Energy Reports 7 (2021) 398–412

Contents lists available at ScienceDirect

Energy Reports
journal homepage: www.elsevier.com/locate/egyr

An adaptive Marine Predators Algorithm for optimizing a hybrid

PV/DG/Battery System for a remote area in China
∗
Guoqing Yu , Zhiyong Meng, Hongtao Ma, Lei Liu
School of Information Science and Engineering, Hebei University of Science and Technology, Shijiazhuang, Hebei, 050018, China

article info a b s t r a c t

Article history: Lack of electricity networks in remote areas and also the high cost of connecting these areas to
Received 20 November 2020 national networks due to the unfavorable geographical situation of the region leads to the use of other
Received in revised form 28 December 2020 energy sources independent of the network in these areas. Today, the use of distributed generation
Accepted 2 January 2021
sources is increasing due to the reduction of fossil fuel sources and heat problems. The output
Available online 9 January 2021
power of distributed generation sources, especially renewable energy sources, fluctuates under the
Keywords: influence of atmospheric conditions. The main purpose of the present study is to present a new optimal
Hoxtolgay configuration for a hybrid photovoltaic/diesel generator/battery system to refine the load demand of
Hybrid renewable energy system a rural area in Hoxtolgay, China. The idea is to minimize the CO2 emissions value, the annualized
Multi-objective optimization cost, and the loss of load probability of the system. To simplify this system, ε -constraint method is
Battery energy storage system performed. The problem has been solved by an adaptive version of Marine Predators Algorithm (AMPA)
Photovoltaic to improve the optimization results. The total ACS for the optimal system is achieved 8224.15 $. Also,
Diesel generator
an optimal value of 451.30 kW for diesel generator, 4266$ for battery storage, and 2670.53 $ for PV
ε-constraint method
Adaptive Marine Predators Algorithm
generator are achieved. The initial capital cost of the system is achieved 47940 $ that is less than the
Sensitivity analysis NPC. In this case, the maximum cost of the system with 43.12% belongs to the PV system of total cost
of the system. Finally, CO2 emission by the suggested AMPA with 1624 kg/year shows the minimum
value with cleanest result among the others. Final results of the proposed technique are compared
with PSO method and HOMER-based optimization to indicate its effectiveness. Sensitivity analysis has
been also performed to determine the system efficiency.
© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).

1. Introduction In recent years, in order to manage electricity consumption

and to solve the problems caused by power outages, many in-
Inaccessibility to the national electricity network, uneconom- centive mechanisms have been considered by the electricity com-
ical construction of transmission lines and lack of reliability to panies to subscribers and owners of capital to produce and sell
connect to the national electricity network are some of the rea- electricity to the companies (Akbary et al., 2019). Although ris-
sons that can affect the system effectiveness separate from the ing electricity purchase prices from private electricity generating
main network (Fan et al., 2020b). At present, diesel generators companies and government facilities to provide cheaper fuel for
(DGs) are the main sources of power for off-grid networks (Ye their power plants have led to further growth in these com-
et al., 2020). High reliability, relatively low cost and ease of panies, greenhouse gas emissions have also increased and their
installation are some of the advantages of using diesel genera- economies are heavily dependent on fossil fuel prices (Eslami
tors (Hamian et al., 2018). Therefore, the use of DGs as the main et al., 2019). If renewable sources are used alongside traditional
source of electricity generation in off-grid remote areas is very generators as a hybrid system, energy production will increase
common (Leng et al., 2018). DGs are used as a backup source significantly due to the reduction of fossil fuel consumption and
in the event of a power outage in areas of the network where the increase of energy injected into the grid (Yu and Ghadimi,
the shutdown rate is high due to low reliability (Mirzapour et al., 2019). These energies are able to produce clean and environmen-
2019; Mohammadi et al., 2018). In most cases, it will not be cost- tally friendly energy and at the same time reduce the use of fossil
effective to meet the demand for energy, either under normal fuels (Ebrahimian et al., 2018). The advantages of using a photo-
circumstances or in the event of a power outage using fossil fuels voltaic (PV) system are the need for very low repairs, quite perfor-
due to their high price (Ebrahimian et al., 2018). mance, easy installation, and the abundance of free solar energy
in the geographical area of China (Yuan et al., 2020). However,
∗ Corresponding author. due to accidental and uncontrolled production of photovoltaic
E-mail address: [email protected] (G. Yu). system and also severe load changes in separate networks from

https://doi.org/10.1016/j.egyr.2021.01.005
2352-4847/© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

maximum to minimum value, the use of PV system along with (PSO) algorithm. The analysis was applied under various electrical
DGs has faced problems. loads. At last, the performance of the DG must be analyzed during
This leads to the need for much larger capacities for the gener- the low value of the ship’s electrical load (Yang et al., 2020).
ator and the photovoltaic system and the operation of the gener- Alturki et al. presented an optimal technique to proper size
ator at the operating point far from the nominal value (Ghadimi, selection of an HRES concerning to long-term cost analysis. The
2015a,b). Therefore, to improve system performance and reduce method was based on supply–demand-based optimization (SDO)
resource capacity, the battery energy storage system is used along algorithm. Effectiveness of the algorithm was verified by com-
with the generator and photovoltaic system. paring with other well-known algorithms. The algorithm was
Battery power storage can be used to reduce fuel consump- performed to the system based on three scenarios concerning to
tion (Liu et al., 2020). When the generator is on, the battery the wind speed, solar radiation, and load profile data. Final results
absorbs energy from the system, causing the generator to operate indicate the prominence of the SDO in optimal size defining of an
at rated power with maximum efficiency (Gheydi et al., 2018). HES with a higher convergence rate (Alturki et al., 2020).
Also, when the battery is fully charged, the generator goes out As can be observed from the literature, different types of
of circuit and the load has been supplied by the photovoltaic sys- metaheuristics are used for optimal configuration of the hybrid
tem and the energy stored in the battery (Ghadimi, 2015b). The systems. However, most of the explained algorithms have differ-
operation of the generator between the nominal load with the ent shortcomings such as trapping into local optimum, premature
highest efficiency and off state, brings the lowest operating hours convergence, and low accuracy.
of the generator and, consequently, the lowest fuel consumption The main idea in this paper is to propose a multi-objective
for the generator (Ghadimi, 2014). optimization arrangement for a hybrid PV/DG/ Battery Energy
In low-consumption load indices, the load power during day- System to give a robust and safe source for a remote area in
light hours, when the sun’s radiant energy is at its highest level, Hoxtolgay, China. To simplification of the multi-objective opti-
creates a power balance between production and consumption
mization problem, ε -constraint technique is utilized. The purpose
(Cai et al., 2019). In this case, it is essential for the energy storage
is to minimize the system capital cost, the CO2 emissions, and the
system to absorb and store the excess power produced in the
unmet load, simultaneously. Using ε -constraint procedure makes
photovoltaic system (Firouz and Ghadimi, 2016). However, the
the problem so simpler with less computational time and also
battery’s energy storage capacity is limited and it is unable to
separate from supplementary actions needed by the Pareto-based
absorb the extra power of the photovoltaic system once it is fully
techniques. The optimization is performed using an adaptive ver-
charged (Khodaei et al., 2018; Aghajani and Ghadimi, 2018). In
sion of Marine Predators Algorithm to cover and refine the before
this case, in order to create power balance, it is necessary to waste
mentioned drawbacks from the literature. Sensitivity analysis is
the productive power of the photovoltaic system at excessive
applied to blend the effect of the components on the hybrid sys-
loads, which leads to non-optimal performance of the system and
tem. Final results are also compared with two other well-known
thus reduces efficiency (Liu et al., 2017).
Bukar et al. worked on the application of the Grasshopper Op- methods from the literature to show its efficiency. Therefore, the
timization Algorithm (GOA) for optimal sizing of an independent main contributions of this paper can be briefly highlighted as
microgrid (MG) system in five residential areas in an off-grid area follows:
in Yobe State, Nigeria. The idea was to determine the optimal – New optimal configuration for a hybrid photovoltaic/diesel
arrangement of the system with concerning to the lack of cost of generator/battery system.
energy (COE) and power supply probability (DPSP). Final results – Minimize the CO2 emissions, the annualized cost, and the
of the proposed system were compared with particle swarm op- loss of load probability.
timization (PSO) algorithms and cuckoo search (CS) optimization – Using ε -constraint to simplify the objective function
to show the method superiority (Saeedi et al., 2019).
– Propose an adaptive version of Marine Predators Algorithm
Makhdoomi and Askarzadeh proposed an optimal strategy for
to improve the optimization results
a hybrid PV/DG/PHS HRES using an improved version of the
– Comparison with some well-known methods.
crow search algorithm (CSA). The main purpose was to minimize
the fuel consumption through the considered period. Simula-
2. System modeling
tions were performed with respect to different solar radiations
to achieve an optimal sharing of deficit power between the DG
and the PHS. Simulation results specified that the methodology The hybrid power generation system including a photovoltaic
has more precise and reliable results than the compared methods array, a diesel generator, a battery bank, an inverter, a controller,
and the original CSA. Besides, optimal power supplying between and other devices and cables. Photovoltaic and diesel arrays work
DG and PHS leads to get minimum operation cost (Makhdoomi together to satisfy the load demand (Meng et al., 2020; Abedinia
and Askarzadeh, 2020). et al., 2019; Gao et al., 2019). While the power of the photo-
Ashtiani et al. presented a new configuration for optimal sizing voltaic and diesel sources is sufficient, the generated power, after
of a grid-connected PV/BESS to minimize the total net present meeting the load demand, has been utilized as battery power to
cost (NPC). The method was based on Teaching-Learning-Based fully charge the battery (Hosseini Firouz and Ghadimi, 2016; Mir
Optimization (TLBO) algorithm. et al., 2019; Hagh et al., 2015). If the battery charge level is high
The paper indicated the effect of PV/battery systems to de- or after a while it is fully charged and the source power is still
crease the electricity bills. The results showed that using this higher than the required load power, the power of the hybrid
configuration compared with the non-renewable case gives better system is injected into the network. In contrast, when the hybrid
NPC and Cost Of Energy (COE) of the grid connected PV/battery power supply is low, the battery releases its stored energy to
system. Final results were compared with some state of the art help the hybrid system meet its load requirements (Gong and
metaheuristics to verify the superiority of the suggested algo- razmjooy, 2020). In order to predict system efficiency, each of
rithm (Najafi Ashtiani et al., 2020). these components must first be individually modeled and then
Yang et al. analyzed a hybrid PV/DG-based ship with 5000 their combination evaluated to meet load demand (Fei et al.,
car spaces. The idea was to optimize the fuel economy and DG 2019; Fan et al., 2020a; Cao et al., 2019). Fig. 1 shows the com-
efficiency as a multi-objective optimization problem. The opti- prehensive model of the studied hybrid renewable energy system
mization was performed based on Partial Swarm Optimization (HRES).
399
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

temperature-power coefficient of the module, ηDA (here 0.9) de-

termines the converter efficiency, Tcell represents is the cell tem-
perature. PMPRef and γPV is given by the manufacturer. Tcell is
mathematically formulated as follows:
G (t )
[ ]
TNOC − 20
Tcell (t ) = Tamb (t ) + × (5)
0.8 1000(W/m2 )
where, TNOC signifies the nominal operation temperature of the
cell, and Tamb (t ) describes the ambient temperature.
If a matrix with the size Np × Ns considering photovoltaic pan-
els, the maximum output power is calculated using the following
formula:
O
PPV (t ) = NP × Ns × PPV (t) × ηoth (6)
Where, ηoth signifies a factor that indicates other losses such
as losses due to cable resistance, dust, etc. Therefore, after the
surface of the solar radiation for the module and the temper-
Fig. 1. The comprehensive model of the studied HRES. ature of the module are determined, the output power of the
photovoltaic system is calculated through equations. Total solar
radiation on a diagonal surface consists of three parts: reflection,
penetration, and radiation. In this research, the reflective part
2.1. The model of diesel generator
has been omitted and the following relations have been used to
calculate the penetration and radiation part.
Diesel Generator (DG) has been employed as a backup energy
source if the generated power by the solar cells does not satisfy 2.3. The model of inverter
the charge level of the energy storage sources is low. In this case,
DG is used and satisfies the lack of power. The fuel consumption As before mentioned, the output of a PV system is DC current.
for a DG depends to its output power and has been formulated Therefore, to convert the DC to the AC current for providing the
by the following (Dufo-Lopez and Bernal-Agustín, 2008): power of the load demand. The output power of an inverter can
be mathematically modeled as follows:
ConsDG (ti ) = BDG × PNDG + ADG × PDG (ti ) (1)
PhD
where, PNDG (kW ) signifies the nominal power, BDG (ti ) describes the P inv = (7)
ηinv
output power of the DG, and ADG and BDG signifies the coeffi-
cients of the fuel consumption curve which are considered 0.0845 where, PhD signifies the hourly demand and ηinv represents the
L/kWh, and 0.246 L/kWh, respectively (Shivaie et al., 2019). The performance of the inverter.
annual cost of diesel fuel consumption is calculated as follows:
2.4. The model of battery storage
8760
∑
Fuel Fuel
CDG = Price × ConsDG (ti ) (2)
In order to alternate the output power of PV modules and
i=1
diesel generators and to adjust any surplus or deficit of power
where, PriceFuel describes the price of fuel consumed is in liters generation, the battery has been used as energy storage sources
per hour. in the hybrid system. When the entire output power of the
Also, the DG efficiency can be obtained by the following: photovoltaic modules and the diesel generator is more than the
required load capacity, the battery bank will start charging.
PDG (ti ) The battery current is directly depends to the output power
ηDG = Fuel
(3)
CDG × LHV of the renewable generators, the SOC, and the control policy (Fan
et al., 2020a). The main technique for improving the efficiency
where, LHV defines the heating value of fuel consumption (Dufo-
in the generation system is to consider the maximum PV power
Lopez and Bernal-Agustín, 2008). Here, LHVf ∈ [10, 11.6] kWh/L.
controlling for the battery charging and discharging (Cao et al.,
2019). The battery energy generation and absorption during the
2.2. The model of photo voltaic time from t − 1 to t can be formulated as follows:
PLO (t )
( )
The proposed model for generating power from solar sources PaBat (t ) = O
PPV (t ) − × ηBat + PaBat (t − 1) × (1 − σ ) (8)
ηinv
consists of two parts: the power generation model of photovoltaic
arrays and the solar radiation model on the surface of the arrays. Such that,
Based on the definition of accumulation factor, the maximum
⎧
⎨Charging : P O (t ) − PLi (t ) > 0 and P Bat (t − 1) < P Batmax
output power of the photovoltaic has been written as follows. PV ηinv a a
i (9)
⎩Discharging : P O (t ) − PL (t ) < 0 and P Bat (t − 1) > P Batmin
G (t ) PV ηinv a a
PPV (t ) = PMPRef × ×[1 + γPV × (Tcell − 25 ◦ C )]×ηDA
1000(W/m2 ) where, PLi (t ) describes the load demand at time t, PaBat (t ) and
(4) PaBat (t − 1) represents the power availability of the BESS at
present (t) and its previous hour (t − 1), respectively, σ signifies
where, PMPRef (kW) describes the maximum power point of the the BESS self-discharge ratio of battery storage, ηBat defines the
module in standard conditions, G (t ) is the sun irradiation over performance of the BESS. Here, σ is set 0.001, ηbat is set 0.7, ηin
Bat Bat
the tilted surface of the PV panels (w/m2 ), γPV signifies the is set 0.92, and PaBat (t ) ∈ [Pa min , Pa max ].
400
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

To enhance the battery storage lifespan, the maximum charg- matrix with equal dimension of Elite was created which is called
ing rate (SOC max ) is set as the upper limitation with overall Prey. The location of the predator is updated by this matrix. This
nominal capacity based on the following equation (Sobu and Wu, matrix is as follows:
2012): ⎡
X1,1 ··· X1,d
⎤

SOC max = PaBat × N Bat ⎢ . .. .. ⎥

(10) P = ⎣ .. . . ⎦ (14)
min Xn,1 ··· X n ,d
And the minimum allowable state of charge (SOC ) as fol- n×d
lows (Sobu and Wu, 2012):
where, Xi,j signifies the jth dimension of the ith prey. Notably, the
SOC min = PaBat × N Bat × 1 − DoDmax
( )
(11) optimization method is associated to these matrixes.
max The MPOA includes three main sections about various speed
where, DoD signifies the maximum depth percentage of dis-
ratio copying the total life of a predator and prey as follows:
charging.
The maximum lifespan in battery will be happened when DOD (a) The prey goes quicker than the predator or in higher speed
is set between 30% and 50%. Therefore, DoDmax = 30%. ratio
The design and optimize of the hybrid DG/PV/BESS are per- (b) The predator goes quicker than the prey or the lower speed
formed in a rural area in Hoxtolgay, China. The system optimiza- ratio
tion is performed using a new developed version of Elephant (c) Both the predator and the prey goes in about the same
Herding Optimization (EHO) algorithm. To validate of the system velocity or in equal speed ratio
efficiency based on the designed algorithm, its results are com-
pared with the results of two other well-known techniques. This For the certain steps, an exact duration of iteration was pro-
design can be also employed to different case studies with the vided. These phases are explained by nature regulations of the
same weather conditions predator and the prey movement when with the inspiration of
The meteorological data of the for the case study is extracted the prey and the predator in the nature. The description of the
from NASA database and the area is selected by collecting data steps are as follows:
from a number of powerless households. The proposed method is (a) In this step, the algorithm exploration is occurred. This uses
performed to the hybrid system based on its meteorological data at the primary iterations’ duration. In higher speed ratio (v ≥ 10),
to optimize the studied hybrid DG/PV/BESS. Sensitivity analysis for the predator, the optimum strategy is to stop moving. This is
also applied to the system for analyzing the effect of variations as follows:
of each component. The results at last compared with PSO algo- 1
rithm, and the HOMER software from the literature to show its While Iter < MaxIter (15)
3 (
superiority. −−−−→ )
⃗ B ⊗ E⃗i − R⃗ B ⊗ P⃗i
stepsizei = R (16)
3. Adaptive Marine Predators Algorithm
i = 1, 2, . . . , n
−−−−→
3.1. Marine Predators Optimization Algorithm P⃗i = P⃗i + P × R
⃗ ⊗ stepsizei (17)

where, R⃗ B explains a vector containing some random amount that

The major idea of the Marine Predators Optimization Algo-
rithm (MPOA) as an effective and clear optimization algorithm are made by Brownian movement (Shezan, 2019) and the nota-
was explained here. This optimization algorithm is a bio-inspired tion ⊗ defines entry-wise products. The prey motion is modeled
by product between prey and R ⃗B .
algorithm. The MPOA begins with random amounts that are dis-
tributed uniformly in the solution space. This can be achieved as P represents a constant quantity amount to 0.5 and R shows
follows: random uniform distributed amounts in the range [0, 1]. Iter
denotes the present iteration when MaxIter explains the highest
X0 = Xmin + rand × (Xmax − Xmin ) (12) amount of iteration.
(b) In this step, the prey and the predator look for their prey.
where, rand defines a uniform random vector between 0 and 1,
This strategy happens in the middle phase of the optimization
and Xmin and Xmax explain the minimum and maximum limits of
process. The exploration term attempts to transmit into the ex-
the parameters.
The best predators have more potentiality of hunting regard- ploitation term. In fact, both exploration and exploitation are
ing the ‘‘survival of the fittest theory’’ (Viswanathan et al., 1999). included here. Correspondingly, the individual separated into two
Correspondingly, a best predator is called ‘‘Elite’’, which is proper sections that one is specified for exploitations and the other one
solution applied to create a matrix. The prey search is monitored for the exploration. These terms are used by the predator and the
by the matrix arrays applying the information on the location of prey. While the predator has a movement in Brownian, the prey
the prey. The definition of the matrix is as follows: assumes to have a movement in Lévy.
1 2
X1I ,1 X1I ,d while MaxIter < Iter <
⎡ ⎤
··· MaxIter (18)
⎢ . .. .. ⎥ 3 3
E = ⎣ .. . . ⎦ (13)
By this strategy, for the exploitation term in the individual,
XnI ,1 ··· XnI ,d n×d −−−−→ ( )
⃗ L ⊗ E⃗i − R⃗ L ⊗ P⃗i
stepsizei = R (19)
where, X I denotes the best predator vector with n simulation to
create the Elite matrix (E), d shows the dimensions’ number, and i = 1, 2, . . . , n/2
the variable n defines the candidates’ number. −−−−→
P⃗i = P⃗i + P × R
⃗ ⊗ stepsizei (20)
The search candidates are both the prey and the predator.
The reason is that when the prey is searching for the food, the where, R⃗ L represents a random-amounted vector by Lévy dis-
predator searches for the prey (Jeong et al., 2020). At the end tribution (Shezan, 2019). The Lévy-based motion of the prey is
of each iteration, when the best predator is substituted with modeled by production between the Prey and R ⃗ L while the prey
a proper one, the Elite value was renewed. Moreover, another motion is modeled by the existence of the step size in the prey
401
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

location. Because of small-scale steps of the Levy distribution in where, ε describes a too small integer to avoid from zero-division-
nearly all parts, it is helpful for the exploitation. Moreover, for the error, P⃗i signifies
( ) the ith individual of the predator at the tth
exploration term in the individual, iteration, f P⃗i represents the cost value of the ith predator
−−−−→ (
⃗ B ⊗ E⃗i − R⃗ B ⊗ P⃗i
) ( )
stepsizei = R (21) at the tth iteration, f P⃗Best defines the optimal cost value of
i = 1, 2, . . . , n/2 predator in the tth iteration.
−−−−→ By considering the above definitions, the adaptive learning
P⃗i = E⃗i + P × CF ⊗ stepsizei (22) factor for the ith predator in the tth iteration is as follows:
where, CF represents an adjusting variable to deal with the preda- 1
tor motion step size which is as follows: αit = (29)
1 + e−ρ
2×Iter
where, ρ ∈ (0, 2].
( ) Max
Itr Iter
CF = 1− (23) Consequently, the updated position of the predator is achieved
MaxIter
by the following:
The Brownian-based motion predator is modeled by the pro-
duction of Elite and R⃗ B so that the location of the prey is modified ⎧ ( )
⎨α t × P⃗i + CF × X⃗min + R⃗ ⊗ (Xmax − Xmin ) ⊗ U⃗ if r ≤ pf
by the predators Brownian-based motion. i
P⃗i = ) ( )
⎩α t × P⃗i + pf × (1 − r ) + r × P⃗r − P⃗r if r > pf
(
(c) In the final stage of the optimization process this strategy
i 1 2
occurs, that is usually allied to higher exploitation term. Lévy is
applied as the optimum strategy for the predator in less-speed (30)
ratio (v = 0.1). the equation is as follows:
Furthermore, to refine the algorithm problem from trapping
2 into the local optimum, DE/best/1 mutation mechanism with
while Iter > MaxIter (24)
3 ( dynamic scaling factor have been employed. This can be math-
−−−−→ ⃗ L ⊗ R⃗ L ⊗ E⃗i − P⃗i
)
ematically formulated as follows Nurunnabi et al. (2019):
stepsizei = R (25)
( )
i = 1, 2, . . . , n qti = P⃗Best + σit × P⃗p1
t t
− P⃗p2 (31)
−−−−→
P⃗i = E⃗i + P × CF ⊗ stepsizei (26) where, qti represents the mutant vector, p1 ̸ = p2 ∈ [1, 2, . . . , n]
The Lévy-based motion of the predator is modeled by the are random integers, and σit signifies the scaling factor and is
production of R⃗ L and Elite while the predator motion was modeled obtained as follows:
( ) ( )
for the prey location renewing when the step size is added to the f P⃗i − f P⃗Best
Elite location. Also, FADs influence and Eddy formation directly σit = σinitial + σfinal − σinitial × (
( )
) ( ) (32)
effect on the algorithm. By the concept of the Fish Aggregating f P⃗Worst − f P⃗Best
Devices (FADs), above 80 percent of the time of the sharks was
spent near the FADs and the rest of it is used for a longer jump ( )
in different dimensions maybe for looking for the position with where, f P⃗Worst signifies the tth worst cost in the population,
various prey distribution. They were used as the local optima. and σinitial and σinitial represent two constants. The crossover
Therefore, noticing the extended jumps avoids from stucking in mechanism has been performed to form a trial vector, Vijt =
the local optima points. The process is achieved as follows: (Vi1t , Vi2t , . . . , Vidt ) as follows:
⎧ {
if r = r0 and rand (0, 1) ≤ cp
( )
⎨P⃗i + CF × X⃗min + R⃗ ⊗ (Xmax − Xmin ) ⊗ U⃗ if r ≤ pf qti,r
P⃗i = Vijt = (33)
) ( P⃗i other w ise
)
⎩P⃗i + pf × (1 − r ) + r × P⃗r − P⃗r if r > pf
(
1 2
where, cp describes the cross probability in the range [0, 1], r
(27)
signifies a d-dimension vector and r0 ∈ [1, 2, . . . , d] represents
where, U ⃗ shows the binary vector with arrays including 0 and a random dimension. Therefore, the population is updated by the
1, pf = 0.2 denotes the FADs impact maybe at the time of following:
the optimization process, r represents a uniform random amount { ( )
if jf Vijt < f P⃗i
( )
between 0 and 1, r1 and r2 explain the prey matrix ‘s random in- Vijt
P⃗i+1 = (34)
dices, and Xmin and Xmax signify the vector involving the minimum P⃗i other w ise
and maximum dimensions’ bounds.
3.3. AMPA validation
3.2. Adaptive marine predators algorithm
To validate the ability of the proposed method, it should
As before mentioned, the Marine Predators Algorithm (MPA)
be analyzed based on some different benchmark functions in-
introduces a new well-organized bioinspired algorithm which can
cluding Rastrigin function, Sphere function, Ackley function, and
be employed to solve different optimization problems. In the
solution space have been randomly distributed while there is Rosenbrock function and the results are compared with some
no adjacent prey near the current predator, an updating policy well-known metaheuristics to show its superiority. The compared
is performed to them. This reduces the convergence tendency algorithms here are: Lion Optimization Algorithm (LOA) (Shezan
and the convergence accuracy under the restricted number of et al., 2018), FOA (DiOrio et al., 2020), COA (Shezan et al., 2015),
iterations. The present study proposes an adaptive learning factor FOA (DiOrio et al., 2020), MPA (Shezan, 2019), and AMPA. Table 1
to resolve this subject. The qualified change rate for the predator tabulates the formulation of the benchmark functions utilized in
cost value is modeled by the following: this study. It is notable that the minimum value for all of the
⏐ ( ) ( )⏐ functions are assumed 0 and the dimension for all functions are
⏐ ⃗
⏐f Pi − f P⃗Best ⏐
⏐
set 50.
ρ= ( ) (28) Fig. 2 shows the three-dimensional diagram of the utilized
f P⃗Best + ε functions.
402
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Table 1
The main information of the benchmark functions utilized in this study.
Function Equation Limitation
d
∑
Sphere f1 (x) = x2i [−5.12, 5.12]
i=1

d
∑
x2i − 10 cos (2π xi ) + 10
( )
Rastrigin f2 (x) = [−1.28, 1.28]
i=1
 ⎛ ⎞ ( )
d d

1 ∑ 1 ∑
Ackley f3 (x) = −20 exp ⎝−0.2√ 2
xi ⎠ − exp cos (2π xi ) + 20 + exp(1) [−32, 32]
D D
i=1 i=1

d−1 ( )
∑ )2
f4 (x) = 100 × xi+1 − x2i + (xi − 1)2
(
Rosenbrock [−10, 10]
i=1

Table 2
The parameters setting information utilized in the comparative algorithms.
Algorithm Parameter Value Algorithm Parameter Value
Number of prides 5 Step size control variable 0.1
COA
Percent of nomad lions 0.3 β 2
Roaming percent 0.4 View 0.3
FOA
LOA Mutate probability 0.1 Predation attempts 12
Sex rate 0.85 P 0.5
Mating probability 0.4 MPA, AMPA α 0.1
Immigrate rate 0.5 pf 0.2

Fig. 2. The sample 2d profile of the utilized functions.

The main parameters setting utilized in the studied optimiza- 4. The ε-constraint method
tion algorithms are tabulated in Table 2.
To validate the results of the studied algorithms, four indica-
Several techniques have been designed and introduced for
tors are used as follows. The first one is the minimum value of
multi-objective optimization algorithms. One of the simplest
the objective function (Min). This indicator shows the accuracy
methods for solving these types of methods is to use ε -constraint.
of each algorithm. the second indicator is the maximum value
In ε -constraint method, all of the objective functions except one
of the cost functions (Max) which shows that how much one
have turned to a limited constraint with target levels, εi while,
algorithm gives an uncertain result. The third indicator shows the
the proper solutions of the problem are obtained by varying the
average value of the cost function for the algorithms to show
levels. For more clarification, consider a multi objective problem
the average precision of each analyzed algorithm. Finally, the
such that:
last indicator declares the standard deviation of cost functions
for each algorithm. In this study, these indicators are achieved Min(f1 (x) , f2 (x) , . . . , fn (x)) (35)
based on 45 independent runs of the algorithms. Table 3 indicates
the simulation results of the algorithms applied to the analyzed where, fi (i = 1, 2, . . . , n) signifies the objective functions, x rep-
functions. resents the optimization parameters, and n describes the number
As can be observed from Table 3, the Min indicator in the pro- of objective functions. By considering x∗ as optimal solution, there
posed algorithm give the smallest value which shows its higher is no better solution x for the problem such that fi (x) ≤ fi (x∗ ).
accuracy toward the others and due to minimum value of the Here, while fk (k = 1, 2, . . . , n) is assumed as considered objec-
Max indicator, its superiority in controlling drop data toward tive function, and fi (i = 1, 2, . . . , n) represents the considered
the other algorithms is determined. Also, the minimum value objective as the constraint (Coppitters et al., 2019):
of the standard deviation of the functions indicate the higher
reliability of the algorithm toward the others. Fig. 3 demonstrates Minfk (x) (36)
the convergence analysis of the comparative algorithms.
Subject to
The results indicated that for all of the benchmark functions,
the proposed AMPA gives the fastest convergence results with fi (x) ≤ εi (37)
a good trade-off within the accuracy which shows its ability in
escaping from the premature convergence. where, i ̸ = k, εi represents a limited restriction.
403
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Fig. 3. The convergence analysis of the compared algorithms.

404
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Table 3
The simulation results of the algorithms performed on the analyzed functions.
Algorithm f1 f2 f3 f4
Min 11.3769 0.0008 0.0012 1.3315
Max 3.2068e+3 0.4857e+2 0.5748e+5 6.3017
LOA Shezan et al. (2018)
Mean 1.0001e+3 3.3694e+2 1.0471e+4 5.3139
std 2.0104e+4 1.4259e+2 2.0549e+4 6.0194
Min 9.3451 0.0206 1.2954e−3 1.0416
Max 387.5175 1.4321 3.5149e−2 34.1249
COA Shezan et al. (2015)
Mean 103.2549 1.0427 2.0845e−2 15.4719
std 71.1948 0.0378 1.3807e−2 3.0716
Min 4.0176 1.4518e−5 2.1207e−5 0.9741
Max 3.0168e+2 0.0068 0.0028 1.0048
FOA DiOrio et al. (2020)
Mean 82.1984 0.0057 0.0026 1.0037
std 234.2493 0.0043 2.0045e−5 1.0041
Min 0.4072 8.3316e−21 4.1342e−11 0.0011
Max 15.0638 1.1031e−19 1.0541e−10 3.1319
MPA Shezan (2019)
Mean 11.4715 1.1650e−19 2.1504e−10 0.0574
std 2.7602 1.0350e−19 3.1514e−11 0.1285
Min 0.0043 2.0948e−22 1.1982e−11 4.1418e−15
Max 2.1537 1.0741e−21 1.3745e−10 1.5218e−14
AMPA
Mean 1.6142 0.1311 e−21 1.0041e−10 1.0210e−14
std 1.0401 0.0184 e−21 4.0652e−11 5.0340e−15

5. The objective function Table 4

Technical and economic information of the components of the desired energy
hybrid system.
5.1. The main objective function
Converter: Initial cost 4850 $/kWpeak
Capacity 25 kW Project lifetime 25 years
Based on ε -constraint technique, the present study considers Initial cost 15000 $ Inflation rate 0.03
the CO2 emission and the annual Loss of Load Probability of the Replacement cost 15000 $ Efficiency 15%
system (LLP) as the limited constraints and the annual cost of the Efficiency 95% MPPT efficiency 93
system (ACS) as the main objective function which is achieved as Project lifetime 20 years BESS:
DG: Technology Lead–acid
follows:
Technology VOLVO-GSW Initial cost 8000 $/kWpeak
ACS = CRa + Cca + CFa + Coa m (38) Nominal power 50 kW Replacement cost 6000 $/kWpeak
Voltage 400 v Project lifetime 5 years
where, Cca
represents the annual capital cost, describes theCFa Frequency 50 Hz Efficiency 80%
a Initial cost 10000 $ Inflation rate 7.84
annual fuel cost, Co&m determines the annual operation and main-
Project lifetime 25 years Charging efficiency 0.7975
tenance cost, and CRa signifies the annual replacement cost.
Fuel cost 0.12 $/L Discharging efficiency 1/0.883
The annual replacement cost (CRa ) determines the battery stor- Emission factor 1.85 kg/L Self-discharge 0.0188
age replacing during the lifetime and is mathematically modeled PV Panel: Power cost coefficient 402 $/kW
as follows Shezan et al. (2016): Technology Polycrystalline Energy cost coefficient 120.6 $/kWh

CRa = CRBat × τs (rir , tR ) (39)

where, CRBat signifies
the battery replacement cost, tR describes a
Also, the annual operation and maintenance cost, i.e. Co&m is
the lifetime of the annual battery storage system, and τs defines
achieved as follows:
the sinking fund factor which is mathematically achieved as
follows Shezan et al. (2016): Coa m = Co&m
1
× (1 + rri )τ (44)
rri
τs = (40) where, 1
Co&m signifies the maintenance cost of the devices in the
(1 + rri )tR − 1 initial process.
where, ri describes the annual real interest rate. Table 4 tabulates the technical and economic information of
The annual capital cost of the system during the lifetime the components for the desired energy hybrid system.
without considering the replacement cost is obtained by the
following: 5.2. The function constraints
Cca = Cd × FcR (i, tl ) (41)
The selected constraints contain the Loss of load probability
where, Cd defines the capital cost of every single devices, τ
(LLP) and the CO2 emissions in the system to characterize the
describes the annual project lifetime, and FcR describes the capital
generation probability which is not satisfactory to supply the
recovery factor and is formulated below:
load demand at the some point during a definite period and is
rri × (1 + rri )τ obtained by the following equation (Bahramara et al., 2020; Feng
FcR = (42)
(1 + rri )τ − 1 et al., 2020):
where, rri describes the annual real interest rate which is formu- 8800
∑ Ul (t )
lated below: LLP = (45)
(rn − F ) De (t )
t =1
rir = (43)
(1 + F ) where, Ul (t) and De (t) represent the unsatisfied load and the
where, rn and F represent the nominal interest and the annual electricity demand during time period t, and LLP ≤ εLLP where
inflation ratios, respectively. εLLP stands for the allowable LLP reliability index.
405
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

the energy flow among different devices have been calculated.

Then, the value of the annual value for the CO2 emission and
the LLP is evaluated. Afterward, the objective function has been
evaluated and the parameters have been updated until the ter-
mination condition has been reached and giving the best value of
the function. The stopping criteria in this study is based on the
number of iterations, i.e. 100.

7. Simulation results

The simulation results were performed with the configuration

of Windows 10 Operating System (64-bit) with Intel⃝ R
CoreTM
i5-8250 CPU @ 1.80 GHz Processor, 4 GB memory and 1 TB
HDD under MATLAB R2017b environment. The experiments are
regarded based on hourly simulation under a certain load. The
assumption is that the generator can satisfy the primary load
demand and the DG is utilized as a backup system when PV and
battery energy storage systems cannot satisfy the load demand.
It is important to note that due to the average wind speed of
Fig. 4. The diagram flowchart of the proposed method. 5.28 m/s in Hoxtolgay, the use of wind turbine in that area in
not very economical. Also, the wind turbine has the following
shortcomings that we do not like to have them: The wind is
The generated CO2 based on the diesel generator is required to unpredictable, so it is not reliable on its own, wind turbines
describe the pollutant emission. This indicator can be formulated produce noise pollution that is not a favorite in the considered
as follows: area. Installing wind turbines requires a high initial cost that is
8800
∑ limited in this case. Wind turbines are very vulnerable to different
ECO2 = Fc (t ) × FE (46) conditions such as lightning strikes, strong winds, etc.
t =1

where, FE signifies the DG fuel consumption which is limited 7.1. Case study
in the range [2.4 kg/L, 2.8 kg/L] (Bahramara et al., 2020), and
the ECO2 ≤ εLLP , where εLLP stands for the allowable emission The presented method is designed to perform a case study
level. There are also some different feasible constraints that are in Hoxtolgay. Hoxtolgay is an oasis town in Hoboksar Mongol
explained in the following. Autonomous County, China with an area of 6684 km2 which is
The number constraints for the photovoltaic modules, DG, and placed at 46◦ 34′ N 85◦ 58′ E. The town’s population consists of,914
batteries is as follows: rural residents. This area is located in Gurbantünggüt Desert, the
farthest point from the sea. The summers in Hoxtolgay are long
N Bat , N PV , PnDG ≥ 0 (47) and warm and dry. The temperature generally varies from −15◦ C
The capacity constraint for battery storage is as follows: to 31◦ C . The analyzed HRES case study includes about 25 rural
households located in the Hoxtolgay rural area. Fig. 5 shows the
Bat
Cmin ≤ C Bat (t ) ≤ Cmax
Bat
(48) Google map of this area (Fu, 2017).
And the constrained for produced energy by the devices (Ej (t)) The load profile of the studied area in Hoxtolgay is con-
is as follows sidered by assuming the traditional electrical tools such as TV,
Fluorescent-based lighting, air conditioner, and refrigerator. This
Ej (t ) ≤ Pj ∆t (49) load profile is provided for winter and summer during 1-hour
period. Fig. 6 shows the load profile of the studied area.
where, ∆t describes the time interval (∆t = 1 h)
Based on Fig. 6, the maximum and the minimum demand of
6. Optimization process the system are 5.2 kW and 1 kW for winter, and 1.3 kW and 4.2
kW for summer, respectively. The synthetic hourly NASA Surface
Based on the explanations before in the previous sections, Meteorological dataset is used for the analysis. These meteorolog-
the main idea in this paper is to apply a multi-objective opti- ical data by considering the clearness index, daily radiation, and
mization by considering the system total cost, the loss of load the daily temperature for the analyzed system.
probability (LPP), and the total CO2 emissions generated by the
DG. For simplifying the problem, it is tuned into a constrained 7.2. Optimization based on AMPA-based methodology
single-objective optimization algorithm based on the ε -constraint
technique and then the problem is solved by the suggested AMPA. The information for parameters setting of the AMPA have been
The decision variables for optimization by the algorithm are the formerly in Table 3. The predators population is set 100 such
size of PV panels, the DG rated capacity, and the battery storage that each predator indicates various arrangement of the system
size, i.e. including a 3 × 1 vector. During the optimization, the minimum
and the maximum capacities for the DG are 2 kW and 5 kW,
z = PDG , N Bat , N PV
[ ]
(50)
respectively and the minimum and the maximum numbers of
Such that each of the decision variables determine differ- battery cells and the PV panels are assumed:5, 55 and 5, 110,
ent arrangement for the system. The diagram flowchart of the respectively. The idea of using the suggested AMPA is to optimal
suggested method is shown in Fig. 4. selection of the aforementioned decision variables to minimize
As can be observed from Fig. 4, after initializing of the data, the annual cost of the system (ACS) subject to CO2 emissions
the algorithm has been initialized. In the following, the value of as constraints and Loss of Load Probability of the system in the
406
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Fig. 5. The google map of Hoxtolgay area (Fu, 2017).

Fig. 6. The load profile of the analyzed area.

Fig. 7. The information about clearness index and the daily radiation.

Table 5 achieved. The simulation results of intra-hourly energy fluctua-

The economical results of the simulation.
a
tion during a definite period for summer (20th of July) and winter
Costca ($) Costo&m ($) CostRa ($) CostFa ($) ACS ($)
(the 6th of January) are indicated in Figs. 9 and 10, respectively.
Converter 409.74 186.14 187.54 0 835.57
Battery bank 1583.4 784.51 1672.5 0 4266.75
During the simulation, in the first step, the PV enters to the
PV subsystem 2581.3 107.62 0 0 2670.53 system to give the needed load demand. While the PV power
Diesel generator 210.33 88.39 0 109 451.30 is insufficient to supply the load demand, the battery energy
Total 4784.77 1166.66 1860.04 109 8224.15
storage system enters as an auxiliary part to provide the power
deficiency. At last, if the combination of the PV and the BESS could
not provide enough power for the demand, the DG enters to the
feasible levels (εCO2 , εLLP ). Table 5 tabulates the economical results system as the final backup system. The monthly power generation
(see Figs. 7 and 8). changes for the DG, the PV system, and the storage capacity of the
It is observed that the total ACS for the optimal system is BESS is shown in Fig. 11.
8224.15 $. Also, an optimal value of 451.30 kW for diesel genera- The results of Fig. 11 observed that the generated power based
tor, 4266$ for battery storage, and 2670.53 $ for PV generator are on the PV gives a direct impact on the BESS with a reverse impact
407
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

sensitivity analysis in this study is based on variation in εLLP , εCO2 ,

and the load consumption. More explanations are given in the
following.

7.3.1. Sensitivity analysis of the load consumption

To do so, the assumption is that the load changes from 1
kWh/day to 40 kWh/day while the other elements such as the
DG, the PV size, and the BESS are fixed. Table 6 illustrates the
results for sensitivity analysis on the load consumption
From Table 6, it is observed that the dependence to the LLP
enhances from 0% to 8.56% and the CO2 emissions increases from
2316.6 kg/yr to 13315 kg/yr. The DG enhances the value of the
ACS from 8440.3 $ to 9187.9 $ (about 1.09% increasing) over the
Fig. 8. The information about the temperature of the Hoxtolgay.
fuel increasing in the consumption. It is clear that the COE reduces
from 0.35 $/kWh to 0.22 $/kWh over the PV penetration reducing
from 94.92% to 61.68%. Decreasing in this value shows that the
on the DG system. The results show that the maximum storage is effect of the fuel is too operative than the PV cost on the COE,
happened in March and October. due to the less cost predictable power production using the PV
system.
7.3. Sensitivity analysis of the system
7.3.2. Sensitivity analysis of the εCO2
For assessment of the reliability of the studied system under Table 7 tabulates the effect of this parameter on the battery,
different conditions, the sensitivity analysis is applied to it. The DG and the PV sizes. The results demonstrate that the εCO2 has an

Fig. 9. The simulation results of intra-hourly energy fluctuation during a definite period for summer.

408
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Fig. 10. The simulation results of intra-hourly energy fluctuation during a definite period for winter.

Table 6
Results for sensitivity analysis on the load consumption.
The surplus load (kWh/day) Co2 (kg/year) COE ($/kWh) PV penetration (%) LLP (%) ACS (US$)
1 2316.6 0.35 94.92 0 8440.3
2 2705.4 0.34 93.55 0 8467.7
3 3214.7 0.34 92.43 0 8508.8
4 3604.5 0.33 90.95 0 8543.6
5 4042.8 0.33 89.76 0 8568.1
6 4418.2 0.33 88.57 0 8602.3
7 4783.9 0.32 87.68 0 8612.5
12 6314.5 0.31 82.98 0 8718
15 6957.7 0.30 80.37 0 8765.6
20 8112.8 0.28 76.74 0.78 8838.8
25 9533 0.26 72.76 2.23 8939
30 10674 0.24 68.62 4.63 9011.2
35 11865 0.23 64.35 6.35 9094.4
40 13315 0.22 61.68 8.56 9187.9

important impact on the BESS size with a such that its size has 7.3.3. Sensitivity analysis of the εLLP
been increased by 1.78% (26.9 kWh to 48 kWh) during variations The results of the variations of this parameter are specified in
of the εCO2 . Table 8. As it is clear in Table 8. The most effect on the εLLP is
409
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Fig. 11. The monthly power generation changes for the DG, the PV system, and the storage capacity of the BESS.

Table 7 Table 8
The effect of this parameter on the battery, DG and the PV sizes. The results of the variations of the εLLP on the system.
εCO2 (kg/ year) PV power (kWp) Battery storage capacity (kWh) DG (kW) εLLP (kg/year) PV power (kWp) Battery storage capacity (kWh) DG (kW)
50 20.5 48 6 0 8 24 5
80 20.4 48 6 0.1 7.8 21 4
100 19.6 47 6 0.2 7 21 4
200 188 47 6 0.3 6.9 21 4
300 17.4 45 6 0.4 6.8 21 4
400 15.8 47 6 0.5 7 19 4
500 15.1 46 6 0.6 7.3 19 4
1000 13.2 46.5 6 0.7 7.3 19 4
2000 12.3 43.9 6 0.8 7.3 19 4
3000 11.8 39.3 6 1 7.2 19 4
4000 10.8 38.7 6 1.5 7.3 19 4
5000 10.5 37.8 6 2 7.3 19 4
6000 10.0 31.5 6
7000 9.6 26.9 6

NPC. In this case, the maximum cost of the system with 43.12%
belongs to the PV system of total cost of the system.
achieved by the BESS so that it reduces from 24 kWh to 19 kWh
At last, Table 10 illustrates the results of the studied algo-
when εLLP changes from 0%–2%.
rithms on the case study. The comparison between the proposed
Table 9 indicates the economic results of the system. Here, the AMPA and the other state of the art methods (PSO (Fodhil et al.,
initial capital cost of the system is 47940 $ that is less than the 2019) and HOMER software) shows that the has best results. It
410
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Table 9
The economic results of the system.
Net present costs
Component Replacement ($) Capital ($) Fuel ($) O&M ($) Salvage ($) Total ($)
PV 0 24540 0 2350 0 26890
Generator 1 0 2000 1085 1850 −140 4795
Hoppecke 10 OPzS 1500 9570 15300 0 1220 −1130 24960
Converter 1520 4150 0 415 −370 5715
System 11090 45990 1050 5835 −1640 62360

Table 10
The results of the studied algorithms on the case study.
Parameter Algorithm
AMPA PSO Fodhil HOMER
et al. (2019)
Total production (kWh/yr) 22713 23260 22840
PV production (kWh/yr) 22232 21839 20832
Input Energy for the batteries (kWh/yr) 9259 9378 9649
Fuel consumption (L/yr) 674 684 709
Output Energy for the batteries (kWh/yr) 9570 9570 8380
CO2 emissions (kg/yr) 1624 1750 2000
DG production (kWh/yr) 470 1311 1990

also can be concluded that the suggested AMPA has the most Declaration of competing interest
renewable penetration.
Furthermore, CO2 emission by the suggested AMPA with The authors declare that they have no known competing finan-
1624 kg/year is the minimum value with cleanest result among cial interests or personal relationships that could have appeared
the others. to influence the work reported in this paper.

8. Conclusions References

Lack of national electricity grid in remote areas, high cost Abedinia, Oveis, et al., 2019. Optimal offering and bidding strategies of renew-
able energy based large consumer using a novel hybrid robust-stochastic
of construction of new transmission line due to long distance,
approach. J. Cleaner Prod. 215, 878–889.
increasing electricity consumption rate was one of the biggest Aghajani, Gholamreza, Ghadimi, Noradin, 2018. Multi-objective energy manage-
problems of electricity companies. In these areas, the required ment in a micro-grid. Energy Rep. 4, 218–225.
electrical power was often satisfied by the DG. The fuel used was Akbary, Paria, et al., 2019. Extracting appropriate nodal marginal prices for all
very expensive and pollutes the carbon dioxide. In this paper, an types of committed reserve. Comput. Econ. 53 (1), 1–26.
Alturki, F.A., Al-Shamma’a, A.A., Farh, H.M., AlSharabi, K., 2020. Optimal sizing of
optimal combined PV/diesel/battery system was investigated to autonomous hybrid energy system using supply–demand-based optimization
provide the required power in remote areas. The studied area was algorithm. Int. J. Energy Res..
Hoxtolgay, China. Three objectives functions, i.e., CO2 emissions Bahramara, S., Mazza, A., Chicco, G., Shafie-khah, M., Catalão, J.P., 2020. Com-
value, the loss of load probability, and the annualized cost of prehensive review on the decision-making frameworks referring to the
distribution network operation problem in the presence of distributed energy
the system were considered. To simplify the mentioned multi-
resources and microgrids. Int. J. Electr. Power Energy Syst. 115, 105466.
objective function, ε -constraint was utilized. The optimization Cai, Wei, et al., 2019. Optimal bidding and offering strategies of compressed
algorithm was optimized using a new adaptive version of Marine air energy storage: A hybrid robust-stochastic approach. Renew. Energy 143,
Predators Algorithm (AMPA) to improve the system efficiency. 1–8.
The system was then simulated based on the proposed method Cao, Y., Wu, Y., Fu, L., Jermsittiparsert, K., Razmjooy, N., 2019. Multi-objective
optimization of a PEMFC based CCHP system by meta-heuristics. Energy Rep.
and the results were analyzed. The total ACS for the optimal sys-
5, 1551–1559.
tem was obtained 8224.15 $. Correspondingly, an optimal value Coppitters, D., De Paepe, W., Contino, F., 2019. Surrogate-assisted robust de-
of 451.30 kW was achieved for diesel generator, along with 4266$ sign optimization and global sensitivity analysis of a directly coupled
for battery storage, and 2670.53 $ for PV generator are achieved. photovoltaic-electrolyzer system under techno-economic uncertainty. Appl.
The initial capital cost of the system was achieved 47940 $ that is Energy 248, 310–320.
DiOrio, N., Denholm, P., Hobbs, W.B., 2020. A model for evaluating the con-
less than the NPC. The maximum cost of the system with 43.12% figuration and dispatch of PV plus battery power plants. Appl. Energy 262,
belongs to the PV system of total cost of the system. Finally, CO2 114465, 2020/03/15/ 2020.
emission by the suggested AMPA with 1624 kg/year indicated Dufo-Lopez, R., Bernal-Agustín, J.L., 2008. Multi-objective design of PV–wind–
minimum value with cleanest result among the others. A perfor- diesel–hydrogen–battery systems. Renew. Energy 33 (12), 2559–2572.
Ebrahimian, Homayoun, et al., 2018. The price prediction for the energy market
mance comparison is also made between the proposed method
based on a new method. Econ. Res. Ekonom. Istraživanja 31 (1), 313–337.
and the PSO-based optimal system and the HOMER software to Eslami, Mahdiyeh, et al., 2019. A new formulation to reduce the number of
indicate the performance of the method. variables and constraints to expedite SCUC in bulky power systems. Proc.
Natl. Acad. Sci. India A 89 (2), 311–321.
CRediT authorship contribution statement Fan, X., Sun, H., Yuan, Z., Li, Z., Shi, R., Razmjooy, N., 2020a. Multi-objective
optimization for the proper selection of the best heat pump technology in
a fuel cell-heat pump micro-CHP system. Energy Rep. 6, 325–335.
Guoqing Yu: Conceptualization, Data curation, Writing - orig- Fan, Xiaochao, et al., 2020b. High voltage gain DC/DC converter using coupled
inal draft, Writing - review & editing. Zhiyong Meng: Conceptu- inductor and VM techniques. IEEE Access 8, 131975-131987.
alization, Data curation, Writing - original draft, Writing - review Fei, X., Xuejun, R., Razmjooy, N., 2019. Optimal configuration and energy
& editing. Hongtao Ma: Conceptualization, Data curation, Writing management for combined solar chimney, solid oxide electrolysis, and fuel
cell: a case study in Iran. Energy Sources A 1–21.
- original draft, Writing - review & editing. Lei Liu: Conceptual- Feng, Y., Hao, W., Li, H., Cui, N., Gong, D., Gao, L., 2020. Machine learning models
ization, Data curation, Writing - original draft, Writing - review & to quantify and map daily global solar radiation and photovoltaic power.
editing. Renew. Sustain. Energy Rev. 118, 109393.

411
G. Yu, Z. Meng, H. Ma et al. Energy Reports 7 (2021) 398–412

Firouz, Mansour Hosseini, Ghadimi, Noradin, 2016. Concordant controllers based Mir, Mahdi, et al., 2019. Application of hybrid forecast engine based intelligent
on FACTS and FPSS for solving wide-area in multi-machine power system. J. algorithm and feature selection for wind signal prediction. Evol. Syst. 1–15.
Intell. Fuzzy Systems 30 (2), 845–859. Mirzapour, Farzaneh, et al., 2019. A new prediction model of battery and wind-
Fodhil, F., Hamidat, A., Nadjemi, O., 2019. Potential, optimization and sensitiv- solar output in hybrid power system. J. Ambient Intell. Humanized Comput.
ity analysis of photovoltaic-diesel-battery hybrid energy system for rural 10 (1), 77–87.
electrification in Algeria. Energy 169, 613–624. Mohammadi, Mohsen, et al., 2018. Small-scale building load forecast based on
Fu, Z., 2017. Dun Huang Caves-a Trove of 1000+ Years of Art and History hybrid forecast engine. Neural Process. Lett. 48 (1), 329–351.
Preserved in the Gobi Desert of Gansu, China. Najafi Ashtiani, M., Toopshekan, A., Razi Astaraei, F., Yousefi, H., Maleki, A., 2020.
Gao, Wei, et al., 2019. Different states of multi-block based forecast engine for Techno-economic analysis of a grid-connected PV/battery system using the
price and load prediction. Int. J. Electr. Power Energy Syst. 104, 423–435. teaching-learning-based optimization algorithm. SoEn 203, 69–82.
Ghadimi, Noradin, 2014. MDE with considered different load scenarios for Nurunnabi, M., Roy, N.K., Hossain, E., Pota, H.R., 2019. Size optimization and sen-
solving optimal location and sizing of shunt capacitors. Natl. Acad. Sci. Lett. sitivity analysis of hybrid wind/PV micro-grids- A case study for Bangladesh.
37 (5), 447–450. IEEE Access 7, 150120-150140.
Ghadimi, Noradin, 2015a. An adaptive neuro-fuzzy inference system for islanding Saeedi, Mohammadhossein, et al., 2019. Robust optimization based optimal
detection in wind turbine as distributed generation. Complexity 21 (1), chiller loading under cooling demand uncertainty. Appl. Therm. Eng. 148,
10–20. 1081–1091.
Ghadimi, Noradin, 2015b. A new hybrid algorithm based on optimal fuzzy Shezan, S.A., 2019. Optimization and assessment of an off-grid photovoltaic-
controller in multimachine power system. Complexity 21 (1), 78–93. diesel-battery hybrid sustainable energy system for remote residential
Gheydi, Milad, Nouri, Alireza, Ghadimi, Noradin, 2018. Planning in microgrids applications. Environ. Progr. Sustain. Energy 38 (6).
with conservation of voltage reduction. IEEE Syst. J. 12 (3), 2782–2790. Shezan, S., Al-Mamoon, A., Ping, H., 2018. Performance investigation of an
Gong, W., razmjooy, N., 2020. A new optimisation algorithm based on OCM and advanced hybrid renewable energy system in indonesia. Environ. Progr.
PCM solution through energy reserve. Int. J. Ambient Energy 1–14. Sustain. Energy 37 (4), 1424–1432.
Hagh, M.T., Ebrahimian, H., Ghadimi, N., 2015. Hybrid intelligent water drop Shezan, S.A., Julai, S., Kibria, M., Ullah, K., Saidur, R., Chong, W., Akikur, R., 2016.
bundled wavelet neural network to solve the islanding detection by Performance analysis of an off-grid wind-PV (photovoltaic)-diesel-battery
inverter-based DG. Front. Energy 9 (1), 75–90. hybrid energy system feasible for remote areas. J. Cleaner Prod. 125 (2016),
Hamian, Melika, et al., 2018. A framework to expedite joint energy-reserve 121–132.
payment cost minimization using a custom-designed method based on Shezan, S.A., Saidur, R., Ullah, K.R., Hossain, A., Chong, W.T., Julai, S., 2015.
mixed integer genetic algorithm. Eng. Appl. Artif. Intell. 72, 203–212. Feasibility analysis of a hybrid off-grid wind-DG-battery energy system for
Hosseini Firouz, M., Ghadimi, N., 2016. Optimal preventive maintenance policy the eco-tourism remote areas. Clean Technol. Environ. Policy 17, 2417–2430.
for electric power distribution systems based on the fuzzy AHP methods. Shivaie, M., Mokhayeri, M., Kiani-Moghaddam, M., Ashouri-Zadeh, A., 2019.
Complexity 21 (6), 70–88. A reliability-constrained cost-effective model for optimal sizing of an au-
Jeong, D., Tyner, W.E., Meilan, R., Brown, T.R., Doering, O.C., 2020. Stochastic tonomous hybrid solar/wind/diesel/battery energy system by a modified
techno-economic analysis of electricity produced from poplar plantations in discrete bat search algorithm. Sol. Energy 189, 344–356.
Indiana. Renew. Energy 149, 189–197. Sobu, A., Wu, G., 2012. Optimal operation planning method for isolated micro
Khodaei, Hossein, et al., 2018. Fuzzy-based heat and power hub models grid considering uncertainties of renewable power generations and load
for cost-emission operation of an industrial consumer using compromise demand. In: IEEE PES Innovative Smart Grid Technologies. IEEE, pp. 1–6.
programming. Appl. Therm. Eng. 137, 395–405. Viswanathan, G.M., Buldyrev, S.V., Havlin, S., Da Luz, M., Raposo, E., Stanley, H.E.,
Leng, Hua, et al., 2018. A new wind power prediction method based on ridgelet 1999. Optimizing the success of random searches. Nature 401 (6756),
transforms, hybrid feature selection and closed-loop forecasting. Adv. Eng. 911–914.
Inform. 36, 20–30. Yang, R., Yuan, Y., Ying, R., Shen, B., Long, T., 2020. A novel energy management
Liu, Yang, Wang, Wei, Ghadimi, Noradin, 2017. Electricity load forecasting by an strategy for a ship’s hybrid solar energy generation system using a particle
improved forecast engine for building level consumers. Energy 139, 18–30. swarm optimization algorithm. Energies 13 (6), 1380.
Liu, Jun, et al., 2020. An IGDT-based risk-involved optimal bidding strategy for Ye, Haixiong, et al., 2020. High step-up interleaved dc/dc converter with high
hydrogen storage-based intelligent parking lot of electric vehicles. J. Energy efficiency. Energy Sources A 1–20.
Storage 27, 101057. Yu, Dongmin, Ghadimi, Noradin, 2019. Reliability constraint stochastic UC by
Makhdoomi, S., Askarzadeh, A., 2020. Optimizing operation of a photo- considering the correlation of random variables with copula theory. IET
voltaic/diesel generator hybrid energy system with pumped hydro storage Renew. Power Gener. 13 (14), 2587–2593.
by a modified crow search algorithm. J. Energy Storage 27, 101040. Yuan, Zhi, et al., 2020. Probabilistic decomposition-based security constrained
Meng, Q., Liu, T., Su, C., Niu, H., Hou, Z., Ghadimi, N., 2020. A single-phase transmission expansion planning incorporating distributed series reactor. IET
transformer-less grid-tied inverter based on switched capacitor for PV Gener. Transm. Distrib. 14 (17), 3478–3487.
application. J. Control Automat. Electr. Syst. 31 (1), 257–270.

412