Energy 154 (2018) 581e591

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Optimizing a hybrid wind-PV-battery system using GA-PSO and

MOPSO for reducing cost and increasing reliability
Narges Ghorbani a, Alibakhsh Kasaeian b, *, Ashkan Toopshekan b, Leyli Bahrami b,
Amin Maghami b
a
Lappeenranta University of Technology, Skinnarilankatu 34, 53850 Lappeenranta, Finland
b
Department of Renewable Energies, Faculty of New Sciences and Technologies, University of Tehran, Tehran, Iran

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

Article history: In this paper, a hybrid genetic algorithm with particle swarm optimization (GA-PSO) is applied for the
Available online 28 December 2017 optimal sizing of an off-grid house with photovoltaic panels, wind turbines, and battery. The GA-PSO is
one of the most powerful single-objective optimization algorithms. In the other hand, the multi-
Keywords: objective PSO (MOPSO) can solve the optimization problems considering all objectives without trans-
Optimization forming them. Minimizing the total present cost including initial cost, operation and maintenance cost,
Wind
and replacement cost with satisfying the load demand is the main goal of this study. In this optimization
Photovoltaic
problem, the considered reliability factor is a loss of power supply probability, which specifies the
Battery
GA-PSO
subtraction of the load power and generated power. The wind velocity, solar irradiance, and load demand
MOPSO are simulated in 12 months of a year by the HOMER software for a suburbs of Tehran. Then, the optimal
size of PV and WT are obtained with both GA-PSO and MOPSO methods, and compared with the HOMER
results. At last, the strengths and weaknesses of each method are explained. The results show that the
proposed approach with 0.502 of the levelized cost of energy for the PV/WT/BAT system has the best
result through the compared methods.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction system to the power generation circuit at any particular time. The
design optimizing and effective sizing of hybrid energy systems are
The technology development, population growth and increasing essential to increase the performance and reliability, meet the
energy demand and consequently increasing the cost and green- external load demand, reduce the energy cost and net present cost
house gas emissions, all resulted from fossil fuels, have led to more (NPC), and minimize the greenhouse gas emissions [6]. Therefore,
attention to renewable energy resources. The wind and solar en- designing a hybrid energy system, which is economically and
ergies are the most accessible and important renewable resources technically justified, requires the multi-objective optimization
which can be integrated together to construct a hybrid energy stages. In general, the sizing methodologies for the optimal
system with higher quality and reliability than the single resources designing of hybrid PV/WT generating systems can be divided into
[1e3]. In fact, the wind and solar energies are the base resources in four categories including probabilistic, analytical, iterative, and
a hybrid energy system. The batteries energy storage, fuel cells (FC) hybrid methods [6]. The literature study indicates that these
and diesel generators can be applied to hybrid systems to increase methods have been developed, as single-objective and multi-
the efficiency and remove the shortcomings. In fact, when the wind objective, in the form of numerical, analytical and optimization
speed or solar radiation decreases or a peak demand occurs, the means or through different commercial software.
existence of these storage units becomes essential [4,5]. Hybrid The studies show that in different single objective problems the
energy systems can operate both stand-alone and grid-connected, total cost is the most applied cost function. Yang et al. [7] optimized
and also, a control unit can be added for connecting the hybrid the design variables (number of the PV modules, number of wind
turbines, number of batteries, the PV module slope angle, and the
wind turbine installation height) of a hybrid Solar/Wind/Battery
system to achieve the desired loss of power supply probability
* Corresponding author.
(LPSP) with minimum annualized cost of system (ACS) concepts
E-mail address: [email protected] (A. Kasaeian).

https://doi.org/10.1016/j.energy.2017.12.057
0360-5442/© 2017 Elsevier Ltd. All rights reserved.
582 N. Ghorbani et al. / Energy 154 (2018) 581e591

Nomenclature PBAT ðtÞ Battery charge at t

PBAT ðtÞ Battery charge at t(kW)
A Wind turbine rotor swept area (m2 ) Pin Power injected by PV and WT (kW)
C Power coefficient of wind turbine PLoad ðtÞ Load demand at t (kW)
c Constant weighting parameter PPV PV panel capacity (kW)
CBAT Per unit cost of battery ($) PPV ðtÞ PV power at t (kW)
CINV Per unit cost of inverter ($) PPV rated PV rated power for G ¼ 1000 w=m2
CM;BAT Maintenance cost of batteries ($/kWh) PWT Wind turbine capacity (kW)
CM;PV Maintenance cost of PV panel ($/kW) PWT ðtÞ Wind turbine power at t (kW)
CM;WT Maintenance cost of wind turbine ($/kW) PWT rated Wind turbine rated power
CPV Per unit cost of PV panel ($/kW) PF Penalty factor
CWT Per unit cost of wind turbine ($/kW) R Random parameter
CI Initial cost ($) T Life cycle of system
CM Maintenance cost ($) TC Terminal criterion
CR Replacement cost ($) V Wind speed (m/s)
CT Total cost V Particle speed vector
Et Total energy generated by the system in one year VCi Cut-in wind speed (m/s)
G solar radiation at the PV surface (w=m2 ) VCo Cut-out wind speed(m/s)
gbest Global best position W Weighting parameter
X Particle position
pbest Particle best position
H Height (m)
r Air density
InfR Inflation rate
a Hellman coefficient
IntR Interest rate
s Battery self-discharge rate
LPS Loss of power supply
hmppt PV module efficiency
LPSP Probability of loss of power supply hREC Rectifier efficiency
LCE Levelized cost of energy ($/kWh) hBAT Charge efficiency of battery system
NBAT Number of batteries hCON Converter efficiency
NINV Number of inverters hINV Inverter efficiency
NPV Number of PV panels GA genetic algorithm
fail EMS Energy management system
NPV Number of failed PV panels
PSO Particle swarm optimization
NWT Number of wind turbines GA-PSO Genetic algorithm particle swarm optimization
fail
NWT Number of failed WT MOPSO Multi-objective particle swarm optimization
NPC Net present cost WT Wind turbine
OF Objective function

through the GA. Kornelakis and Marinakis [8] used PSO as a single emissions for optimizing the system design parameters and mini-
objective optimization algorithm for optimizing different design mizing the life cycle cost. Bhattacharjee and Acharya [14] carried
parameters of PV grid-connected systems, such as the PV modules out a techno-economic analysis by the HOMER software for a PV/
number, in order to achieve to the maximum net economic benefit WT hybrid unit in an educational building located in the north-east
during the period of the system usage. Khoury et al. [9] designed a Indian with a low wind topography. Baghdadi [15] evaluated the
PV/Battery system as a backup during the power outage times in a possibility of a PV/WT/Diesel/Battery hybrid unit under the climate
residential building and minimized the cost of the unit via the GA of southern Algeria and optimized it by the HOMER software taking
and PSO optimization algorithms. Maleki et al. [10] investigated into account renewable resources potential and energy demand;
different optimization algorithms in order to optimize the size of a while maximizing renewable electricity use and fuel saving are the
PV/WT/FC hybrid system for achieving the minimal total annual purpose Singh et al. [16] performed a detailed simulation by
cost. HOMER considering manufacturing cost and efficiency for opti-
Also, the literature survey shows that the HOMER program, for mizing the size and location of a PV/WT/Battery system. This
the use of optimization algorithms and the ability to analyze the optimal hybrid unit was able to provide some advantages like lower
sensitivity of stand-alone and grid-connected hybrid systems, is the emission, lower cost of energy and elimination of dummy load over
most common software in this field. Kamel and Dahl [11] compared the existing system. Amrollahi et al. [17] investigated the ability of
the economic potential and greenhouse gas emissions reduction of demand response programming in the case of component size
the use of a stand-alone hybrid Solar/Wind/Diesel generator case optimization of hybrid PV/WT/Battery system in a stand-alone
with an alone-diesel generator, using a hybrid optimization model micro-grid. For this purpose, in order to reduce the number of
in the HOMER software. After that, Bernal-Agustín et al. [12] required batteries, the required inverter and the photovoltaic cells
investigated the Strength Pareto Evolutionary Algorithm using the capacity, the optimization program was performed by the HOMER
HOMER program to the optimal design of a PV/WT/Diesel system software together with the GAMS software via the CPLEX solver.
for reducing the total cost and greenhouse gas emissions. Also Mohammadi et al. [18] provided an optimal planning approach, by
Hafez and Bhattacharya [13] investigated and compared several examining different scenarios, in order to select a hybrid power
cases containing a diesel-only, a fully renewable-based, a diesel generation system based on 100% renewable energy system (RES)
renewable mixed, and an external grid-connected microgrid for a residential house located in Tehran, Iran. This study, by using
configuration by the HOMER software in terms of economy and HOMER software, aimed to reduce the energy supply cost and
 N. Ghorbani et al. / Energy 154 (2018) 581e591 583

increase the reliability. Halabi et al. [19] studied and compared the for the optimal system sizing. Since the GA-PSO is a single-objective
use of different power generation systems including all possible algorithm, and cost is the main objective in the single-objective
standalone diesel generators, hybrid PV/diesel/battery, and 100% optimizations, therefore the reliability factor is converted to cost
PV/battery scenarios for two decentralized power stations in Sabah, by an optional penalty factor. Then, in order to achieve satisfying
Malaysia by HOMER. Recently, Hemmati [20] has already con- load demand, the sizing results of these two methods are
ducted an economic study for home energy management system compared, and their strengths and weaknesses are determined.
(HEMS), by applying wind turbine and battery energy storage. The
results show that installing wind turbine with battery storage al-
lows home to make profit through selling energy to the grid, when 2. Methodology
the energy price is high.
The studies on the use of various optimization algorithms for This paper aims at providing an optimal cost/reliability sizing for
optimization of hybrid energy systems indicate that minimizing the a PV/WT/Battery system. The variables in this optimization method
costs of energy are common objectives in the multi-objective are the power of PVs, the wind turbines power, and the batteries
optimization. Bilal et al. [21] optimized a hybrid Solar/Wind/Bat- capacity. Table 1 shows the estimated cost of the components. It
tery system by multi-objective GA for an isolated site, located in the shall be notified that since the problem is solved for a small scale of
northern coast of Senegal, and investigated the influence of the residential application, the maintenance cost of a wind turbine
load profiles on the optimal configuration. The objective functions would be higher than the large-scale turbines [29].The total cost
(OF) in this study were minimization of the annualized cost and includes all the costs throughout the useful life of the system (20
minimization of LPSP. Also, in another study, Bilal et al. [22] opti- years), which are translated to the initial moment of the
mized the size of a hybrid PV/WT/Diesel/Battery unit through investment.
different data such as wind speed, air temperature, and solar ra-
diation. They conducted the optimization for the Levelized Cost of
Energy (LCE) and CO2 emission by using the Multi-Objectives GA 2.1. System modeling
approach. Daud et al. [23] presented a backup control unit for a
grid-connected hybrid PV/battery system and optimized the con- The system consists of a wind turbine, a PV source and a battery
trol parameters through the input data of Malaysia for mitigating for energy storage (Fig. 1). There are many different configurations
PV farm output power fluctuations by the GA-based multi-objec- for an electrical generator, but the generator which can be used in a
tive optimization. Dufo-Lo
pez et al. [24] optimized a PV/WT/Diesel/ wind power system is the Permanent Magnet DC Generator or
Battery unit in for minimizing the energy cost and CO2 emissions PMDC Generator. The same PMDC machine may be driven electri-
using a multi-objective optimization. The analysis of different cally as a motor to move a mechanical load, or it may be driven
Pareto fronts revealed that applying the diesel generator to the mechanically as a simple generator to generate an output voltage.
hybrid system is better than gasoline generator, in aspects of the This case makes the PMDC generator ideal to be used as a simple
economy and reducing CO2 emissions. Also, the best Pareto fronts wind turbine generator. The load profile, solar radiation, and wind
showed that the PV panel is necessary for reducing the emissions speed data are simulated by the HOMER for 8760 h in a year. The PV
and also the wind system without PV panel is not optimal. Tahani and wind turbine supply the load power. If there is extra power, it
et al. [25] applied a hybrid FPA/SA (Flower Pollination Algorithm must be stored in the battery bank and, in the case of lacking
and Simulated Annealing algorithm) algorithm as a new optimi- generated power; the batteries must supply the power shortage. A
zation technique to maximize the reliability and minimize the costs charge regulator regulates the voltage and current going to the
of a hybrid PV/WT/Battery unit for a three-floor building, located in batteries and keeps them from overcharging. Since electrical de-
Tehran. Mohamed et al. [26] applied an Artificial Bee Colony algo- vices mostly need AC power to work, the inverter converts the DC
rithm (ABC) to maximize the output power and minimize the life to the AC so that the renewable system can supply the electrical
cycle cost of a PV system including photovoltaic, a battery bank, a load demand.
battery charger controller, and inverter. The ABC algorithm results
for the Helwan city (Egypt) were compared with the results of the  Photovoltaic array
use of GA for another case in the Zagazig city, and it was found that
the ABC was more efficient than the GA. Clarke et al. [27] analyzed a The power, generated by a PV, depends on the cell temperature
stand-alone desalination renewable energy system with the aim of and solar radiation. The power generation of a PV array is shown by
sizing and power management, by comparing the MOPSO and the Eq. 1:
HOMER results. Finally, it was found that the optimization by
MOPSO showed better results in reducing the NPC and CO2 emis- G
PPV ¼ PPV rated   hmppt (1)
sions. Baghaee et al. [28] have designed a hybrid PV/WT/FC 1000
generating system and presented a MOPSO algorithm to minimize
the three objective functions namely the annualized cost of the where G is solar radiation at the PV surface (w=m2 ), PPV rated is the
system, the loss of load expected and the loss of energy expected, PV rated power for G ¼ 1000w=m2 , and hmppt is the module effi-
with maximum reliability. The results reveal that the cost and ciency. Usually, the PV generation stays around maximum power
reliability are under the direct effect of three main components of point, because of a charge controller which uses the maximum
the unit including a wind turbine, PV array, and AC/DC converter. power point tracking system [30].
In this paper, the main objectives are minimizing the total
present cost and maximizing the reliability of a hybrid PV-wind  Wind turbine generator
turbine system with energy management (EMS) or battery bank,
for a place in Tehran, Iran. Based on the literature, this is the first The wind turbine generates power when the wind velocity
time that such an optimization is conducted. In this work, the (wind speed) is within the cut-in and cut-out velocity, and it de-
hybrid method of GA-PSO is applied as the single-objective algo- pends on the wind velocity, air density, and windmill area. The
rithm; while the MOPSO is utilized as the multi-objective algorithm power, generated by a wind turbine, is shown by Eqs. (2a) and (2b)
[30e32].
584 N. Ghorbani et al. / Energy 154 (2018) 581e591

Table 1
Cost of the components [29,49].

Item PV Panel Wind Turbine Battery Bank Inverter, Charge regulator

Rated power 1 (kW) 1 (kW) 1 (kWh) 1(kW)

Lifetime (year) 20 20 5 10
Investment cost ($) 2000 3200 100 700
Maintenance Cost ($/year) 33 100 5 0

efficiency is supposed to be unity. Then, the power of the battery in

the discharge state is shown by Eq. (4) [30]:

PLoad ðtÞ
PBAT ðtÞ ¼ PBAT ðt  1Þ  ð1  sÞ   ðNPV  PPV ðtÞ
hINV

 hCON þ NWT  PWT ðtÞ  hREC Þ hBAT (4)

Fig. 1. Simple diagram of the PV/WT/Battery system.

 Energy management strategy

8
< 0 V < VCi ; V > VCo The PV power and wind power depend on the solar radiation,
PWT ¼ 1 (2a) wind velocity, and temperature, so these are uncertain and changed
: rCAV 3 V  V  V hour by hour. The load demand also varies according to the user
Ci Co
2
energy demand. The power injected by PV and WT is defined by Eq.
where, r is the air density (Kg=m3 ), A is the windmill area (5) [34]:
perpendicular to the wind (m2 ), C is the power coefficient of the      
fail fail fail fail
wind turbine, and V is the wind velocity (m/s) at the height of Pin NPV ; NWT ¼ NPV  NPV  PPV þ NWT  NWT  PWT
turbine hub, which is calculated according to: (5)
 
V2 H2 a where, NPV and NWT are the total installed PV panels, and WTs, Nfail
¼ (2b) PV ,
V1 H1
and Nfail
WT are the failed PV panels and WTs; the failures indicate the
Having V1 as the wind velocity at a reference height denoted by unavailability of power sources. Then, the output power is
H1, and a is the Hellman coefficient. In this study; r ¼ 1:08 , distributed between the load and the battery storage system
C ¼ 0.42, a ¼ 0.25, H1 ¼ 10 m and H2 ¼ 40 m. through the following strategy:

 Battery energy storage system  If P in is equal to the power demand, all the load demand is
supplied by the power sources, and there is no storage and
Energy management system is widely used for renewable en- extraction from the batteries.
ergy generators, especially in PV and WTs, because of the unpre-  If P in is higher than the required power demand, all the load
dictable nature of the wind and solar irradiation. The extra demand is fulfilled by the power sources, and the excess power
electricity, generated by PV and WT, is stored in the battery bank is stored by the storage system. If the excess power is more than
and, in the case of electricity deficiency; it supplies the required the batteries rated power, a portion of the power will be lost.
load demand. The power of the battery in charge and discharge  If P in is less than the required power demand, the storage sys-
state is obtained by Eq. (3) [33]: tem supplies the shortage power. If the shortage power is more
than the batteries rated power, a part of the load will be lost.

PBAT ðtÞ ¼ PBAT ðt  1Þ  ð1  sÞ þ ðNPV  PPV ðtÞ  hCON þ NWT

P ðtÞ 2.2. The system cost and reliability model
 PWT ðtÞ  hREC Þ  Load  hBAT
hINV
(3) The total cost (CT) of the system includes the initial cost (CI),
maintenance cost (CM) and replacement cost (CR), which are ob-
where PBAT ðtÞ is the battery charge at timet, and PBAT ðt  tained from eqs. (6e9) [35e37]:
1Þrepresents the charge amount at time t  1. Also, s is the battery
self-discharge rate, hCON is the converter efficiency, hREC is the CT ¼ CI þ CM þ CR (6)
rectifier efficiency, hINV is the inverter efficiency, hBAT is the charge
efficiency of the battery system, and PLoad ðtÞis the load demand. NPV CI ¼ ðNPV CPV Þ þ ðNWT CWT Þ þ ðNBAT CBAT Þ þ ðNINV CINV Þ (7)
and NWT are the numbers of PV panels and wind turbines.
If the generated power by renewable resources is less than the where, Ci is per unit cost of the ith component and Ni is the number
power demand, discharging begins. In this paper, the discharge of ith component.
 N. Ghorbani et al. / Energy 154 (2018) 581e591 585

after each iteration, and after the initial evaluation, the iterative
T 
X 
1 þ InfR t loop begins. The algorithm calculates the objective function for all
CM ¼ NPV CM;PV þ NWT CM;WT þ NBat CM;BAT
t¼1
1 þ IntR particles and updates the pbest and gbest. If the termination cri-
terion (TC) is satisfied, the optimal solution is gbest, and the algo-
(8)
rithm is ended, otherwise the loop repeats. In this study, TC is the
specific number of iterations (100 iteration). Also, a population size
where, CM;i is per unit maintenance cost of the ith component, InfR of 50 particles is defined for PSO approach. Each particle has three
is the inflation rate, IntR is the interest rate, and t is the life cycle of variables as the PV power capacity, WT power capacity and battery
the components which is 20 years. energy storage capacity. The search space for the variables is
In this study, it is assumed that only batteries and inverter need 0e1000 kW for the PV and WT, and 0 to 1000 kWh for the battery.
replacement, so the replacement cost is obtained by: In the initialization step, random positions are assigned to each
particle. Then, in the next 100 iterations, the positions and veloc-
T 
X 
1 þ InfR t ities are updated according to Eqs. 11 and 12 to obtain the lowest
CM ¼ ðNBAT CBAT þ NINV CINV Þ (9)
t¼1
1 þ IntR cost function.

The subtraction of the load power at each hour and the gener-  Genetic algorithm
ated power from the renewables, expressed by the loss of power
supply (LPS) and LPSP, would the probability of the LPS's occurrence The genetic algorithm is a method, inspired by natural selection,
in a complete year (8760 h), as shown in Eqs. (10.a,b): to find the solution for optimization problems [43]. As Tomassini
summarized [44], the idea is to adapt a population to environ-
X
8760
LPS ¼ ½PLoad ðtÞ  ðPPV ðtÞ þ PWT ðtÞ þ PBAT ðtÞÞ (10a) mental conditions like what happens for genes in nature. Each in-
t¼1 dividual of the population has its own characteristic, and in this
technique, the bad and ineffective traits must be eliminated, while
where Pi ðtÞ is the power quantity of ith resource at time t. the helpful behaviors must be improved. The genetic algorithm is
an evolutionary algorithm with evolution strategies and evolu-
P8760
LPSt tionary genetic programming. The solutions, found by this way, are
t¼1
LPSP ¼ (10b) similar to the real populations that are adapted to the natural
P
8760
PLoad ðtÞ environment. The algorithm starts with a randomly-generated in-
t¼1 dividual. In this study, 50 randomly initialized individuals are
considered as the population. Each individual has three variables
where LPSk is the loss of power supply at each hour, and PLoad ðtÞ is
with a search space similar to the proposed PSO approach. In each
the load demand at each hour.
iteration, new populations are generated by applying certain sto-
chastic operators through a loop, called the generation. These op-
2.3. Optimization algorithms erators are as crossover and mutation. Crossover is the process of
taking two parents and producing two children solution from them.
 Particle swarm optimization (PSO) The crossover rate of 0.8 assigns the number of parents which are
included in the crossover operation, and it means that in each
PSO is a population-based stochastic approach for solving iteration, 40 parents are taken in order to generate 40 children. The
continuous and discrete optimization problems which arose from other operator, called mutation, is a randomly change in one gene
swarming behaviors were observed in schools of fish, flocks of value of an individual from its initial state. Mutation is a kind of
birds, a swarm of bees and even human social behavior [38e40]. In error, and if the mutation rate is set low, it may lead GA to a better
PSO, a population of particles moves in the search space of an solution. Here, the mutation rate is 0.1 which means five in-
optimization problem. The position of a particle represents a dividuals meet mutation operation in each iteration. After the
candidate solution to the optimization problem at hand. Each generation, 50 best solutions are selected to continue the algorithm
particle searches for better positions in the search space according for 100 iterations.
to its local best position and global best position [41,42]: This iterative process can be summarized as followings [44]:

vk ði þ 1Þ ¼ wðiÞ  vk ðiÞ þ c1  r1 ðpbestk ðiÞ  xk ðiÞÞ þ c2 I. Iteration ¼ 0

II. Seed population
 r2 ðgbestðiÞ  xk ðiÞÞ (11)
III. While not (termination condition) do
IV. Iteration ¼ Iterationþ1
xk ði þ 1Þ ¼ vk ði þ 1Þ þ xk ðiÞ (12) V. Calculate fitness
VI. Selection
k ¼ 1; 2; …; Np and i ¼ 1; 2; …; imax VII. Crossover
VIII. Mutation
where, xk and vk represent the particle's position and velocity, k is IX. End while
the particle's index, i is the iteration number, r1 and r2 are the
randomly generated numbers between 0 and 1 and c2 are learning Fig. 2 shows the optimization and simulations algorithm. The
factors. process of optimization by the GA-PSO algorithm is described at the
First of all, the PSO generates a random population and a GA-PSO algorithm subsection in the following.
random initial velocity for each particle in the search space. Then,
the algorithm calculates the objective function for all particles.  GA-PSO algorithm
Here, pbest is the best experience of each particle, and gbest is the
best experience of all particles in the population. The position and In this study, a hybrid GA-PSO method is proposed which exe-
value of the objective function for gbest and pbest must be stored cutes the genetic algorithm and particle swarm optimization
586 N. Ghorbani et al. / Energy 154 (2018) 581e591

The parameters of the algorithms used in this paper are sum-

marized as follows:
PSO: N ¼ 50; c1 ¼ 2; c2 ¼ 2; b ¼ 0.99; w ¼ 1; maximum
iteration ¼ 100;
GA: N ¼ 50; Mutation rate ¼ 0.1; crossover rate ¼ 0.8 maximum
iterations ¼ 100; The search space for variables is 0e1000 kW for
PV and WT and 0 to 1000 kWh for battery.
Since the GA-PSO is a single objective algorithm and the cost
function is the final objective, the reliability factor must be con-
verted to cost. The weighted summation method counts the pri-
mary cost function in addition to weighted cost, obtained from LPS,
multiplied by an optional penalty factor. The total cost of the
components plus the capacity shortage penalty factor, yield in the
total net present cost (NPC). In this study, the penalty is considered
as 5.6$/kWh, which is an average capacity shortage penalty [29].

Fig. 2. Optimization and simulation flowchart. NPC ¼ CT þ PF  LPS ð$Þ (13)

Moreover, the levelized cost of energy (LCE) term is used to find
simultaneously, selects the best-evaluated population from the GA, the cost of the generated unit of energy. This term is the ratio of the
and optimizes it by the PSO in any iteration [40]. The first step is total yearly cost of the system components to the total yearly en-
initializing the GA and PSO subsystems. Then, both algorithms are ergy provided by the system [45]. The LCE can be calculated as:
executed simultaneously, and the best solution is memorized. After
executing a specified number of iterations (termination criterion), NPC
LCE ¼ ð$=kWhÞ (14)
the program running stops and the best solution are announced as Et
the final solution. In this paper, the percentage of using GA and PSO
in the hybrid system is equal. Fig. 2 shows the GA-PSO flowchart, where Et (kWh/year) is the total energy, generated by the system in
and the GA-PSO algorithm processes are summarized as followings one year.
[36]: In this study, the cost optimization has been implemented for
2%, 5% and 10% of the maximum LPSP. The PPV rated , PWT rated and the
Step-1 Set population size (2*Ps), maximum number of generations number of batteries are the first positions in the PSO. So that, after
(Mg), the probability of crossover (PC), the probability of few iterations, the optimal cost of these variables are obtained.
mutation (Pm) and the bounds of decision variables. Also, the minimum and maximum bounds for the PV panels and
Step-2 Set t ¼ 0. [t presents the generation/iteration number] wind turbines are 0e200, while these are 0e300 for the batteries.
Step-3 Initialize the chromosomes/particles of the population P(t). Three different scenarios are considered for the simulation:
Step-4 Compute the fitness function for each chromosome of P(t).
Step-5 Find the global best chromosome/particles (Pg) having the - PV and battery bank system
best fitness value. - WT and battery bank system
Step-6 Divide the chromosome/particles into two groups, viz. - PV, WT and battery bank system
PGA(t) and PPSO(t) with equal population size.  Multi-objective particle swarm optimization (MOPSO)
Step-7 Repeat the following until the termination criterion is
satisfied: Many real world problems have two or more objectives to be
i. Increase the value of t by unity. obtained. In these problems, objectives may be in conflict, and a
ii. Apply GA for population PGA(t). tradeoff between possible solutions is the optimal point [46e48]. In
iii. Apply crossover & mutation operators on PGA(t) to this study, the system cost and risk are the objectives. The PSO al-
produce new population PGA(t). gorithm is an inherent single-objective optimizer while, in multi-
iv. Find the best chromosome (P0 g) from the current objective problems, an absolute global optimum does not often
population PGA(t). exist. Thus, to make the PSO algorithm capable of dealing with
v. Compare P0 g with earlier best chromosome Pg and store multi-objective optimization problems, some modifications
better one in P0 g. become necessary. The personal best performance (pbest) of each
vi. Set t ¼ t þ 1. individual particle is replaced with the new solution, if and only if,
vii. Select the population PGA(t) from the population it dominates the former pbest. Also, two major issues should be
PGA(t  1) of (t  1)- th generation using tournament considered in the updating process of the global best performance
selection. (gbest).
viii. Apply PSO for PPSO(t). Firstly, the fitness assignment and selection should be
ix. Improve the best position of each particle by addressed, such that a search can move towards the Pareto optimal
comparing the position of all chromosomes of PGA(t). set. Fitness is proportional to the dominance rank of solutions. The
x. Compute the velocity of each particle. MOPSO, which is used in this study, is a dominance-based method.
xi. Obtain the new position of each particle. Secondly, the diversity of the swarm should be maintained to
xii. Improve the position of each particle and also find the prevent premature convergence and obtain an evenly distributed
global best particle (Pg). Pareto optimal front [28]. In this paper, an archiving mechanism
Step-8 Print the position and fitness of global best particle. was employed to form a repository, which may contain only a
Step-9 End. limited number of solutions. Compared to the conventional
archiving, this approach shows a better performance in terms of
both diversity and optimality and less complexity.
 N. Ghorbani et al. / Energy 154 (2018) 581e591 587

3. Results and discussion

3.1. Irradiance/wind/load data

The simulated radiation for a house in Tehran by the HOMER

software is shown in Fig.3. The latitude and the longitude are
considered as 35 68920 S and 51 38900 E, respectively. As expected,
the solar radiation varies in each hour, day and month. The average
solar radiation for Tehran is 5.2kWh⁄ ðm2⁄^ dayÞ; and the maximum
occurs in June with nearly 7.5 kWh
m2
: Also, the month with minimum
day

solar radiation is December with nearly 2.5kWh⁄ ðm2⁄ ^ dayÞ.

Fig. 3(a,b) shows the monthly radiations and the distribution of
solar radiation in hours for one day in each month, respectively.
The wind velocity (v) data, obtained from the Renewable Energy
Organization of Iran's anemometric stations, is shown in Fig. 4(a,b).
The average wind speed in Tehran is 7.5 m/s which can lead to
generate power at nights, unlike solar radiation.
The load demand is simulated by the HOMER for a residential
section with an average demand of 210 kWh per day. As shown in
Fig. 5(a, b), the electricity consumption in summer is more than
other seasons. Two peaks occur in the load demands between 12 to
18 and 18 to 24 in summer. Also, the load demand in August and
February is maximum and minimum, respectively.

3.2. The GA-PSO results Fig. 4. The mean monthly (a) and hourly (b) wind velocity [50].

The GA-PSO was the first algorithm which was used to optimize
the problem, and the simulation results of GA-PSO are shown in
Figs. 6e8. The algorithm starts with an initial evaluation and con-
tinues until the termination criterion (TC) is satisfied. The LPSP with
three levels of 2%, 5%, and 10% is considered to have wider choices.
Since the using a hybrid renewable energy system has a capital cost
more than using PVs or wind turbines individually, the

Fig. 5. Monthly (a) and hourly (b) load demand diagram for a residential section.

optimization is done for three different systems. The WT/Battery is

considered as the first system and Fig. 6 (a, b, c) shows the opti-
mization results for the first system with three different levels of
LPSP. As Fig. 6 (a, b, c) shows, the optimization is done for 100000
numbers of function evaluations (NFE). Approximately after 20 it-
erations, the optimal point for 2% and 5% of LPSP is obtained. For
10% of the LPSP, the optimal point is obtained after nearly 100 it-
Fig. 3. The mean monthly (a) and hourly (b) solar radiation in one year in Tehran, Iran.
erations. The detailed information is reported in Table 2(a).
588 N. Ghorbani et al. / Energy 154 (2018) 581e591

Fig. 8. Single objective optimum solution for the hybrid WT/Battery system with 2%
(a), 5% (b), and 10% (c) of LPSP.
Fig. 6. Single objective optimum solution for the hybrid PV/WT/Battery system with
2% (a), 5% (b), and 10% (c) of LPSP.

batteries for 20 years, plus minor items like charge controller and
installation costs.
Fig. 8 (a,b,c) shows the optimization results for three different
levels of the LPSP for this case. As Fig. 9 shows, nearly after 20000
NFEs, the optimal point for 2% and 5% of LPSP is obtained. For 10% of 3.3. The MOPSO results
the LPSP, the optimal point is obtained after about 100 iterations.
The detailed information is represented in Table 2(c). In the last section, the optimum cost function is evaluated for
The first system with WT and battery cannot use suitable solar three systems with different penalty factors. In this part, both
irradiance of the area, and the second system with PV and battery reliability and cost objectives are considered simultaneously. It is
needs too much storage for the night since PV is not able to provide obvious that each point with lower cost and higher reliability
electricity to meet the load demand. The third system, with (lower risk) is a better point than the others, but some points have
employing PV, WT, and battery, applies all the energy production lower cost and higher risk, while some points are vice-versa. Since
capacity and obtains the minimum total cost. there is no advantage for any objective, all points that make a curve
The results of cost, power, and the capacity of the components combined with the best costs and reliabilities, are considered as the
are summarized in Table 2 (a,b,c). The total cost includes the in- best points. This method's name is the none-dominated sorting
vestment cost, the operation and maintenance (O&M) and optimization, and the curve's name is the Pareto front.
replacement cost of PVs, also the costs of WTs, inverter, and The red stars in Fig.10 are the Pareto front points and the black
circles are the non-optimum points which are dominated by the
Pareto front. Similar to the single objective optimization, three
scenarios are considered and simulated. It is evident that the op-
timum results of the single objective optimization are procurable
from the multi-objective optimization. The blue rectangles are the
optimum points with 2%, 5%, and 10% of LPSP, which are obtained
from single objective optimization too.
The multi-objective optimization, compared to the single
objective optimization method, gives a wider choice of optimal
point. By this method, it is possible to have the cost of the system
for any value of LPSP. For the first hybrid system, consisting PV/WT/
Battery with 0% of LPSP, the total cost is nearly 800000$. If a lower
cost is desired, the lower values of LPSP must be chosen, and there
would be a tradeoff between the cost and the LPSP. For example, for
20% of LPSP, the total cost is 250000$, and it is about one-third of
the previous choice, as shown in Fig. 9. Figs. 10 and 11 show the
other scenarios' results. The black circles are the initial population,
and the red stars indicate the Pareto front in Figs. 9e11. Also, the 2%,
5% and 10% of LPSP are marked with blue rectangles.
The MOPSO method was also applied to the hybrid WT/Battery
system, and the Pareto front was obtained. If the loss of the load is
not acceptable (LPSP ¼ 0), the total cost would be nearly 1400000$,
so 20% of the LPSP costs 500000$.
Fig. 7. Single objective optimum solution for the hybrid PV/Battery system with 2% (a), Fig.11 represents the Pareto front of the hybrid PV/Battery sys-
5% (b), and 10% (c) of LPSP.
tem. The costs of this case are less than that of the WT/Battery
 N. Ghorbani et al. / Energy 154 (2018) 581e591 589

Table 2
The optimum cost of components for PV/WT/Battery (a), WT/Battery (b), and.PV/Battery (c).

MAX LPSP PV (kW) WT (kW) Battery (KWh) Inverter (kW) Total NPC ($)

A
2% 82 25 190 33 787193
5% 62 25 148 33 644235
10% 47 25 98 33 508540
B
2% 0 166 370 33 1361403
5% 0 120 230 33 919202
10% 0 92 150 33 708695
C
2% 144 0 241 33 950412
5% 104 0 224 33 773241
10% 81 0 192 33 637476

sizing. As shown in Table 3, the GA-PSO results have been

compared with the HOMER software. Also, with two cities of ref.
[29], the comparison has been done for the LCE factor (shown in
Table 4). The researchers in Ref. [29], have applied an optimization
in four different cities, among those, the Moaleman and Nikouye
cities have earned the best results for the hybrid PV/WT/battery
unit, and the maximum LPSP of 2%. It is shown that the cost of
components is the same with [29]; nevertheless, the load data is
different. For the proposed approach, ref. [29] and the HOMER
optimization average daily loads are 210 kWh, 220kwh, 400kwh,
for the GA-PSO method, HOMER optimization, and ref. [29]. The
Nikouye and Moaleman cities are 5.2 kWh/m2 =day, 5.2kwh/
m2 =day, 5 kWh/m2 =day, and 5.77 kWh/m2 =day, respectively. The
average wind velocity for the GA-PSO method and HOMER opti-
mization is 7.5 m/s; this velocity is 7.2 m/s and 6.1 m/s for ref. [29].
Fig. 9. MOPSO simulation results for hybrid PV/WT/Battery system. As shown in Table 3, even though the wind and solar resources in
the GA-PSO and HOMER are exactly equal and approximately
equal to ref. [29], the GA-PSO has the best result among different
system. For 0% of LPSP, it costs 1100000$, and the total cost for 20%
methods. The deviation percentage between the proposed
of LPSP is 300000$. As expected from the GA-PSO results, the
method's LCE and the LCE of [29] for Moaleman and Nikouye is 6%
hybrid PV/WT/Battery has a lower cost compared to the other two
and 10%, respectively. The difference between the algorithms like
systems.
GAPSO and the HOMER optimization is that the HOMER must
calculate all modes of variables without any history of variables'
3.4. Comparison of the results best global and personal results. It is actually like a try and error
process. But, in the GAPSO method, the relationship between the
For evaluating the validity of the proposed approach, the LCE variables cause more convergence rate and better result.
factor is considered. Based on the average solar radiation and To show the impact of cost function, from the electrical load
wind speed maps, the techno-economically optimized systems data, the LCE for different load data are also plotted in Fig.12.
are designed by simulating the behavior of various combinations As shown in Fig.12 for different electrical load data between 150
of renewable energy systems with different sizing. This is the (kWh/d) to 350 (kWh/d), the LCE is about 0.5e0.7 ($/kWh). The red
same method by which, the HOMER software obtains the optimal

Fig. 10. MOPSO simulation results for hybrid WT/Battery system. Fig. 11. MOPSO simulation results for hybrid PV/Battery system.
590 N. Ghorbani et al. / Energy 154 (2018) 581e591

Table 3
Comparison with HOMER software.

Method PV (kW) WT (kW) BAT (kWh) Wind Vel. Irradiance (kWh/m2 =d) Inverter (kW) LCE ($/kWh) NPC ($) LPSP

GA-PSO 82 25 190 7.5 5.2 33 0.508 787193 0.02

HOMER 165 54 350 7.5 5.2 65 0.511 1652300 0.02

Table 4
Comparison results.

Method PV(kW) WT(kW) BAT(kWh) Wind Vel. Irradiance LCE

GA-PSO 82 25 190 7.5 5.2 0.508

[29] Moaleman 61 22 434 6.1 5.8 0.540
[29] Nikouye 35 30 360 7.2 5 0.560

Fig. 12. LCE for different electrical loads.

foursquare is the load data, considered for simulations in this paper is 28.24% less than the WT/BAT and 20.22% less than the PV/BAT
(210 kWh/d). system. Through the MOPSO, the cost and reliability are optimized
simultaneously, as the two objectives. Also, three Pareto fronts are
obtained for the three systems, which can be used in order to find
4. Conclusion the sizing strategy with the lowest cost in every level of maximum
LPSP. The comparison results show that the proposed approach
In this paper, a hybrid PV-WT generating unit with a battery with 0.508 of LCE, represents a deviation between 7% and 10%.
bank for storing extra electricity, generated by PV and WT, was
designed and the load profile, solar radiation, and wind speed data
References
were simulated by the HOMER for 8760 h in a year in an off-grid
house in Tehran, Iran. The GA-PSO and MOPSO methods were [1] Dalton G, Lockington D, Baldock T. Case study feasibility analysis of renewable
applied for the optimal system sizing with two objectives including energy supply options for small to medium-sized tourist accommodations.
maximizing the system reliability and minimizing the total present Renew Energy 2009;34(4):1134e44.
[2] Aghahosseini A, et al. Analysis of 100% Renewable energy for Iran in 2030:
cost. Then, the sizing results were compared to achieve the satis- integrating solar PV, wind energy and storage. Int. J. Environ, Sci, Technol.
fying load demand. Three different scenarios were considered for 2017:1e20. https://doi.org/10.1007/s13762-017-1373-4.
the simulation, including PV panel and battery bank system, wind [3] Fathabadi H. Novel grid-connected solar/wind powered electric vehicle
charging station with vehicle-to-grid technology. Energy 2017;132:1e11.
turbine and battery bank, and PV panel-wind turbine with battery [4] Celik AN. Techno-economic analysis of autonomous PV-wind hybrid energy
bank system. In the optimization through GA-PSO, as a single systems using different sizing methods. Energy Convers Manage 2003;44(12):
objective algorithm, the LPSP was considered as the reliability 1951e68.
[5] Dufo-Lo
pez R, Bernal-Agustín JL. Design and control strategies of PV-Diesel
factor, and 5.6 $/kWh penalty factor was determined for the ca- systems using genetic algorithms. Sol Energy 2005;79(1):33e46.
pacity shortage. [6] Luna-Rubio R, et al. Optimal sizing of renewable hybrids energy systems: a
The results show that the PV/WT/BAT system has lower cost, review of methodologies. Sol Energy 2012;86(4):1077e88.
[7] Yang H, et al. Optimal sizing method for stand-alone hybrid solarewind
compared to the PV/BAT and WT/BAT systems in all three levels of system with LPSP technology by using genetic algorithm. Sol Energy
LPSP. So, the PV/WT/BAT system is the best choice for this case. The 2008;82(4):354e67.
best-obtained NPC for the PV/WT/BAT system was 787193$ for 2% [8] Kornelakis A, Marinakis Y. Contribution for optimal sizing of grid-connected
PV-systems using PSO. Renew Energy 2010;35(6):1333e41.
of the maximum LPSP which is 42.17% less than the WT/BAT system
[9] Khoury J, et al. Optimal sizing of a residential PV-battery backup for an
and 17.17% less than the PV/BAT system. For 5% of the maximum intermittent primary energy source under realistic constraints. Energy Build
LPSP, the total NPC of 644235$ is obtained as the best result for the 2015;105:206e16.
PV/WT/BAT system; which is 29.91% less than the WT/BAT, and [10] Maleki A, Askarzadeh A. Comparative study of artificial intelligence tech-
niques for sizing of a hydrogen-based stand-alone photovoltaic/wind hybrid
16.68% less than the PV/BAT system. For 10% of the maximum LPSP, system. Int J Hydr Energy 2014;39(19):9973e84.
the best total NPC was 508540$ for the PV/WT/BAT system; which [11] Kamel S, Dahl C. The economics of hybrid power systems for sustainable
 N. Ghorbani et al. / Energy 154 (2018) 581e591 591

desert agriculture in Egypt. Energy 2005;30(8):1271e81. 2015;36(3):315e24.

[12] Bernal-Agustín JL, Dufo-Lo
pez R, Rivas-Ascaso DM. Design of isolated hybrid [31] Yang H, Wei Z, Chengzhi L. Optimal design and techno-economic analysis of a
systems minimizing costs and pollutant emissions. Renew Energy hybrid solarewind power generation system. Appl Energy 2009;86(2):163e9.
2006;31(14):2227e44. [32] Pishgar-Komleh S, Keyhani A, Sefeedpari P. Wind speed and power density
[13] Hafez O, Bhattacharya K. Optimal planning and design of a renewable energy analysis based on Weibull and Rayleigh distributions (a case study: Fir-
based supply system for microgrids. Renew Energy 2012;45:7e15. ouzkooh county of Iran). Renew Sustain Energy Rev 2015;42:313e22.
[14] Bhattacharjee S, Acharya S. PVewind hybrid power option for a low wind [33] Maleki A, Pourfayaz F. Optimal sizing of autonomous hybrid photovoltaic/
topography. Energy Convers Manage 2015;89:942e54. wind/battery power system with LPSP technology by using evolutionary al-
[15] Baghdadi F, et al. Feasibility study and energy conversion analysis of stand- gorithms. Sol Energy 2015;115:471e83.
alone hybrid renewable energy system. Energy Conver Manage 2015;105: [34] Nomura S, et al. Wind farms linked by SMES systems. IEEE Transactions on
471e9. applied Superconductivity 2005;15(2):1951e4.
[16] Singh G, et al. Optimal sizing and location of PV, wind and battery storage for [35] Abedi S, et al. A comprehensive method for optimal power management and
electrification to an island: a case study of Kavaratti, Lakshadweep. J Energy design of hybrid RES-based autonomous energy systems. Renew Sustain En-
Storage 2017;12:78e86. ergy Rev 2012;16(3):1577e87.
[17] Amrollahi MH, Bathaee SMT. Techno-economic optimization of hybrid [36] Dufo-Lo
pez R, Bernal-Agustín JL. Multi-objective design of
photovoltaic/wind generation together with energy storage system in a PVewindedieselehydrogenebattery systems. Renew Energy 2008;33(12):
stand-alone micro-grid subjected to demand response. Appl Energy 2559e72.
2017;202:66e77. [37] Kaviani AK, Riahy G, Kouhsari SM. Optimal design of a reliable hydrogen-
[18] Mohammadi M, et al. Optimal planning of renewable energy resource for a based stand-alone wind/PV generating system, considering component out-
residential house considering economic and reliability criteria. Int J Electrical ages. Renew Energy 2009;34(11):2380e90.
Power Energy Systems 2018;96:261e73. [38] Shukla A, Singh S. Advanced three-stage pseudo-inspired weight-improved
[19] Halabi LM, et al. Performance analysis of hybrid PV/diesel/battery system crazy particle swarm optimization for unit commitment problem. Energy
using HOMER: a case study Sabah, Malaysia. Energy Convers Manage 2016;96:23e36.
2017;144:322e39. [39] Zhang J, et al. A hybrid particle swarm optimization with small population size
[20] Hemmati R. Technical and economic analysis of home energy management to solve the optimal short-term hydro-thermal unit commitment problem.
system incorporating small-scale wind turbine and battery energy storage Energy 2016;109:765e80.
system. J Clean Prod 2017;159:106e18. [40] Sahoo L, et al. An efficient GAePSO approach for solving mixed-integer
[21] Bilal BO, et al. Optimal design of a hybrid solarewind-battery system using nonlinear programming problem in reliability optimization. Swarm Evol
the minimization of the annualized cost system and the minimization of the Comput 2014;19:43e51.
loss of power supply probability (LPSP). Renew Energy 2010;35(10):2388e90. [41] Das S, Abraham A, Konar A. Particle swarm optimization and differential
[22] Bilal BO, et al. Methodology to size an optimal stand-alone PV/wind/diesel/ evolution algorithms: technical analysis, applications and hybridization per-
battery system minimizing the levelized cost of energy and the CO2 emis- spectives. In: Advances of computational intelligence in industrial systems.
sions. Energy Procedia 2012;14:1636e47. Springer; 2008. p. 1e38.
[23] Daud MZ, Mohamed A, Hannan M. An improved control method of battery [42] Ahmed H, Glasgow J. Swarm intelligence: concepts, models and applications.
energy storage system for hourly dispatch of photovoltaic power sources. School of Computing, Queens University Technical Report; 2012.
Energy Convers Manage 2013;73:256e70. [43] Holland J, Goldberg D. Genetic algorithms in search, optimization and ma-
[24] Dufo-Lo
pez R, et al. Multi-objective optimization minimizing cost and life chine learning. Reading, MA: Addison-Wesley; 1989.
cycle emissions of stand-alone PVewindediesel systems with batteries stor- [44] Tomassini M. Parallel and distributed evolutionary algorithms: A review.
age. Appl Energy 2011;88(11):4033e41. 1999.
[25] Tahani M, Babayan N, Pouyaei A. Optimization of PV/Wind/Battery stand- [45] Ibrahim IA, Khatib T, Mohamed A. Optimal sizing of a standalone photovoltaic
alone system, using hybrid FPA/SA algorithm and CFD simulation, case system for remote housing electrification using numerical algorithm and
study: Tehran. Energy Convers Management 2015;106:644e59. improved system models. Energy 2017;126:392e403.
[26] Mohamed AF, Elarini MM, Othman AM. A new technique based on artificial [46] Doagou-Mojarrad H, et al. Optimal placement and sizing of DG (distributed
bee colony algorithm for optimal sizing of stand-alone photovoltaic system. generation) units in distribution networks by novel hybrid evolutionary al-
J Adv Res 2014;5(3):397e408. gorithm. Energy 2013;54:129e38.
[27] Clarke DP, Al-Abdeli YM, Kothapalli G. Multi-objective optimisation of [47] Khorasaninejad E, Hajabdollahi H. Thermo-economic and environmental
renewable hybrid energy systems with desalination. Energy 2015;88:457e68. optimization of solar assisted heat pump by using multi-objective particle
[28] Baghaee H, et al. Reliability/cost-based multi-objective Pareto optimal design swam algorithm. Energy 2014;72:680e90.
of stand-alone wind/PV/FC generation microgrid system. Energy 2016;115: [48] Li X, Malkawi A. Multi-objective optimization for thermal mass model pre-
1022e41. dictive control in small and medium size commercial buildings under summer
[29] Hosseinalizadeh R, et al. Economic sizing of a hybrid (PVeWTeFC) renewable weather conditions. Energy 2016;112:1194e206.
energy system (HRES) for stand-alone usages by an optimization-simulation [49] Ahmadi S, Abdi S. Application of the Hybrid Big BangeBig Crunch algorithm
model: case study of Iran. Renewable and Sustainable Energy Reviews for optimal sizing of a stand-alone hybrid PV/wind/battery system. Sol Energy
2016;54:139e50. 2016;134:366e74.
[30] Shanbedi M, et al. Thermal performance prediction of two-phase closed [50] Website of renewable energy and energy efficiency organization at http://
thermosyphon using adaptive neuro-fuzzy inference system. Heat Transf Eng www.satba.gov.ir/fa/regions/tehran.