Received April 2, 2022, accepted April 15, 2022, date of publication April 26, 2022, date of current version

May 10, 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3170547

Performance Improvement of Multiobjective

Optimal Power Flow-Based Renewable Energy
Sources Using Intelligent Algorithm
TRUONG HOANG BAO HUY 1 , TRI PHUOC NGUYEN 2 , NURSYARIZAL MOHD NOR 3,

IRRAIVAN ELAMVAZUTHI 3 , (Member, IEEE), TAIB IBRAHIM3 , AND DIEU NGOC VO 2,4
1 Institute
of Engineering and Technology, Thu Dau Mot University, Thu Dau Mot, Binh Duong 75109, Vietnam
2 Department of Power Systems, Ho Chi Minh City University of Technology (HCMUT), Ho Chi Minh City 72506, Vietnam
3 Department of Electrical and Electronics Engineering, Universiti Teknologi PETRONAS, Seri Iskandar, Perak 32610, Malaysia
4 Vietnam National University Ho Chi Minh City (VNU-HCM), Thu Duc, Ho Chi Minh City 71308, Vietnam

Corresponding author: Nursyarizal Mohd Nor ([email protected])

This work was supported by the Ministry of Higher Education (MOHE) through the Fundamental Research Grant Scheme (FRGS) and the
Universiti Teknologi PETRONAS under Grant FRGS/1/2018/TK07/UTP/02/3.

ABSTRACT Producing energy from a variety of sources in a power system requires an optimal schedule
to operate the power grids economically and efficiently. Nowadays, power grids might include thermal
generators and renewable energy sources (RES). The integration of RES adds complexity to the optimal
power flow problem due to intermittence and uncertainty. The study suggests a Multi-Objective Search
Group Algorithm (MOSGA) to deal with multi-objective optimal power flow integrated with a stochastic
wind and solar powers (MOOPF-WS) problem. Weibull and lognormal probability distribution functions
(PDFs) are respectively adopted to describe uncertainties in wind speed and solar irradiance. The MOSGA
incorporates crowding distance strategies, fast non-dominated sorting, and an archive selection mechanism
to define and preserve the best non-dominated solutions. The total cost, real power loss, and emission
were defined as the objectives for the MOOPF-WS problem. In the economic aspect, formulated cost
modelling includes both overestimation and underestimation situations related to wind and solar power
prediction. Further, uncertainty in load demand is represented by a normal PDF and is considered as a special
study case due to its novelty. The effectiveness of MOSGA was validated on the IEEE 30-bus and 57-bus
systems considering various combinations of objective functions as well as different loading scenarios. Its
performance was comprehensively compared with the other three well-regarded multi-objective optimization
algorithms including NSGA-II, MOALO, and MOGOA in terms of Spread metric, Hypervolume metric,
and the best compromise solutions for all scenarios. The comparisons showed MOSGA was capable of
obtaining well-distributed Pareto fronts and producing better quality solutions compared to the others in
all tested scenarios. In addition, MOSGA also obtained better solution quality than significant research in
the literature for all the comparable cases. These show the superiority of the MOSGA in dealing with the
MOOPF-WS problem.

INDEX TERMS Multi-objective search group algorithm, multi-objective optimal power flow, renewable
energy sources, wind power, solar power.

I. INTRODUCTION systems [2]. Generation cost minimization is the primary

The optimal power flow (OPF) was developed by objective of traditional OPF. However, environmental con-
Carpentier [1], and the OPF problem has become a significant cerns and technical issues have led to the consideration of
function in the production and operation of modern energy different objectives, including emissions, real power loss,
voltage stability, and voltage profile. Thus, multi-objective
The associate editor coordinating the review of this manuscript and OPF (MOOPF) was formulated to optimize a group of
approving it for publication was Akshay Kumar Saha . objectives simultaneously in power systems. In general,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 10, 2022 48379
 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

the MOOPF problem considers fossil fuel-powered thermal differential evolution (MDE) technique was implemented by
generators. Renewable energy sources (RES), which are Shaheen et al. [15] for the MOOPF problem on 118-bus
environmentally friendly and sustainable, provide a better and 57-bus systems. Pulluri et al. [16] presented a mixed
solution, reducing the environmental impact and limiting crossover operator and self-adaptive strategy to improve the
fossil fuel consumption. Studies of OPF integrating RES have DE for the MOOPF problem. In [17], the authors proposed
become necessary due to their increasing application in power three different strategies to improve the strength Pareto evolu-
systems. tionary algorithm (ISPEA2). The outcomes showed that these
In the beginning stage of problem exploration, the MOOPF strategies enhanced the uniformity and distribution of Pareto
problem was handled by several conventional methods like fronts. Recently, the multi-objective dimension-based firefly
gradient-based method [3], linear programming [4], quadratic algorithm (MODFA) was suggested by Chen et al. [18] for
programming [5], Newton-based method [6], and interior the MOOPF problem with nine cases in three test systems.
point method [7]. Despite having good convergence char- Fuzzy affiliation was also used to extract the best compro-
acteristics and effectively handling inequality constraints, mise solution. Warid et al. [9] introduced a new method by
such conventional methods cannot ensure global optimum combining a Jaya algorithm with a quasi-oppositional (QO)
result [8]. In addition, conventional approaches are designed strategy. The new algorithm, QOMJaya, was implemented for
specifically for each type of MOOPF problem. In other the MOOPF problem.
words, each variant of the MOOPF problem is related to In the past decade, research studies mainly focused on
a certain solution approach. Further, these approaches face handling the OPF and MOOPF problems with thermal gener-
many difficulties when coping with the attributes of the ators. Recently, the OPF problem integrated with RES has
problem such as integer and discrete decision variables; non- attracted the attention of researchers due to the increasing
differentiable objective function; and large range of search penetration of RESs into power systems. Henerica et al. [19]
space. Moreover, the application of these methods to solving proposed an OPF solution for an off-grid hybrid diesel-
the MOOPF problem is very complex [9]. solar PV-battery system. In [20], the MBFA technique was
In the later stage of exploration, with the advent of a applied to handle the OPF framework considering hydro-
new field of computation known as soft computing, a large thermal-wind (HTW) systems. Jabr et al. [21] introduced
number of optimization algorithms belonging to the meta- into the OPF a stochastic model of wind power, forecast-
heuristic class have been proposed. The main advantages of ing power with a probability or relative frequency his-
such methods are that they can deal with large-scale search togram. To address the same problem, Mishra et al. [22]
spaces and are less dependent on problem characteristics. developed the bacterial foraging algorithm (BFA) for the
Moreover, these algorithms are capable of estimating multi- OPF to consider wind generation. Stochastic wind speed
ple points in the search domain simultaneously due to their and power distribution were defined using the Weibull dis-
population-based nature. Therefore, metaheuristic methods tribution. In [23], the authors introduced a wind-generated
are more effective in finding optimal solutions for OPF and cost model that included the opportunity cost of excesses
MOOPF problems in comparison with conventional meth- and shortfalls in wind power. The proposed method was
ods [9]. In the literature, the MOOPF problem has been implemented using the IEEE New England system. In [24],
studied extensively, and many powerful algorithms have been a modified bacteria foraging algorithm (MBFA) was applied
applied with remarkable results. In an attempt to deal with the to simulate uncertainty in wind energy generation in the
MOOPF problem, Bouchekara et al. [10] proposed improved OPF framework. In [25], the MBFA was implemented for
colliding bodies optimization (ICBO). The performance of the OPF problem considering combined thermal-wind gen-
ICBO was improved through iterations with increasing num- eration. Moreover, the voltage security aspect was improved
bers of colliding bodies. Chaib et al. [11] studied a backtrack- using shunt FACTS (STATCOM) devices to compensate
ing search optimization algorithm (BSA) considering several for reactive power. In [26], the author formulated the OPF
objective functions: emissions, voltage profile, voltage sta- problem for a hybrid thermal-wind-solar-storage system.
bility, and fuel costs. In [12], the MOOPF problem was for- A two-point estimating method and the GA method were
mulated as multi-objective problems by assigning weighting implemented to handle the proposed OPF strategy. In [27], the
factors for power loss, voltage stability, deviation of voltage, authors solved a security-constrained OPF with wind plant
and emissions. A differential search algorithm (DSA) was by using a fuzzy adaptive artificial physics algorithm. Roy
then proposed to deal with the MOOPF problem, and a moth and Jadhav [28] studied the OPF with wind generation by
swarm algorithm (MSA) was also applied, with fourteen way of an Improved artificial bee colony (ABC) method.
different cases based on the weighting factor approach [13]. In [29], a hybrid algorithm (HA) was applied for the OPF
Additionally, Biswas et al. [14] applied differential evolu- problem with thermal-wind generators; a modified BFA tech-
tion (DE) to handle the MOOPF problem with numerous case nique and mutation techniques of the genetic algorithm (GA)
studies considering different objectives. The authors dealt were combined to develop the HA method. A technique
with the operational constraints of the system by using a com- based on the DE was suggested in [30] for the OPF
bination of self-adaptive penalty and superiority of feasible problem considering solar and wind powers. Some recent
solution as constraint handling methods. A multi-objective applications of metaheuristic methods for OPF problem in

48380 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

presence of RES such as constrained multi-objective popula- The second one named crowding-distance calculation is to
tion extremal optimization (CMOPEO) [31], a combination preserve the diversity of non-dominated solutions in a spe-
of fitness–distance balance and adaptive guided differential cific front. The last one named Pareto archive selection is
evolution (FDBAGDE) [32], levy interior search algorithm used to store non-dominated solutions in an archive. The
(LISA) [33], grey wolf optimization (GWO) [34], barnacles archive is updated via a selection mechanism after each
mating optimizer (BMO) [35], differential evolutionary par- iteration to avoid erroneous rejection of potential candidate
ticle swarm optimization (DEEPSO) [36], and Levy coyote solutions during the optimization process. Further, a decision-
optimization algorithm (LCOA) [37]. making process is implemented to obtain the best compro-
A review of the literature has shown that the research on mise solution. The proposed MOSGA is implemented to find
OPF incorporating RESs is indeed encouraging. Although the optimal operation points of the thermal, wind, and solar
some attempts have been made to address the OPF with the generators with different case studies of objective functions.
incorporation of RES into power grids, previous research This study contributes to the literature as follows:
typically only investigated the problem with the single- • The MOOPF-WS problem was formulated as a con-
objective function of optimal generation cost [21]–[27] or strained multi-objective optimization problem consid-
the two-objective function of generation cost and emission ering the scheduling of thermal power generation and
[28]–[37]. In addition to economic and environmental factors, stochastic wind and solar power generation, whereas the
technical aspect also needs to be considered when examining references [38]–[43] considered either solar power or
the effect of RES on power grids. For this regard, genera- wind power as a renewable source. Three vital objec-
tion cost, real power loss, and emission should be optimized tive functions of total operational cost, power loss, and
simultaneously in the MOOPF problem. Further, from vari- emission were concurrently optimized, whereas only a
ous combinations of objective functions under different sce- single objective or two objectives were optimized in the
nario studies, the conflicts between objective functions can references [21]–[37]. The control variables of the prob-
be judged. To our knowledge, solving the MOOPF problem lem included most of the important parameters related
with the stochastic RESs using multi-objective algorithms to the operation of a power transmission network such
have not been documented in the literature. In addition, as the active power output of thermal generators, the
many research [38]–[43] considered only solar power or wind active power output of wind and solar generators, volt-
power as a renewable source. Very few research [30]–[36] age magnitudes at generation buses, settings of trans-
examined both stochastic wind and solar power when solving former tap, and reactive power output of shunt VAR
the MOOPF problem. Further, most of the previous research compensators. Most of the previous relevant research in
considered only the invariable load case, which did not reflect [30]–[36] examined the problem with a fewer number of
the characteristics of the actual load demand. Further study control variables that did not reflect the actual operating
is therefore needed to investigate the MOOPF problem in condition of the transmission network.
a system comprising thermal, wind, and solar power gen- • A new MOSGA was developed to handle the MOOPF-
eration with the variable loading condition; this constitutes WS problem. Three improvement techniques are inte-
significant motivation for proposing a new technique to deter- grated into the search mechanism of the original SGA
mine trade-off solutions for the MOOPF problem considering to convert it into a true multi-objective optimization
stochastic wind and solar power along with uncertain load algorithm. The development of MOSGA aims to pro-
scenario. vision Pareto optimal front and respective trade-offs
The search group algorithm (SGA) is a robust optimization for conflicting objectives. Meanwhile, previous sig-
method proposed in [44]. Notably, SGA has the advantage of nificant research [30]–[36] used a weighted factors
striking a good balance between exploitation and exploration, method for transforming from a multi-objective func-
providing powerful searchability for finding the optimum tion into a weighted sum function and applied a
solution. Several recent studies have been done to verify the single-objective optimization algorithm as the optimizer.
SGA applicability for various optimization problems such This workaround only offered a unique optimal solution
as truss optimal voltage regulation in power systems [45], after a run, which did not reflect the degree of con-
automatic generation control [46], networked control sys- flict between objectives as well as not provide multi-
tem [47], steel frames optimization [48], and structure opti- ple options in terms of various operating solutions for
mization [49]. Moreover, SGA has not been applied to deal decision-makers.
with the MOOPF problem with the stochastic RESs. • A practical case study relative to uncertainty in load
Inspired by the above motivations, this paper aims to sug- demand was examined to the MOOPF-WS problem
gest a novel multi-objective SGA (MOSGA) for MOOPF where a normal probability distribution function was
incorporating the stochastic wind and solar power (MOOPF- adopted to describe the variable situation of load
WS) problem. Three approaches are combined into the orig- demand. Notably, this case study has been investigated
inal SGA to create the proposed MOSGA. The first one for the first time in the MOOPF-WS problem.
named the fast non-dominated sorting technique is applied for • The effectiveness of MOSGA was tested on adapted
determining different non-dominated fronts in a population. 30-bus and 57-bus systems incorporating RES with the

VOLUME 10, 2022 48381

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

consideration of various scenarios. Specifically, in the The direct cost of a wind power plant can be expressed
two first scenarios of each network, the MOOPF-WS as [30]:
problem was investigated in the fixed load situation with
Cw,j (Pws,j ) = gj Pws,j (2)
two simultaneously optimized objectives, whereas the
third scenario was studied in the same load condition where gj denotes the direct cost coefficient and Pws,j the
for the simultaneous optimization of three objectives. scheduled power for the jth wind power plant.
For the last scenario of each network, the problem was As above mentioned, there might be another two costs
solved in the variable load condition while minimizing caused by overestimation and underestimation situations.
three objectives concurrently. Precisely, the overestimation situation will lead to the reserve
• The performance of MOSGA was comprehensively cost, whereas the underestimation one will lead to the penal
compared with three other multi-objective optimization cost. The wind power plant’s reserve cost can be determined
algorithms including NSGA-II, MOALO, and MOGOA as [30]:
for all scenarios. Comparisons were made in terms of
Spread metric, Hypervolume metric as well as the best CRw,j (Pws,j − Pwav,j ) = KRw,j (Pws,j − Pwav,j )
P
Zws,j
compromise solution. In all scenarios, statistical results
showed the MOSGA achieved better convergence and = KRw,j (Pws,j − pw,j )fw (pw,j )dpw,j
distribution of Pareto optimal solutions than other multi- 0
objective algorithms. Further, when comparing with pre- (3)
vious significant research [30]–[34], MOSGA acquired
better solution quality in all the comparable cases. where Pwav,j , KRw,j , and fw (pw,j ) represent the actual available
Section 2 introduces mathematical models for the power, reserve cost constant, and probability density function
generation costs of thermal, wind, and solar genera- for the jth wind power plant, respectively.
tors. Section 3 formulates the MOOPF-WS problem. The wind power plant’s penalty cost can be deter-
Section 4 outlines the proposed MOSGA and its implemen- mined as [30]:
tation for the MOOPF-WS problem. Section 5 details the CPw,j (Pwav,j − Pws,j ) = KPw,j (Pwav,j − Pws,j )
simulation outcomes, and Section 6 presents the conclusions. P
Zwr,j
= KPw,j (pw,j − Pws,j )fw (pw,j )dpw,j
II. MATHEMATICAL MODELS
A. COST OF THERMAL GENERATION Pws,j

The generation cost of thermal generators can be modeled as (4)

follows: where Pwr,j and KPw,j denote rated output power and
the penalty cost constant for the jth wind power plant,
NTG
X respectively.
CT (PTG ) = (ai + bi PTG,i + ci P2TG,i ) (1) The wind speed distribution is defined using the Weibull
i=1 probability density function (PDF) with two parameters
[25], [50]. Hence, the probability of wind speed (v) is
where ai , bi , and ci are the cost constants of the ith thermal expressed as:
generator that generates output power PTG,i , and NTG denotes   
k v k−1 −( v )k
the total number of thermal generators. fv (v) = e c ; for0 ≤ v ≤ ∞ (5)
c c
B. COST OF WIND GENERATION where k is the shape coefficient and c is the scale coefficient
Being a renewable energy source, wind power has inherent for the PDF.
uncertainty related to variable wind power. Due to insuffi- The mean of the Weibull distribution can be stated as:
(
ciency and unavailability, wind turbines cannot effectively Mwbl = c × 0(1 + k −1 )
provide schedule power as expected in some cases. As a
R∞ (6)
0(x) = 0 e−t t x−1 dt
spinning reserve control unit, the independent system oper-
ator (ISO) is in charge of handling the deficit amount. Such The wind turbine power output can be defined as:
a situation is referred to as the overestimation of power from
 v < vin and v > vout

 0 
renewable sources and should be included in the generation 
 v − vin
cost because of keeping the spinning reserve. In the contrast, pw (v) = pwr vin ≤ v ≤ vr (7)
 vr − vin
if available power produced by wind turbines is greater than 
vr < v ≤ vout

pwr
schedule power, the ISO should pay a penalty cost for the
surplus amount. This situation is known as underestimation. where pwr represents the rated output power of the wind
Therefore, the cost of wind power generation relates to three turbine; vin , vr , and vout respectively denote the cut-in speed,
components of direct cost, reserve cost, and penalty cost [30]. rated speed, and cut-out speed of wind turbine.

48382 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

According to Eq. (7), in a couple of zones of wind speeds, of a turbine in Eq. (7) and the mathematical equations of
the variable wind power is discrete. Specifically, the wind Weibull distribution, the available wind power probabilities
power is zero for wind speed v smaller than vin or for v greater corresponding to different wind speed zones can be defined
than vout . The wind power is equal to the rated value when v by Eqs. (8) – (10). Lastly, thanks to the probability functions
is between vr and vout . Probabilities for these discrete zones for available wind power estimation at different operating
can be calculated based on Weibull cumulative distribution zones, we can compute the cost of stochastic wind power
function (cdf) as follows [51]: using Eqs. (11) and (12).
Overall, the generation cost of wind generators can be
fw (pw ){pw = 0}
defined as:
= cdf (vin ) + (1 − cdf (vout )) N
X WG
       CWG = [Cw,j (Pws,j ) + CRw,j (Pws,j − Pwav,j )
vin k vout k
= 1 − exp − + exp − (8) j=1
c c
+ CPw,j (Pwav,j − Pws,j )] (13)
fw (pw ){pw = pwr }
where NWG represents the number of wind generators.
= cdf (vout ) + (1 − cdf (vr ))
       C. COST OF SOLAR GENERATION
vr k vout k
= exp − − exp − (9) Besides wind source, solar PV is another well-known renew-
c c able source. Generation of a solar PV plant also includes
For wind speed being between vin and vr , the wind power a summation of direct cost, reserve cost, and penalty cost
is a continuous variable. The continuous zone is related to the components. The direct cost of the solar PV plant can be
probability as follows: modeled as in Eq. (14) [30]:
 k−1
k(vr − vin ) pw Cs,k (Pss,k ) = hk Pss,k (14)
fw (pw ) = k vin + (vr − vin )
c × pwr pwr
 !k  where hk and Pss,k denote the direct cost constant and sched-
vin + ppwrw (vr − vin ) uled power for the k th solar PV plant.
× exp −  (10)
c The solar PV plant’ reserve cost can be obtained using the
below relation [30]:
Given the probabilities of available wind power at different
operating zones, Eqs. (3) and (4) can be expanded to include CRs,k (Pss,k − Psav,k )
the PDF of wind power as in Eqs. (11) and (12), as shown at = KRs,k (Pss,k − Psav,k )
the bottom of the page, respectively. = KRs,k × fs (Psav,k < Pss,k ) × [Pss,k − E(Psav,k < Pss,k )]
Remark 1: The process of handling the stochastic wind (15)
power can be summarized as follows: Firstly, the wind
speed distribution according to the Weibull probability den- where Psav,k and KRs,k are the actual available power and the
sity function (PDF) [41], [52]–[54] is established using reserve cost constant of the k th solar PV plant, respectively;
Eqs. (5) and (6). With the known parameters of Weibull fs (Psav,k < Pss,k ) denotes the occurrence probability of
shape (k) and scale (c), the Monte-Carlo method is adopted solar power shortage from the schedule power (Pss,k ); and
for generating the wind speed frequency distribution and E(Psav,k < Pss,k ) signifies the expectation of solar power
Weibull fitting. Next, based on the power output function being less than Pss,k .

CRw,j (Pws,j − Pwav,j )

P
Zws,j

k−1

pw !k 
vin + pwr (vr − vin )

(Pws,j − pw,j ) × k(v r − vin ) pw
= KRw,j k
vin + (vr − vin ) × exp − 
c × pwr pwr c
0
+ (Pws,j − 0) × fw (pw,j ){pw,j = 0} (11)
CPw,j (Pwav,j − Pws,j )
P
Zwr,j

k−1

pw !k 
vin + pwr (vr − vin )

(pw,j − Pws,j ) × k(vr − vin ) p w
= KPw,j v in + (vr − vin ) × exp − 
ck × pwr pwr c
Pws,j
+ (pwr − Pws,j ) × fw (pw,j ){pw,j = pwr } (12)

VOLUME 10, 2022 48383

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

The solar PV plant’ penalty cost is as follows [30]: PDF having characteristics as in Eqs. (17) and (18) is used
to express the distribution of solar irradiance. Via Monte
CPs,k (Psav,k − Pss,k )
Carlo simulation, we can obtain the lognormal fitting and
= KPs,k (Psav,k − Pss,k ) frequency distribution of solar irradiance. Next, we calculate
= KPs,k × fs (Psav,k > Pss,k ) × [E(Psav,k > Pss,k ) − Pss,k ] the solar irradiance to energy conversion for solar PV plant
(16) using Eq. (19). Given the distribution of actual available solar
power, the reserve and penalty costs related to stochastic solar
where KPs,k are the penalty cost constant relative to the power can be computed using Eqs. (20) and (21), respectively.
k th solar PV plant; fs (Psav,k > Pss,k ) denotes the probability Overall, the generation cost of solar generators is as
that solar power is more than the schedule power (Pss,k ); and follows:
E(Psav,k > Pss,k ) signifies the expectation of solar power
NSG
being more than Pss,k . X
CSG = [Cs,k (Pss,k ) + CRs,k (Pss,k − Psav,k )
Lognormal PDF is used to represent the solar irradiance
k=1
(Gs ) distribution. Hence, the probability of solar irradiance
+ CPs,k (Psav,k − Pss,k )] (22)
(Gs ) is defined by the following equation [55]:
−(ln x − µ)2 where NSG represents the number of solar PV generators.
 
1
fG (Gs ) = √ exp forGs > 0 (17)
Gs σ 2π 2σ 2
III. PROBLEM FORMULATION
where µ is the mean and σ is the standard deviation for the The MOOPF-WS problem is set out to optimize prede-
PDF. fined objectives while meeting several constraints by defin-
The mean of the lognormal PDF can be determined as ing the optimum values of control variables. Therefore, the
follows: MOOPF-WS problem may be expressed accordingly:
σ2
 
Minimize:
Mlgn = exp µ + (18)
2
F(u, x) = [F1 (u, x), F2 (u, x), . . . , Fm (u, x)] (23)
The solar irradiance (Gs ) to energy conversion for solar PV
can be determined as follows [56]: Subject to:
G2s
  
Psr for0 < Gs < Rc h(u, x) = 0 (24)


Ps (Gs ) =  Gstd Rc (19) g(u, x) ≤ 0 (25)
Gs
Psr forGs ≥ Rc


Gstd
where Fm (u,x) is the mth objective function; u and x are the
where Psr denotes the rated output power of solar PV plant, vectors of control and state variables, respectively; h(u,x)
Rc is the certain irradiance point, and Gstd denotes solar denotes the set of equality constraints and g(u,x) represents
irradiance in the standard environment. the inequality constraints.
The reserve cost in Eq. (15) is rewritten as follows [30]:
CRs (Pss − Psav ) = KRs (Pss − Psav ) A. PREDEFINED OBJECTIVES
N− 1) TOTAL COST MINIMIZATION
X
= KRs [Pss − Psn− ] × fsn− (20) The total cost includes generation costs for the thermal, wind,
n=1 and solar PV generators, which is defined as a summation of
Eqs. (1), (13), and (22):
where Psn− signifies that available power falls short of sched-
uled power, fsn− denotes the relative frequency of occurrence F1 = CT (PTG ) + CWG + CSG (26)
of Psn− , and N − signifies the number of pairs (Psn− , fsn− )
created for the PDF. 2) POWER LOSS MINIMIZATION
Similarly, the penalty cost in Eq. (16) is defined as [30]:
Real power loss can be expressed by the following equation:
CPs (Psav − Pss ) = KPs (Psav − Pss ) Nl
+
h i
N
X
X F2 = Ploss = Gq(ij) Vi2 + Vj2 − 2Vi Vj cos(δi − δj )
= KPs [Psn+ − Pss ] × fsn+ (21)
q=1
n=1
(27)
where Psn+ signifies that available power is greater than
scheduled power, fsn+ denotes the relative frequency of where Nl denotes the number of transmission lines, Gq(ij)
occurrence of Psn+ , and N + signifies the number of pairs denotes the transfer conductance between buses i and j; Vi and
(Psn+ , fsn+ ) created for the PDF. Vj represent the voltage magnitudes at buses i and j, respec-
Remark 2: Similarly, the process of handling the stochastic tively; δi and δj represent voltage angles at buses i and j,
solar power can be summarized as follows: The lognormal respectively.

48384 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

3) EMISSION MINIMIZATION in which Nb denotes the number of buses; PD,i and QD,i rep-
Producing electricity using conventional fossil fuels would resent real and reactive power demand at bus i, respectively;
release harmful emissions into the environment. The total Gij represents the transfer conductance and Bij the suscep-
emission caused by thermal generators is defined as follows: tance between bus i and bus j.
NTG h
X i 2) INEQUALITY CONSTRAINTS
F3 = (αi + βi Pgi + γi P2gi ) + ωi e(µi Pgi ) (28)
Generator constraints:
i=1

TG,i ≤ PTG,i ≤ PTG,i ; i = 1, . . . , NTG

Pmin max
where αi , βi , γi , ωi , and µi denote the emission constants of (33)
the ith thermal generator. Pmin
WG,j ≤ WG,j ; j = 1, . . . , NWG
PWG,j ≤ Pmax (34)
Pmin
SG,k ≤ PSG,k ≤ PSG,k ; k = 1, . . . , NSG
max
(35)
B. CONTROL VARIABLES
1) CONTROL VARIABLES
Qmin
TG,i ≤ TG,i ; i = 1, . . . , NTG
QTG,i ≤ Qmax (36)
The set of control variables used is as follows: Qmin
WG,j ≤ WG,j ; j = 1, . . . , NWG
QWG,j ≤ Qmax (37)
- PTG,i : real power output of the ith thermal generator, Qmin ≤ QSG,k ≤ QSG,k ; k = 1, . . . , NSG
max
(38)
SG,k
except for the slack bus. min
VG,i ≤ max
VG,i ≤ VG,i ; i = 1, . . . , Ng (39)
- PWG,i : real power output of the ith wind generator.
- PSG,i : real power output of the ith solar PV generator. Shunt VAR compensator constraints:
- VG,i : voltage magnitude at the ith generation bus.
C,i ≤ QC,i ≤ QC,i ; i = 1, . . . , Nc
Qmin max
- Ti : setting of the ith transformer tap. (40)
- QC,i : reactive power output of the ith shunt VAR Transformer constraints:
compensator.
The vector of control variables (u) is stated as: Timin ≤ Ti ≤ Timax ; i = 1, . . . , Nt (41)

u = [PTG,2 , . . . , PTG,NTG , PWG,1 , Security constraints:

. . . , PWG,NWG , PSG,1 , . . . , PSG,NSG , min
VL,i max
≤ VL,i ≤ VL,i ; i = 1, . . . , Nd (42)
VG,1 , . . . , VG,Ng , T1 , . . . , TNt , Q1 , . . . , QC,Nc ]
T
(29) SL,i ≤ SL,i ; i = 1, . . . , Nl
max
(43)

in which Ng denotes the total number of generation buses, IV. MULTI-OBJECTIVE SEARCH GROUP ALGORITHM
Nt denotes the number of transformers, and Nc denotes the The MOSGA is created by integrating fast non-dominated
number of switchable capacitors. sorting, crowding distance strategies, and Pareto archive
selection into the SGA. The original SGA, the new integration
2) STATE VARIABLES strategies, and the proposed MOSGA are detailed as below.
The set of state variables used is as follows:
- PG,1 : real power output at the slack generator. A. SGA
- QG : reactive power output of generation buses. SGA is a metaheuristic method developed in [44]. An intro-
- VL : voltage magnitudes at the load buses. duction of the basic SGA is described as follows. To start
- SL : power flow in the transmission lines. the optimization algorithm, a population P can be generated
Thus, the vector of state variables (x) is expressed as: randomly as the following equation:
x=[PG,1 , QG,1 , . . . , QG,Ng , V1 , . . . , VL,Nd , SL,1 , . . . , SL,Nl]T Pij = xj,min + (xj,max − xj,min )U [0, 1];
(30) i = 1, . . . , npop , j = 1, . . . , n (44)

in which Nd signifies the number of load buses. where npop is the population size and n denotes the number of
design variables; Pij denotes the jth control variable of the ith
C. SYSTEM CONSTRAINTS individual of P, U [0, 1] denotes a random number ranging
1) EQUALITY CONSTRAINTS from 0 to 1; xj,min and xj,max respectively denote the lower
and upper limits of the jth control variable.
Nb
X The objective function value for each individual is com-
 
PG,i − PD,i = Vi Vj Gij cos(δij ) + Bij sin(δij ) ; puted after population initialization. A number of good indi-
j=1 viduals are then extracted from population P to create a search
i = 1, . . . , Nb (31) group R using a standard tournament selection. Further infor-
Nb
X mation on the tournament selection is given in [57]. The
 
QG,i + QC,i − QD,i = Vi Vj Gij cos(δij ) − Bij sin(δij ) ; search group R is mutated at every iteration to improve global
j=1 exploration. Depending on the rank in search group R, the
i = 1, . . . , Nb (32) worst member is chosen for mutation. The aim is to explore

VOLUME 10, 2022 48385

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 1. Pseudocode of SGA.

TABLE 2. Pseudocode of MOSGA.

FIGURE 1. Crowding distance computation.

FIGURE 2. Pareto archive selection procedure.

family. To exploit the region of the best current individual,

new areas in the search domain. Therefore, new individuals
new search group is formed from the best ng individuals
are mutated as follows:
from all the families in the local phase. Algorithm 1 in
xj,mut = E[R:,j ] + tεσ [R:,j ]; j = 1, . . . , n, (45) Table 1 describes the SGA’s pseudocode.

in which xj,mut denotes the jth control variable of a specified B. FAST NON-DOMINATED SORTING TECHNIQUE
mutated individual, R:,j signifies the jth column of the search For each solution p of a set S, two entities are computed [58]:
group matrix, E denotes the mean value, t is the coefficient - Domination count np : the number of solutions that dom-
that determines the position of a newly created individual, inate the solution p.
ε denotes a random variable, and σ denotes the standard - Sp : the set of solutions dominated by solution p.
deviation.
The solutions with a zero domination count (np ) are con-
In SGA, a set comprising each search group member and
tained in the first non-dominated rank (P1 ). Afterwards, for
the respectively generated individuals is a family. Therefore,
each solution p with zero domination count, the process visits
after the search group is created, each member generates a
each solution q of set Sp and decreases the domination count
family by perturbation as follows:
of each solution q by one. If a member q has a zero domination
xj,new = Rij + αε; j = 1, . . . , n, (46) count, it is located in a different list Q. These members are
also placed in the second front (P2 ). This process continues
where α denotes the perturbation constant, which is reduced with each member of Q to determine the third rank (P3 ) and
by the iteration: all non-dominated ranks.
α k+1 = bα k (47)
C. CROWDING DISTANCE COMPUTATION
in which b means the coefficient, which is defined by a Crowding distance is computed to estimate the density
combination of the linear function. of solutions around a specific solution i in a specific
SGA comprises global and local phases. The principal aim non-dominated front. The population needs to be sorted
in the global phase is to explore the entire design domain by by the values of the objective function in ascending order.
creating a new search group from the best member of each Then, an infinite distance value for each objective function is

48386 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

assigned to the solutions with minimum and maximum objec- in the global phase. Meanwhile, the selection mechanism is
tive values, i.e., the boundary solutions. A distance value is adjusted in the local phase so that the best ng individuals from
assigned to all other intermediate solutions; this value is the the new Pareto archive is extracted to generate the new search
absolute normalized difference between the objective values group. The MOSGA process executes until the stopping con-
of two neighboring solutions (i+1 and i – 1). This computa- dition is met. The MOSGA’s pseudocode (Algorithm 2) is
tion is employed for other objectives. The value for crowding presented in Table 2.
distance is computed as the total of the individual distance
values for each objective. Before calculating the crowding F. IMPLEMENTATION OF THE MOSGA FOR THE
distance, each objective function is normalized. Fig. 1 shows MOOPF-WS PROBLEM
a schematic of the crowding distance computation. 1) POPULATION INITIALIZATION
The crowded-comparison operator (≺n ) is used based on To implement the MOSGA for the MOOPF-WS problem,
non-dominated level (r) and crowding distance (d) to define each individual in the population P represents a vector of
the better solution: control variables as follows:
i ≺n j if (ri < rj ) or ((ri = rj ) and (di > dj )) (48)
Pd = [PTG,1 , . . . , PTG,NTG , PWG,1 , . . . , PWG,NWG , PSG,1 ,
According to Eq. (48), the solution that belongs to the . . . , PSG,NSG , VG,1 , . . . , VG,NG ,
better non-dominated level is prioritized; if both solutions are
T1 , . . . , TNt , Q1 , . . . , QC,NC ]T
in the same level, the one with the better value for crowding
distance is prioritized. d = 1, . . . , npop (49)

D. PARETO ARCHIVE SELECTION 2) OBJECTIVE FUNCTION VALUE

The Pareto archive is utilized with the best non-dominated In the MOOPF-WS problem, each objective function value is
solutions to maintain a non-dominated set. In the MOSGA, calculated as follows:
all generated solutions are placed in the advanced archive. NG
X
Two archives are obtained (current and advanced), and these Fm0 = Fm + Kp (PG,1 − Qlim 2
G,1 ) + Kq (QG,i − Qlim
G,i )
2
are combined. The combined archive exceeds the fixed size i=1
(npop ), and so this archive is truncated using the selection Nd
X N
Xl
strategy suggested by Deb et al. [58] for the next step of the lim 2 max 2
+ Kv (VL,i −VL,i ) +Ks (SL,i − SL,i ) (50)
algorithm. Fig. 2 outlines this procedure. i=1 i=1
Fast non-dominated sorting is used first to separate the
combined archive into non-dominated ranks (P1 , P2 , . . . , Pn ). in which Fm denotes the mth objective for the MOOPF-WS;
The solutions in the first non-dominated front (P1 ) are chosen Kp , Kq , Kv , and Ks denote the penalty constants. In this study,
first for the new archive. If the first front (P1 ) is smaller than penalty constants are set to 106 .
the archive, the remainder of the archive is selected from The limitation values of the state variables are defined
other non-dominated fronts in rank order (P2 , . . . , Pn ). The according to Eq. (51):
process is implemented until the Pareto archive reaches its 
limit, assuming that front Fk is the final non-dominated front, xmax if x > xmax

after which no further sets can be added. Solutions from the x lim = xmin if x < xmin (51)

last front (Pk ) are chosen in descending order of the crowding x otherwise

distance to select the number of solutions for the new Pareto
archive accurately. where x lim and x denote the limited values and calculated
values of QG , VL , and SL , respectively.
E. THE PROPOSED MOSGA The power flow problem was addressed using the
The MOSGA begins the procedure by randomly generating Matpower 6.0 toolbox [59].
an initial population P. Objective values are calculated for
all individuals of P. Based on the non-dominated ranks of P, 3) BEST COMPROMISE SOLUTION
a tournament selection is later employed to choose ng best A decision-making method based on a fuzzy membership
individuals from P to form an initial search group R for the function is used to define the best compromise solution. The
subsequent two processes (mutation and generation of the membership function µij represents the degree of satisfaction
families). Afterwards, the advanced archive is yielded, and of the ith solution for the jth objective [60]:
the current and advanced archives are combined. The selec- 
tion strategy is implemented to choose npop best individuals 
 1 if Fij ≤ min(Fj )
 max(F ) − F
j ij
for the new Pareto archive. Finally, a crucial stage of the µij = if min (Fj ) ≤ Fij ≤ max(Fj )
MOSGA is to select a new search group. This process in both 
 max(Fj ) − min(Fj )

stages is carried out using a tournament selection. The new 0 if Fij ≥ max(Fj )
search group is generated by the best member of each family (52)

VOLUME 10, 2022 48387

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

Step 9: Create families (Fi ) for each search group member

using Eq. (46);
Step 10: Store all newly generated solutions into an
advanced archive. Combine the advanced and current
archives;
Step 11: Choose the best N solutions from the combined
archive as the new Pareto archive;
Step 12: Select the new search group in two stages:
- Global stage: select the best member of each family to
create search group Rk+1 ;
- Local stage: select ng best solutions from the archive to
create search group Rk+1 .
Step 13: Determine α k+1 according to Eq. (47);
Step 14: If k < Itermax , return to Step 7; or else, go to
Step 15;
Step 15: Solutions achieved: non-dominated solutions in
the archive set;
FIGURE 3. The modified IEEE 30-bus system incorporating RES.
Step 16: Define the best compromise solution based on the
defined approach;
where min(Fj ) and max(Fj ) are the minimum and maximum G. PERFORMANCE METRICS
values of the jth objective, respectively. The normalized mem-
1) SPREAD INDICATOR
bership function of the ith solution can be defined as:
The diversity of non-dominated solutions of a specified algo-
nP
obj rithm is provided by the spread (1) indicator, which shows
µij the spread of extreme solutions and the distribution of solu-
j=1
µi = npf nP
obj
(53) tions in the generated non-dominated set. This parameter is
P
µij defined in the following equation [61]:
i=1 j=1 m
d(Ei , ) + d(X , ) − d̄
P P
X ∈
where npf denotes the number of objective functions and nobj i=1
1= m (54)
denotes the number of non-dominated solutions, respectively.
d(Ei , ) + (|| − m) d̄
P
The best compromise solution is the one with a maximum i=1
value of the normalized membership function.
where
4) OVERALL PROCEDURE d(X , ) = min kF(X ) − F(Y )k (55)
The MOSGA implementation for the MOOPF-WS is as Y ∈,Y 6 =X

follows: 1 X
d̄ = d(X , ) (56)
Step 1: Define detailed data of the test system and accept- || X ∈
able ranges of control variables; where  is the generated Pareto optimal front, Ei is the
Step 2: Set the controlling parameters for MOSGA, includ- ith extreme solutions in true Pareto front, and m represents
ing population size (npop ), number of search group members the number of objective functions. The smaller value of the
(ng ), number of mutations (nmut ), perturbation factor (α), 1 indicator provides better diversity (i.e., the better extent of
Pareto archive size (N ), maximum number of iterations spread and distribution) in a non-dominated set.
(Itermax ), and global iteration ratio (GIR);
Step 3: Create the initial population P as in Eq. (44); 2) HYPERVOLUME
Step 4: Execute a power flow calculation using Matpower This metric computes the volume related to members of
6.0 to evaluate objective function values (F1 , F2 , F3 ) for each a non-dominated set in the objective space. The hypervol-
individual of P; ume (HV) can be computed [62]:
Step 5: Sort all the individuals in the initial population P
||
using fast non-dominated sorting and crowding distance com- [
putations and place them into a Pareto archive; HV = vi (57)
Step 6: Choose the best ng solutions from P to generate the i=1
initial search group Rk . Set k = 0; where vi is a hypercube for each solution i ∈  formed with
Step 7: Begin the main loop, k = k+ 1; a reference point W . The HV values after normalization lie
Step 8: Perform the mutation phase for nmut individuals within the range [0,1]. It is desirable that the algorithm has a
using Eq. (45); large HV value.

48388 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 3. Different scenarios of the MOOPF-WS problem.

TABLE 4. Summary of the IEEE 30-bus system incorporating RESs. TABLE 6. PDF parameters and cost constants for wind generators for
modified IEEE 30-bus system [30].

TABLE 7. PDF parameters and cost constants for solar PV generator for
modified IEEE 30-bus system [30].
TABLE 5. Cost constant and emission constant of thermal generators for
modified IEEE 30-bus system [2].

respectively. NSGA-II used a mutation probability of 0.5,

a crossover probability of 0.9, and distribution indexes for
mutation and crossover operators of ηm = 20 and ηc = 20,
V. SIMULATION RESULTS respectively.
The MOSGA was implemented to deal with the MOOPF-WS
problem on the adapted IEEE 30-bus and 57-bus systems A. IEEE 30-BUS SYSTEM
incorporating RES. Table 3 summarizes the scenarios consid- Table 4 summarizes the modified IEEE 30-bus system incor-
ered in this study. The initial parameters of MOSGA, includ- porating the RES used in this study; its topology diagram is
ing npop , ng , nmut , α k , and GIR, were chosen as 100, 20, 5, displayed in Fig. 3. Of note, locations of RESs were chosen
2, and 0.3, respectively, for all test systems. The optimization similar to the references [30] to make a fair results com-
process of the MOSGA was independently run thirty times parison. In this context, some conventional generators were
for each scenario. The number of function evaluations (NFEs) replaced by respective renewable generators. Table 5 details
was set as 30,000 and 50,000 for the 30-bus and 57-bus the fuel cost and emission factors for thermal generators.
systems, respectively. The Pareto archive size was kept at Tables 6-7 give PDF parameters and cost constants for wind
100 for all trials. and solar PV generators. After performing 8,000 Monte-
Moreover, MOSGA was compared with multi-objective Carlo scenarios, Figs. 4-5 depict wind frequency distributions
ant lion optimizer (MOALO) [64], multi-objective grasshop- for two wind farms. Similarly, Fig. 6 displays the frequency
per optimization algorithm (MOGOA) [65], and non- distribution of solar irradiance for a solar PV plant. Stochastic
dominated sorting genetic algorithm II (NSGA-II) [58]. power output from the solar PV plant is presented in Fig.
The population size for all algorithms was fixed at 100. 7. For the initial scenario, the system has a total cost of
In MOGOA, the maximum and minimum values of decreas- 834.6627 $/h, a real power loss of 5.7866 MW, and an emis-
ing coefficient (cmax and cmin ) were set to 1 and 0.00004, sion of 0.2694 ton/h.

VOLUME 10, 2022 48389

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 8. Simulation results obtained by multi-objective algorithms for scenarios 1 and 2.

FIGURE 4. Wind speed distribution for wind generators #1 at bus 5. FIGURE 5. Wind speed distribution for wind generators #2 at bus 11.

1) SCENARIO 1: SIMULTANEOUS OPTIMIZATION OF TOTAL were 4.2514% and 43.0223%, respectively, in comparison
COST AND ACTIVE POWER LOSS with the initial scenario. Moreover, it showed that the best
The MOSGA was executed to optimize the total cost and real compromise solution obtained by MOSGA (799.1777 $/h
power loss concurrently in scenario 1. Fig. 8 portrays the best and 3.2970 MW) was better than those obtained by NSGA-II
Pareto optimal front associated with the best HV metric for (801.4558 $/h and 3.3097 MW), MOALO (800.1741 $/h and
scenario 1, which shows a trade-off between the two objec- 3.6280 MW) and MOGOA (800.4808 $/h and 3.3484 MW).
tives considered. Table 8 reports the simulation results found Therefore, MOSGA is capable of finding a better optimal
by MOSGA, NSGA-II, MOALO, and MOGOA, including solution for this case study.
optimal values of design variables and objective functions for
the best compromise solution. Table 8 also indicates that the 2) SCENARIO 2: SIMULTANEOUS OPTIMIZATION OF TOTAL
MOSGA offered a total cost of 799.1777 $/h and a real power COST AND EMISSION
loss of 3.2970 MW for the best compromise solution. Thus, In this scenario, the MOSGA minimized the total cost and
total cost and real power loss reductions for this scenario emission concurrently. Fig. 9 demonstrates the Pareto front

48390 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

FIGURE 6. Solar irradiance distribution for solar PV at bus 13.

FIGURE 9. Pareto front obtained by MOSGA for scenario 2.

FIGURE 7. Available solar PV power (MW) at bus 13.

FIGURE 10. Pareto front obtained by MOSGA for scenario 3.

FIGURE 8. Pareto front obtained by MOSGA for scenario 1.

FIGURE 11. Load bus voltage profiles for scenarios 1, 2, and 3.

obtained by MOSGA for this scenario. As stated above,

the best Pareto optimal front is the one that corresponds to MOSGA acquired the best compromise solution with lower
the best value for the HV metric. The curve in Fig. 9 shows the total cost and higher emission. For a comprehensive compar-
conflicting nature of total cost and emission due to the equa- ison, MOSGA not only found a better result than MOALO
tion characteristics of objective functions. Table 8 displays and MOGOA in both objectives, but it also obtained the best
simulation output obtained from the MOSGA, NSGA-II, value for the total cost objective among other algorithms for
MOALO, and MOGOA for the best compromise solution. best compromise solution. Hence, MOSGA can provide a
The total cost and emission for the best compromise solu- good-quality best compromise solution for scenario 2.
tion yielded by MOSGA were 784.7058 $/h and 0.2788 To further investigate its performance, MOSGA was
ton/h, respectively. These correspond to a reduction in total also compared with published results in the literature
cost of 5.9853% and an increase in emissions of 3.4754%, for scenario 2. Table 9 shows comparative results of
as compared to the initial scenario. Compared with NSGA-II, MOSGA and other prominent algorithms in the previous

VOLUME 10, 2022 48391

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 10. Simulation results obtained by multi-objective algorithms for

scenario 3.

FIGURE 12. Representation of uncertainty in load by a normal

distribution with µl = 70 and σl = 10.

TABLE 9. Comparative results of MOSGA and previous studies for

scenario 2.

TABLE 11. Scenarios of loading and corresponding probabilities.

studies, including success history based adaptive differ-

ential evolutionary with superiority of feasible solutions
(SHADE-SF) [30], CMOPEO [31], constrained NSGA-II
(CNSGA-II) [31], FDBAGDE [32], ISA [33], LISA strategy
I [33], LISA strategy II [33], GA [34], PSO [34], crow search
algorithms (CSA) [34], ABC [34], and GWO [34]. As for
scenario 2, MOSGA yielded the best minimum value of total
cost and emission objectives in best compromise solution
among published results. Hence, MOSGA can provide a 3.2319%, a real power loss reduction of 47.3919%, and an
good-quality best compromise solution for scenario 2. emission reduction of 50.1767%, respectively, in comparison
with the initial scenario. It can be seen that MOSGA domi-
3) SCENARIO 3: SIMULTANEOUS OPTIMIZATION OF TOTAL nated MOGOA in all three objectives in the best compromise
COST, ACTIVE POWER LOSS, AND EMISSION solution. Moreover, MOSGA outperformed NSGA-II and
The total cost, real power loss, and emission were minimized MOALO in two out of three objectives. Therefore, MOSGA
concurrently by way of the MOSGA in this scenario. Fig. 10 had great potential in finding a better compromise solution.
portrays the Pareto optimal front found by the MOSGA, According to the results obtained from this scenario, all
indicating the relations between the three objectives. The best three objective functions indicate a remarkable improvement
compromise solution was also extracted through the decision- using the MOSGA. Precisely, all the objectives were met
making method. Table 10 represents simulation outcomes simultaneously to the maximum extent, an optimal result for
for the best compromise solutions yielded by MOSGA, the best compromise solution. Moreover, an essential con-
NSGA-II, MOALO, and MOGOA. Table 10 also shows straint in the MOOPF-WS problem is that load bus voltage
that the best compromise solution provided a total cost of must be maintained within a specific range; Fig. 11 shows
807.6874 $/h, a power loss of 3.0442 MW, and emission voltage profiles of load buses for scenarios 1-3, and it can be
of 0.1342 ton/h, corresponding to a total cost reduction of seen that the voltages are all within specified boundaries.

48392 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 12. Simulation results obtained by proposed MOSGA algorithm for

loading scenarios 4.1-4.4.

FIGURE 14. Pareto front obtained by MOSGA for scenario 4.2.

FIGURE 15. Pareto front obtained by MOSGA for scenario 4.3.

FIGURE 16. Pareto front obtained by MOSGA for scenario 4.4.

FIGURE 13. Pareto front obtained by MOSGA for scenario 4.1.

The color-filled areas manifest four different scenarios of net-

4) SCENARIO 4: SIMULTANEOUS OPTIMIZATION OF TOTAL
work load demand (Pl ) considered. For a certain scenario, the
COST, ACTIVE POWER LOSS, AND EMISSION (UNCERTAIN
mean loading and its occurrence probability can be estimated
LOAD DEMAND CASE STUDY)
by the following equations [66].
This case study considers a practical variable situation of
Probability of the ith loading scenario is calculated by:
load demand. Normal PDF is utilized to reflect the uncer-
up
tain nature of load [66]. Next, we apply the scenario-based
ZPl,i " #
method to implement the process of optimization at some 1 (Pl − µl )
δsce,i = √ exp − dPl (58)
discrete load levels. Fig. 12 portrays the load demand uncer- σl 2π 2σl2
tainty in relation to the graphic form of normal distribution, Plo
l,i
in which the horizontal axis deputizes the percentage of
up
network loading. The PDF parameters include a mean (µl ) in which Plo l,i and Pl,i are the lower and upper bounds
and a standard deviation (σl ) of 70 and 10, respectively. of the ith loading scenario. Next, mean of the ith loading

VOLUME 10, 2022 48393

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 13. Comparison of network performance profiles optimized by different algorithms for scenarios 4.1-4.4.

FIGURE 17. Load bus voltage profiles for scenarios 4.1-4.4.

TABLE 14. Summary of the IEEE 57-bus system incorporating RESs.

FIGURE 18. The modified IEEE 57-bus system incorporating RES.

the percent loading, given the nominal loading of the network

of 100%. At each scenario, the proposed MOSGA algo-
rithm optimizes the multi-objective function as formulated
in Eq. (50). Hence, the control variables of scheduled power
from all generators, voltage magnitudes at generation nodes,
transformer tap settings, and VAR compensators’ output are
scenario is computed: optimized in each scenario of loading. Of note, the case study
up with the implementation of OPF at certain time intervals
ZPl,i " #
for various objectives optimization is similar to the practical
1 1 (Pl − µl )
Pmean = √ exp − dPl (59) operation condition of an electrical network.
l,i
δsce,i σl 2π 2σl2
Plo
Regarding the boundaries of control variables, they were
l,i
set the same as in the fixed loading case. Simulation results
Table 11 provides information on the estimated means and related to various scenarios of loading of the 30-bus net-
probabilities for all loading scenarios. For a specific scenario, work are tabulated in Table 12 using MOSGA optimizer.
the demands in all nodes are adjusted by multiplying them by Figs. 13-16 depict the Pareto optimal fronts found by the

48394 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 15. Cost constant and emission constant of thermal generators for
modified IEEE 57-bus system [2].

TABLE 16. PDF parameters and cost constants for wind generators for
modified IEEE 57-bus system [36]. FIGURE 19. Wind speed distribution for wind generators #1 at bus 6.

TABLE 17. PDF parameters and cost constants for solar PV generator for
modified IEEE 57-bus system [36].

FIGURE 20. Wind speed distribution for wind generators #2 at bus 9.

power generators. Specifically, the optimal schedule power

of generators increased when the load demand of network
MOSGA for scenarios 4.1-4.4, respectively. As for each increased. To satisfy the increasing load demand, solar PV
loading scenario, the optimized values of the total cost and wind generators were preferred to be scheduled. This
(TCsce ), active power loss (PLsce ), and emission (Esce ) are may contribute to the notable improvement level of network
also reported in Table 12. Afterwards, the expected values performance profiles in examined discrete loading scenarios
of the total cost (ETC), active power loss (EPL), and emis- as well as of expected performance profiles over all scenarios.
sion (EE) over all scenarios can be computed according to Compared with the fixed loading case, the uncertain loading
Eqs. (60)-(62) [67]: case proposed a more flexible and effective scheduling of
nsce
X generators that led to lower total cost and power loss, and
ETC = δsce × TCsce (60) emission. Besides MOSGA, three multi-objective optimiza-
sce=1 tion algorithms including NSGA-II, MOALO, and MOGOA
nsce
X were also applied for results comparison in this case study.
EPL = δsce × PLsce (61) A comparison of results optimized by the four methods
sce=1 is summarized in Table 13. Clearly, MOSGA dominated
nsce
X MOALO and MOGOA in two out of three objectives in
EE = δsce × Esce (62) the best compromise solutions for all considered discrete
sce=1 loading scenarios. These helped MOSGA achieve lower three
in which, nsce and δsce are the total numbers of load scenarios expected values (i.e., total cost, power loss and emission)
and the likelihood of a scenario, respectively (Table 11). over all scenarios than those of MOALO, whereas MOSGA
From the simulated outcomes in Tables 10 and 12, the achieved lower expected values of total cost and emission
total operational cost, power loss, and emission achieved over all scenarios than those of MOGOA. Compared with
by the uncertain loading case were justifiably lower than NSGA-II, MOSGA obtained the best compromise solution
those achieved by the fixed loading case. The change in with the better two out of three objectives for the second and
network loading resulted in a significant change in sched- third discrete loading scenarios while it excelled at only one
ule power of generators, especially in solar PV and wind objective for the first and fourth loading scenarios. Notably,

VOLUME 10, 2022 48395

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 18. Simulation results obtained by multi-objective algorithms for scenarios 5 and 6.

FIGURE 21. Solar irradiance distribution for solar PV at bus 2. FIGURE 22. Available solar PV power (MW) at bus 2.

when considering the overall performance across all loading voltage profiles for all loading scenarios while ensuring the
scenarios, MOSGA was superior in two out of three expected permissible limits of bus voltage.
performance profiles over NSGA-II. Further, the voltage pro-
files at buses of the 30-bus network for all scenarios after the B. IEEE 57-BUS SYSTEM
optimization by MOSGA are plotted in Fig. 17. The solutions Table 14 describes the 57-bus system incorporating RES; its
obtained by MOSGA led to a significant improvement in bus topology diagram is shown in Fig. 18. Notably, locations of

48396 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 19. Simulation results obtained by multi-objective algorithms for TABLE 20. Simulation results obtained by proposed MOSGA algorithm for
scenario 7. loading scenarios 8.1-8.4.

RESs were selected the same as the references [36] by replac-

ing some conventional generators with respective renewable
generators. Table 15 details the fuel cost and emission factors
of thermal generators. Tables 16-17 provide the PDF param-
eters and cost constants of wind and solar PV generators. cost and real power loss. Table 18 shows the optimization
Figs. 19-20 depict wind frequency distributions for two wind values of the best compromise solution from the MOSGA,
farms after Monte-Carlo simulations with 8000 scenarios NSGA-II, MOALO, and MOGOA. The best compromise
obtained for each. Fig. 21 displays the frequency distribution solution from MOSGA provided a total cost of 30212.29 $/h
of solar irradiance for a solar PV plant. Stochastic power (i.e., 41.6705 % total cost reduction), along with a real
output from the solar PV plant is depicted in Fig. 22. For the power loss of 11.8540 MW (i.e., 56.9926 % real power loss
initial scenario, the system has a total cost of 51795.8561 $/h, reduction) compared to their initial scenario values. From
a real power loss of 27.5627 MW, and an emission Table 18, MOSGA (30212.29 $/h and 11.8540 MW) obtained
of 2.6474 ton/h. lower total cost and real power loss values than MOALO
(30944.81 $/h and 12.3851 MW) and MOGOA (30359.67 $/h
1) SCENARIO 5: SIMULTANEOUS OPTIMIZATION OF TOTAL and 11.9031 MW), which indicated that MOSGA domi-
COST AND ACTIVE POWER LOSS nated those of MOALO and MOGOA for best compro-
In this scenario, the MOSGA method simultaneously mise solution. In addition, MOSGA outperformed NSGA-II
optimized total cost and real power loss objectives. in one out of two objectives optimized. Hence, MOSGA
Fig. 23 presents the Pareto optimal front obtained by the was able to highly compete with MOALO, MOGOA,
MOSGA for scenario 5. Similar to scenario 1, scenario and NSGA-II in finding the high-quality best compromise
5 also shows the conflicting relationship between the total solution.

VOLUME 10, 2022 48397

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 21. Comparison of network performance profiles optimized by different algorithms for scenarios 8.1-8.4.

FIGURE 25. Pareto front obtained by MOSGA for scenario 7.

FIGURE 23. Pareto front obtained by MOSGA for scenario 5.

FIGURE 24. Pareto front obtained by MOSGA for scenario 6. FIGURE 26. Load bus voltage profiles for scenarios 5, 6, and 7.

2) SCENARIO 6: SIMULTANEOUS OPTIMIZATION OF TOTAL to 40.9697% and 62.0652% reductions. Therefore, both
COST AND EMISSION objectives were significantly improved upon compared to the
The total cost and emission were simultaneously optimized initial scenario. In this case, MOSGA obtained better results
using the MOSGA in this scenario. The Pareto optimal front than MOALO and MOGOA for the total cost and emission
yielded by MOSGA is described in Fig. 24. Table 18 shows objectives. Meanwhile, the optimal solution of MOSGA was
the simulation outputs from four algorithms for scenario 6. better than NSGA-II in the objective of total cost only. Hence,
The total cost and emission were reduced to 30575.27 $/h and MOSGA had great potential in finding better compromise
1.0043 ton/h for the best compromise solution, corresponding solution.

48398 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

FIGURE 27. Pareto front obtained by MOSGA for scenario 8.1. FIGURE 30. Pareto front obtained by MOSGA for scenario 8.4.

FIGURE 28. Pareto front obtained by MOSGA for scenario 8.2.

FIGURE 31. Load bus voltage profiles for scenarios 8.1-8.4.

was not able to directly dominate NSGA-II, MOALO, and

MOGOA, MOSGA obtained better total cost and real power
loss values than the others in the best compromise solution.
Hence, MOSGA was more advantageous in handling the
MOOPF-WS problem with the large-scale system.
It can be seen that the implementation of the MOSGA for
the MOOPF-WS problem offers a notable improvement for
all three objective function values. Fig. 26 depicts the voltage
profiles of load buses for scenarios 5-7, which shows that the
voltage limits were all within acceptable limitations.
FIGURE 29. Pareto front obtained by MOSGA for scenario 8.3.
4) SCENARIO 8: SIMULTANEOUS OPTIMIZATION OF TOTAL
COST, ACTIVE POWER LOSS, AND EMISSION (UNCERTAIN
3) SCENARIO 7: SIMULTANEOUS OPTIMIZATION OF TOTAL LOAD DEMAND CASE STUDY)
COST, ACTIVE POWER LOSS, AND EMISSION Similarly, experiment outcomes for the uncertain load case
In scenario 7, the MOSGA simultaneously minimized the of the 57-bus network obtained by MOSGA are presented
total cost, real power loss, and emission. Fig. 25 portrays in Table 20. Figs. 27-30 show the Pareto optimal fronts
the Pareto optimal front found by MOSGA. Table 19 gives found by the MOSGA for scenarios 8.1-8.4, respectively.
the simulation results generated by the MOSGA, NSGA-II, Comparing the results between Table 19 and Table 20 shows
MOALO, and MOGOA for the best compromise solution in the uncertain case of load demand offered justifiably lower
this scenario. For MOSGA, the total cost, real power loss, network performance profiles of total cost, power loss, and
and emission were reduced to 30663.58 $/h, 11.8439 MW, emission than those of the fixed loading case. Again, the
and 1.0846 ton/h, respectively. Hence, MOSGA found a total scheduling of generators from the uncertain load case was
cost reduction of 40.7992% and a power loss reduction of more suitable with the variable load demand. Solar PV and
57.0292%, along with a reduction of 59.0327% in emis- wind power generators were preferable to schedule due to
sion for the best compromise solution. Although MOSGA their positive impacts in reducing network operation cost,

VOLUME 10, 2022 48399

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

TABLE 22. Comparisons of multi-objective methods using 1 indicator for scenarios 1-8.

TABLE 23. Comparisons of multi-objective methods using HV indicator for scenarios 1-8.

FIGURE 32. Box plots of 1 metric of multi-objective algorithms for scenarios 1-8.

power loss, and emission. To confirm the effectiveness of NSGA-II, MOALO, and MOGOA (Table 21). Obviously,
MOSGA, the obtained best compromise solutions related MOSGA showed its domination in two of three objectives
to four scenarios of loading were compared with those of when compared with NSGA-II, MOALO, and MOGOA for

48400 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

FIGURE 33. Box plots of HV metric of multi-objective algorithms for scenarios 1-8.

all discrete loading scenarios. As a result, MOSGA acquired TABLE 24. Comparisons of multi-objective methods in terms of
computational time in seconds.
lower three expected values of the total cost, power loss,
and emission over all scenarios than MOALO. In comparison
with NSGA-II and MOGOA, MOSGA obtained lower two
expected values of total cost and emission over all scenarios.
Moreover, for all scenarios, bus voltage profiles of the 57-bus
network were significantly enhanced and completely satisfied
with the limits after the optimization by MOSGA (Fig. 31).

C. STATISTICAL COMPARISONS AND ANALYSIS

This section demonstrates the comparisons among MOSGA,
NSGA-II, MOALO, and MOGOA based on 1 and HV met-
rics. The initial parameters of all algorithms were retained at
the same values as in the preceding section.

1) SPREAD (1) METRIC

Table 22 shows statistical results attained by the MOSGA,
NSGA-II, MOALO, and MOGOA for the 1 indicator for Therefore, MOSGA found a better quality Pareto optimal
scenarios 1-8. Boxplots of the statistical analysis for 1 metric front in comparison to NSGA-II, MOALO, and MOGOA.
are depicted in Fig. 32. As can be seen, the MOSGA provided As seen in Figs. 33, MOSGA was also superior to the other
the best solutions for the 1 metric, with the best distribution methods in terms of stability and robustness.
for Pareto optimal fronts among the four methods for all
scenarios. Hence, the proposed MOSGA showed superiority 3) COMPUTATIONAL TIME
over the others for scenarios 1-8. In this study, MOSGA, NSGA-II, MOALO and MOGOA
were executed on a 2.6 GHz 4-core computer with 16 GB of
2) HYPERVOLUME (HV) METRIC RAM. Table 24 presents the average computational times of
Table 23 gives the optimization results of four methods for the MOSGA, NSGA-II, MOALO and MOGOA for all scenarios.
HV metric for scenarios 1-8. Figs. 33 shows the boxplots of The computational times of MOSGA were significantly faster
the HV metric for scenarios 1-8. From Table 23 and Fig. 33, than those of NSGA-II, MOALO, and MOGOA for all case
MOSGA was better than the other methods for the average studies. Hence, MOSGA outperformed the other methods for
value of the HV metric, which indicated that solutions yielded the solution quality, robustness, and computational times in
by the MOSGA had the best convergence and diversity. all cases.

VOLUME 10, 2022 48401

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

VI. CONCLUSION [3] R. C. Burchett, H. H. Happ, D. R. Vierath, and K. A. Wirgau,

This study suggested a new MOSGA for the MOOPF-WS ‘‘Developments in optimal power flow,’’ IEEE Trans. Power App.
Syst., vol. PAS-101, no. 2, pp. 406–414, Feb. 1982, doi: 10.1109/TPAS.
problem, in which thermal, wind and solar PV generators 1982.317121.
were integrated into the grid. Different PDFs were used to [4] K. Zehar and S. Sayah, ‘‘Optimal power flow with environmental con-
model uncertainties relative to RES and load demand as straint using a fast successive linear programming algorithm: Application
to the Algerian power system,’’ Energy Convers. Manage., vol. 49, no. 11,
well. The MOSGA integrated fast non-dominated sorting and pp. 3362–3366, Nov. 2008, doi: 10.1016/j.enconman.2007.10.033.
crowding distance methods in defining the non-dominated [5] G. P. Granelli and M. Montagna, ‘‘Security-constrained economic dispatch
ranks and densities of the solutions obtained. Furthermore, using dual quadratic programming,’’ Electr. Power Syst. Res., vol. 56, no. 1,
pp. 71–80, 2000, doi: 10.1016/S0378-7796(00)00097-3.
the Pareto archive selection strategy was applied for the [6] O. Crisan and M. A. Mohtadi, ‘‘Efficient identification of binding inequal-
distribution maintenance of the non-dominated solutions. ity constraints in the optimal power flow Newton approach,’’ IEE Proc. C
The MOSGA was implemented successfully to the adapted Gener., Transmiss. Distrib., vol. 139, no. 5, pp. 365–370, Sep. 1992, doi:
10.1049/ip-c.1992.0053.
30-bus and 57-bus systems incorporating RES where various [7] J. A. Momoh and J. Z. Zhu, ‘‘Improved interior point method for OPF prob-
combinations of objective functions and different loading sce- lems,’’ IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1114–1120, Aug. 1999,
narios were examined. The findings showed that system per- doi: 10.1109/59.780938.
[8] S. Frank, I. Steponavice, and S. Rebennack, ‘‘Optimal power flow: A
formance is significantly improved for the best compromise bibliographic survey I,’’ Energy Syst., vol. 3, no. 3, pp. 221–258, Sep. 2012,
solution in both fixed and variable loading scenarios. Notably, doi: 10.1007/s12667-012-0056-y.
in scenario 7, the total cost, real power loss, and emis- [9] W. Warid, H. Hizam, N. Mariun, and N. I. A. Wahab, ‘‘A novel quasi-
oppositional modified Jaya algorithm for multi-objective optimal power
sion were decreased by 40.7992%, 57.029%, and 59.0327%, flow solution,’’ Appl. Soft Comput., vol. 65, pp. 360–373, Apr. 2018, doi:
respectively, in comparison with the initial scenario. For the 10.1016/j.asoc.2018.01.039.
variable loading case, the expected performance profiles over [10] H. R. E. H. Bouchekara, A. E. Chaib, M. A. Abido, and R. A. El-Sehiemy,
‘‘Optimal power flow using an improved colliding bodies optimization
all discrete loading scenarios were significantly lower than algorithm,’’ Appl. Soft Comput., vol. 42, pp. 119–131, May 2016, doi:
the respective ones found by the fixed loading case. This 10.1016/j.asoc.2016.01.041.
proved that solving OPF in the variable load situation offered [11] A. E. Chaib, H. R. E. H. Bouchekara, R. Mehasni, and M. A. Abido,
‘‘Optimal power flow with emission and non-smooth cost functions using
more flexible and effective scheduling of generators than in backtracking search optimization algorithm,’’ Int. J. Elect. Power Energy
the fixed load situation, resulting in lower expected perfor- Syst., vol. 81, pp. 64–77, Oct. 2016, doi: 10.1016/j.ijepes.2016.02.004.
mance profiles. Moreover, the performance of the MOSGA [12] K. Abaci and V. Yamacli, ‘‘Differential search algorithm for solving multi-
objective optimal power flow problem,’’ Int. J. Electr. Power Energy Syst.,
was compared with NSGA-II, MOALO, and MOGOA using vol. 79, pp. 1–10, Jul. 2016, doi: 10.1016/j.ijepes.2015.12.021.
1 and HV indicators. In all cases, the MOSGA showed [13] A.-A. A. Mohamed, Y. S. Mohamed, A. A. M. El-Gaafary, and
superiority in convergence and diversity of Pareto optimum A. M. Hemeida, ‘‘Optimal power flow using moth swarm algorithm,’’
Electr. Power Syst. Res., vol. 142, pp. 190–206, Jan. 2017, doi: 10.1016/j.
solutions compared with the other three methods. As a result, epsr.2016.09.025.
the MOSGA could detect a broader range of non-dominated [14] P. P. Biswas, P. N. Suganthan, R. Mallipeddi, and G. A. J. Amaratunga,
solutions for all objective functions. Furthermore, when com- ‘‘Optimal power flow solutions using differential evolution algo-
rithm integrated with effective constraint handling techniques,’’ Eng.
paring with significant research in the literature, MOSGA Appl. Artif. Intell., vol. 68, pp. 81–100, Feb. 2018, doi: 10.1016/
obtained better solution quality in all the comparable cases. j.engappai.2017.10.019.
These revealed the strengths of the MOSGA and validated its [15] A. M. Shaheen, S. M. Farrag, and R. A. El-Sehiemy, ‘‘MOPF solution
methodology,’’ IET Gener., Transmiss. Distrib., vol. 11, no. 2, pp. 570–581,
potential in dealing with the MOOPF-WS problem. In future 2016, doi: 10.1049/iet-gtd.2016.1379.
studies, the MOOPF-WS problem can be expanded in some [16] H. Pulluri, R. Naresh, and V. Sharma, ‘‘An enhanced self-adaptive dif-
specific aspects such as connection consideration of small- ferential evolution based solution methodology for multiobjective optimal
power flow,’’ Appl. Soft Comput., vol. 54, pp. 229–245, May 2017, doi:
hydro source, electric vehicles, and flexible AC transmission
10.1016/j.asoc.2017.01.030.
system (FACTS) devices. [17] X. Yuan, B. Zhang, P. Wang, J. Liang, Y. Yuan, Y. Huang, and X. Lei,
‘‘Multi-objective optimal power flow based on improved strength Pareto
ACKNOWLEDGMENT evolutionary algorithm,’’ Energy, vol. 122, pp. 70–82, Mar. 2017, doi:
10.1016/j.energy.2017.01.071.
The authors acknowledge the support of the Min- [18] G. Chen, X. Yi, Z. Zhang, and H. Wang, ‘‘Applications of multi-objective
istry of Higher Education (MOHE), Malaysia to pro- dimension-based firefly algorithm to optimize the power losses, emission,
vide financial assistance under Fundamental Research and cost in power systems,’’ Appl. Soft Comput., vol. 68, pp. 322–342,
Jul. 2018, doi: 10.1016/j.asoc.2018.04.006.
Grant Scheme (FRGS) Reference number: FRGS/1/2018/ [19] H. Tazvinga, B. Zhu, and X. Xia, ‘‘Optimal power flow management for
TK07/UTP/02/3 and Universiti Teknologi PETRONAS to distributed energy resources with batteries,’’ Energy Convers. Manage.,
provide the laboratory facilities to conduct this research work. vol. 102, pp. 104–110, Sep. 2015, doi: 10.1016/j.enconman.2015.01.015.
[20] A. Panda, M. Tripathy, A. K. Barisal, and T. Prakash, ‘‘A modified bac-
teria foraging based optimal power flow framework for hydro-thermal-
REFERENCES wind generation system in the presence of STATCOM,’’ Energy, vol. 124,
[1] J. Carpentier, ‘‘Contribution a l’Etude du Dispatching Economique,’’ pp. 720–740, Apr. 2017, doi: 10.1016/j.energy.2017.02.090.
Bull. de la Societe Francaise des Electriciens, vol. 3, pp. 431–474, [21] R. A. Jabr and B. C. Pal, ‘‘Intermittent wind generation in optimal power
1962. flow dispatching,’’ IET Generation, Transmiss. Distrib., vol. 3, no. 1,
[2] A. E. Chaib, H. R. E. H. Bouchekara, R. Mehasni, and M. A. pp. 66–74, Jan. 2009, doi: 10.1049/IET-GTD:20080273.
Abido, ‘‘Optimal power flow with emission and non-smooth cost func- [22] S. Mishra, Y. Mishra, and S. Vignesh, ‘‘Security constrained economic
tions using backtracking search optimization algorithm,’’ Int. J. Elect. dispatch considering wind energy conversion systems,’’ in Proc. IEEE
Power Energy Syst., vol. 81, pp. 64–77, Oct. 2016, doi: 10.1016/ Power Energy Soc. Gen. Meeting, Jul. 2011, pp. 1–8, doi: 10.1109/
j.ijepes.2016.02.004. PES.2011.6039544.

48402 VOLUME 10, 2022

T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

[23] L. Shi, C. Wang, L. Yao, Y. Ni, and M. Bazargan, ‘‘Optimal power [41] Y. Zhu, J. Wang, and B. Qu, ‘‘Multi-objective economic emission dispatch
flow solution incorporating wind power,’’ IEEE Syst. J., vol. 6, no. 2, considering wind power using evolutionary algorithm based on decompo-
pp. 233–241, Jun. 2012, doi: 10.1109/JSYST.2011.2162896. sition,’’ Int. J. Electr. Power Energy Syst., vol. 63, pp. 434–445, Dec. 2014,
[24] A. Panda and M. Tripathy, ‘‘Optimal power flow solution of wind inte- doi: 10.1016/j.ijepes.2014.06.027.
grated power system using modified bacteria foraging algorithm,’’ Int. [42] R. Azizipanah-Abarghooee, T. Niknam, A. Roosta, A. R. Malekpour,
J. Elect. Power Energy Syst., vol. 54, pp. 306–314, Jan. 2014, doi: and M. Zare, ‘‘Probabilistic multiobjective wind-thermal economic emis-
10.1016/j.ijepes.2013.07.018. sion dispatch based on point estimated method,’’ Energy, vol. 37, no. 1,
[25] A. Panda and M. Tripathy, ‘‘Security constrained optimal power flow pp. 322–335, Jan. 2012, doi: 10.1016/j.energy.2011.11.023.
solution of wind-thermal generation system using modified bacteria [43] N. A. Khan, A. B. Awan, A. Mahmood, S. Razzaq, A. Zafar, and
foraging algorithm,’’ Energy, vol. 93, pp. 816–827, Dec. 2015, doi: G. A. S. Sidhu, ‘‘Combined emission economic dispatch of power sys-
10.1016/j.energy.2015.09.083. tem including solar photo voltaic generation,’’ Energy Convers. Manage.,
[26] S. S. Reddy, ‘‘Optimal power flow with renewable energy resources includ- vol. 92, pp. 82–91, Mar. 2015, doi: 10.1016/j.enconman.2014.12.029.
ing storage,’’ Electr. Eng., vol. 99, no. 2, pp. 685–695, Jun. 2017, doi: [44] M. S. Gonçalves, R. H. Lopez, and L. F. F. Miguel, ‘‘Search group
10.1007/s00202-016-0402-5. algorithm: A new Metaheuristic method for the optimization of truss
[27] K. Teeparthi and D. M. V. Kumar, ‘‘Security-constrained optimal power structures,’’ Comput. Struct., vol. 153, pp. 165–184, Jun. 2015, doi:
flow with wind and thermal power generators using fuzzy adaptive artifi- 10.1016/j.compstruc.2015.03.003.
cial physics optimization algorithm,’’ Neural Comput. Appl., vol. 29, no. 3, [45] G. Acampora, D. Caruso, A. Vaccaro, and A. Vitiello, ‘‘A search group
pp. 855–871, Feb. 2018, doi: 10.1007/s00521-016-2476-4. algorithm for optimal voltage regulation in power systems,’’ in Proc. IEEE
[28] R. Roy and H. T. Jadhov, ‘‘Optimal power flow solution of power system Congr. Evol. Comput. (CEC), Jul. 2016, pp. 3662–3669, doi: 10.1109/
incorporating stochastic wind power using Gbest guided artificial bee CEC.2016.7744253.
colony algorithm,’’ Int. J. Elect. Power Energy Syst., vol. 64, pp. 562–578, [46] U. K. Rout, R. K. Sahu, and S. Panda, ‘‘Gravitational search algorithm
Jan. 2015, doi: 10.1016/j.ijepes.2014.07.010. based automatic generation control for interconnected power system,’’ in
Proc. Int. Conf. Circuits, Power Comput. Technol. (ICCPCT), Mar. 2013,
[29] A. Panda and M. Tripathy, ‘‘Solution of wind integrated thermal generation
pp. 557–568, doi: 10.1109/iccpct.2013.6528998.
system for environmental optimal power flow using hybrid algorithm,’’
J. Electr. Syst. Inf. Technol., vol. 3, no. 2, pp. 151–160, Sep. 2016, doi: [47] S. F. H. Noorbin and A. Alfi, ‘‘Adaptive parameter control of search group
10.1016/j.jesit.2016.01.004. algorithm using fuzzy logic applied to networked control systems,’’ Soft
Comput., vol. 22, no. 23, pp. 7939–7960, Dec. 2018, doi: 10.1007/s00500-
[30] P. P. Biswas, P. N. Suganthan, and G. A. J. Amaratunga, ‘‘Optimal
017-2742-0.
power flow solutions incorporating stochastic wind and solar power,’’
[48] F. Carraro, R. H. Lopez, and L. F. F. Miguel, ‘‘Optimum design of planar
Energy Convers. Manage., vol. 148, pp. 1194–1207, Sep. 2017, doi:
steel frames using the search group algorithm,’’ J. Brazilian Soc. Mech.
10.1016/j.enconman.2017.06.071.
Sci. Eng., vol. 39, no. 4, pp. 1405–1418, Apr. 2017, doi: 10.1007/s40430-
[31] M.-R. Chen, G.-Q. Zeng, and K.-D. Lu, ‘‘Constrained multi-objective pop- 016-0628-1.
ulation extremal optimization based economic-emission dispatch incorpo-
[49] R. L. Pedro, J. Demarche, L. F. F. Miguel, and R. H. Lopez, ‘‘An efficient
rating renewable energy resources,’’ Renew. Energy, vol. 143, pp. 277–294,
approach for the optimization of simply supported steel-concrete compos-
Dec. 2019, doi: 10.1016/j.renene.2019.05.024.
ite I-girder bridges,’’ Adv. Eng. Softw., vol. 112, pp. 31–45, Oct. 2017, doi:
[32] U. Guvenc, S. Duman, H. T. Kahraman, S. Aras, and M. Katı, ‘‘Fitness– 10.1016/j.advengsoft.2017.06.009.
distance balance based adaptive guided differential evolution algorithm for [50] P. P. Biswas, P. N. Suganthan, B. Y. Qu, and G. A. J. Amaratunga, ‘‘Mul-
security-constrained optimal power flow problem incorporating renewable tiobjective economic-environmental power dispatch with stochastic wind-
energy sources,’’ Appl. Soft Comput., vol. 108, Sep. 2021, Art. no. 107421, solar-small hydro power,’’ Energy, vol. 150, pp. 1039–1057, May 2018,
doi: 10.1016/j.asoc.2021.107421. doi: 10.1016/j.energy.2018.03.002.
[33] N. Karthik, A. K. Parvathy, R. Arul, and K. Padmanathan, ‘‘Multi-objective [51] H. M. Dubey, M. Pandit, and B. K. Panigrahi, ‘‘Hybrid flower pollina-
optimal power flow using a new heuristic optimization algorithm with the tion algorithm with time-varying fuzzy selection mechanism for wind
incorporation of renewable energy sources,’’ Int. J. Energy Environ. Eng., integrated multi-objective dynamic economic dispatch,’’ Renew. Energy,
vol. 12, no. 4, pp. 641–678, Dec. 2021, doi: 10.1007/s40095-021-00397-x. vol. 83, pp. 188–202, Nov. 2015, doi: 10.1016/j.renene.2015.04.034.
[34] I. U. Khan, N. Javaid, K. A. A. Gamage, C. J. Taylor, S. Baig, and [52] J. Hetzer, D. C. Yu, and K. Bhattarai, ‘‘An economic dispatch model
X. Ma, ‘‘Heuristic algorithm based optimal power flow model incor- incorporating wind power,’’ IEEE Trans. Energy Convers., vol. 23, no. 2,
porating stochastic renewable energy sources,’’ IEEE Access, vol. 8, pp. 603–611, Jun. 2008, doi: 10.1109/TEC.2007.914171.
pp. 148622–148643, 2020, doi: 10.1109/ACCESS.2020.3015473. [53] X. Liu and W. Xu, ‘‘Economic load dispatch constrained by wind power
[35] M. H. Sulaiman and Z. Mustaffa, ‘‘Solving optimal power flow problem availability: A here-and-now approach,’’ IEEE Trans. Sustain. Energy,
with stochastic wind–solar–small hydro power using barnacles mating vol. 1, no. 1, pp. 2–9, Apr. 2010, doi: 10.1109/TSTE.2010.2044817.
optimizer,’’ Control Eng. Pract., vol. 106, Jan. 2021, Art. no. 104672, doi: [54] X. Liu, ‘‘Economic load dispatch constrained by wind power availabil-
10.1016/j.conengprac.2020.104672. ity: A wait-and-see approach,’’ IEEE Trans. Smart Grid, vol. 1, no. 3,
[36] S. Duman, S. Rivera, J. Li, and L. Wu, ‘‘Optimal power flow of power sys- pp. 347–355, Dec. 2010, doi: 10.1109/TSG.2010.2057458.
tems with controllable wind-photovoltaic energy systems via differential [55] T. P. Chang, ‘‘Investigation on frequency distribution of global radiation
evolutionary particle swarm optimization,’’ Int. Trans. Electr. Energy Syst., using different probability density functions,’’ Int. J. Appl. Sci. Eng., vol. 8,
vol. 30, no. 4, Apr. 2020, Art. no. e12270, doi: 10.1002/2050-7038.12270. no. 2, pp. 99–107, Sep. 2010, doi: 10.6703/IJASE.2010.8(2).99.
[37] E. Kaymaz, S. Duman, and U. Guvenc, ‘‘Optimal power flow solution with [56] S. S. Reddy, P. R. Bijwe, and A. R. Abhyankar, ‘‘Real-time economic
stochastic wind power using the Lévy coyote optimization algorithm,’’ dispatch considering renewable power generation variability and uncer-
Neural Comput. Appl., vol. 33, no. 12, pp. 6775–6804, Jun. 2021, doi: tainty over scheduling period,’’ IEEE Syst. J., vol. 9, no. 4, pp. 1440–1451,
10.1007/s00521-020-05455-9. Dec. 2015, doi: 10.1109/JSYST.2014.2325967.
[38] A. R. Abul’Wafa, ‘‘Optimization of economic/emission load dispatch [57] D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine
for hybrid generating systems using controlled elitist NSGA-II,’’ Elec- Learning, 1st ed. Reading, MA, USA: Addison-Wesley, 1989.
tric Power Syst. Res., vol. 105, pp. 142–151, Dec. 2013, doi: 10.1016/j. [58] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, ‘‘A fast and elitist
epsr.2013.07.006. multiobjective genetic algorithm: NSGA-II,’’ IEEE Trans. Evol. Comput.,
[39] B. Y. Qu, J. J. Liang, Y. S. Zhu, Z. Y. Wang, and P. N. Suganthan, ‘‘Eco- vol. 6, no. 2, pp. 182–197, Jan. 2002, doi: 10.1109/4235.996017.
nomic emission dispatch problems with stochastic wind power using sum- [59] R. D. Zimmerman and C. E. Murillo-Sanchez, MATPOWER-Manual-
mation based multi-objective evolutionary algorithm,’’ Inf. Sci., vol. 351, 4.1.Pdf. Cornell University, Ithaca, NY, USA: Power Systems Engineering
pp. 48–66, Jul. 2016, doi: 10.1016/j.ins.2016.01.081. Research Center, 2011. Accessed: Sep. 19, 2020. [Online]. Available:
[40] T. C. Bora, V. C. Mariani, and L. D. S. Coelho, ‘‘Multi-objective https://matpower.org/docs/MATPOWER-manual-4.1.pdf
optimization of the environmental-economic dispatch with reinforce- [60] W. Sheng, K.-Y. Liu, Y. Liu, X. Meng, and Y. Li, ‘‘Optimal placement
ment learning based on non-dominated sorting genetic algorithm,’’ and sizing of distributed generation via an improved nondominated sorting
Appl. Thermal Eng., vol. 146, pp. 688–700, Jan. 2019, doi: 10.1016/j. genetic algorithm II,’’ IEEE Trans. Power Del., vol. 30, no. 2, pp. 569–578,
applthermaleng.2018.10.020. Apr. 2015, doi: 10.1109/TPWRD.2014.2325938.

VOLUME 10, 2022 48403

 T. H. Bao Huy et al.: Performance Improvement of Multiobjective OPF-Based RES Using Intelligent Algorithm

[61] Y.-N. Wang, L.-H. Wu, and X.-F. Yuan, ‘‘Multi-objective self-adaptive NURSYARIZAL MOHD NOR received the M.Sc.
differential evolution with elitist archive and crowding entropy-based degree in electrical power engineering from The
diversity measure,’’ Soft Comput., vol. 14, no. 3, p. 193, Jan. 2009, doi: University of Manchester Institute of Science and
10.1007/s00500-008-0394-9. Technology (UMIST), U.K., in 2001, and the
[62] K. Deb, Multi-Objective Optimization Using Evolutionary Algorithms. Ph.D. degree in electrical engineering from Uni-
Hoboken, NJ, USA: Wiley, 2001. versiti Teknologi PETRONAS (UTP), Malaysia,
[63] O. Alsac and B. Stott, ‘‘Optimal load flow with steady-state security,’’ in 2009. His areas of specialization are ‘analysis
IEEE Trans. Power App. Syst., vol. PAS-93, no. 3, pp. 745–751, May 1974,
and optimization of large-scale power systems’
doi: 10.1109/TPAS.1974.293972.
and ‘state estimation’. He has several publications
[64] S. Mirjalili, P. Jangir, and S. Saremi, ‘‘Multi-objective ant lion optimizer: A
multi-objective optimization algorithm for solving engineering problems,’’ at his credit. His research interests include power
Appl. Intell., vol. 46, no. 1, pp. 79–95, Jan. 2017, doi: 10.1007/s10489-016- system state estimation, power system analysis, renewable energy, and elec-
0825-8. trical machine.
[65] S. Z. Mirjalili, S. Mirjalili, S. Saremi, H. Faris, and I. Aljarah, ‘‘Grasshop-
per optimization algorithm for multi-objective optimization problems,’’
Appl. Intell., vol. 48, no. 4, pp. 805–820, 2018, doi: 10.1007/s10489-017-
1019-8.
[66] S. M. Mohseni-Bonab, A. Rabiee, and B. Mohammadi-Ivatloo, ‘‘Volt-
age stability constrained multi-objective optimal reactive power dis-
patch under load and wind power uncertainties: A stochastic approach,’’
Renew. Energy, vol. 85, pp. 598–609, Jan. 2016, doi: 10.1016/j. IRRAIVAN ELAMVAZUTHI (Member, IEEE)
renene.2015.07.021. received the Ph.D. degree from the Department of
[67] P. P. Biswas, P. N. Suganthan, R. Mallipeddi, and G. A. J. Amaratunga, Automatic Control & Systems Engineering, The
‘‘Optimal reactive power dispatch with uncertainties in load demand University of Sheffield, U.K., in 2002. He is cur-
and renewable energy sources adopting scenario-based approach,’’ rently an Associate Professor at the Department
Appl. Soft Comput., vol. 75, pp. 616–632, Feb. 2019, doi: 10.1016/j. of Electrical and Electronic Engineering, Univer-
asoc.2018.11.042. siti Teknologi PETRONAS (UTP), Malaysia. His
research interests include control, robotics, mecha-
tronics, power systems, and bio-medical applica-
tions. He is also the Chair of the IEEE Robotics
and Automation Society (Malaysia Chapter).

TAIB IBRAHIM was born in Kedah, Malaysia,

TRUONG HOANG BAO HUY received the
in 1972. He received the B.Eng. degree (Hons.) in
B.Eng. degree in electrical and electronics engi-
electrical and electronics engineering from Coven-
neering from the Ho Chi Minh City University
try University, U.K., in 1996, the M.Sc. degree in
of Technology (HCMUT), VNU-HCM, Vietnam,
electrical power engineering from the University
in 2017, and the M.S. degree in electrical and
of Strathclyde, U.K., in 2000, and the Ph.D. degree
electronics engineering from Universiti Teknologi
in electrical machine design from The University
PETRONAS (UTP), Malaysia, in 2020. He is cur-
of Sheffield, U.K., in 2009. His employment expe-
rently a Research Officer with the Institute of
rience includes Airod (M) Sdn Bhd and Universiti
Engineering and Technology, Thu Dau Mot Uni-
Teknologi PETRONAS (UTP). He is currently the
versity, Binh Duong, Vietnam. His research inter-
Leader for power and energy cluster and the Co-Leader for mission-oriented
ests include renewable energy power generation, artificial intelligence-based
research (energy) with the UTP. His research interests include motion control
algorithms, and their application in optimization problems.
to electromagnetic devices and their associated drives.

DIEU NGOC VO received the B.Eng. and M.Eng.

degrees in electrical engineering from the Ho Chi
Minh City University of Technology (HCMUT),
VNU-HCM, Ho Chi Minh City, Vietnam, in
TRI PHUOC NGUYEN received the B.Eng. 1995 and 2000, respectively, and the D.Eng.
degree in electrical engineering from Can Tho degree in energy from the Asian Institute of Tech-
University, Can Tho City, Vietnam, in 2014, and nology (AIT), Pathum Thani, Thailand, in 2007.
the M.Eng. degree in electrical engineering from He is currently an Associate Professor with the
the Ho Chi Minh City University of Technol- Department of Power Systems, Faculty of Elec-
ogy, Ho Chi Minh City, Vietnam, in 2017. His trical and Electronic Engineering, HCMUT. His
research interest includes applications of meta- research interests include applications of AI in power system optimization,
heuristic methods to optimization of distribution power system operation and control, power system analysis, and power
systems. systems under deregulation.

48404 VOLUME 10, 2022