Journal of Cleaner Production 241 (2019) 117920

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

A heuristic-based approach for optimizing a small independent solar

and wind hybrid power scheme incorporating load forecasting
Weiping Zhang a, Akbar Maleki b, c, *, Marc A. Rosen d
a
Department of Electronic Information Engineering, Nanchang University, Nanchang, China
b
Faculty of Mechanical Engineering, Shahrood University of Technology, Shahrood, Iran
c
Department of Renewable Energies, Faculty of New Science & Technologies, University of Tehran, Tehran, Iran
d
Faculty of Engineering and Applied Science, University of Ontario Institute of Technology, Oshawa, Ontario, L1H 7K4, Canada

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

Article history: Small independent hybrid power schemes (IHPSs) are promising alternatives for load supply for remote
Received 17 April 2019 areas. In IHPS optimization, load data are generally the main input. In this paper, forecasting strategies
Received in revised form are proposed for load related parameters and tested on real data. Also, an efficient method based on the
16 July 2019
heuristic procedure (tabu search) is presented for optimization of an IHPS based on solar and wind
Accepted 3 August 2019
Available online 5 August 2019
energy along with a battery. The effect of using forecast load information instead of past information on
the IHPS performance is investigated. In the optimization, there are three main decision variables:
Handling Editor: Panos Seferlis number of batteries, surface area of the PV system, and wind turbine swept area. The optimization is
done to satisfy continually the load demand and to minimize the IHPS life cycle cost while respecting
Keywords: relevant limitations. To ensure the scheme's reliability, the probability of loss of power supply is
Small autonomous hybrid power scheme determined. The performance of the proposed algorithm-based load forecasting approach is compared
Solar energy with the harmony search algorithm-based load forecasting and simulated annealing algorithm-based
Wind energy load forecasting. The simulation results clearly demonstrate the advantages of utilizing load fore-
Optimization
casting in an IHPS optimization problem, and confirm that the tabu search method earnings more
Heuristic algorithm-based load forecasting
promising results than the harmony search and simulated annealing methods.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction storage (lead acid batteries) can be advantageous (Deveau et al.,

2015; Giaouris et al., 2018). Hybrid systems often exhibit better
The gradual reduction of fossil fuels and increase in climate reliability than single-renewable energy schemes (Merei et al.,
change impacts attributable to conformist energy resources have 2013).
increased interest in exploiting sources of renewable energy, which In order to have a hybrid scheme with minimal cost and
can satisfy current global energy requirements and incorporate maximal reliability, optimum design is essential. In this regards, for
various benefits. Renewable energy sources can enhance security of predicting energy production, in other words, for optimization of
energy supply, expand energy supply markets, and reduce global hybrid renewable energy scheme, information from many load and
warming impacts (Kashefi Kaviani, Riahy et al., 2009). In particular, weather stations are essential.
wind turbines and photovoltaic technology can provide reliable Electricity supply is greatly influenced by demand conditions. In
electrical power for remote regions (Maleki, 2018). Because of the the other words, load is among the most influential factor in
changeable and intermittent nature of wind and solar energy re- designing electrical systems. Thus, load forecasting is important for
sources and their dependences on environmental situations, hybrid supplying the load demand in remote areas with hybrid energy
renewable energy schemes based on these technologies and energy systems. Since, accurate load forecasting can improve hybrid en-
ergy system operational efficiency; an accurate method is needed
for load demand forecasting. The forecasting of load related pa-
rameters can be assisted significantly by the use of computational
* Corresponding author. Faculty of Mechanical Engineering, Shahrood University intelligence, which includes neural networks and fuzzy systems
of Technology, Shahrood, Iran.
(Mellit and Pavan, 2010; Palma-Behnke et al., 2011; Wide
n et al.,
E-mail addresses: [email protected], [email protected], a_
[email protected] (A. Maleki). 2015; Hazir et al., 2018).

https://doi.org/10.1016/j.jclepro.2019.117920
0959-6526/© 2019 Elsevier Ltd. All rights reserved.
2 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

In the recent past, many significant investigations of autono- et al., 2018), and others (Sinha and Chandel, 2015; Peng et al.,
mous hybrid power schemes based on wind turbines (WTs) and 2018), are methods for sizing hybrid schemes.
photovoltaic (PV) panels have been reported (Li et al., 2013; Kolhe Ekren and Ekren, (2010) used the well-known heuristic methods
et al., 2015; Fathabadi, 2017; Al-Ghussain et al., 2018; Anoune et al., based on simulated annealing (SA) for size optimization of a
2018) (Table 1). It can be seen that small independent hybrid power standalone PV-wind electrical system including a battery with
schemes (IHPSs), especially solar and wind power systems, are minimum cost. The considered decision variables were as follows:
promising alternatives for load supply for remote areas compared capacity/size of WT, PV and battery (BAT). The authors found that
with single-type power systems. Table 1 also shows that HOMER the proposed method (SA) provides superior results relative to the
software is a traditional tool for designing, optimizing and perfor- other method. Similarly, Maleki et al. (Maleki and Askarzadeh,
mance assessment of the hybrid power schemes. Nonetheless, 2014; Maleki and Pourfayaz, 2015) used harmony search (HS)
there are some limitations in using this software which necessitate based on well-known heuristic methods for optimization of a
developing new approaches, and several optimization methods standalone PV-wind power scheme including a battery and found
have been reported for sizing hybrid power schemes. Utilizing that it provides more accurate results than SA. In this study,
artificial intelligence is an appropriate approach to enhance the consequently, a robust optimization method based on heuristic
optimization process. Also, to overcome the problems related to methodologies for hybrid optimization is proposed; the results
using one method for optimization, hybrid optimization algorithms from the proposed method are then compared to those from the
can be advantageous. well-known algorithm.
Hatata et al. (Hatata et al., 2018) used a Clonal selection tech- In the present investigation, optimal sizing is investigated for
nique to optimize a hybrid solar and wind scheme with a battery to three hybrid schemes, WT/PV/BAT, WT/BAT, and PV/BAT, consid-
utilize its output with minimum cost and a small fluctuation rate, in ering load demand forecasting, to satisfy all energy requirements
Troyes Barbery Station, France. Belouda et al. (Belouda et al., 2018) with minimal cost (as represented by total life cycle cost (TLCC))
utilized an optimization method for a hybrid power scheme based and maximal reliability index (in the form of loss of power supply
on solar/wind/battery technology in a remote region in Borj Cedria, probability (LPSP)) for a stand-alone remote area in Iran. A
Tunisia. Moghaddam et al. (Moghaddam et al., 2018) used an computationally intelligent technique, namely artificial neural
intelligent flower pollination algorithm to size optimally a hybrid network (ANN), is presented for load forecasting and tabu search
system (solar/wind/hydrogen), minimizing the total life cycle cost. (TS), an effective heuristic algorithm, is proposed for optimal design
Singh and Fernandez, (2018) investigated the effectiveness of of a small independent hybrid scheme, based on wind and solar
Cuckoo Search in using a hybrid solar/wind/battery scheme design technology and a battery (WT/PV/BAT), in northeastern Iran. The
problem in an isolated area in India (Almora district of Uttarak- simulation results are contrasted with results from harmony search
hand). Khiareddine et al. (Khiareddine et al., 2018) used a techno- and simulated annealing approaches to assess the performance
economic optimization model based on an energy management efficiency of the TS algorithm. The effect is illustrated of increased
strategy for optimization of an autonomous hybrid scheme (solar/ reliability on three hybrid system component quantities, TLCC,
wind/battery/hydrogen) with a pumping system in Sahline, Tunisia. LCOE, and LPSP of the optimized system.
Lorestani and Ardehali, (2018) used a simulation model for opti- To further demonstrate the potential of a small independent
mizing an off-grid solar/wind hybrid system with particle swarm solar and wind hybrid power scheme, this paper focuses on two
optimization algorithm. Sarhan et al. (Sarhan et al., 2018) used an main issues: a hybrid optimization approach is developed for the
optimization technique to size an autonomous PV/battery scheme load demand forecasting and sizing of a system consisting of PVs,
in Yemen for the minimum levelized cost of energy rather than net wind turbines, and battery storage for a stand-alone remote area in
present value. Acuna et al. (Acun ~ a et al., 2018) presented an opti- Iran. The spatial model has the ability of forecasting the load de-
mization routine for the size of standalone PV-wind-battery mand by considering environmental and technical criteria. Also, an
scheme. Giallanza et al. (Giallanza et al., 2018) used a high- enhanced optimal sizing approach using TLCC minimization and
resolution analysis for optimizing a hybrid solar-wind-battery maximization reliability (a decrease in LPSP) is proposed for three
system for supplying the load demand while minimizing total off-grid hybrid renewable energy systems based on solar and wind.
cost with an iterative technique. Nafeh, (2011) used a genetic Section 2 of this study defines the improved system's configu-
optimization algorithm for a WT, PV, and battery scheme based ration being studied along with its mode of operation and a theo-
using minimum total cost and power reliability. retical modeling, while Section 3 presents the effective heuristic
Although various facets of small independent hybrid power algorithm model. A case study is included in Section 4 along with
systems, based on wind and solar systems, have been studied, the main simulation and optimization results, while section 5
worthwhile models and accurate techniques for optimization have provides conclusions.
not been extensively utilized for optimizing the design and techno-
economic behaviour of such systems. For instance, such tools have
not been utilized for the forecasting of time series. Also, complete 2. Problem formulation
modeling, sizing and costing studies of independent hybrid power
schemes based on PV and WTs, considering forecasting of load The structure of the small autonomous hybrid power scheme is
demand, to satisfy all energy needs in remote regions, are rarely presented in Fig. 1, which forms the focus of this study, and which
found. involves renewable energy (wind and solar) as the main power
Nonetheless, several optimization methods have been reported sources as well as auxiliary supplies like energy storage (batteries
for sizing hybrid power schemes (Khatib et al., 2012; Ahmadi and (BAT)). The small autonomous hybrid power system complemen-
Abdi, 2016; Edalati et al., 2016; Chaurasia et al., 2017) HOMER tarily integrates the renewable energy sources. In hybrid power
(Fazelpour et al., 2014; Sen and Bhattacharyya, 2014) is a common system optimization, after load forecasting, component modeling is
software/method for optimally designing hybrid schemes. Heuris- performed, and then these are joined to analyse the performance of
tic methods, like genetic algorithms (Mellit et al., 2010; Merei et al., hybrid power system for meeting electrical loads in remote areas.
2013), simulated annealing (Ekren and Ekren, 2010; El-Naggar et al., Fig. 2 shows the overall process for size optimization of indepen-
2012; Garlík and Krivan, 2013), harmony search (HS) (Guangqian dent hybrid power schemes (IHPSs).
Table 1
Summary of recent studies of hybrid energy systems based on solar and wind.

Authors/year Hybrid system Method/technology Main objective Site System type Strategy Remarks

Gholami and Dehnavi/2019 Solar/wind/diesel/CHP PSO and harmony Minimizing cost Iran Grid-connected Cost minimization Particle swarm optimization (PSO)
(Gholami and Dehnavi, 2019) search with and without algorithm is modified for the
uncertainty optimization
Kaabeche and Bakelli/2019 Solar/wind/battery Grey wolf optimizer Minimizing energy Southwest Algeria Stand-alone Loss of power Overall presentation of IHPS with
(Kaabeche and Bakelli, 2019) product cost supply probability various electrochemical energy storage
(LPSP) technologies
Samy et al., /2019 (Samy et al., Solar/wind/hydrogen Firefly algorithm Minimizing total cost Egypt Stand-alone Levelized cost of Solar/wind/hydrogen combination
2019) energy (LCOE) provides good performance
Krishan and Suhag/2019 Solar/wind/battery HOMER Selecting most Yamunanagar, Stand-alone Net present cost Techno-economic analysis and
(Krishan and Suhag, 2019) economically viable Haryana, India and cost of energy optimum sizing of system is proposed
option
Ben Ali et al., /2018 (Ali et al., Solar/wind/RO desalination Genetic algorithm Supplying fresh water Djerba-Midoun, Stand-alone Power Energy management strategy based on
2018) demand Tunisia management fuzzy logic theory is proposed
strategy
Mandal et al., /2018 (Mandal Solar/wind/diesel/battery HOMER Minimizing cost of Northern region of Stand-alone Net present cost Optimized system is economically
et al., 2018) energy and CO2 Bangladesh viable relative to solar home systems

W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

emission
Eteiba et al., /2018 (Eteiba et al., Solar/biomass/battery Harmony search Minimizing cost Monshaet Taher Stand-alone LPSP Optimization of IHPS with several
2018) village, Egypt battery technologies
Singh and Fernandez/2018 Solar/wind/battery Cuckoo search Minimizing total cost Almora district of Stand-alone Cost of energy Modeling of IHPS components
(Singh and Fernandez, 2018) Uttarakhand, India considering hardware status of PV
Vishnupriyan and Manoharan/ Solar/wind/diesel/battery HOMER Minimizing total cost Tamil Nadu, India Stand-alone Net present cost Common model is presented for daily
2018 (Vishnupriyan and global solar radiation estimation for six
Manoharan, 2018) climatic locations
Hatata et al., /2018 (Hatata Solar/wind/battery Genetic algorithm Minimizing total cost Qena Al-Gadida, Stand-alone LPSP Proposes technique for optimum size of
et al., 2018) Egypt IHPS
Giallanza et al., /2018 (Giallanza Solar/wind/battery Modeling Minimizing total cost Three locations in Stand-alone Seasonal loss of Defines reliability parameter (seasonal
et al., 2018) Sicily, Italy load probability loss of load probability ratio) that
ratio accounts for seasonal variability of data
Shi et al., /2017 (Shi et al., 2017) Solar/wind/diesel/battery Simulated annealing Improving utilization of China Stand-alone Power Improved power management strategy
battery bank and management is proposed
avoiding loss of power strategy
supply
Brandoni and Bosnjakovi
c/2017 Solar/wind/biodiesel/battery HOMER Minimizing cost Sb Saharan, Africa Grid-connected Cost of energy Benefits of integrating renewables in
(Brandoni and Bosnjakovi
c, wastewater treatment plants are
2017) studied

pez et al., /2016 (Dufo-Lo
Lo
pez Solar/wind/diesel/battery Genetic algorithm Minimizing cost Navarra, Spain Stand-alone Net present cost Probabilistic approach by Monte Carlo
et al., 2016) simulation combined with GA is
presented
Fathy/2016 (Fathy, 2016) Solar/wind/hydrogen Mine blast Minimizing annual cost Helwan city, Egypt Stand-alone Power Mine blast algorithm is used to solve
management optimal sizing of IHPS
strategy
Chauhan and Saini/2016 Solar/wind/diesel/battery Harmony search Minimizing total cost Uttarakhand state Stand-alone LPSP Size optimization of IHPS is performed
(Chauhan and Saini, 2016) in India
Ahmadi and Abdi/2016 Solar/wind/battery Big BangeBig Crunch Minimizing total cost Qazvin, Iran Stand-alone Net present cost Reliability index of energy not supplied
(Ahmadi and Abdi, 2016) and HS is considered for system reliability
Shezan et al., /2016 (Shezan Solar/wind/diesel/battery HOMER Minimizing total cost Selangor, Malaysia Stand-alone Net present cost Analyzes performance of IHPS
et al., 2016) and CO2 gas and cost of energy
Velik and Nicolay/2016 (Velik Solar/storage Simulated annealing Optimizing energy Austria Grid-connected Power Modified SA optimization is presented
and Nicolay, 2016) management management
strategy
Tahani et al., /2015 (Tahani Solar/wind/battery Simulated annealing Maximizing system Tehran, Iran Stand-alone LPSP LPSP considered as objective function
et al., 2015) reliability and
minimizing system cost

3
4 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

parameters on subsequent days. Each node is assigned a weight

vector to connect it to all nodes in the subsequent layer. Each node
in the MLP-ANN model has the characteristic of receiving inputs,
processing and providing outputs. All nodes of the input layer are
summed and provide input for hidden layer nodes (Toghyani et al.,
Battery bank
Wind turbine 2016; Ozoegwu, 2019).
For forecasting variability, the input information is considered as
a matrix of load demands for the prior hour, which is defined as:
2 3
Load DC Bus 6 Lt1500 ::: Lt2 Lt1 7
PV arrays
6 Lt1501 ::: Lt3 Lt2 7
Input Variables ¼ 6
6
7
Fig. 1. Small autonomous hybrid power scheme. 4 « 1 « « 7 5
Lt1699 ::: Lt201 Lt200

Load demand?
where Lt-1 is the load demand for 1 h ago. Each row is a variable. The
Forecaster important variables are selected in the matrix of the load (i.e.,
feature selection) as a means of neural network training, because
Metrological inputs Forecasted load Specification of system these variables do not need to train the NN. As a result, two ap-
proaches are used in this study. Statistical correlation and mutual
Configuration of proposed independent hybrid information are used in this study, so an input to forecaster 2 is the
power system output of forecaster 1 in the iterative neural networks level and vice
Optimal configuration with maximum power
versa. The stopping criterion is based on the errors being almost the
reliability and minimum cost same for both forecasters. This criterion is monitored, with the
iterative procedure terminated when it is met. Otherwise, the
iterative process repeats. Note that the load demand is utilized for
No
Optimized neural network training. Here, the data set is randomly divided into
combinations?
training, testing and validation sets, with 86% of total dataset
Yes
regarded as the training set and the remaining data as testing and
validation sets.
Select result

Fig. 2. Overall process for size optimization of IHPSs.

2.1.1. Statistical correlation
Statistical correlation is only briefly discussed here since it is
2.1. Load forecasting described extensively elsewhere (Zhu et al., 2010; Ashkezari et al.,
2014; Maleki et al., 2016a,b). For variables B and A having respec-
Hourly data of wind speed, air temperature, solar irradiation tive expected values mB and mA along with standard deviations sB
at particular locations, and load demand (in 2013e2014) for the and sA, the population correlation coefficient corr(A, B) can be
studied site, Davarzan, Khorasan, Iran (latitude ¼ 36.35 , longi- written as:
tude ¼ 56.83 ), are plotted in Fig. 3 for a one year period (IREO,
2018). covðA; BÞ E½ðA  mA ÞðB  mB Þ
corrðA; BÞ ¼ ¼ (1)
Iterative and direct forecasting methods are proposed for day sA sB sA sB
ahead forecasting (Reis and Da Silva, 2005; Pao, 2007). Since direct
forecasting frequently incurs significant rounding errors (Pao, Here, cov denotes covariance and E the expected value of the
2007; Zhang et al., 2019), in the present investigation an iterative operator. Both standard deviations must be non-zero and limited
forecasting approach is utilized (see Fig. 4). Normalized data are for the Pearson correlation to be valid, due to the CauchyeSchwarz
used for the feature selection process, before entering the iterative inequality, which requires the absolute value of the correlation not
neural networks stage. For artificial neural networks (ANNs), to exceed unity. The correlation coefficient is symmetrical, i.e.,
multilayer perceptron (MLP) is the most typical approach corr(A, B) ¼ corr(B, A). The value of the Pearson correlation is be-
employed (Hemmati-Sarapardeh et al., 2018), as it is a common tween 1 and 1. A value of 1 implies an increasing linear rela-
approach for both classification and regression problems (Qazi tionship (i.e., a positive correlation), and of 1 implies a decreasing
et al., 2015; Niu et al., 2016; Renno et al., 2016). In the present linear relation (i.e., a negative or anti correlation). The relationship
study, the multilayer perceptron artificial neural network (MLP- between the variables is stronger as the correlation coefficient
ANN) method is used, which is shown in its general form in Fig. 5. It shifts towards either 1 or 1. Less correlation is evident when the
can be seen that the MLP-ANN method has several nodes in three Pearson correlation value approaches zero, but when the correla-
layers. The input layer provides input data. Several inter-layers tion coefficient is 0, the variables are independent or uncorrelated,
(hidden layers) exist, with a number of neurons. The final but the variables may exhibit some dependence since only linear
(output) layer gives the results forecasted by the ANN. The numbers dependencies are assessed by the correlation coefficient.
of neurons in the input and output layers respectively are normally The expected value can be written for a random variable as:
determined based on the desired input and output variables. Here,
X
n
the number of hidden neurons is 12, based on historical load data, E½A ¼ at pa ðat Þ (2)
the number of days per month and weather data as input factors. t¼1
The output gives the forecast load (FL), i.e., the hourly load
and the standard deviation as:
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 5

20
Wind speed

15

(m/s)
10

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000

1200
Solar irradiation
1000

800
(W/m2)

600

400

200

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000

50
Ambient temperature
40

30
(oC)

20

10

-10
0 1000 2000 3000 4000 5000 6000 7000 8000 9000

8
Load demand
7

5
(kW)

1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time (h)

Fig. 3. From top to bottom, wind speed, solar insolation, ambient temperature, and load profile over one year.
6 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

Load 2.1.2. Mutual information

Mutual information is described briefly here, as it is discussed
extensively elsewhere (Tsai et al., 2013; Shu and Qian, 2014). The
mutual information (I) for two random discrete variables B and A
can be expressed formally as follows (Peng et al., 2005):

XX  
Preprocessing pða; bÞ
IðA; BÞ ¼ pða; bÞlog2 (4)
pðaÞpðbÞ
b2Ba2A

Here, p(a, b) denotes the joint probability distribution function

Feature selection for A and B, and p(b) and p(a) denote respectively the marginal
probability distribution functions for B and A. When mutual infor-
mation is presented as continuous random variables, the summa-
tion becomes the following definite double integral:
Normalization
ðð  
pða; bÞ
IðA; BÞ ¼ pða; bÞlog2 (5)
pðaÞpðbÞ
BA
Forecaster 1
In this case, p(a, b) denotes the joint probability density function
for A and B, while p(b) and p(a) denote respectively the marginal
probability density functions for B and A.
Forecaster 2 The mutual information is a measure of the information that B
and A share. It indicates how much uncertainty reduction occurs in
Iterative Neural Networks one of these variables by knowing the other. For instance, mutual
information of A and B is zero if they are independent, since
knowing A does not provide details about B and vice versa, but A
Forecast load shares information with B if they are similar, so in that case
knowing A specifies the value of B and vice versa. When A and B are
independent random variables and the mutual information is zero,
Fig. 4. Block diagram of forecaster.
p(a, b) ¼ p(a) p(b). Then, part of equation (4) or (5) can be written
as:

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
u n pða; bÞ
uX log ¼ log 1 ¼ 0 (6)
s ¼ t ðat  aÞ2 (3) pðaÞpðbÞ
t¼1
The mutual information can be written equivalently using en-
tropies as:

Fig. 5. Structure of MLP-ANN

 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 7

Here, Tref and hr-PV are the cell temperature (normally 25  C) and
IðA; BÞ ¼ HðAÞ  HðAjBÞ the efficiency of the reference module, and hpc denotes the power
¼ HðBÞ  HðBjAÞ conditioning efficiency. Also, NT denotes the efficiency temperature
(7)
¼ HðAÞ þ HðBÞ  HðA; BÞ factor of the photovoltaic collector. This term takes on a value
¼ HðA; BÞ  HðAjBÞ  HðAjBÞ of 3.7  103  C1 for poly and monocrystalline silicon (Ismail
Here, for A and B respectively, H(A) and H(B) are marginal en- et al., 2013). Finally, Tc denotes the PV cell temperature ( C),
tropies, and H(A|B) and H(A|B) conditional entropies. Also, H(A,B) is expressible as (Bakelli et al., 2011):
the joint entropy of the parameters. Details on the entropy func-  
tions are provided elsewhere (Kwak and Choi, 2002). In Fig. 6,
NOCT  20
Tc ¼ TA þ Rt (12)
mutual information functions are presented along with their re- 800
lations to H functions. Here, TA denotes ambient air temperature (often manufacturer
Here, we check for errors by using the mean absolute percentage specified), while NOCT denotes nominal operating cell tempera-
error (ME). This can be expressed as: ture; we assume the photovoltaic collectors use a maximum power
" # point tracking system and NOCT ¼ 43  C.
1X n
jPðtÞ  AðtÞj
ME ¼  100 (8)
n t¼1 AðtÞ 2.3. WT generator
Here, P(t) and A(t) denote respectively the forecast and actual
The electrical power generation of a wind turbine (WT) is
data for time t.
expressible as (Caballero et al., 2013):
The mutual information expressions are seen in equations (4)
and (5) to incorporate base 2 logarithms (Peng et al., 2005), sug- 8
>
> SðtÞ  SI
gesting it is appropriate to apply binomial distributions to inputs >
> 0 if
<
and outputs. To start, therefore, the target variable and candidate c  S3 ðtÞ  d  Pr if SI < SðtÞ < SR
PWT ðtÞ ¼ (13)
inputs are normalized linearly within the range 0e1. This step is >
> Pr if SR  SðtÞ < SO
>
>
represented in the data preparation part of Fig. 4 by the heading : 0 if SðtÞ  SO
“Normalization”. Here, data are normalized as follows (Amjady,
2002; LIU, WANG et al., 2009): where

train 8 . 
ntrain ¼ (9) >
< c ¼ Pr S3R  S3I
MaxðtrainÞ .  (14)
>
: d ¼ S3I S3R  S3I
Here, ntrain denotes a normalized train matrix while train de-
notes an input matrix, generated via the process of feature
selection. and the rated power Pr-WT is expressible as

2.2. Photovoltaic collector 1

Pr ¼ AWT ,CP ,ra ,hWT ,S3R (15)
2
The electric power generated by a photovoltaic collector can be
written as (Bakelli et al., 2011, Akbar Maleki et al., 2016a,b): where AWT and ƞWT are the swept area and the efficiency of WT,
respectively, S, SR, SO, and SI are the wind speed, rated wind speed,
PPV ðtÞ ¼ Rt hPV APV (10) cut-off and cut-in speed, CP is the power coefficient, and ra is the air
density.
Here, Rt is solar irradiation (in kW/m ), hPV and APV are efficiency
2

and area of the PV panels (in m2). The latter term can be written as:
2.4. Battery
h  i
hPV ¼ hrPV hpc 1  NT Tc  Tref (11)
According to the time consumption, the state of charge (SOC)
l ðtÞ
can be expressed for the battery. WhenEG ðtÞ  ED
hinv , the charge
level can be expressed as
 
EDl ðtÞ
H(A) SOCðtÞ ¼ SOCðt  1Þ,ð1  sÞ þ EG ðtÞ  hbc (16)
H(B) hinv
l ðtÞ
When EG ðtÞ  ED
hinv , the discharging level can be written as
follows:
 
H(A|B) I(A;B) H(B|A) EDl ðtÞ
SOCðtÞ ¼ SOCðt  1Þ,ð1  sÞ   EG ðtÞ hbd (17)
hinv

where EG, EDl, are the produced energies and energy demand,
respectively, hinv is the inverter efficiency, s is the hourly self-
discharge rate, hbd andhbc are the discharging and charging effi-
ciencies for the storage system.

H(A,B) 2.5. LCC

Fig. 6. Demonstration of mutual information and several entropies. The cost analyses of the hybrid WT/PV/BAT, WT/BAT, and PV/
8 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

BAT schemes use the life cycle cost (LCC), expressible as

X AWT  0 (27)
LCC ¼ AC þ AM (18)
APV  0 (28)
where AM and AC denote the annual maintenance and capital costs.
NBAT  0 (29)
2.5.1. LCC of WTs
The maintenance (AMWT) and capital cost (ACWT) of wind tur- PWT ðtÞ þ PPV ðtÞ þ PBAT ðtÞ  PT ðtÞ (30)
bines is formulated as follows:
LPSPDesired  LPSP (31)
ð1 þ ir Þn ir
ACWT ¼ AWT ,CWT , (19)
ð1 þ ir Þn  1
SOCmax  SOCðtÞ (32)
AMWT ¼ AWT ,CMntWT (20)
SOCmin  SOCðtÞ (33)
where AWT, CWT, and CMnt-WT are the swept area of the wind turbine
Here, SOCmax and SOCmin denote the maximum and minimum
blades, and the unit and annual maintenance costs respectively of
states of charge. We can write
each WT, n and ir are life of the system and the interest rate.
SOCmin ¼ SBAT ð1  DODÞ (34)
2.5.2. LCC of PV panels When SOC(t) < SOCmin, the storage system does not meet part of
The capital and annual maintenance cost of the photovoltaic the load. Then, SOC(t) ¼ SOCmin, and the we can write the loss of
panels is written as: power supply (LPS) as follows:

ir ð1 þ ir Þn EDl ðtÞ
ACPV ¼ APV ,CPV , (21) LPSðtÞ ¼  ½SOCðt  1Þ , ð1  sÞ  SOCmin hbd  EG ðtÞ
ð1 þ ir Þn  1 hinv
(35)
AMPV ¼ APV ,CMntPV (22)
The loss of power supply probability (LPSP) needs to be
where CMnt-PV, CPV, and APV are the annual maintenance cost and considered for hybrid renewable energy schemes to have good
unit cost of each solar panel, and the area for the PV schemes, reliability. LPSP ranges from 0 to 1, with LPSP ¼ 0 implying the load
respectively. always is met and LPSP ¼ 1 implying it is never met. The re-
quirements of power reliability differ for systems with different
load characteristics. The maximum allowable LPSP (LPSPDesired) is
2.5.3. LCC of batteries (BAT) approximately 0.02 for rural and stand-alone applications
The capital and annual maintenance cost of the storage system (Caballero et al., 2013). As is the case in this study, we assume
(battery) can be expressed as: LPSPDesired ¼ 0.02 here, i.e., LPSP <0.02 is applied as a reliability
constraint in the optimization of the power generation system. LPSP
ir ð1 þ ir Þn can be calculated for a quantified time, T, as follows:
ACBAT ¼ NBAT ,PWBAT , (23)
ð1 þ ir Þn  1
P
T
LPSðtÞ
AMBAT ¼ NBAT ,CMntBAT (24) t¼1
LPSP ¼ (36)
P
T
where ELoad ðtÞ
t¼1
X 1
PWBAT ¼ CBAT (25) where DOD, SBAT, PPV, PWT, PBAT, PT denote maximum depth of
k¼0;5;10;15ð1 þ i r Þk
discharge, nominal capacity of the storage system, the solar and
Here, PWBAT, NBAT, CMnt-BAT, and CBAT are the present worth of wind power, total load demand, and discharged/charged battery
battery, the number of batteries, the annual maintenance cost and power, respectively. Fig. 7 shows the operating strategy of the in-
unit cost per battery, respectively. dependent hybrid power schemes at each hour during the year.

2.6. Objective function and constraints 2.7. LCOE

The design aim is reduction of the LCC or first cost based on LPSP A significant economic system parameter, the levelized cost of
and other constraints, this optimization ability to significantly energy (LCOE) represents the average unit cost for useful electricity
minimize system costs and energy usage. This can be expressed as: developed by the scheme, and can be written as follows:
X
Minimize LCCðAWT ; APV ; NBAT Þ ¼ ACm þ AMm (26) Total Life Cycle Cost
m2WT;PV;BAT LCOE ¼ (37)
Total Lifetime Energy Production
and the constraints are:
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 9

Start

Historical data

Solar radiation Load demand

Temperature Forecaster

Wind speed Forecasted load

j=1

Configuration j

Calculate the total power production

t =1

Yes No
Over production?

Battery is charging Battery is discharging

Calculate SOC Calculate SOC

Yes No No Yes

SOC > SOC max SOC < SOC min

SOC = SOC max LPSP = 0 SOC = SOC min

Yes
t=t+1 t < t-max Calculate LPSP

No

Yes

j < j-max

No

End

Fig. 7. Operating strategy of IHPS for each hour during the year.

3. Tabu search (TS) a great family of derivative methods that present memory ar-
rangements in Meta-heuristics, such as parallel TS and reactive TS.
Tabu search is a metaheuristic or Meta-strategy in other words, The TS method proposed here is the same as that presented in (Wu
a global optimization method for monitoring an embedded heu- et al., 2018). The objective of the TS method is to limit an embedded
ristic algorithm, which proposed by Glover, (1986), TS is a parent for heuristic from returning to newly visited areas of the search space.
10 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

The algorithm strategy is to preserve a short-term memory of the Askarzadeh, 2014; Maleki and Pourfayaz, 2015) and the simulated
particular changes of new moves within the search space and annealing method (Ekren and Ekren, 2010; Velik and Nicolay,
stopping upcoming moves from undoing those changes. The steps 2016), 30 independent runs are implemented and the maximum
in TS are as follows (Abido, 2002; Wu et al., 2018): TLCC values (Worst), the average of the TLCC values (Mean), the
minimum TLCC values (Best), the standard deviation of the values
1 Iteration index (iter) set to 0 and initial random solution (xinitial) of TLCC (STD), as well as the mean, minimum, and maximum
generated, which becomes current and best solution xbest simulation time indexes are reported. The factors related to the
(xinitial ¼ xcurrent ¼ xbest). components and the heuristic methods are set in Tables 2 and 3.
2 Set of trial solutions (ntrial) generated near current solution. Each Fig. 11 shows the forecasted information from the suggested tech-
trial solution (xtrial) is input to objective function and quality is nique, i.e. hourly values of the load demand for Davarzan. The mean
evaluated. Trial solutions sorted based on values of objective absolute percentage error is 4.77, the number of hidden neurons is
function. xjtrial defined as jth trial solution in sorted set, where 12, and the thresholds of the first and second filters are 0.3, and 0.8,
respectively.
1  j  ntrial , so x1trial is best trial solution based on objective
function values.
3 Parameter j set to 1. If CT ðxjtrial Þ > CT ðxbest Þ, proceed to Step 4. 4.2. Comparison of TSLF with HSLF and SALF
j j
Otherwise set xbest ¼ xtrial xbest ¼ xtrial before Step 4. Heuristic methods optimize the design of PV panels, WTs, and
4 Tabu status of xjtrial
is checked. If in tabu list, solution proceeds to batteries for the IHPSs in order to minimize the TLCC while satis-
Step 5. Otherwise, value is added to tabu list, and set xcurrent ¼ fying constraints for LPSPDesired. The optimization problem consists
of three elements, which are integer (NBAT) and continuous (AWT
xjtrial before Step 7.
and APV) decision variables.
5 Criterion of xjtrial is checked. If satisfied, tabu restrictions are The results for the investigated algorithms-based load fore-
j
overridden and aspiration level is updated to xcurrent ¼ xtrial casting for several small autonomous hybrid power schemes are
before Step 7. Otherwise, set j ¼ j þ 1 and proceed to Step 6. given in Table 4. It is seen that the minimal TLCC (Best index) with
6 If j > ntrial, proceed to Step 7. Otherwise, return to Step 4. forecast information of the WT/PV/BAT scheme is $13,400, as ob-
7 Stopping criterion is checked. tained with the TSLF method followed by the HSLF ($18,300) and
SALF ($25,600). The relative error among the Best indexes for TSLF
The computational flow of TS algorithm is shown in Fig. 8. For and HSLF, BestTSLF BestHSLF
 100, is 37%. The relative error among the
BestTSLF
validation of the TS algorithm, harmony search (HS) and simulated
annealing (SA) are used. HS is a heuristic method that tries to mimic Best indexes for TSLF and SALF algorithm is 91%. Also, TSLF performs
musician improvisation. Details on the HS algorithm are provided better e in term of the mean simulation time e than HSLF and SALF.
elsewhere (Maleki and Askarzadeh, 2014; Maleki and Pourfayaz, It is shown that TSLF has the lowest Best, Worst, Mean, and STD
2015). The steps in HS are given in Fig. 9. SA is a probabilistic TLCC values for the PV/WT/BAT system compared to the other
method for approximating the global optimum of a specified methods. Also, TSLF has the lowest mean, minimum, and maximum
function, i.e., it is a heuristic method to estimate the global opti- simulation time indexes for the PV/WT/BAT system compared to
mum in a large search space for an optimization problem. SA is a HSLF and SALF. In the hybrid WT/BAT system, the electrical load is
heuristic algorithm including heating and controlled cooling of a supplied using wind energy. It can be seen that the Best index
material that grows the size of its crystals and decreases their faults. obtained with TSLF is similar to that obtained with SALF and better
Details on the HS algorithm are provided elsewhere (Ekren and than that obtained with HSLF. The TSLF method is more advanta-
Ekren, 2010; Velik and Nicolay, 2016). The steps in SA are given in geous than the HSLF and SALF methods according to the Best,
Fig. 10. Worst, STD, and Mean values, and simulation time indexes (Mean,
Min., and Max.). Finally, the value of TLCC for the optimal WT/BAT
system is seen to be $58,100, and the minimum convergence time is
4. Results and discussion
found to 0.22 s, using TSLF. In the hybrid PV/BAT scheme, the power
is derived from solar energy only. The Best index obtained by TSLF
The advantages are illustrated of the suggested methodology
($12,700) is less than that obtained by HSLF ($13,600) and SALF
based on load forecast data with a case study. It is conducted to
($14,600), and the relative error between the Min. index of TSLF and
examine an off-grid hybrid scheme which supplies the residential
HSLF is 7%, and the relative error between the Min. index of TSLF
electrical demand in the northeastern Iran. The analysis is per-
and SALF is 15%. It is seen that the mean convergence time (Mean
formed for three hybrid systems based on wind energy, solar en-
index) with forecast information of the PV/BAT scheme is 0.25 s, as
ergy and battery energy storage. Matlab/Simulink is used to
obtained with the TSLF method followed by the HSLF (0.32 s) and
implement the suggested approach.
SALF (0.33 s). Also, the TSLF performance is more promising than
that for HSLF and SALF, in terms of other indexes. As a result, the
4.1. Confirmation of reliability and validity of TSLF proposed algorithm (TSLF) has the lowest Worst, Mean, Best, and
the STD values of TLCC compared to the HSLF and SALF methods.
To confirm the reliability and validity of the TS algorithm-based Also, the TSLF algorithm has the lowest Mean, Min., and Max. values
load forecasting (TSLF) to meet all the energy needs with the hybrid of simulation time for all hybrid system compared to the HSLF and
solarewind schemes, the harmony search algorithm-based load SALF methods with 30 independent runs. As a result, the TSLF al-
forecasting (HSLF) and simulated annealing algorithm-based load gorithm not only succeeds in finding the ideal TLCC of the hybrid
forecasting (SALF) are programmed and applied to optimize the scheme but also converges to the ideal result faster than the other
proposed systems. For performance comparisons of the suggested methods. Fig. 12 shows the TSLF method convergence for optimi-
method with the harmony search (HS) method (Maleki and zation of the hybrid schemes, namely PV/WT/BAT, PV/BAT, and WT/
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 11

Initialization Step

Set iter = 0

Generate xinitial
Calculate CT (xinitial)

Set xinitial = xcurrent = xbest

Generate a set of trial solution S (xcurrent)

Sort S in ascending order based on J of its
elements. Set j = 1

No
j
CT x trial CT x best ?

j
Yes Set x best x trial

No
j
Is x trial tabu ?

j
Set j = j + 1 Yes Add x trial to the tabu list
j
Set x current x trial
No Yes
Aspiration criterion
satisfied?

No Update the aspiration level

j
Yes Set x current x trial
j > ntrial ?

Yes Stopping criterion No

Stop Set iter = iter + 1
satisfied?

Fig. 8. Computational flow of TS algorithm.

12 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

BAT. It can be seen that the ideal objective function value for each
Step 1: Initialize the problem and algorithm parameters
iteration of the TSLF method. Fig. 12 corresponds to the ideal TSLF
method performance over 30 runs for each case. For PV/BAT, the
Step 2: Initialize the harmony memory objective function value is $53,000 at the start. As the method
proceeds, the value of the TLCC decreases, i.e., new solutions are
found. Consequently, much of the TLCC is saved relative to the first
Step 3: Improvise a new harmony iteration.
To test the reliability of the proposed method, the investigation
methods in terms of optimal solution and cost saving after 30 in-
Step 4: Update the harmony memory dependent runs are compared in Table 5. The largest TLCC, which is
obtained with the SALF algorithm for the PV/WT/BAT and PV/BAT
systems and obtained with the HSLF algorithm for the WT/BAT
Step 5: Check the stopping criterion
system, is selected as a base case and the cost saving for the other
algorithms are compared to it. For the PV/BAT system, TSLF exhibits
Fig. 9. Steps in HS.
a cost saving of about 13% compared to the SALF algorithm, 48%
compared to the SALF algorithm for the PV/WT/BAT system, and
0.3% compared to the HSLF algorithm for the WT/BAT system. Thus,
Step 1: Set the initial temperature and create a random initial solution the TSLF method is reliable, valid and efficient for solving the
optimization problem of the hybrid system based on forecast data.
To support this finding, the minimum TLCC values for the hybrid
Step 2: Start looping until a stop condition is met (calculate objective
function) systems found with the investigated methods are shown in Fig. 13.
To ensure the reliability of the proposed algorithm based on
TSLF, the performance of each algorithm is compared after
Step 3: Generate a random neighboring solution numerous independent runs for the PV/BAT system. The compari-
son of the optimal solutions obtained via the investigated methods
Step 4: Calculate the new solutions is shown in Table 6. It can be seen that the Best index obtained with
TSLF after 30 independent runs ($12,700) is similar to that obtained
Step 5: Compare them: with 40e200 independent runs ($12,700). The performance of the
If new solution < cold: move to the new solution HSLF algorithm in terms of the Best index for 200 independent runs
If new solution > cold: maybe move to the new solution ($12,900) is better than for 40 independent runs ($13,600). Simi-
larly, the Best solution found by SALF for 200 independent runs
Step 6: Repeat steps 3-5 above until an acceptable solution is found or ($12,700) is smaller than that found for 40 independent runs
reach some maximum number of iterations ($14,600). Finally, the performance of the TSLF algorithm in terms
of the minimum TLCC value ($12,700) is better than for the SALF
Fig. 10. Steps in SA ($12,700) and HSLF ($12,900) algorithms. Thus, the performance

Table 2
Simulation input data.

Interest rate (ir) 10%

Project lifetime (n) 20 years

Cut-in wind speed (SI) 3 m/s
Cut-out wind speed (SO) 20 m/s
Nominal wind speed (SR) 9 m/s
Wind turbine cost (CWT) 574.5 $/m2
Annual maintenance costs of wind turbine (CMnt-WT) 0.05CWT $/m2/year
Power coefficient (CP) 0.59
Air density (ra) 1.225 kg/m3
Wind turbine reference efficiency (ƞWT) 85%
Wind turbine lifetime 20 years
PV initial cost (CPV) 285.3 $/m2
Annual operation and maintenance cost of PV system (CMnt-PV) 0.03  CPV $/m2/year
PV panel reference efficiency (hr-PV) 15.8%
PV system lifetime 20 years
Nominal capacity of battery bank (SBAT) 2.1 kWh
Charge efficiency of battery bank (ƞbc) 85%
Discharging efficiency of battery bank (ƞbd) 100%
Battery price (CBAT) $310
Battery lifetime 5 years
Annual maintenance costs of battery (CMnt-BAT) 10 $/year
Maximum depth of discharge (DOD) 80%
Hourly self-discharge rate (s) 0.02%
Nominal converter/inverter power 3 kW
Converter/inverter efficiency (hinv) 95%
Converter/inverter lifetime 10 years
Converter/inverter price CConv/Inv $1583
Annual maintenance costs of converter/inverter (CMnt-Inv) 15 $/year
Maximum allowable value of loss of power supply probability (LPSPDesired) 0.02
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 13

Table 3 over 0e200 independent runs is better for the TSLF algorithm than
Parameters for heuristic algorithms. the other algorithms.
Tabu search (TS)
Trial solutions (ntrial) 10
Maximum number of iterations (itermax) 100 4.3. Comparison of three hybrid systems (PV/WT/BAT, PV/BAT, WT/
Harmony search (HS) BAT)
Harmony memory considering rate (hmcr) 0.9
Maximum pitch adjusting rate (parmax) 1
Table 7 summarizes the results obtained with the studied
Minimum pitch adjusting rate (parmin) 0.1
Maximum bandwidth of generation (BGmax) 1 methods (TSLF, HSLF, and SALF) for the PV/WT/BAT schemes. It can
Minimum bandwidth of generation (BGmin) 0.01 be seen that TLCC is $13,400 and the values of LCOE and LPSP
Maximum number of iterations (itermax) 1000 respectively are 0.209 $/kWh and 1.65%. Moreover, the optimal
Simulated annealing (SA) values of PV surface area, WT swept area, and number of batteries
Initial temperature (T0) 100
Maximum number of iterations (itermax) 1000
respectively are 149.2 m2, 6.3 m2, and 60. Also, the optimal values of
Constant (wf( 5 PV cost, WT cost, and cost of batteries are $5912, $624, and $5507,
Step size (s) 0.97 respectively. Therefore, TSLF achieves a cost saving of about $12,200
compared to the SALF algorithm and $4900 compared to the HSLF
algorithm. Fig. 14 shows the contribution of each device to the total
LCC of the optimized PV/WT/BAT system.

10
9 Target
Forecast
8
Load (KW)

7
6
5
4
3
100 200 300 400 500 600 700
Time (h)
Fig. 11. Load forecasting, April 21 to May 21, 2013.

Table 4
Results from investigated methods for small autonomous hybrid power systems.

Method Proposed hybrid system TLCC ($) Simulation time (sec)

Best Worst Mean STD Mean Min. Max.

TSLF PV/WT/BAT 13,400 39,800 21,690 9100 0.28 0.22 0.31

PV/BAT 12,700 47,200 22,860 10,000 0.25 0.22 0.30
WT/BAT 58,100 72,500 60,960 3920 0.24 0.22 0.27
HSLF PV/WT/BAT 18,300 62,600 35,200 13,860 0.35 0.31 0.47
PV/BAT 13,600 44,800 24,220 6870 0.32 0.31 0.34
WT/BAT 58,300 88,700 69,530 8850 0.33 0.31 0.38
SALF PV/WT/BAT 25,600 79,900 61,990 15,290 0.34 0.31 0.38
PV/BAT 14,600 85,700 58,410 20,540 0.33 0.30 0.50
WT/BAT 58,100 101,300 84,350 13,240 0.32 0.30 0.38

Fig. 12. Convergence characteristic of TSLF for the hybrid schemes.

14 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

Table 5
Comparison of investigated methods in terms of optimal solution and cost saving
after 30 runs.

Hybrid system Method TLCC ($) Cost saving ($) % cost saving

PV/WT/BAT TSLF 13,400 12,200 48

HSLF 18,300 7300 29
SALF 25,600 N/A N/A
PV/BAT TSLF 12,700 1900 13
HSLF 13,600 1000 7
SALF 14,600 N/A N/A
WT/BAT TSLF 58,100 200 0.3
HSLF 58,300 N/A N/A
SALF 58,100 200 0.3

Fig. 14. Contribution of each device to total LCC of optimized PV/WT/BAT system.

hybrid WT/BAT scheme for load supply for remote areas. The effects
of varying LPSP on the optimal characteristics of the optimized WT/
BAT system by TSLF are presented in Table 9. It can be seen that the
optimal values of WT swept area and number of batteries respec-
tively for LPSPDesired ¼ 0% are 147.6 m2 and 505, and for LPSPDe-
2
sired ¼ 10% are 135.1 m and 345. As a result, the number of
components of the optimized WT/BAT system decreases with
increasing LPSPDesired (Fig. 15). The variations of LPSP, LCOE, and
TLCC of the optimized system with LPSPDesired of the WT/BAT sys-
tem are illustrated in Fig. 16. It can be seen that LCOE and TLCC
decrease as LPSP increases. For LPSPDesired ¼ 0%, LCOE is 0.918
$/kWh, while TLCC is $62,100 and LPSP is 0%, while the values of
LCOE, TLCC, and LPSP are 0.746 $/kWh, $46,200, and 8.68%,
respectively, for LPSPDesired ¼ 10%. This variability is mainly due to
the difference in sizes of components of the optimized WT/BAT
Fig. 13. Best TLCC of hybrid system by investigated methods. system, particularly WT swept area and the number of batteries.
Table 10 lists the results obtained with the proposed methods
for the PV/BAT schemes considered. It can be seen that TLCC is
Table 6 $12,700 and the optimal sizing has APV ¼ 157.4 m2 and NBAT ¼ 61,
Comparison of optimal solution obtained via investigated methods after several
while the LCC of the PV panels and batteries represents 49% and
runs for PV/BAT system.
44% of the system TLCC respectively. The optimal values of photo-
No. of runs TLCC ($) voltaic panels and batteries with the HSLF algorithm are found to be
TSLF HSLF SALF 157.4 m2 and 71, respectively. Also the ideal values of PV panels and
40 12,700 13,600 14,600
batteries with the SALF algorithm are found to be 159.1 m2 and 81,
50 12,700 13,600 14,600 respectively. The proposed method achieves a cost saving of about
70 12,700 13,300 14,600 $1900 compared to the SALF algorithm and $900 compared to the
80 12,700 13,300 14,600 HSLF algorithm; also, the HSLF algorithm achieves a cost saving of
100 12,700 12,900 14,300
about $1000 compared to the SALF algorithm. Compared to the WT/
120 12,700 12,900 14,300
150 12,700 12,900 12,700 BAT system, the optimal number of batteries is lower. Also, the
200 12,700 12,900 12,700 optimal values of LCOE and LPSP respectively are 0.195 $/kWh and
1.59%. Table 11 shows the effect of varying LPSP on the optimal
characteristics of the optimized PV/BAT system by the best algo-
The characteristics of WT/BAT system optimized by heuristic rithm, TSLF. For LPSPDesired ¼ 1%, the TLCC is $13,400 and the values
algorithms are listed in Table 8. It can be seen that the TLCC, the of LCOE and LPSP respectively are 0.204 $/kWh and 0.79%. More-
optimal value of WT swept area and the number of batteries are over, the optimal values of PV surface area and number of batteries
$58,100, 144.4 m2, and 465, respectively. Also, the LPSP and LCOE respectively are 159.1 m2 and 68, and for LPSPDesired ¼ 5% the cor-
values respectively are 0.91% and 0.878 $/kW, and that the WT and responding values are 152.5 m2 and 47. The hybrid system
battery contributions to the TLCC are 25% and 73%, respectively. As a component sizes are increased as the reliability of system increases.
result, the hybrid WT/PV/BAT scheme is more beneficial than the Fig. 17 shows the variations of LPSP, LCOE and TLCC of the optimized

Table 7
Optimal characteristics of optimized PV/WT/BAT system by heuristic algorithms.

Method APV (m2) AWT (m2) NBAT PV cost ($) Wind cost ($) BAT cost ($) TLCC ($) LPSP (%) LCOE ($/kWh)

TSLF 149.2 6.3 60 5912 624 5507 13,400 1.65 0.209

HSLF 132.8 18.8 107 5263 1871 9820 18,300 0.99 0.289
SALF 154.2 3.1 194 6107 312 17,805 25,600 0 0.395
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 15

Table 8
Optimal characteristics of optimized WT/BAT system by heuristic algorithms.

Method AWT (m2) NBAT Wind cost ($) BAT cost ($) TLCC ($) LPSP (%) LCOE ($/kWh)

TSLF 144.4 465 14,347 42,676 58,100 0.91 0.878

HSLF 144.4 467 14,347 42,860 58,300 0.89 0.880
SALF 144.4 465 14,347 42,676 58,100 0.91 0.878

Table 9
Effect of variation of LPSP on the optimal characteristics of optimized WT/BAT system by TSLF.

LPSPDesired (%) AWT (m2) NBAT Wind cost ($) BAT cost ($) TLCC ($) LPSP (%) LCOE ($/kWh)

0 147.6 505 14,659 46,348 62,100 0 0.918

1 144.4 466 14,347 42,768 58,200 0.88 0.879
2 144.4 465 14,347 42,676 58,100 0.91 0.878
5 141.3 425 14,035 39,005 54,100 3.58 0.836
10 135.1 345 13,412 31,663 46,200 8.68 0.746

system vs. LPSPDesired for the PV/BAT system, while Fig. 18 shows the
variation of number of components and their costs for the opti-
mized system vs. LPSPDesired of the PV/BAT system.
To improve the concept, the difference between the state of
charge for the battery, the load demand and the produced power
for the optimal design of the PV/BAT scheme are shown in Fig. 19 for
three days of the year. During hours when the generated electrical
power exceeds load, the storage system is in a charging state and
when the load exceeds the generated electrical power, the storage
system is in a discharging state and supplies the load.
Several hybrid systems are compared. The results suggest that
the WT/BAT scheme exhibits a higher overall cost relative to the PV/
BAT system, and demonstrates that adding solar energy is an
effective policy. As a result, using the PV/BAT system is an effective
system since its life cycle cost (Fig. 20). The difference between the
Fig. 15. Variation of number of components and their cost for optimized system vs.
LPSP of WT/BAT system.
TLCC of the PV/WT/BAT with and without the PV increases signif-
icantly with increasing AWT. It is detected that the hybrid energy
scheme without wind scheme raises the installed PV scheme size.
In terms of the portion that each component contributes to the LCC,
a
the highest cost component for the optimum solution for the PV/
BAT scheme obtained with forecast information is the PV system,
which comprises 49% of the TLCC; the next highest capital cost is
for the batteries (44%). In the PV/WT/BAT scheme and the WT/BAT
scheme respectively, the highest cost components of the optimum
scheme determined using forecast information are the batteries,
which are responsible for 54% and 73% of the TLCC. It is also
demonstrated that the PV part of the TLCC (for the PV/WT/BAT
scheme) and wind (of the WT/BAT system) is 29% and 25%,
respectively.

5. Conclusions
b
In small autonomous hybrid power systems, the electric power
produced with renewable sources directly depends on the load
demand. Photovoltaic panels and wind turbines can be combined
with energy storage system (battery) to totally supply the elec-
tricity demand while providing high reliability. This study assesses
the performance of tabu search optimization for optimally
designing a hybrid renewable energy scheme (WT/PV/BAT) to
minimize TLCC of the scheme subject to relevant limits by modi-
fying three decision variables, namely, the swept area of the wind
turbines, the PV surface area, and the number of batteries. The
reliability index of LPSP is measured. The effects of increased reli-
ability on the hybrid system component sizes, the TLCC, the LCOE,
Fig. 16. (a) Variation of LPSP and TLCC of optimized system vs. LPSPDesired; and (b) and the LPSP of the optimized system are illustrated. The results are
variation of LCOE and TLCC of optimized system vs. LPSPDesired of WT/BAT system. presented of the operation simulation and optimal design for the
studied hybrid renewable energy schemes using load forecast data,
16 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

Table 10
Optimal characteristics of optimized PV/BAT system by heuristic algorithms.

Method APV (m2) NBAT PV cost ($) BAT cost ($) TLCC ($) LPSP (%) LCOE ($/kWh)

TSLF 157.4 61 6237 5598 12,700 1.59 0.195

HSLF 157.4 71 6237 6516 13,600 1.45 0.210
SALF 159.1 81 6302 7434 14,600 0.61 0.222

Table 11
Effect of variation of LPSP on optimal characteristics of optimized PV/BAT system by TSLF.

LPSPDesired (%) APV (m2) NBAT PV cost ($) BAT cost ($) TLCC ($) LPSP (%) LCOE ($/kWh)

0 160.7 94 6367 8627 15,800 0 0.239

1 159.1 68 6302 6241 13,400 0.79 0.204
2 157.4 61 6237 5598 12,700 1.59 0.195
5 152.5 47 6042 4314 11,200 4.61 0.178
10 144.3 39 5718 3579 10,700 9.78 0.170

Fig. 17. (a) Variation of LPSP and TLCC of optimized system vs. LPSPDesired; and (b) variation of LCOE and TLCC of optimized system vs. LPSPDesired of PV/BAT system.

and tested on real data. Note that, for the hybrid renewable energy optimization problem. The precision of the optimal design results
scheme optimal sizing process, the use of forecast data of the load improves when forecast information are used instead of data for
demand affects the optimal design results. The optimization the earlier year. It is also confirmed that tabu search algorithm-
method is tested and verified by considering three hybrid systems based load forecasting (TSLF) can find the optimal design and it
using real data from Iran. The tabu search algorithm based on load exhibits better performance than the harmony search algorithm-
forecast data is compared with the harmony search algorithm based load forecasting (HSLF) and simulated annealing algorithm-
based on load forecast data and the simulated annealing algorithm based load forecasting (SALF). The results show the relative error
based on load forecast data. The results of the simulation demon- to be 15% between the best performance of TSLF and HSLF and the
strate that load demand forecasts can be reliable data for relative error among the best indexes for the TSLF and SALF
 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920 17

autonomous hybrid power schemes based on load forecast data

and can obtained the system objectives.

References

Abido, M., 2002. Optimal power flow using tabu search algorithm. Electr. Power
Compon. Syst. 30 (5), 469e483.
Acun ~ a, L.G., Lake, M., Padilla, R.V., Lim, Y.Y., Ponzo

n, E.G., Too, Y.C.S., 2018. Modelling
autonomous hybrid photovoltaic-wind energy systems under a new reliability
approach. Energy Convers. Manag. 172, 357e369.
Ahmadi, S., Abdi, S., 2016. Application of the Hybrid Big BangeBig Crunch algorithm
for optimal sizing of a stand-alone hybrid PV/wind/battery system. Sol. Energy
134, 366e374.
Al-Ghussain, L., Ahmed, H., Haneef, F., 2018. Optimization of hybrid PV-wind sys-
tem: case study Al-Tafilah cement factory, Jordan. Sustain. Energy Technol.
Assess. 30, 24e36.
Ali, I.B., Turki, M., Belhadj, J., Roboam, X., 2018. Optimized fuzzy rule-based energy
Fig. 18. Variation of number of components and their costs for optimized system vs. management for a battery-less PV/wind-BWRO desalination system. Energy
LPSPDesired of PV/BAT system. 159, 216e228.
Amjady, N., 2002. Introduction to Intelligent Systems. Semnan University Press,
Semnan, Iran.
Anoune, K., Bouya, M., Astito, A., Abdellah, A.B., 2018. Sizing methods and optimi-
zation techniques for PV-wind based hybrid renewable energy system: a re-
view. Renew. Sustain. Energy Rev. 93, 652e673.
Ashkezari, A.D., Ma, H., Saha, T.K., Cui, Y., 2014. Investigation of feature selection
techniques for improving efficiency of power transformer condition assess-
ment. IEEE Trans. Dielectr. Electr. Insul. 21 (2), 836e844.
Bakelli, Y., Hadj Arab, A., Azoui, B., 2011. Optimal sizing of photovoltaic pumping
system with water tank storage using LPSP concept. Sol. Energy 85 (2),
288e294.
Belouda, M., Hajjaji, M., Sliti, H., Mami, A., 2018. Bi-objective optimization of a
standalone hybrid PVeWindebattery system generation in a remote area in
Tunisia. Sustain. Energy Grids Netw. 16, 315e326.
Brandoni, C., Bosnjakovi
c, B., 2017. HOMER analysis of the water and renewable
energy nexus for water-stressed urban areas in Sub-Saharan Africa. J. Clean.
Prod. 155, 105e118.
Caballero, F., Sauma, E., Yanine, F., 2013. Business optimal design of a grid-
connected hybrid PV (photovoltaic)-wind energy system without energy stor-
age for an Easter Island's block. Energy 61, 248e261.
Chauhan, A., Saini, R., 2016. Discrete harmony search based size optimization of
integrated renewable energy system for remote rural areas of Uttarakhand state
in India. Renew. Energy 94, 587e604.
Fig. 19. Sample of hourly operation of the PV/BAT scheme.
Chaurasia, G.S., Singh, A.K., Agrawal, S., Sharma, N., 2017. A meta-heuristic firefly
algorithm based smart control strategy and analysis of a grid connected hybrid
photovoltaic/wind distributed generation system. Sol. Energy 150, 265e274.
Deveau, J., White, C., Swan, L.G., 2015. Lead-acid battery response to various for-
mation levelsepart A: recommended formation levels for off-grid solar and
conventional applications. Sustain. Energy Technol. Assess. 11, 1e10.
Dufo-Lo
pez, R., Cristo
bal-Monreal, I.R., Yusta, J.M., 2016. Stochastic-heuristic
methodology for the optimisation of components and control variables of PV-
wind-diesel-battery stand-alone systems. Renew. Energy 99, 919e935.
Edalati, S., Ameri, M., Iranmanesh, M., Tarmahi, H., Gholampour, M., 2016. Technical
and economic assessments of grid-connected photovoltaic power plants: Iran
case study. Energy 114, 923e934.
Ekren, O., Ekren, B.Y., 2010. Size optimization of a PV/wind hybrid energy conver-
sion system with battery storage using simulated annealing. Appl. Energy 87
(2), 592e598.
El-Naggar, K., AlRashidi, M., AlHajri, M., Al-Othman, A., 2012. Simulated Annealing
algorithm for photovoltaic parameters identification. Sol. Energy 86 (1),
266e274.
Eteiba, M., Barakat, S., Samy, M., Wahba, W.I., 2018. Optimization of an off-grid PV/
Biomass hybrid system with different battery technologies. Sustain. Cities Soc.
40, 713e727.
Fathabadi, H., 2017. Novel standalone hybrid solar/wind/fuel cell power generation
system for remote areas. Sol. Energy 146, 30e43.
Fathy, A., 2016. A reliable methodology based on mine blast optimization algorithm
for optimal sizing of hybrid PV-wind-FC system for remote area in Egypt.
Fig. 20. Total life cycle cost of optimized hybrid system. Renew. Energy 95, 367e380.
Fazelpour, F., Soltani, N., Rosen, M.A., 2014. Feasibility of satisfying electrical energy
needs with hybrid systems for a medium-size hotel on Kish Island, Iran. Energy
algorithm is 35%, and among HSLF and SALF is 16%. Finally, TSLF is 73, 856e865.
seen to be an efficient method for optimizing small autonomous Garlík, B., Krivan, M., 2013. Renewable energy unit commitment, with different
acceptance of balanced power, solved by simulated annealing. Energy Build. 67,
hybrid power systems while after TSLF, the HSLF exhibits better
392e402.
performance than SALF. It is also demonstrated that, of the small Gholami, K., Dehnavi, E., 2019. A modified particle swarm optimization algorithm
independent hybrid power schemes considered, the hybrid PV/BAT for scheduling renewable generation in a micro-grid under load uncertainty.
scheme is the most cost effective and reliable for meeting loads for Appl. Soft Comput. 78, 496e514.
Giallanza, A., Porretto, M., Puma, G.L., Marannano, G., 2018. A sizing approach for
the case study investigated. Also, with increased reliability (a stand-alone hybrid photovoltaic-wind-battery systems: a Sicilian case study.
decrease in LPSP), the TLCC, the LCOE, and the hybrid system J. Clean. Prod. 199, 817e830.
component sizes increase. The proposed optimization method Giaouris, D., Papadopoulos, A.I., Patsios, C., Walker, S., Ziogou, C., Taylor, P.,
Voutetakis, S., Papadopoulou, S., Seferlis, P., 2018. A systems approach for
provides a useful tool and effective for the design of small management of microgrids considering multiple energy carriers, stochastic
18 W. Zhang et al. / Journal of Cleaner Production 241 (2019) 117920

loads, forecasting and demand side response. Appl. Energy 226, 546e559. Nafeh, A.E.-S.A., 2011. Optimal economical sizing of a PV-wind hybrid energy sys-
Glover, F., 1986. Future paths for integer programming and links to artificial intel- tem using genetic algorithm. Int. J. Green Energy 8 (1), 25e43.
ligence. Comput. Oper. Res. 13 (5), 533e549. Niu, M., Sun, S., Wu, J., Yu, L., Wang, J., 2016. An innovative integrated model using
Guangqian, D., Bekhrad, K., Azarikhah, P., Maleki, A., 2018. A hybrid algorithm based the singular spectrum analysis and nonlinear multi-layer perceptron network
optimization on modeling of grid independent biodiesel-based hybrid solar/ optimized by hybrid intelligent algorithm for short-term load forecasting. Appl.
wind systems. Renew. Energy 122, 551e560. Math. Model. 40 (5e6), 4079e4093.
Hatata, A., Osman, G., Aladl, M., 2018. An optimization method for sizing a solar/ Ozoegwu, C.G., 2019. Artificial neural network forecast of monthly mean daily
wind/battery hybrid power system based on the artificial immune system. global solar radiation of selected locations based on time series and month
Sustain. Energy Technol. Assess. 27, 83e93. number. J. Clean. Prod. 216, 1e13.
Hazir, E., Erdinler, E.S., Koc, K.H., 2018. Optimization of CNC cutting parameters Palma-Behnke, R., Benavides, C., Aranda, E., Llanos, J., S
aez, D., 2011. Energy man-
using design of experiment (DOE) and desirability function. J. For. Res. 29 (5), agement system for a renewable based microgrid with a demand side man-
1423e1434. agement mechanism. In: Computational Intelligence Applications in Smart Grid
Hemmati-Sarapardeh, A., Varamesh, A., Husein, M.M., Karan, K., 2018. On the (CIASG), 2011 IEEE Symposium on. IEEE.
evaluation of the viscosity of nanofluid systems: modeling and data assess- Pao, H.-T., 2007. Forecasting electricity market pricing using artificial neural net-
ment. Renew. Sustain. Energy Rev. 81, 313e329. works. Energy Convers. Manag. 48 (3), 907e912.
IREO, 2018. http://www.suna.org.ir/fa/home. Iran Renewable Energy Organization Peng, H., Long, F., Ding, C., 2005. Feature selection based on mutual information
(SUNA). criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans.
Ismail, M., Moghavvemi, M., Mahlia, T., 2013. Techno-economic analysis of an Pattern Anal. Mach. Intell. 27 (8), 1226e1238.
optimized photovoltaic and diesel generator hybrid power system for remote Peng, W., Maleki, A., Rosen, M.A., Azarikhah, P., 2018. Optimization of a hybrid
houses in a tropical climate. Energy Convers. Manag. 69, 163e173. system for solar-wind-based water desalination by reverse osmosis: compari-
Kaabeche, A., Bakelli, Y., 2019. Renewable hybrid system size optimization consid- son of approaches. Desalination 442, 16e31.
ering various electrochemical energy storage technologies. Energy Convers. Qazi, A., Fayaz, H., Wadi, A., Raj, R.G., Rahim, N., Khan, W.A., 2015. The artificial
Manag. 193, 162e175. neural network for solar radiation prediction and designing solar systems: a
Kashefi Kaviani, A., Riahy, G., Kouhsari, S., 2009. Optimal design of a reliable systematic literature review. J. Clean. Prod. 104, 1e12.
hydrogen-based stand-alone wind/PV generating system, considering compo- Reis, A.J.R., Da Silva, A.P.A., 2005. Feature extraction via multiresolution analysis for
nent outages. Renew. Energy 34 (11), 2380e2390. short-term load forecasting. IEEE Trans. Power Syst. 20 (1), 189e198.
Khatib, T., Mohamed, A., Sopian, K., 2012. Optimization of a PV/wind micro-grid for Renno, C., Petito, F., Gatto, A., 2016. ANN model for predicting the direct normal
rural housing electrification using a hybrid iterative/genetic algorithm: case irradiance and the global radiation for a solar application to a residential
study of Kuala Terengganu, Malaysia. Energy Build. 47, 321e331. building. J. Clean. Prod. 135, 1298e1316.
Khiareddine, A., Salah, C.B., Rekioua, D., Mimouni, M.F., 2018. Sizing methodology Samy, M., Barakat, S., Ramadan, H., 2019. Techno-economic analysis for rustic
for hybrid photovoltaic/wind/hydrogen/battery integrated to energy manage- electrification in Egypt using multi-source renewable energy based on PV/
ment strategy for pumping system. Energy 153, 743e762. wind/FC. Int. J. Hydrogen Energy. Available online 3 May 2019, (in press).
Kolhe, M.L., Ranaweera, K.I.U., Gunawardana, A.S., 2015. Techno-economic sizing of Sarhan, A., Hizam, H., Mariun, N., Ya'acob, M., 2018. An improved numerical opti-
off-grid hybrid renewable energy system for rural electrification in Sri Lanka. mization algorithm for sizing and configuration of standalone photo-voltaic
Sustain. Energy Technol. Assess. 11, 53e64. system components in Yemen. Renew. Energy 134, 1434e1446.
Krishan, O., Suhag, S., 2019. Techno-economic analysis of a hybrid renewable energy Sen, R., Bhattacharyya, S.C., 2014. Off-grid electricity generation with renewable
system for an energy poor rural community. J. Energy Storage 23, 305e319. energy technologies in India: an application of HOMER. Renew. Energy 62,
Kwak, N., Choi, C.-H., 2002. Input feature selection for classification problems. IEEE 388e398.
Trans. Neural Netw. 13 (1), 143e159. Shezan, S.A., Julai, S., Kibria, M., Ullah, K., Saidur, R., Chong, W., Akikur, R., 2016.
Li, C., Ge, X., Zheng, Y., Xu, C., Ren, Y., Song, C., Yang, C., 2013. Techno-economic Performance analysis of an off-grid wind-PV (photovoltaic)-diesel-battery
feasibility study of autonomous hybrid wind/PV/battery power system for a hybrid energy system feasible for remote areas. J. Clean. Prod. 125, 121e132.
household in Urumqi, China. Energy 55, 263e272. Shi, B., Wu, W., Yan, L., 2017. Size optimization of stand-alone PV/wind/diesel hybrid
LIU, H.-m., WANG, H.-q., LI, X., 2009. A study on data normalization for target power generation systems. J. Taiwan Inst. Chem. Engi. 73, 93e101.
recognition based on RPROP algorithm. Mod. Radar 5, 017. Shu, W., Qian, W., 2014. Mutual information-based feature selection from set-
Lorestani, A., Ardehali, M., 2018. Optimization of autonomous combined heat and valued data. In: Tools with Artificial Intelligence (ICTAI), 2014 IEEE 26th Inter-
power system including PVT, WT, storages, and electric heat utilizing novel national Conference on. IEEE.
evolutionary particle swarm optimization algorithm. Renew. Energy 119, Singh, S.S., Fernandez, E., 2018. Modeling, size optimization and sensitivity analysis
490e503. of a remote hybrid renewable energy system. Energy 143, 719e731.
Maleki, A., 2018. Design and optimization of autonomous solar-wind-reverse Sinha, S., Chandel, S., 2015. Review of recent trends in optimization techniques for
osmosis desalination systems coupling battery and hydrogen energy storage solar photovoltaicewind based hybrid energy systems. Renew. Sustain. Energy
by an improved bee algorithm. Desalination 435, 221e234. Rev. 50, 755e769.
Maleki, A., Askarzadeh, A., 2014. Optimal sizing of a PV/wind/diesel system with Tahani, M., Babayan, N., Pouyaei, A., 2015. Optimization of PV/Wind/Battery stand-
battery storage for electrification to an off-grid remote region: a case study of alone system, using hybrid FPA/SA algorithm and CFD simulation, case study:
Rafsanjan, Iran. Sustain. Energy Technol. Assess. 7, 147e153. Tehran. Energy Convers. Manag. 106, 644e659.
Maleki, A., Pourfayaz, F., 2015. Sizing of stand-alone photovoltaic/wind/diesel sys- Toghyani, S., Ahmadi, M.H., Kasaeian, A., Mohammadi, A.H., 2016. Artificial neural
tem with battery and fuel cell storage devices by harmony search algorithm. network, ANN-PSO and ANN-ICA for modelling the Stirling engine. Int. J.
J. Energy Storage 2, 30e42. Ambient Energy 37 (5), 456e468.
Maleki, Akbar, Pourfayaz, F., Ahmadi, M.H., 2016a. Design of a cost-effective wind/ Tsai, Y.-S., Yang, U.-C., Chung, I.-F., Huang, C.-D., 2013. A comparison of mutual and
photovoltaic/hydrogen energy system for supplying a desalination unit by a fuzzy-mutual information-based feature selection strategies. In: Fuzzy Systems
heuristic approach. Sol. Energy 139, 666e675. (FUZZ), 2013 IEEE International Conference on. IEEE.
Maleki, A., Khajeh, M.G., Rosen, M.A., 2016b. Weather forecasting for optimization Velik, R., Nicolay, P., 2016. Energy management in storage-augmented, grid-con-
of a hybrid solar-windepowered reverse osmosis water desalination system nected prosumer buildings and neighborhoods using a modified simulated
using a novel optimizer approach. Energy 114, 1120e1134. annealing optimization. Comput. Oper. Res. 66, 248e257.
Mandal, S., Das, B.K., Hoque, N., 2018. Optimum sizing of a stand-alone hybrid Vishnupriyan, J., Manoharan, P., 2018. Prospects of hybrid photovoltaicediesel
energy system for rural electrification in Bangladesh. J. Clean. Prod. 200, 12e27. standalone system for six different climate locations in Indian state of Tamil
Mellit, A., Pavan, A.M., 2010. A 24-h forecast of solar irradiance using artificial Nadu. J. Clean. Prod. 185, 309e321.
neural network: application for performance prediction of a grid-connected PV Wide
n, J., Carpman, N., Castellucci, V., Lingfors, D., Olauson, J., Remouit, F.,
plant at Trieste, Italy. Sol. Energy 84 (5), 807e821. Bergkvist, M., Grabbe, M., Waters, R., 2015. Variability assessment and fore-
Mellit, A., Kalogirou, S.A., Drif, M., 2010. Application of neural networks and genetic casting of renewables: a review for solar, wind, wave and tidal resources.
algorithms for sizing of photovoltaic systems. Renew. Energy 35 (12), Renew. Sustain. Energy Rev. 44, 356e375.
2881e2893. Wu, B., Maleki, A., Pourfayaz, F., Rosen, M.A., 2018. Optimal design of stand-alone
Merei, G., Berger, C., Sauer, D.U., 2013. Optimization of an off-grid hybrid reverse osmosis desalination driven by a photovoltaic and diesel generator
PVeWindeDiesel system with different battery technologies using genetic al- hybrid system. Sol. Energy 163, 91e103.
gorithm. Sol. Energy 97, 460e473. Zhang, W., Maleki, A., Rosen, M.A., Liu, J., 2019. Sizing a stand-alone solar-wind-
Moghaddam, M.J.H., Kalam, A., Nowdeh, S.A., Ahmadi, A., Babanezhad, M., Saha, S., hydrogen energy system using weather forecasting and a hybrid search opti-
2018. Optimal sizing and energy management of stand-alone hybrid photo- mization algorithm. Energy Convers. Manag. 180, 609e621.
voltaic/wind system based on hydrogen storage considering LOEE and LOLE Zhu, Q., Lin, L., Shyu, M.-L., Chen, S.-C., 2010. Feature selection using correlation and
reliability indices using flower pollination algorithm. Renew. Energy 135, reliability based scoring metric for video semantic detection. In: Semantic
1412e1434. Computing (ICSC), 2010 IEEE Fourth International Conference on. IEEE.