm related to the exploration of the search space. As WOA
for hunting their prey. Since its inception, WOA has been used has better exploration ability in different uncertain and small space
region, many other researchers hybridized this WOA to overcome
⇑ Corresponding author. the time constraints and accuracy in obtaining the optimum solu-
E-mail addresses: quazinafees@iut-dhaka.edu (Q.N.U. Islam), ashik123@iut- tion of the problem. In reference [10] WOA was hybridized with
dhaka.edu (A. Ahmed), saadabdullah@iut-dhaka.edu (S.M. Abdullah).

https://doi.org/10.1016/j.asej.2021.01.035
2090-4479/Ó 2021 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah, Optimized controller design for islanded microgrid using non-
dominated sorting whale optimization algorithm (NSWOA), Ain Shams Engineering Journal, https://doi.org/10.1016/j.asej.2021.01.035
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

PSO, and in [11] WOA was hybridized with local search strategy, (WOA). The proposed NSWOA is being used to obtain an optimum
where, they showed the ability of this hybridized WOA in solving solution for optimizing the controller parameters for an islanded
various real-life problems and found the best optimum solution microgrid consisting of both static load and dynamic load. The
for those situations with better accuracy, computation time and aim is to develop the proposed algorithm in obtaining stable
efficiency. Recently, a enhanced whale optimization algorithm dynamic performance of the system during load variation. The
(EWOA) has been developed for maximum power point tracking non-dominated sorting technique has been used to develop the
(MPPT) of variable-speed wind generators [12]. algorithm to be capable of solving multi-objective practical prob-
The algorithms mentioned above were initially developed as lems. The performance of the proposed NSWOA has been analyzed
single-objective optimization algorithms to optimize any single by comparing it with the existing, well recognized, Non-dominated
objective function. All of these proposed algorithms efficiently Sorting Genetic Algorithm-II(NSGA-II) and Strength Pareto Evolu-
handled single-objective optimization problems. However, in most tionary Algorithm (SPEA) [25]. The contribution of the paper are
cases, the problems do not remain limited to only a single objective represented below:
function; instead, there appear multiple functions that may con-
flict with each other, but all of them are required to be optimized (1) Develop a hybridized multi-objective optimization algo-
simultaneously. rithm NSWOA and observe its performance and ability to
In single-objective optimization, generally the aim is to search obtain stable operation of microgrid with optimized con-
for the best design or decision, which is usually the optimum solu- troller parameters.
tion of the optimization problem. Nevertheless, in the case of mul- (2) Carry out a comparative analysis of the hybridized NSWOA
tiple objectives, the best solution (minimum or maximum) is technique with existing and well recognized NSGA-II and
determined after fulfilling the conditions of all the objective func- SPEA techniques.
tions [13]. Usually, the objective functions used in multi-objective
optimization are contradictory. Generally, a set of solutions can be The remaining of the paper has been organized as follows: in
obtained for this type of problem, which may be better than the Section 2, the microgrid model has been developed, which consists
rest of the solutions in the search space. Multi-objective and of both static and dynamic load. Section 3 shows the problem that
single-objective optimization techniques have a different method- has been formulated and the objective functions which need to be
ology of ranking the generated population. Multi-objective opti- satisfied. Moreover, the complete description about the proposed
mization techniques have the advantage of selecting the best NSWOA technique is done in Section 3. The results and analysis
optimum solution from a set of solutions arranged in a Pareto- of NSWOA along the comparative study between NSWOA, NSGA-
front using a hyper volumetric measures. They are ranked accord- II, and SPEA are presented in Section 4. Finally, Section 5 shows
ing to their fitness value and after satisfying the constraints they the concluding remarks.
find the optimal solution for the multiple objective functions
[14]. On the other hand, nature-inspired single-objective optimiza- 2. Description of the microgrid model
tion techniques rank the fitness function using a bio-type ranking.
They consider a single optimal solution as the optimum solution An islanded microgrid consisting of both static and dynamic
[15]. So, it is inevitable that multi-objective optimization tech- load has been used for this study. The Microgrid model shown
niques also provide the flexibility in demonstrating real-life chal- in Fig. 1 consists of two DG units where static load has been
lenges and provide the methodology in optimizing them. installed on one unit and induction motor as a dynamic load on
In the world of increasing energy demand with increased black- the other unit. In this section, the model of the aforementioned
outs, burnouts; microgrid came as a competent solution due to it islanded microgrid will be discussed where the necessary study
being a pollution-free, energy-saving and smart energy system has been done from [22,26] for the development of the complete
with diversified energy sources [16]. Microgrid evolved as a solu- microgrid model. Modeling of a microgrid includes three main
tion to the universal energy crisis, and at the same time, it became segments: inverter, loads, and network. Inverters were designed
a talking point in between the researchers for its advancement and in their individual ðdij  qij Þ reference frame, and power-sharing
development [17,18]. Intending to enhance the performance, effi-
controller was being used to fix their angular frequency xij . Here,
ciency, and cost-effectiveness of microgrid, optimization of the th th
controller parameters, load sharing, cost, etc. has become the i and j indicates i and j inverter, respectively. In this study, the
prime concern for the researchers [19,20]. Among the controller reference frame of one of the inverters was taken as the common
parameters, tuning the controller gains play a vital role in the reference frame ðD  Q Þ, and the load and network state equa-
proper operation of the microgrid. Optimization techniques play tions were transformed into the same ðD  Q Þ frame, including
a significant role in tuning the controller gains. By selecting of opti- the other inverters. The inverter structure used for both DGs are
mized controller parameters, the system performance can be
improved, and the quality of power developed can be assured
[21]. Furthermore, if there exists any type of disturbance in the sys-
tem, the selection of proper controller parameters and tuning them
at optimized value ensures stable operation of the system [22]. For
instance, in order to control the frequency variation of islanded
microgrid, droop controller gains were optimized using artificial
fish swarm algorithm (AFSA) [23]. For the microgrid system,
NSGA-II, which is a multi-objective based optimization algorithm,
was used to optimize the controller parameters of the system but
did not mention any comparative analysis with other algorithms
[24].
In this paper, a new meta-heuristic multi-objective optimiza-
tion algorithm has been developed using the Non-dominated Sort-
ing technique with Whale Optimization Algorithm (WOA) to form
hybridized Non-dominated Sorting Whale Optimization Algorithm Fig. 1. DGs with both static and dynamic load.

2
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 2. Block Diagram of Different parts of Inverter.

the same, and the inverter structure used in this study is taken An individual inverter model consists of thirteen states. In this
from Ref. [22], which is shown in Fig. 2. The detailed power, volt- study, two inverters have been used as there are two DGs. In
age, and current controller used in the DG along with the LCL fil- total, there will be twenty-six states for two inverters. The com-
ter design is developed in Ref. [22]. Fig. 2 shows the inverter ponents of AINVi ; BINVi ; Bcom , and C INVi are given in matrix form in
block diagram in which the power controller block at first Eqs. (4)–(6)

2 3
0 mp 0 0 0 0 0 0 0 0 0 0 0
6 7
6 0 xc 0 0 0 0 0 0 0 1:5xc Iodi 1:5xc Ioqi 1:5xc V odi 1:5xc V oqi 7
6 7
6 0 0 xc 0 0 0 0 0 0 1:5xc Ioqi 1:5xc Iodi 1:5xc V oqi 1:5xc V odi 7
6 7
6 7
6 0 0 nq 0 0 0 0 0 0 1 0 0 0 7
6 7
6 0 0 0 0 0 0 0 0 0 0 1 0 0 7
6 7
6 7
6 0 0 ni K pv K ic 0 0 0 1 0 K pv xnoload F 0 7
6 7
6 0 0 0 0 K ic 0 0 0 1 xnoload K pv 0 F 7
6 7
6 7
6 7
6 7
6 0 mp Ilqi
nq K pc K pv K iv K pc
0 K ic
0
K pc Rf
xo  xnoload K pc K pv 1 xnoload C f K pc K pc F
0 7
6 7
6 Lf Lf Lf Lf Lf Lf Lf
7
AINVi ¼6
6
7
7 ð4Þ
6 K iv K pc K ic
xo  xnoload K pc Rf xnoload C f K pc K pc K pv 1 K pc F 7
6 0 mp Ildi 0 0 0 0 7
6 Lf Lf Lf Lf Lf Lf
7
6 7
6 7
6 mp V oqi 1
xo 1 7
6 0 0 0 0 0 0 0 0 0 7
6 Cf Cf
7
6 7
6 7
6 1 7
6 0 mp V odi 0 0 0 0 0 0 1
xo 0 0 7
6 Cf Cf
7
6 7
6 7
61
6 Lc ðV bDi sin di  V bQi cos di Þ mp Ioqi 0 0 0 0 0 0 0 1
0 Rc
xo 7 7
6 Lc Lc 7
6 7
4 5
Rc
1
Lc
ðV bDi cos di þ V bQi sin di Þ mp Iodi 0 0 0 0 0 0 0 0 1
Lc
xo Lc

"  cos di  sin di


#T
calculates the instantaneous real power pðtÞ and reactive power 0 ...: 0 Lc Lc
qðtÞ from output voltage v odq and current iodq , which is then BINVi ¼ sin di  cos di
Bcom ¼ ½ 1 0 0 0 . . . :: 0 T131
0 ...: 0 Lc Lc 132
passed through a low-pass filter to obtain average real power P
ð5Þ
and reactive power Q. Eventually, power controller provides fre-
quency and magnitude of the output voltage reference. Next,  
the voltage controller compares the actual and reference values ðIodi sin di  Ioqi cos di Þ 0 0 . . . : 0 0 cos di  sin di
C INVi ¼
of output voltage and provides reference inductor output cur- ðIodi cos di  Ioqi sin di Þ 0 0 . . . : 0 0 sin di cos di 213
rents, which are fed as input to the current controller. Finally, ð6Þ
the current controller provides the switching signal, which is used
as a gate pulse of the transistor to trigger the inverter. The com- Line current flows from one node to another node of the line con-
plete inverter state-space model is given in the Eqs. (1)–(3). necting the buses. The line and static load model for this study
are taken from Ref. [22]. The line model, as well as the static load
Dx_ INVi ¼ AINVi DxINVi þ BINVi Dv bDQi þ Bcom Dxcom ð1Þ
model, consists of 2 states each.
In our study, an induction motor is being used as a dynamic
DioDQi ¼ C INVi DxINVi ð2Þ
load. The induction motor model has been developed taking [26]
as a reference. The induction motor consists of 5 states where
DxINVi ¼ ½Ddi DPi DQ i Dudqi Dcdqi Dildqi Dv odqi Diodqi  ð3Þ
DiQs ; DiDs ; DiQr ; DiDr ; Dsl are the state variables. DiQs and DiDs are

3
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

the state variable for stator current of Q and D axis, respectively. troller of inverter-2 respectively. The main objective of this work is
Similarly, DiQr and DiDr are the state variable for rotor current of to fine-tune these controller parameter gains in order to obtain a
Q and D axis respectively, and Dsl is the state variable for rotor slip. stable microgrid consisting of both static and dynamic load. The
The complete microgrid model of our system is developed by objective functions based on the damping ratio and eigenvalue
combining the individual inverter, line network, load, and induc- are stated below in Eq. (11) where both are minimization functions
tion motor state-space model from Ref. [22,26]. The complete but contradictory.
microgrid model is represented in Eq. (7).
J 1 ¼ minðrdesired  minðrN ÞÞ;
DX_ MG ¼ AMG DX MG ðtÞ ð7Þ J 2 ¼ minðfdesired  minðfN ÞÞ ð11Þ
where DX MG is the integrated states of the microgrid model. The Here, r and f represents the real part of eigenvalues and damping
components of DX MG and AMG are given in Eqs. (8) and (9) ratio, respectively. Here, N represents the number of states. For this
DX MG ¼ ½DxINV1 DxINV2 DiLineDQ DiLoadDQ DX IM 135 ð8Þ study, N ¼ 1; 2; 3; . . . . . . 35 as there are 35 states in the system as
per the mathematical model. The objective function is chosen in

2 3
½AINV þ BINV RN M INV C INV 2626 ½BINV RN MNET 264 ½BINV RN MIM 265
6 ½B1Line RN M INV C INV þ B2Line C INV x 226 ½AINV þ B1Line RN M NET 24 ½B1Line RN MIM 25 7
6 7
AMG ¼6 7 ð9Þ
4 ½B1Load RN M INV C INV þ B2Load C INV x 226 ½B1Load RN M NET 24 ½ALoad þ B1Load RN M IM 25 5
½F IM T½RN M INV C INV  þ GIM C INV x þ HIM C INV x AINV  ½F IM T½RN M NET  EIM þ ½F IM T½RN M IM  3535

C INV x com ¼ ½0  mp 0 . . . : 0113 C INV x ¼ ½C INV x com 0 . . . : 0126 such a way that the minimum f and the minimum r value found
ð10Þ from all the eigenvalues remain as close as possible to the desired
damping ratio and desired r respectively. Hence, minðrN Þ and
M INV ; MNET , and M IM are used for mapping the connection points minðfN Þ will return scalar values. The two objective functions con-
between nodes of the network and line. In this study, the system sidered for this study are contrary to each other, which can be
has 2 nodes, 2 inverters connected by a single line. The dimension understood from Eq. (12) where it can be seen that if the magnitude
of MINV is ½ð2  number of nodesÞ -by- ð2  number of inv ertersÞ. of rN decreases, then fN becomes less negative similarly if rN
Similarly, dimension of M IM is ½ð2  number of nodesÞ -by- ðnumber increases then reverse situation occurs.
of induction motor connectedÞ. For this model, there are 5 connec- rN
tion points between induction motor and nodes. The dimension of fN ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð12Þ
MNET is ½ð2  number of nodesÞ -by- ð2  number of LinesÞ, and RN r2N þ x2N
is ½ð2  number of nodesÞ -by- ð2  number of nodesÞ. The diagonal
rdesired and fdesired are taken as reference values to limit the objective
elements of RN are equal to Rn . Components of MNET are either +1 or
functions within the desired boundary [27].
1 depending upon the direction of line current means if the current
The problem bounds for the developed model are given below
is entering the node, then +1 is considered and, if leaving the node,
in Eq. (13). In order to obtain a stable system with less oscillations
then 1 is considered. For linking the stator voltage dynamics
and improved damping performance, the controller gains are lim-
of induction motor with the microgrid system matrix, T is used
ited as follows
whose dimension is ½ðnumber of induction motor connection pointsÞ
-by- ð2  number of nodesÞ. 0 < K pv 1;2 ; K pi1;2 ; K iv 1;2 ; K ic1;2 < 500 ð13Þ

Here, the controller parameters are aimed to be kept within the


3. Problem formulation given range in order to obtain stable performance of the microgrid
system. As the system model was developed following [26]; consid-
Stable operation of a microgrid in an islanded mode in the pres- ering their values of controller gains, the range of controller gains
ence of both static and dynamic load is a challenge and it is mentioned in Eq. (13) was selected accordingly and justified by root
addressed in this work. In the sections as mentioned above it is locus analysis. Thus, J 1 and J 2 are the objective functions that need
mentioned that for efficient, fast, and stable operation of microgrid, to be optimized using optimizing algorithms in order to obtain opti-
the role of controller parameters is noteworthy. In the autonomous mized controller parameters for the stable operation of the
mode of operation of a microgrid controller parameter settings microgrid.
play an essential role in system stability.
3.2. Optimization process of non-dominated sorting whale
3.1. Objective functions optimization algorithm (NSWOA)

In the previous sections, the mathematical model of the micro- Non-dominated sorting based optimization technique has been
grid was developed where there were separate PI regulators for incorporated with the Whale optimization algorithm (WOA) to
voltage and current controllers. The mathematical model used in form a nature-inspired hybridized algorithm. Multi-objective opti-
this study consists of two inverters, as there are two DGs. There mization problems often require various conflicting objectives to
are individual voltage and current controllers for each inverter sep- be solved, so it imposes the challenge of satisfying each of the con-
arately where K pv 1 ; K iv 1 , and K pc1 ; K ic1 are the PI gains for voltage flicting objectives and optimize them [13]. At times there exist dif-
and current controller of inverter-1, respectively. Similarly, ferent discordant objective functions, so non-dominated sorting
K pv 2 ; K iv 2 and K pc2 ; K ic2 are the PI gains for voltage and current con- technique at first sorts a set of Pareto-optimal solutions that can
4
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

be considered as the optimal solutions. In order to differentiate Step 6: Now, there are two cases for updating the position of the
between dominated and non-dominated sets of data, Pareto- whales: one is by waning or reducing the encircling procedure, and
optimal value is computed from the Pareto-optimal solutions the other one is using spiral model [3]. So the probability is indi-
[28]. As mentioned earlier, the sorting is done by moving the opti- cated by p, which is a random number between [0, 1]. In this step,
mal solutions towards Pareto-front by non-dominated sorting of an if-else condition is run to satisfy the objective. If random num-
their rank values. The obtained non-dominated sorted solutions ber p < 0:5 and j~ Aj < 1 then the position of search agent is
are then optimized using the WOA technique in order to obtain a updated using Eqs. (14) and (15).
better, efficient result with less computational time. As a non-
Alternately, if p < 0:5 and j~
Aj P 1, then a random search agent
dominated sorting technique is hybridized with WOA, so this pro-
is selected, and X~r denotes the position of that search agent in
posed technique is named as the Non-dominated Sorting Whale
the current population. For that random search agent, the distance
Optimization Algorithm (NSWOA).
between prey and search agent is determined using Eq. (19). Fur-
The whale optimization algorithm consists of three stages. In
thermore, the position of the search agent is updated using Eq.
the first stage, the humpback whales encircle their preys, and in
(20).
the next stage, they use bubble hunt methods to attack their prey.
Finally, the whales again begin to search for their prey randomly D ¼ j~
~ C:X~r  ~
Xj ð19Þ
depending upon the position of each other [3,29]. A flowchart is
given in Fig. 3, showing a detailed procedure of NSWOA. The ~ Xr  ~
Xðt þ 1Þ ¼ ~ A:~
D ð20Þ
step-wise procedure of NSWOA is described as follow:
Step 1: The first step is to initialize the system parameters and Step 7: If the conditions of step 6 are not satisfied and if p P 0:5,
define the number of iteration and generate the initial population then the position of the search agent is updated using Eq. (18).
for the controller parameters, which are to be taken into consider- Step 8: Fitness function values of the search agent of the whales
ation for optimization. In this work, size of the initial population of are evaluated in this step, which is the new fitness value N f .
whales, i.e., search agent was taken as 50. Step 9: Initial generation of fitness function values generated in
Step 2: In this step, the fitness of each search agent is evaluated step 3 i.e., O is merged with the WOA optimized new fitness values
using fitness function i.e., the objective functions J 1 and J 2 of this generated in step 8 i.e., N f to generate a solution vector of length
study. O þ Nf .
Step 3: In this step, initially generated search agents are sorted Step 10: Here, non dominated sorting of the new solution vector
according to their fitness value. The non-dominated sorting tech- is carried out. A loop is run from step 4 to step 10 till the maximum
nique was used as the sorting method to sort the search agents. number of iteration is reached, and stopping criterion is fulfilled.
Next crowding distance and ranking of the current population is Step 11: The best possible position of the search agent is deter-
calculated. mined after the loop ends, and that is the optimal solution.
Step 4: In this step, a loop is run to update the position of the Thus, in this way the NSWOA optimization technique is being
search agents so that they can encircle their prey. Humpback implemented.
whales, i.e., the search agents considers the current solution as
the best solution and detect that as the target prey. Now, if ~ X  is 4. Results and discussion
considered as the position of current best solution, then encircling
behavior of the whales can be represented by Eqs. (14) and (15) [3]. The mathematical model of the microgrid as mentioned above
is being represented in MATLAB software in order to carry out
D ¼ j~
~ ~
C:X ðtÞ  ~
XðtÞj ð14Þ
the performance analysis of the microgrid. In this section, the sys-
tem model is used to carry out the stability and performance anal-
Xðt þ 1Þ ¼ X~ ðtÞ  ~
~ A:~
D ð15Þ ysis of the system by determining the optimized controller
parameters using the proposed NSWOA algorithm.
Here ~
X is the position vector, j j represents absolute value, 0: 0 is
used for the element by element multiplication, t represents the
4.1. Simulation setup
number of iteration; ~
A; ~
C are co-efficient vectors which are updated
to adjust the latest position of the prey around the search agent fol- The simulating parameters for all the optimizing algorithms,
lowing Eqs. (16) and (17) [3]. namely: NSWOA, NSGA-II, and SPEA, are kept the same due to
~ the comparative study that is to be made amongst these algo-
A ¼ 2~ r ~
a:~ a ð16Þ
rithms in order to compare their performance. The initial popula-
tion size for NSWOA, NSGA-II, and SPEA is taken as 50, and the
~
C ¼ 2:~
r ð17Þ maximum number of iterations for each case is 100. As all these
algorithms consist of a common trait that is randomness, so micro-
Here, ~
a decreases linearly from 2 to 0 as the iteration progresses,
grid performance using each algorithm was simulated for 30 inde-
which makes ~ ~ ~
A vary in ½a; a range, and ~
r is a random vector pendent runs to find the average performance of each of the
between [0, 1] [3]. algorithms and compare their performance. The search space was
Step 5: The next stage of the WOA technique is spiral position- kept the same for all the aforementioned algorithms as the con-
ing update of the humpback whales, which is done to replicate troller parameters were kept within 0 to 500. Initial population
their helix-shaped movement. Positions of the search agent are was selected randomly within the mentioned limit and all the sim-
updated using Eq. (18) as given in [3]. ulations were done using MATLAB. The simulations were per-
~
Xðt þ 1Þ ¼ ~
D0 :ebl : cosð2plÞ þ ~
X  ðtÞ ð18Þ formed using a laptop of CPU of 8th generation, Intel Core i7-
8550U at 1.8 GHz with 12 GB installed RAM.
where, b defines the shape of the logarithmic spiral movement of Now for NSWOA, the random vector ~ r is defined between
the whales and it is a constant value, l is a random number between
a linearly decreases from 2 to 0, which results in ~
½0; 1; ~ A to be lim-
[-1, 1], and ~
D0 ¼ j~
X  ðtÞ  ~
XðtÞjindicates the distance between the ited in the range of ½a; a. b is taken as 1 to define the logarithmic
position of prey (X~ ) and the position (~
X) of search agent. shape of spiral movement, and p is defined as a random number
5
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 3. Flow chart depicting NSWOA.

between ½0; 1. All these parameters are adopted from following first eigenvalue analysis is done to ensure the ability of NSWOA,
Ref. [3]. NSGA-II and SPEA in stabilizing the performance of the dynamic
Similarly, for NSGA-II, the rate of mutation was defined as 0.02. system. Later on, Time Domain Simulation, and Statistical Tests
Percentage of crossover and percentage of mutation was taken as were performed where the performance of each of the algorithms
0.7 and 0.4, respectively [28]. has been analyzed.
Finally, for SPEA, the archive size was defined as 50, the per-
centage of crossover is the same as NSGA-II i.e., 0.7, and the per-
4.2.1. Eigenvalue analysis
centage of mutation is 0.3 [25].
Eigenvalue analysis has been done to determine the capability
of the proposed NSWOA algorithm in stabilizing the proposed sys-
4.2. Comparative assessment tem containing both static and dynamic loads. The eigenvalue of
the state variables of the proposed system has been reflected in
In this portion, a comparative analysis is performed between Fig. 4, which shows the pre-optimization condition. It is noticed
proposed NSWOA and well known NSGA-II, SPEA algorithms. At from Fig. 4 and Table 1 that there exists instability in the system
6
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 4. Eigenvalue of the system before optimization. Fig. 5. Eigenvalue of the system after optimization using NSWOA.

as the real parts of the eigenvalues of states Duq2 ; Dcd2 ; Dioq2 and
DiLineD are situated on the right half portion of s plane. As a result,
the proposed NSWOA algorithm is being used to stabilize the sys-
tem in a optimized manner.
Fig. 5–7 shows the eigenvalue of the states after optimizing the
controller parameters of the system using NSWOA, NSGA-II and
SPEA algorithm respectively. The dominant eigenvalues i.e., the
values nearer to the imaginary axes are depicted in Fig. 4–7 as
other values are well situated in the left half of s plane indicating
those states being stable. The eigenvalues of the system before
and after optimization are mentioned in Table 1 and Table 2. It is
observable that the unstable positive eigenvalues of the states
Duq2 ; Dcd2 ; Dioq2 , and DiLineD have moved from right half of the s
plane to the left half of s plane for all the mentioned algorithms.
For Duq2 and Dioq2 there is no significant difference between Fig. 6. Eigenvalue of the system after optimization using NSGA-II.

Table 1
Eigenvalue Analysis of NSWOA.

Index States Eigen Value of the System Before Optimization Eigen Value of the System After Using NSWOA Optimization
1 Dd1 2909410.542 + 12209388.904i 2909410.508 + 12209388.839i
2 DP 1 2909410.542–12209388.904i 2909410.508–12209388.839i
3 DQ 1 3261123.439 + 8309127.704i 3261123.425 + 8309127.501i
4 Dud1 3261123.439–8309127.704i 3261123.425–8309127.501i
5 Duq1 55.194 + 45315.607i 28863.554 + 384659.375i
6 Dcd1 55.194–45315.607i 28863.554–384659.375i
7 Dcq1 352.417 + 44476.812i 28820.165 + 384322.746i
8 Dild1 352.417–44476.812i 28820.165–384322.746i
9 Dilq1 4107.969 + 31524.405i 12450.226 + 172830.502i
10 Dv od1 4107.969–31524.405i 12450.226–172830.502i
11 Dv oq1 5629.972 + 29764.098i 12497.721 + 172390.083i
12 Diod1 5629.972–29764.098i 12497.721–172390.083i
13 Dioq1 8720.541 + 8365.113i 8197.709 + 21329.295i
14 Dd2 8720.541–8365.113i 8197.709–21329.295i
15 DP 2 6328.127 + 8624.675i 12535.996 + 19463.657i
16 DQ 2 6328.127–8624.675i 12535.996–19463.657i
17 Dud2 1291.429 + 0i 2217.887 + 515.833i
18 Duq2 213.426 + 784.754i 2217.887–515.833i
19 Dcd2 213.426–784.754i 22.813 + 194.338i
20 Dcq2 81.284 + 376.280i 22.813–194.338i
21 Dild2 81.284–376.280i 138.088 + 0i
22 Dilq2 162.677 + 0i 4.848 + 52.592i
23 Dv od2 70.227 + 1.578i 4.848–52.592i
24 Dv oq2 70.227–1.578i 70.683 + 0i
25 Diod2 67.497 + 0i 67.556 + 2.059i
26 Dioq2 22.647 + 0i 67.556–2.059i
27 DiLineD 1.315 + 0i 5.740 + 0.974i
28 DiLineQ 2.394 + 0i 5.740–0.974i
29 DiLoadD 2.393 + 0i 2.801 + 0i
30 DiLoadQ 0.018 + 0.045i 1.715 + 0i
31 DiQs 0.018–0.045i 0.188 + 0i
32 DiDs 0.021 + 0i 1.472 + 1.870e-06i
33 DiQr 0.329 + 0i 1.472–1.870e-06i
34 DiDr 0.200 + 0i 0.405 + 0i
35 Dsl 0.202 + 0i 0.405 + 0i

7
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

dominated solutions for both the objectives. The optimized set of


controller parameters resulting in this Pareto-front are K pv 1 ¼
60:0229; K iv 1 ¼ 339:6462; K pc1 ¼ 105:3193; K ic1 ¼ 155:0871;
K pv 2 ¼ 129:8271; K iv 2 ¼ 227:7560; K pc2 ¼ 240:2978; K ic2 ¼ 97:3373:

4.2.2. Time domain simulation analysis


In this section, the results of proposed NSWOA algorithms are
compared with both NSGA-II and SPEA algorithms to differentiate
their performance. The comparison of performance was made
based on the amount of peak overshoot, and oscillation frequency
of real power, reactive power, inductor current (d  q), and output
voltage (d  q) of DG-1 and DG-2. Conventional approach was fol-
lowed to calculate the oscillation frequency from the imaginary
Fig. 7. Eigenvalue of the system after optimization using SPEA. part of eigenvalues[30]. Peak overshoot was calculated from the
step response of the mentioned parameters in MATLAB. Peak over-
shoot is basically the deviation of the peak value of the states from
its steady state point. The comparison is presented inTable 3. The
NSWOA, NSGA-II and SPEA. However, for Dcd2 and DiLineD it is step response of the above-mentioned parameters for DG-1 and
observable that NSGA-II is comparatively less negative to NSWOA DG-2 is represented inFig. 9–11 where a step disturbance is given
and SPEA. Moreover, the eigenvalues nearer to the imaginary axes at 1 sec. For the real power of DG-1, it is visible from Table 3 and
i.e., of states DiLoadQ ; DiQs ; DiDs ; DiQr ; DiDr , and Dsl have moved further Fig. 9(a) that SPEA has a peak overshoot of 72:2e  5 compared
away from the imaginary axes towards into the left-half portion of to zero peak overshoot in NSGA-II and NSWOA. However, oscilla-
s plane which indicates that system has become more stable after tion frequency is comparable for all these algorithms. In the case
using NSWOA technique. of DG-2, all the algorithms namely NSGA-II, SPEA and NSWOA
In non-dominated sorting technique basically pareto- gives zero peak overshoot, as shown in Fig. 9(b). Similarly, in the
dominance criterion is used to sort the solutions obtained from case of reactive power, as shown in Fig. 9c) and Fig. 9(d) and tab-
the generated population. Non-dominated solutions are the solu- ulated data inTable 3 it is seen that NSWOA and NSGA-II gives zero
tions where one objective function does not dominate other objec- peak overshoot for both the DGs. But, SPEA produces a high peak
tive function. Fig. 8 represents the most optimum Pareto front overshoot of 2398 for DG-1 and 4:4 peak overshoot for DG-2. The
among the 30 independent runs, which shows the non- oscillation frequency of reactive power is comparable for all the

Table 2
Eigenvalue Analysis of NSGA-II and SPEA.

Index States Eigen Value of the System Using NSGA-II Optimization Eigen Value of the System Using SPEA Optimization
1 Dd1 2909410.547 + 12209388.859i 2909410.529 + 12209388.805i
2 DP 1 2909410.547–12209388.859i 2909410.529–12209388.805i
3 DQ 1 3261123.505 + 8309127.596i 3261123.517 + 8309127.438i
4 Dud1 3261123.505–8309127.596i 3261123.517–8309127.438i
5 Duq1 31079.667 + 748441.857i 59787.105 + 486045.741i
6 Dcd1 31079.667–748441.857i 59787.105–486045.741i
7 Dcq1 31097.864 + 748806.691i 59850.414 + 486371.655i
8 Dild1 31097.864–748806.691i 59850.414–486371.655i
9 Dilq1 25183.414 + 123925.053i 42180.717 + 316294.455i
10 Dv od1 25183.414–123925.053i 42180.717–316294.455i
11 Dv oq1 25572.932 + 123361.294i 42190.873 + 316724.617i
12 Diod1 25572.932–123361.294i 42190.873–316724.617i
13 Dioq1 7630.771 + 30869.799i 8158.994 + 20798.415i
14 Dd2 7630.771–30869.799i 8158.994–20798.415i
15 DP 2 12648.788 + 29168.383i 12454.773 + 18925.813i
16 DQ 2 12648.788–29168.383i 12454.773–18925.813i
17 Dud2 2476.499 + 526.494i 2178.097 + 516.632i
18 Duq2 2476.499–526.494i 2178.097–516.632i
19 Dcd2 11.526 + 134.882i 23.435 + 200.800i
20 Dcq2 11.526–134.882i 23.435–200.800i
21 Dild2 7.757 + 47.245i 137.294 + 0i
22 Dilq2 7.757–47.245i 7.810 + 49.700i
23 Dv od2 80.298 + 0i 7.810–49.700ii
24 Dv oq2 71.040 + 5.850i 70.767 + 0i
25 Diod2 71.040–5.850i 67.341 + 2.142i
26 Dioq2 69.908 + 0i 67.341–2.142i
27 DiLineD 0.850 + 1.034i 7.426 + 0i
28 DiLineQ 0.850–1.034i 4.239 + 0i
29 DiLoadD 1.197 + 0i 2.637 + 0.274i
30 DiLoadQ 0.395 + 0i 2.637–0.274i
31 DiQs 1.058 + 0i 0.207 + 0i
32 DiDs 0.730 + 0.0001i 1.381 + 1.122e-05i
33 DiQr 0.730–0.0001i 1.381–1.122e-05i
34 DiDr 0.682 + 0i 0.873 + 0i
35 Dsl 0.682 + 0i 0.873 + 0i

8
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 8. Pareto Front of the objective functions.

Table 3
NSGA-II, SPEA and NSWOA comparison on the basis of peak overshoot and oscillation frequency.

Criterion Parameters NSGA-II SPEA NSWOA


Real Power of DG-1 Peak Overshoot 0 72.2e 5 0
Oscillation Frequency (Hz) 1943184.58924875 1943184.58908575 1943184.58590787
Real Power of DG-2 Peak Overshoot 0 0 0
Oscillation Frequency (Hz) 3093.62629126010 5025.72911114868 3097.73723900745
Reactive Power of DG-1 Peak Overshoot 0 2398 0
Oscillation Frequency (Hz) 1322438.73204884 1322438.73035667 1322438.71460419
Reactive Power of DG-2 Peak Overshoot 0 4.4 0
Oscillation Frequency (Hz) 3093.62629126010 5025.72911114868 3097.73723900745
Output Voltage of DG-1 (d-axis) Peak Overshoot 0 0 0
Oscillation Frequency (Hz) 42789.3365339348 18305.6577760513 27506.8287713625
Output Voltage of DG-2 (d-axis) Peak Overshoot 0 0 0
Oscillation Frequency (Hz) 8.10924153674564 2.04792679461399 8.37029981802671
Output Voltage of DG-1 (q-axis) Peak Overshoot 5.1e - 3 73.1e - 3 0
Oscillation Frequency (Hz) 42858.5317697674 18218.3019399697 27436.7338495822
Output Voltage of DG-2 (q-axis) Peak Overshoot 0 12.2 0
Oscillation Frequency (Hz) 0.931166398517253 2.04792679461399 0
Inductor Current of DG-1 (d-axis) Peak Overshoot 0 0 0
Oscillation Frequency (Hz) 64349.5728503008 75598.1848439340 61166.8648946795
Inductor Current of DG-2 (d-axis) Peak Overshoot 0 4.5 0
Oscillation Frequency (Hz) 0 6.78536409154396 0
Inductor Current of DG-1 (q-axis) Peak Overshoot 5.52 32.29 0
Oscillation Frequency (Hz) 42789.3365339348 18305.6577760513 27506.8287713625
Inductor Current of DG-2 (q-axis) Peak Overshoot 1.91 1.43 1.52
Oscillation Frequency (Hz) 8.10924153674564 6.78536409154396 8.37029981802671

algorithms in the case of DG-1, whereas for DG-2, SPEA provides 32:29 respectively compared to zero peak overshoot of NSWOA.
more oscillation frequency i.e., 5025:729 Hz compared to Moreover, SPEA has the least oscillation frequency of
3093:626 Hz of NSGA-II and 3097:737 Hz of NSWOA. Now, for 18305:6577 Hz compared to NSGA-II and NSWOA. For DG-2,
the inductor current (d  axis), it is observed fromFig. 10(a) that Fig. 10(d) shows that the oscillation frequency and peak overshoot
in case of DG-1, all optimizing algorithms provide zero peak over- values of NSGA-II,SPEA and NSWOA are comparable. Now for out-
shoot, among which NSWOA has the least oscillation frequency of put voltage (d  axis) of DG-1 and DG-2, as shown in Fig. 11a) and
61166:864 Hz compared to 64349:572 Hz of NSGA-II and Fig. 11(b), it is seen that all algorithms produce zero peak over-
75598:184 Hz of SPEA. Similarly, for DG-2 shown inFig. 10(b), shoot for both the DGs. Now, SPEA and NSWOA have a relatively
NSGA-II and NSWOA gives zero peak overshoot compared to 4:5 lower oscillation frequency for the output voltage (d  axis) for
peak overshoot of SPEA where, the oscillation frequency of both DGs. Similarly, for output voltage (q  axis) represented
NSGA-II and NSWOA is zero compared to more significant oscilla- inFig. 11(c), Fig. 11(d), and Table 3, NSWOA provides zero peak
tion frequency of SPEA, which is 6:785 Hz. Now, inductor current overshoot for both DGs with better oscillation frequency response.
(q  axis), it is seen fromFig. 10(c) and tabulated data that for But, NSGA-II and SPEA gives peak overshoot values of 5:1e  3 and
DG-1, NSGA-II and SPEA give peak overshoot values of 5:52 and 73:1e  3 respectively for DG-1 where as for DG-2 the peak over-

9
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 9. Step Response of the system for Real and Reactive Power.

Fig. 10. Step Response of the system for Inductor Current.

shoot value of SPEA is 12:2. The time-domain analysis of the sys- and improve damping characteristics of the system, where the pro-
tem using the algorithms, namely NSGA-II, SPEA, and NSWOA, posed NSWOA algorithm provides better step response compared
shows that though all the optimizing techniques can optimize to the other mentioned algorithms.

10
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 11. Step Response of the system for Output Voltage.

Table 4
NSGA-II and NSWOA comparison based on F-test and t-test.

Parameters Levene’s Test for Equality Equality Means Test (t - test)


of Variances (F- test)
F Sig (p) t df Sig. (2- tailed)(p) Mean Difference
Total Number of iterations 5.069276 .028 4.040 52.273 .000 7.50000
Total summation of Eigenvalues (Real) 8.947109 .004 1.795 42.743 .078 45741.56900
Total time required 2.758497 .102 3.473 58 .001 3.80446

4.2.3. Statistical tests ered, which indicates the mean values of the algorithms’ group
In order to ratify the competence of the algorithms mentioned data are equal. However, if Sigmað2  tailedÞðpÞ is less than 0:05,
above, a statistical assessment was enacted using SPSS software, then alternative hypothesis, H1 will be taken into consideration,
where an independent sample t  test was carried out for determi- which indicates that the mean value of group data is not equal.
nation of the uniqueness of each of the algorithms. In this test, the This independent sample t  test helps in determining the homo-
parameters taken under consideration were the total number of geneity of the algorithms.
iterations needed by each of the algorithms to satisfy the objective Table 4 shows the F  test and t  test analysis of NSWOA and
functions, the total time required for completion of the iteration, NSGA-II. It is observable from the above-mentioned table that in
and total summation of the real part of eigenvalues for determina- F  test; SigmaðpÞ value is higher than 0:05 for the total time
tion of the algorithm providing most negative eigenvalues. The required for completion of iteration only, but for all other cases,
data sets included here are for 30 independent runs from, which it is less than 0:05. So, it concludes that the variance of NSWOA
the average data are being tabulated. The total time required for and NSGA-II is not spread equally. Now, for equality means
completion of the iteration indicates the time required by the algo- t  test, it is seen from Table 4 that equality means exists only
rithm from initialization to reach the optimized controller param- for total summation of the real part of eigenvalues where
eters which is actually algorithm run time. Sigmað2  tailedÞðpÞ is higher than 0:05. For the total number of
Lev ene0s test for equality of v ariances [31] i.e., F  test, was also iterations and total time required, it is less than 0:05, which indi-
carried out alongside the t  test. F  test determines whether cates that the null hypothesis is rejected, and the difference in
there exist equal variances among the algorithm depending upon means between NSGA-II and NSWOA is statistically significant.
the condition that the SigmaðpÞ value is higher than 0:05 or not. Similarly, Table 5, shows the analysis between NSWOA with
If the value is higher than 0:05, then equal variances between the SPEA. From Table 5 it is seen that the SigmaðpÞ value of the
data sets of the algorithms are assumed. Similarly, in t  test, the F  test is less than 0:05 for the total summation of the real part
assumption of null hypothesis ðHo Þ and alternative hypothesis H1 of eigenvalues though for the total number of iterations, it is
are determined based on Sigmað2  tailedÞðpÞ value. If the value is slightly higher than 0:05. Moreover, from the t  test, it is observed
higher than 0:05, then the null hypothesis ðHo Þ needs to be consid- that Sigmað2  tailedÞðpÞ is significantly lower than 0:05 for the

11
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

Table 5
SPEA and NSWOA comparison based on F-test and t-test.

Parameters Levene’s Test for Equality Equality Means Test (t - test)


of Variances (F- test)
F Sig(p) t df Sig. (2- tailed)(p) Mean Difference
Total Number of iterations 3.092900 .084 2.287 58 .029 2.56667
Total summation of Eigenvalues (Real) 8.527837 .005 .942 42.896 .350 24030.48633
Total time required 1.622513 .208 3.603 58 .001 2.74787

Table 6
Group Statistical Data of NSGA-II, SPEA and NSWOA.

Parameters Algorithm Mean Standard Deviation Standard Error Mean


Total Number of iterations NSGA-II 11.5333 8.29513 1.51448
SPEA 10.1333 4.24047 .77420
NSWOA 4.0333 5.88091 1.07370
Total Summation of Eigenvalues (Real) NSGA-II 12734141.7097 62613.54069 11431.61621
SPEA 12709722.4533 60017.17312 10957.58652
NSWOA 12688400.1407 124729.01128 22772.29768
Total Time Required(sec) NSGA-II 6.7245 4.57376 .83505
SPEA 4.2842 1.88320 .34382
NSWOA 2.9201 3.88320 .70897

total number of iterations and total summation of the real part of NSWOA can be useful is optimizing controller gains for grid
eigenvalues. Thus from Table 5, it can be analyzed that there exists connected mode of operation of microgrid provided the system
significant difference in means between NSWOA, and SPEA model is developed for grid connected mode. As NSWOA is not
statistically. based on any unique problem, instead it is compatible with other
Table 6 shows the group data of NSGA-II, SPEA and NSWOA systems depending on system parameters, so in future, this hybri-
from which it is clearly visible that for NSWOA, even though total dized NSWOA technique might be used in other aspects of micro-
summation of eigenvalues is least negative, but NSWOA takes a grid such as, load flow analysis in order to determine steady state
significant lesser number of iterations and less time to achieve con- operating point of microgrid, optimal sizing of hybrid renewable
vergence compared to NSGA-II and SPEA. Eigenvalue analysis energy systems (HRES), etc.
proved the ability of NSWOA to obtain a much stable system,
which ensures that despite having the least negative total summa-
tion of the real part of eigenvalues, NSWOA is much efficient and Declaration of Competing Interest
much faster in obtaining optimized values for the controller.
None.

5. Conclusion
References
In this work, a new hybridized optimization algorithm entitled,
[1] S.M. Abdullah, A. Ahmed, Hybrid bare bones fireworks algorithm for load flow
Non-dominated Sorting Whale Optimization Algorithm (NSWOA), analysis of islanded microgrids, in: Handbook of Research on Fireworks
has been developed by combining nature-inspired swarm intelli- Algorithms and Swarm Intelligence, IGI Global, 2020, pp. 283–314.
gence based Whale Optimization Algorithm (WOA) with multi- [2] Q.N.U. Islam, A. Ahmed, Optimized controller design for islanded microgrid
employing nondominated sorting firefly algorithm, in: Nature-Inspired
objective based non-dominated sorting technique. Moreover, a
Computation and Swarm Intelligence, Elsevier, 2020, pp. 247–272.
mathematical model of typical microgrid was developed with a [3] Mirjalili S, Lewis A. The whale optimization algorithm. Adv Eng Software
static load and induction motor as a dynamic load. The controller 2016;95:51–67.
parameters of the developed microgrid system were optimized [4] Hasanien HM. Whale optimisation algorithm for automatic generation control
of interconnected modern power systems including renewable energy sources.
using NSGA-II, SPEA, and the new NSWOA technique, and detailed IET Gener, Transmiss Distrib 2017;12(3):607–14.
performance analysis of each of the algorithms along with the sta- [5] Hasanien HM. Performance improvement of photovoltaic power systems using
tistical analysis was carried out. The performance analysis and sta- an optimal control strategy based on whale optimization algorithm. Electric
Power Syst Res 2018;157:168–76.
tistical analysis concludes to the fact that the proposed NSWOA [6] Elazab OS, Hasanien HM, Elgendy MA, Abdeen AM. Parameters estimation of
technique can obtain a stable system with less overshoot and oscil- single-and multiple-diode photovoltaic model using whale optimisation
lation frequency in most cases compared to the well recognized algorithm. IET Renew Power Gener 2018;12(15):1755–61.
[7] El-Fergany AA, Hasanien HM, Agwa AM. Semi-empirical PEM fuel cells model
NSGA-II, and SPEA technique. Furthermore, the computation speed using whale optimization algorithm. Energy Convers Manage
of NSWOA is much faster than NSGA-II and SPEA. It takes less num- 2019;201:112197.
ber of iteration for reaching convergence, which adds to the bene- [8] Qais MH, Hasanien HM, Alghuwainem S. Whale optimization algorithm-based
Sugeno fuzzy logic controller for fault ride-through improvement of grid-
ficial characteristics of NSWOA. For instance, the total number of connected variable speed wind generators. Eng Appl Artif Intell
iteration required by NSWOA to reach convergence is 4:033 ’ 4 2020;87:103328.
compared to 11:533 ’ 12 iterations of NSGA-II and 10:13 ’ 10 [9] Laskar NM, Guha K, Chatterjee I, Chanda S, Baishnab KL, Paul PK. Hwpso: A new
hybrid whale-particle swarm optimization algorithm and its application in
iterations of SPEA. Similarly, the time to reach convergence is
electronic design optimization problems. Appl Intell 2019;49(1):265–91.
2:92 s for NSWOA, which is much faster than 6:72 s of NSGA-II, [10] I.N. Trivedi, P. Jangir, A. Kumar, N. Jangir, R. Totlani, A novel hybrid pso–woa
and 4:28 s of SPEA. From the statistical F  test and t  test, it is algorithm for global numerical functions optimization, in: Advances in
also proved that NSWOA, NSGA-II, and SPEA techniques are differ- Computer and Computational Sciences, Springer, 2018, pp. 53–60.
[11] Abdel-Basset M, Manogaran G, El-Shahat D, Mirjalili S. A hybrid whale
ent from each other and there are some unique features in each of optimization algorithm based on local search strategy for the permutation
these algorithms. flow shop scheduling problem. Future Gener Comput Syst 2018;85:129–45.

12
Quazi Nafees Ul Islam, A. Ahmed and Saad Mohammad Abdullah Ain Shams Engineering Journal xxx (xxxx) xxx

[12] Qais MH, Hasanien HM, Alghuwainem S. Enhanced whale optimization [26] Kahrobaeian A, Mohamed YA-RI. Analysis and mitigation of low-frequency
algorithm for maximum power point tracking of variable-speed wind instabilities in autonomous medium-voltage converter-based microgrids with
generators. Appl Soft Comput 2020;86:105937. dynamic loads. IEEE Trans Industr Electron 2013;61(4):1643–58.
[13] Prabakar K, Li F, Xiao B. Controller hardware-in-loop testbed setup for multi- [27] Mudaliyar SR, Sahoo SS. Comparison of different eigenvalue based multi-
objective optimization based tuning of inverter controller parameters in a objective functions for robust design of power system stabilizers. Int J Electr
microgrid setting. In: 2016 Clemson University power systems conference Electron Eng Telecommun 2015;1(2):329–42.
(PSC), IEEE. p. 1–8. [28] Deb K, Agrawal S, Pratap A, Meyarivan T. A fast elitist non-dominated sorting
[14] Ishibuchi H, Nojima Y, Doi T. Comparison between single-objective and multi- genetic algorithm for multi-objective optimization: Nsga-ii. In: International
objective genetic algorithms: Performance comparison and performance conference on parallel problem solving from nature. Springer; 2000. p.
measures. In: 2006 IEEE International conference on evolutionary 849–58.
computation, IEEE. p. 1143–50. [29] Reddy PDP, Reddy VV, Manohar TG. Optimal renewable resources placement
[15] El-Fergany AA, Hasanien HM. Single and multi-objective optimal power flow in distribution networks by combined power loss index and whale
using grey wolf optimizer and differential evolution algorithms. Electric Power optimization algorithms. J Electr Syst Inf Technol 2017;28:669–78.
Comp Syst 2015;43(13):1548–59. [30] Nise NS. Control systems engineering. John Wiley & Sons; 2020.
[16] Basu AK, Chowdhury S, Chowdhury S, Paul S. Microgrids: Energy management [31] Brown MB, Forsythe AB. Robust tests for the equality of variances. J Am Stat
by strategic deployment of ders–a comprehensive survey. Renew Sustain Assoc 1974;69(346):364–7.
Energy Rev 2011;15(9):4348–56.
[17] Al-Awami A, Sortomme E, El-Sharkawi M. Optimizing economic,
Quazi Nafees Ul Islam completed his B.Sc. and M.Sc. degrees from the Electrical and
environmental dispatch with wind and thermal units. In: IEEE Power &
Electronic Engineering (EEE) department of Islamic University of Technology (IUT),
energy society general meeting, IEEE, 2009. p. 1–6.
[18] Hsu C-T, Korimara R, Cheng T-J, Tsai L-J, Huang H-M. Cost power curtailment Gazipur, Bangladesh on 2016 and 2020 (January) respectively. His undergrad
analysis for optimum PV size and the energy potential for the desalination research was focused on solar cell material and improvement of efficiency of thin
plant on the island distribution system. J Clean Energy Technol 2017;5 film solar cell. Later on, he did his M.Sc. on optimization of Microgrid controller
(3):199–205. parameters. He is working as a Lecturer in the Electrical and Electronic Engineering
[19] Hemamalini S, Simon S. Maclaurin series-based lagrangian method for (EEE) department of IUT since 2017. Currently his research interest includes: neural
economic dispatch with valve-point effect. IET Gener, Transmiss Distrib network, optimization of microgrid controller parameters, nature inspired opti-
2009;3(9):859–71. mization algorithm.
[20] Coelho LS, Mariani VC. Combining of chaotic differential evolution and
quadratic programming for economic dispatch optimization with valve-point Dr. Ashik Ahmed completed his MSc and PhD in Electrical and Electronic Engi-
effect. IEEE Trans Power Syst 2006;21(2):989–96. neering (EEE) from King Fahd University of Petroleum and Minerals, Dharhan, Saudi
[21] Iyer SV, Belur MN, Chandorkar MC. A generalized computational method to Arabia, and Islamic University of Technology (IUT), Gazipur, Bangladesh in the years
determine stability of a multi-inverter microgrid. IEEE Trans Power Electron 2010 and 2016, respectively. Earlier, he completed his undergraduate level of study
2010;25(9):2420–32.
from the EEE department of IUT, Gazipur, Bangladesh in 2003. Currently, he is
[22] Pogaku N, Prodanovic M, Green TC. Modeling, analysis and testing of
serving as a Professor in the EEE department of IUT. His current fields of research
autonomous operation of an inverter-based microgrid. IEEE Trans Power
interests include Swarm and Evolutionary Algorithms, Nonlinear Control of
Electron 2007;22(2):613–25.
[23] Ibrahim A, Jibril Y, Haruna Y. Determination of Optimal Droop Controller dynamic system, Power System Stability and Control, Flexible AC Transmission
Parameters for an Islanded Microgrid System Using Artificial Fish Swarm Systems, Demand Side Management, Hybrid Renewable Energy Systems, etc.
Algorithm (AFSA). Int J Scient Eng Res 2017;8(3):959–65.
[24] Wang R, Wu S, Wang C, An S, Sun Z, Li W, Xu W, Mu S, Fu M. optimized Saad Mohammad Abdullah received his B.Sc. and M.Sc. in Electrical and Electronic
operation and control of microgrid based on multi-objective genetic Engineering (EEE) from Islamic University of Technology (IUT), Gazipur, Bangla-
algorithm. In: 2018 International conference on power system technology desh, in the years 2016 and 2019, respectively. He is working as a lecturer in the
(POWERCON), IEEE. p. 1539–44. department of EEE, IUT since 2017. His research interests include microgrids,
[25] Zitzler E, Laumanns M, Thiele L. SPEA2: Improving the strength Pareto engineering optimization, power system stability and control, and wireless power
evolutionary algorithm. TIK-report 2001;103:1–21. transfer.

13

You might also like