Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

Contents lists available at SciVerse ScienceDirect

Renewable and Sustainable Energy Reviews

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

A comprehensive method for optimal power management and design of hybrid

RES-based autonomous energy systems
S. Abedi a , A. Alimardani b,∗ , G.B. Gharehpetian b , G.H. Riahy a , S.H. Hosseinian b
a
Renewable Energy Lab, National Center of Excellence in Power Engineering, Department of Electrical Engineering, Amirkabir University of Technology (Tehran Polytechnic), P.O.
Box: 15875-4413, 424 Hafez Ave., Tehran, Iran
b
National Center of Excellence in Power Engineering, Department of Electrical Engineering, Amirkabir University of Technology (Tehran Polytechnic), P.O. Box: 15875-4413, 424
Hafez Ave., Tehran, Iran

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

Article history: The power management strategy (PMS) plays an important role in the optimum design and efficient
Received 22 May 2011 utilization of hybrid energy systems. The power available from hybrid systems and the overall lifetime of
Accepted 25 November 2011 system components are highly affected by PMS. This paper presents a novel method for the determina-
Available online 18 January 2012
tion of the optimum PMS of hybrid energy systems including various generators and storage units. The
PMS optimization is integrated with the sizing procedure of the hybrid system. The method is tested on a
Keywords:
system with several widely used generators in off-grid systems, including wind turbines, PV panels, fuel
Hybrid energy systems
cells, electrolyzers, hydrogen tanks, batteries, and diesel generators. The aim of the optimization problem
Optimum power management strategy
Differential evolution algorithm
is to simultaneously minimize the overall cost of the system, unmet load, and fuel emission considering
Fuzzy multi-objective optimization the uncertainties associated with renewable energy sources (RES). These uncertainties are modeled by
Resource uncertainty using various possible scenarios for wind speed and solar irradiation based on Weibull and Beta proba-
bility distribution functions (PDF), respectively. The differential evolution algorithm (DEA) accompanied
with fuzzy technique is used to handle the mixed-integer nonlinear multi-objective optimization prob-
lem. The optimum solution, including design parameters of system components and the monthly PMS
parameters adapting climatic changes during a year, are obtained. Considering operating limitations of
system devices, the parameters characterize the priority and share of each storage component for serving
the deficit energy or storing surplus energy both resulted from the mismatch of power between load and
generation. In order to have efficient power exploitation from RES, the optimum monthly tilt angles of
PV panels and the optimum tower height for wind turbines are calculated. Numerical results are com-
pared with the results of optimal sizing assuming pre-defined PMS without using the proposed power
management optimization method. The comparative results present the efficacy and capability of the
proposed method for hybrid energy systems.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction due to their direct dependence on climatic conditions. In addition,

the variations of the wind and solar energy may not match with load
Distributed generation (DG) has been recently nominated as changes [2]. The reliability of the system is important to both plan-
one of the solutions to reliable, less costly, and more efficient ning and utilization stages. Designing energy systems including
energy supply systems. Specifically, small DG systems with power solar and wind energies together, to some extent, reduces the depth
level ranging from 1 kW to 10 MW [1], located near the loads, of the problem. Since, the advantage of one source can overcome
are extensively utilized both in grid-connected and stand-alone the disadvantage of the other one and vice versa [3]. In addition,
modes. Among available DGs, renewable-based systems (RES) taking into account the intermittency and uncertainty associated
such as photovoltaics and wind turbines have attained the most with solar and wind energy sources, improves the adaptation of
popularity due to ever increasing concerns about depletion of design results with practical and realistic conditions. Another con-
fossil fuels and global warming. They have also been getting more ventional solution is appropriate incorporation of energy storage
cost-effective during recent years. devices such as batteries and hydrogen storage systems compris-
However, a significant drawback associated with solar and wind ing electrolyzers, hydrogen tanks and fuel cells into the system
energy systems is their intermittent and unpredictable behavior [3]. Hence, hybrid energy systems are becoming more attractive
to power engineers.
One of the most prominent issues regarding hybrid energy
∗ Corresponding author. systems is to determine their optimum design and operation
E-mail address: arash.alimardani@ieee.org (A. Alimardani). modes taking account of regional conditions and load demand

1364-0321/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.rser.2011.11.030
1578 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

characteristics. Therefore, many efforts have been made for design- into three main gas components and values for each gas has been
ing and planning hybrid DG systems [1–18]. These works can be separately presented.
reviewed from various points of view such as system modeling, the The cost minimization of a hybrid energy system including
applied data in the solution algorithm, objective functions to be hydrogen storage has been presented in [10]. Using the classic
optimized and the mathematical tool used to handle the optimiza- linear programming optimization tool, a comparison between an
tion problem. In [2], the overall cost of the introduced hybrid system optimized dispatch strategy and a fixed one has been performed so
throughout the estimated lifespan of the installation has been that the latter strategy can be reformed using the values obtained
minimized. The reliability of serving energy and proper functioning for the dispatch variables of the former one. Ref. [11] has incorpo-
of components has been involved in the applied algorithm as a con- rated the effect of the number of battery charge/discharge cycles
stant constraint not to be violated. Solar radiation and wind speed relative to their depth of discharge (DOD) on their lifespan, oper-
are also assumed deterministic. Particle Swarm Optimization (PSO) ation strategy and consequently the system cost. A similar work
algorithm has been applied to handle the single-objective problem. has been addressed in [12], which has modeled the uncertainty
In [3], the cost analysis of a hybrid wind/PV/fuel cell system has involved in operating and design characteristics of the system tak-
been focused on through a residential consumer case study. ing the advantage of Stochastic Annealing optimization algorithm.
In [4–10], the cost has been minimized by selecting the corre- This work has referenced the classification of uncertainties pre-
sponding system configuration and/or operation strategy. The load sented in [13] as model-inherent, system inherent and external
demand has been supposed to be completely or partially fulfilled uncertainties existing in the system.
with a fixed user-defined value for unmet load. In this case, the In [14], a method has been introduced, in which the optimized
solution method has no choice to reach a compromise between configuration and operation of the system has been achieved
cost and unmet load. In [9,10], pollutant emissions have been rep- based on twelve parameters defined relative to operational limits
resented in the form of cost and limited by being economically of different storage devices. This work has been rearranged in
involved in the cost function, and thus the results depend on the [15] considering a more complex hybrid system and a couple of
cost coefficient assigned to the emissions. More realistic optimal objective functions including cost, emission and unmet load.
configuration may be achieved by involving pollutant emissions As seen in [14], the power available from such systems and
measurement in their relative unit (kg per liters of fuel) instead of the overall lifetime of system components are highly affected by
converting it to the cost. From the aspect of the system modeling the applied power management strategy; hence, various strate-
accuracy, Ref. [5], has taken a fixed PV panels tilt angle into account gies result in various designs for the system aiming to meet load
using Hay and Devis (1980) and Orgin and Holland (1977) meth- requirements within the estimated lifespan of the system.
ods, whereas [6] has imported the angle value as an optimization In this work, a general method for optimal PMS of autonomous
variable in the solution algorithm. In the latter work, commercially systems including various generators and storage media, com-
available types for each device and their costs have been taken bined with optimal design of system components is proposed.
into account, and optimization results representing size and type To do so, appropriate model of system components for the plan-
of each device have been demonstrated. The load demand has been ning problem is reviewed. To solve the introduced problem, the
assumed to be entirely fulfilled. The maximum power point tracker differential evolution algorithm (DEA), which is nominated to be
(MPPT) has been included in both PV and wind turbines systems. capable of solving various nonlinear optimization problems, is uti-
The presented approach in [7] has considered the influence of the lized. The algorithm is accompanied with the fuzzy technique to
uncertainty in solar irradiation on the sizing process of an off-shore handle the multi-objective optimization problem in a time effi-
PV power system using three probabilistic models such as Markov cient manner. In this case, there is no need to find appropriate
Chain. A size optimization procedure has been demonstrated in coefficients as penalty factors of constraints violations and as
[8], where the results have been compared by using the optimiza- objective function weights. A novel method, involved in the size
tion software called HOMER [19]. It has been stated that because of optimization problem, is proposed to obtain values for the param-
the nonlinearity and complexity of hybrid systems, the application eters of PMS by which the optimum performance and minimum
of evolutionary algorithms like genetic algorithm generates better cost, emission and unmet load are achieved. Considering operat-
results than the application of classical optimization methods [8]. ing limitations of system devices, these parameters characterize
On the case of PMS [5], as one of the earliest works, has the priority and share of each storage component for serving
introduced two parameters named SDM and SAR, representing the deficit energy or storing surplus energy both resulted from
the minimum and maximum State of Charge (SOC) of batter- the mismatch of power between load and RES. Optimal values
ies, respectively. The optimum values of the SOCs have been for design parameters and PMS parameters are simultaneously
calculated within the context of the size optimization problem. attained.
Although [6] has applied an accurate model for PV and wind The RES uncertainties are applied to the optimization procedure
turbines systems representing installation parameters, the oper- by scenario generation based on Weibull [20] and Beta distribu-
ation control has been fixedly defined by the user and did not tion functions for wind speed and solar irradiation, respectively
conform with the system conditions. In [8], the effect of the min- [21]. Numerical results, including type and number of each com-
imum and maximum limit of SOC on the system operation and ponent, monthly values for PV panels tilt angle, the height of wind
cost has been evaluated. The algorithm suffers a computation bur- turbine towers along with the PMS parameters ensure the capabil-
den since a sub-algorithm for finding optimum values of control ity of the proposed method to achieve the aim of optimization. In
parameters has been executed for any single control vector gener- order to focus on the significance of PMS and show the influence of
ated in the main optimization algorithm as a nested optimization strategy variations on the performance and objectives of the sys-
loop. tem, a comparison is presented, too. In addition, determining the
In [9], a grid-connected hybrid energy system, capable of PMS parameters in the optimization procedure is confirmed as an
mutual exchange of power with the grid, comprising wind turbine, efficacious concept.
micro-turbine, diesel generator, photovoltaic array, fuel cell and In the next section, the description of the system under study
battery storage has been analyzed and on-line optimal power and the model of components are presented. Section 3 illustrates
management has been attained. The problem has been treated as the problem statement including objective functions, constraints
a single objective problem considering all objectives such as fuel and the proposed method for the power management. In Section
emissions in terms of cost. The fuel emission has been factorized 4, DEA and fuzzy multi-objective technique are explained. Finally,
 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587 1579

Fig. 1. Configuration of stand-alone hybrid energy system under study.

The solar elevation angle (), which is the angle between the direc-
tion of the sun and the horizon is then estimated, as follows:

 = sin−1 [sin ϕ sin ı + cos ϕ cos ı cos (LMST − )] (2)

where LMST stands for Local Mean Sidereal Time and ϕ is the geog-
raphy of the longitude [17]. The diffuse and reflected radiations are
neglected. Using Fig. 2, the incident radiation on the tilted surface
(Gi ) can be expressed in terms of horizontal component of solar
irradiation (Gh ), as follows:
Fig. 2. Representation of solar elevation angle and PV panel tilt angle.
Gh
Gi = (3)
the results and conclusions are represented in two subsequent sin 
sections. Gp = Gi sin( + ˇ) (4)

where Gp is the effective component of the solar radiation perpen-

2. Model of system components dicular to the panel. Based on the calculated Gp , the power available
from PV panels at time step t considering the temperature effect is
In this work, a stand-alone hybrid energy system comprising determined using the following equations:
wind turbines, PV panels, fuel cells, electrolyzers, hydrogen tanks,
batteries and diesel generators is studied. A simplified diagram of NCOT − 20
TC (t) = TA (t) + Gp (t, ˇ) (5)
the system is depicted in Fig. 1. The available power from renew- 800
able sources is directly delivered to the load in order to provide the Gp (t)
load demand. The excess or deficient power from RES is saved in or ISC (t) = [ISC,STC + KI (TC (t) − 25◦ )] (6)
1000
supplied from other equipments in the system based on the pro-
VOC (t) = VOC,STC − KV · TC (t) (7)
posed dispatch strategy explained in Section 3.2. The model of each
system component is briefly described in the following subsections. PM (t) = Ns · Np · VOC · ISC (t) · FF(t) (8)

where NCOT is the nominal cell operating temperature (◦ C),

TA (t)
2.1. Wind turbine
is the ambient temperature at t, TC (t) is the cell temperature at t,
VOC,STC and ISC,STC are the module open-circuit voltage and short-
The output power of the wind turbine, Pw , is calculated on the
circuit current under STC, with KI and KV as their corresponding
basis of its power curve given by the manufacturer. The effect of
temperature coefficients; PM (t) is the power of array with NP mod-
the wind turbine installation height and the roughness of the land
ules in parallel and Ns modules in series and FF(t) is the fill factor
surface of the towers on the output power is also considered [2,6].
[6].

2.2. PV panel
2.3. Fuel cell

To achieve the maximum output power from a PV panel, it is

The formulations, which represent the balance between the
essential to consider the influence of the solar position and panel tilt
hydrogen as input and the electric power as output in steady state
angle. In [16], it is shown that such effects can change the total avail-
condition, is the required fuel cell model used in the design pro-
able power up to 9.94% of the total maximum power. The panels
cedure. However, for the sake of ensuring maximal efficiency of
are assumed to have rotation on one axis. The following equations
the system and preventing degradation of fuel cells due to abun-
lead to the incident irradiation on a tilted panel. Eq. (1) is employed
dance of start-up and shut-down cycles, a hysteresis band is also
to calculate the longitude at the equator (ı), with respect to solar
embedded into the model, as in [12].
ecliptic longitude and latitude (, ) and earth’s axis inclination to
The fuel cell hydrogen consumption, ConsFC (kg/h), relative to
the plane of its orbit (I = 23.439 ◦ ).
its output electric power is computed by the following equation
[15]:

BFC · PN + AFC · PFC ≤ Pmax
FC   P  for P/PN FC ef
ConsFC = (9)
BFC · PN FC + AFC · PFC 1 + Fef − Pmax ef otherwise
PN FC

where PFC is the output power of fuel cell in kW, PN FC is the nomi-
ı = sin−1 (sin  cos I + cos  sin  sin I) (1) nal output power, AFC and BFC (kg/kWh) are the coefficients of the
1580 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

consumption curve. Pmax ef , in terms of percentage of PN FC , is the where AG and BG (l/kW) are fuel consumption curve coefficients
output power, at which the efficiency of the fuel cell is maxi- provided by the manufacturer, PG (kW) is the output power and
mum and Fef is a factor to consider the high consumption above PN G (kW) is the nominal output power of the diesel generator.
Pmax ef . The hydrogen consumption parameters for all fuel cells are The values assigned to AG and BG are 0.246 and 0.08145 l/kWh,
assumed to be AFC = 0.05 and BFC = 0.004 kg/kWh, Pmax ef = 0.2 and respectively, for all diesel generators, which have been used in [15],
Fef = 1 [15]. By using these values, the fuel cell efficiency is 46% at too.
the maximum and 31% at rated powers.
3. Problem statement
2.4. Electrolyzer and hydrogen tank
3.1. Objective functions and constraints
The input electrical energy dependence on the hydrogen mass
flow is modeled, as follows [15]: The objective functions, considered in the optimization prob-
ConsE = BE · QN + AE · Q (10) lem, are as follows:
E

where QN E is the nominal hydrogen mass flow (kg/h), Q is the • The overall cost of the system discounted to the year of
hydrogen mass flow (kg/h), AE and BE are the coefficients of the installation (NPC) including the investment cost, operation and
consumption curve in kW/kg/h. The parameters of the electrical maintenance cost during the lifespan of the system, replacement
energy consumption for all electrolyzers are assumed to be the cost, and fuel cost of diesel generators. All of the cost terms asso-
same as parameters given in [15], i.e., AE = 40 and BE = 20 both in ciated with the system components are based on the data in [15].
kW/kg/h. • The total fuel emissions produced by diesel generators during the
The maximum hydrogen tank capacity is assumed to be equal to total lifespan of the system. Thanks to the employed fuzzy multi-
10 kg and the total capacity of the hydrogen storage is determined objective optimization technique, the model implemented for the
by the design program as the number of hydrogen tanks (Ntank ). fuel emission is adequately accurate and there is no requirement
The minimum allowed level of the hydrogen (in tanks) is also a for taking it into account in terms of cost. As the most significant
parameter determined by the proposed PMS. gas existing in the diesel generator exhaust is CO2 , which also
causes green house effects, the produced CO2 is considered to
2.5. Battery represent the fuel emission.
• The total Loss of Load Probability (LPSP), which represents the
The battery bank with the total nominal capacity of Cn (Ah), can total unmet energy. This index is the most commonly used index
serve energy to the load until the maximum depth of discharge in related works [2]. LPSP is proportional to the unmet load and
(DOD) or SOCmin is reached. SOCmin of battery bank along with is written, as follows:
SOCmax are taken into account as two of the parameters of PMS,
which should be optimized by the developed program and should LOEE
not exceed the values introduced by the manufacturer. LPSP = H (14)
h−1
D(h)
The SOC of battery bank, in each simulation time step, is calcu-
lated by applying the following equation: where D(h) is the load demand in kWh at time step h, and LOEE,
PB (t) standing for Loss of Energy Expected, is defined by the following
SOC(t) = SOC(t − 1) + nB · 100 (11) equation:
Cn
where SOC(t) is the batteries’ SOC at time step t, nB is the battery
H
LOEE = E[LOE(h)] (15)
round-trip efficiency and PB (t) is the power charged in or dis- h=1
charged from battery bank at time step t [6]. nB is approximately 
equal to 80% in charging and 100% in discharging modes [6]. PB (t) E[LOE] = Q (s) · f (s) (16)
is determined due to the mismatch of power between load and s∈S
RES and the share associated with batteries to supply the defi-
where Q(s) represents the amount of the unmet energy when the
cient power or to save the excess power. The mentioned share is a
system experiences the state s and f(s) is the probability associated
parameter of PMS and optimized in the proposed algorithm.
with the occurrence of the state s.
The number of series connected batteries depends on the DC bus
The control variables, which should be used in simulations to
voltage (VBUS ) and the nominal voltage of each battery, i.e.:
calculate the objective functions, as well as their boundaries, are
VBUS listed, as follows:
nSB = (12)
VB

The total number of batteries in battery bank is equal to nBS multi- - A k × 2 matrix [[Nk ], [Tk ]] of size parameters of the system com-
plied by nPB (the number of batteries in parallel), which should be ponents shown in Fig. 1, where, Nk is the number and Tk is the
determined in the optimization algorithm. type of component k. Both are positive integers.
In order to model the effect of the number of charge/discharge - A vector of installation parameters including monthly tilt angles
cycles on the lifetime of batteries, the model used in HOMER soft- of PV panels (0◦ < ˇk < 90◦ , k = 1,. . .,12) and wind turbine tower
ware [22] is used. This model is based on the battery “cycles to height (h > 0),
failure” curve, which is more explained in detail in [11]. - A vector of PMS parameters (described in the coming section)
including:
- Monthly charge shares associated with n − 1 types of storage
2.6. Diesel generator
media (0 ≤ CSi ≤ 1, where, n is the total types of the storage
media existing in the system),
The fuel consumption of the diesel generator, ConsG , in terms of
- Monthly discharge shares associated with m − 1 types of backup
its output power is written, as follows:
devices (0 ≤ DSi ≤ 1, where, m is the total number of backup
ConsG = BG · PN G + AG · PG (13) devices capable of feeding power to the load),
 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587 1581

- Upper and lower limits on the stored energy level The value of SOCmax is obtained within the context of the optimiza-
of all the storage media (SOCi,min,rated ≤ SOCi,min ≤ 0.5, tion procedure.
0.5 ≤ SOCi,max ≤ SOCi,max,rated , 0 ≤ SOC ≤ 1). Case (1-3): Pi (t) is beyond the HBR. Hence, the power to be drawn
from the device does not fall within the allowable range and no
In addition to the boundaries that control variables should not power can be delivered, i.e.:
take values beyond them, there are operational constraints to be
mentioned as follows: Pi∗ (t) = 0 (21)

Case (1-4): The operating limitations of the device considering its

- The specified maximum and minimum SOC of batteries (SOCmin , dynamics are not satisfied. For instance, the change in the power
SOCmax ), share of the device from time step t to t + 1 is more than the maxi-
- The rated current or power of all devices, mum tolerable rate of the power change due to the system inertia
- The minimum power injected to the electrolyzers and the min- (i.e., the power ramp rate, represented by r). This inertia implies the
imum supplied power from fuel cells and diesel generators, at impossibility of timely tracking very rapid input power fluctuations
which they can start-up. This constraint is considered to avoid [23].
the multitude of the start-up/shut-down cycles or short periods
of frequent utilization of these devices, and enhance their life- Pi∗ (t) = r · t (22)
time. This operation limit has been introduced as Hysteresis Band
Range (HBR) in [12]. Other imposed constraints are illustrated in In above mentioned cases, the power assigned to other devices,
the next section. Pj (t), should also be modified because the constraint did not allow
the device i to fulfill the storage of its initially assigned Pi (t), calcu-
3.2. Proposed method for power management of system lated in (18). Therefore, a portion of Pi (t) is still remained, namely
Prem,i (t), which other devices may be able to absorb it. The modifi-
In general, the PMS is so important that the operation, relia- cation procedure is named Charge Share Correction (CSC), written
bility, cost and lifetime of the system is affected with even minor as follows:
alterations in the strategy. In the proposed dispatch strategy, to effi-
ciently utilize the RES, the available power from RES, i.e., PRES (t), is Prem,i (t) = Pi (t) − Pi∗ (t) (23)
directly delivered to the load in order to provide the load demand,
CSi∗ (t) = 0 (24)
Pload (t). Other equipments are considered without any predefined
priorities and there is not any preference to some components. The  
CSi
priorities are determined by the optimization process. The excess CSj∗ (t) = CSj 1+ n for j =
/ i, j = 1, . . . , n (25)
or deficient power of RES, i.e., Prem (t), is saved in or supplied from k=1,k =
/ i
CSk
other equipments in the system, i.e.:

Prem (t) = PRES (t)– Pload (t) (17)


i−1

Pj∗ (t) = Pj (t) + CSj∗ (t) · Prem,k (t) (26)

where Prem (t) might be a negative or positive value. k=1
For Prem (t) > 0: In this case, PRES (t) is more than the load demand
where CSj is the charge share of all devices except device i, and Pj∗ (t)
at time step t, and the remaining power should be delivered to the
storage devices namely electrolyzers and battery bank. Therefore, is the modified power assumed for the storage device j. Eq. (24)
based on the proposed method, the optimization procedure should means that the device i has absorbed all possible amount of power
assign Charge Shares (CS) for battery bank and electrolyzers so that it could, and there is no more power stored in it during the current
Prem (t) is optimally shared among them considering constraints. time step t. If one of the constraints corresponding to this device is
The power, to be stored in storage device I, i.e., Pi (t), is written, as violated, the same modification procedure is then performed, for
follows: that constraint.
For Prem (t) < 0: In this case, the load demand is partially supplied
Pi (t) = CSi (k) · Prem (t) i = 1, . . . , n (18) by RES. Hence, Discharge Shares (DS) should be determined and
assigned to the back-up devices. Utilizing these back-up devices,
where n is the total number of storage media and CSi (k) is the CS the effort is made to fully compensate the remaining power by the
of storage device i in month k (k = 1,. . .,12). It should be mentioned power from suppliers other than the RES (Pbackup ), and try to achieve
that all CSs fall within the range (0,1). Pbackup = |Prem |.
Now, the status of all storage devices is checked one after The power to be drawn from backup device i, Pi (t), is written as
another (as seen in Fig. 3), considering their constraints and lim- follows:
itations on the storable power. If any constraint is violated, Pi (t)
of the device should be modified so that the violated constraint is Pi (t) = DSi (k) · Prem (t) i = 1, . . . , m (27)
satisfied. These constraints violations or incompatibilities and the
necessary modification to Pi (t) in each case are, as follows: where m is the total number of back-up devices capable of feeding
Case (1-1): The power assigned to device i, Pi (t), exceeds its rated power to the load and DSi (k) is the discharge share of the back-up
power. In this case, it is set to the rated power (Prated,i ) of the device, device i in month k (k = 1,. . .,12). All DSs fall within the range (0,1).
i.e.: As shown in Fig. 3, the status of all back-up devices is checked
one after another considering their constraints and limitations on
Pi∗ (t) = Prated,i (19) the power that they are able to serve in the current time step. If
Case (1-2): Pi (t) exceeds the remaining storage capacity of device i. any constraint is violated, Pi (t) of the device should be modified, as
In this case, the power is injected to the device until the maximum follows, so that the violated constraint is satisfied.
allowable level of the stored energy is reached (SOCmax ): Case (2-1): Pi (t) exceeds the rated power of device i. (Eq. (21)
should be applied.)
Prated,i Case (2-2): The stored energy in the device in the current sim-
Pi∗ (t) = (SOCmax,i − SOCi (t − 1)) (20)
100 ulation time step is not as adequate as to supply Pi (t). In this case,
1582 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

Start

t=1

Calculate Prem(t)

CSi(k), Yes No DSi(k),

Eq. (18) Prem>0 Eq. (27)

t=t+1 i=1 i=1

Case (1-1) Yes Yes Case (2-1)

occurred occurred
Eq. (19), CSC Eq. (19), DSC
No No

i=i+1 Yes Case (1-2) Case (2-2) i=i+1

Yes
occurred occurred
Eq. (20), CSC Eq. (28), DSC
No No

Case (1-3) Yes Yes Case (2-3)

occurred occurred
Eq. (21), CSC Eq. (21), DSC
No No

Yes Case (1-4) Case (2-4) Yes

occurred occurred
Eq. (22), CSC Eq. (22), DSC
No No

Calculate stored and Calculate supplied and

discarded energy unsupplied energy

No No
i=n i=m

Yes Yes

No
t=8760
Yes

End

Fig. 3. Flowchart of proposed PMS.

the power is drawn from the device until the minimum allowable not allow device i to fulfill the supply of its initially assigned Pi (t),
level of the stored energy is reached (SOCmin,i ): calculated in (27), and a portion of Pi (t) is still remained, namely
Prem,i (t), which other devices may be able to supply it. The modifica-
Prated,i
Pi∗ (t) = (SOCi (t − 1) − SOCmin,i ) (28) tion procedure is named Discharge Share Correction (DSC), written
100 the same as CSC (Eqs. (23)–(25)), with only replacing all CS variables
The value of SOCmin is obtained within the context of the opti- by DS, and n by m.
mization procedure. While any of the above situations take place and CSC or DSC is
Cases (2-3) and (2-4): These cases are the same as cases (1-3) performed, it is possible that again, the modified dispatch shares are
and (1-4), respectively. incompatible with system operation constraints. In this case, the
In the above mentioned cases, the power assigned to other CSC or DSC is performed for the second time so that the constraints
devices, Pj (t), should also be modified because the constraint did are satisfied. This is the last effort for the determination of dispatch
 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587 1583

Fig. 4. Membership functions of objective functions and constraints.

shares. Then the remaining excess or deficient power is considered functions are initially combined by “and” operator (minimum). This
as discarded or unmet power in the current simulation time step, procedure can be expressed by the following equation:
respectively (as presented in Fig. 3).
D (x) = min(f 1 (x), f 2 (x), . . . , c1 (x), c2 (x), . . .) (29)

4. Optimization algorithm where D (x) represents the membership function of the optimal
decision function and is treated as the evaluation value in the opti-
The multi-objective optimization algorithm proposed in this mization algorithm.
paper, takes the advantage of the differential evolution algorithm The membership values express the degree of satisfaction for
accompanied with fuzzy-multi objective technique. each objective. Highly satisfied objective is given a low value,
though lowly satisfied one is assigned a high value. Hence, the
multi-objective problem can be transformed into the following
4.1. Multi-objective optimization using fuzzy technique maximization problem subject to a crisp constraint set:

The multi-objective problem is generally solved by three types max D (x, u)s.t. H(x, u) = 0, C(x, u) ≤ 0 (30)
of methods. The first one is pareto-based approach to get a set
where D (x,u) is called the fuzzy index.
of non-dominated solutions in the process of optimization [24].
The second one is the coefficient method and the last method is
4.2. Differential evolution algorithm
transforming the multiple objective function into a single objec-
tive model and optimizing it through single objective strategies
Differential Evolution Algorithm (DEA), introduced by Storn
[24]. In this method, the determination of appropriate values for
and Price [27], is a simple population based stochastic parallel
the coefficients is of very important. In addition, the results are
search evolution algorithm for global optimization and is capable
highly dependent on and sensitive to selected values for coeffi-
of handling non-differentiable, nonlinear and multi-modal objec-
cients. In the method developed by Bellman and Zadeh [25], the
tive functions [28]. In DEA, the population consists of real-valued
single objective problem is achieved by maximizing the minimum
vectors with dimension D, which is equal to the number of control
degree of satisfaction among membership functions.
variables. The initial population with the size Np, is uniformly dis-
The basic idea in fuzzy optimization is to optimize objective
tributed in the search space, falling within variables’ boundaries.
function and constraints simultaneously [26]. The fuzzy decision
The procedure of this algorithm is shown in Fig. 5. The algorithm is
is marked out due to the intersection of fuzzy objectives and fuzzy
described in the following steps:
constraints. The first operation is the fuzzification process of the
Step (1), Initialization
merged objective function and the constraints. In this procedure, k
For each control variable k with lower bound xmin and upper
two types of function (x) are defined for each objective function
k
bound xmax , initial values are randomly and uniformly chosen in
and constraint, as shown in Fig. 4. In this figure, the minimum value
k , xk
the interval [xmin
for each objective is obtained by the single objective optimization max ]:
and the maximum value is specified by the initial set of control k k k k
xi,G = xmin + rand [0, 1] × (xmax − xmin ) i ∈ [1, Np ], k ∈ [1, D](31)
variables in the optimization algorithm. According to this method,
it is possible to change the effectiveness of any objective function Step (2), Evaluation
by reducing or increasing its specified maximum value. In other In DEA, after generation of the initial population XG , all of its
words, when the specified maximum value of an objective function consisting vectors are evaluated by the calculation of the objective
decreases, it is considered to be more important in the optimiza- function in its subprogram, and then the vectors are sorted accord-
tion than the previous state and vice versa. These membership ing to their objective function value. In the present work, DEA is
1584 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

The elements of the mutant vector enter the trial vector with prob-
ability CR as follows:

Vj,i,G+1 if randj,i ≤ CR or j = Irand

Uj,i,G+1 = (33)
Xj,i,G if randj,i > CR and j =/ Irand

where randj,i ∈ [0,1], Irand is an integer randomly selected from

[1,2,. . ., D]. Irand ensures that one of the trial vectors is selected
among the mutant vectors.
Step (3-3), Modification and constraints verification:
This step prepares the trial vectors for the next step by veri-
fying the variables whether they meet their constraints, given in
Sections 3.2 and 4.4. Otherwise, for the vectors containing vari-
ables with unsatisfied constraints, mutation and crossover steps
are repeated for a specified number of times or until the problem
is fixed. Nevertheless, these vectors will be regenerated, similar to
step (1).
Step (3-4), Selection:
The vectors generated in the previous step are evaluated and
sorted as described in the step (2). Then, the trial vector Ui,G+1 is
compared with the corresponding vector Xi,G and the one with the
better fitness value is admitted to the next iteration as Xi,G+1 :

Ui,G+1 if f (Ui,G+1 ) < f (Xi,G )

Xi,G+1 = i ∈ [1, NP ] (34)
Xi,G otherwise

4.3. Representation of uncertainty

Natural characteristics of wind and solar energy impose uncer-

tainty in their design and operation. Hence, considering various
possible scenarios in the model of these resources can lead to
more realistic decisions. The uncertain parameters are expressed
by probability distributions, showing the range of values that a
parameter could take, and also accounting for the probability of the
occurrence of each value in the considered range [12]. In this paper,
Weibull and Beta PDFs are used for the wind speed and the solar
irradiation, respectively [21]. Using this solution, seven scenarios
Fig. 5. Procedure of the used DEA. are generated, to model the uncertainty of these resources.

4.4. Implementation
accompanied with fuzzy technique to handle the multi-objective
problem. Hence, this step has some differences, described in Section A program in MATLAB environment is developed to determine
4.4. the best set of design and PMS parameters. Furthermore, the con-
Step (3), Iteration vergence and computational time for the algorithm is evaluated.
This step is the main procedure of DEA to conduct the final The same system is simulated via two other optimization methods
results, which is performed iteratively, until either an acceptable widely applied to renewable energy [29], namely Genetic Algo-
solution has been reached, or for at least a specified number of iter- rithm (GA) [30] and Linearly Decreasing Inertia Particle Swarm
ations, no further improvement in the solution is obtained, or a Optimization (LDI-PSO) [31,32] and their corresponding results are
predefined number of iterations are completed [28]. compared in the Section 5.2. The performance of the proposed PMS
Step (3-1), Mutation: is also evaluated by comparison with another PMS.
In DEA, the perturbations applied to the current population to There are two main modules in the developed program, the
generate new population are derived from the current population optimization module and the system simulation module. The first
itself, and no predefined probability density function is considered. module generates and modifies a valid vector of control variables
For a given vector of control variables, Xi,G , in the population, two to be evaluated by the system simulation module in each iteration.
vectors (Xr2 ,G and Xrs ,G ) are randomly selected such that the indices When the latter module is invoked, the input vector is processed.
i, r2 and r3 are mutually different. Then, the weighted difference The module simulates the operation of the system for each scenario,
between two vectors is added to the vector with the best fitness calculates the operating point of each system component and deter-
function, Xb,G , written as follows: mines f,i (x) and ci (x), and the resultant D (x), using the resultant
values for objective functions. These values are transformed into a
Vi,G+1 = Xb,G + F · (Xr2 ,G − Xr3 ,G ) r2 , r3 ∈ {1, 2, . . . , NP } (32) single value using the described fuzzy technique and become the
evaluation value for the considered scenario. The final evaluation
where F, a constant from (0, 2), characterizes the amount of the value for an input vector is the average of values obtained by using
movement of vectors within the search space. data of all generated scenarios [12], which is then admitted back to
Step (3-2), Crossover: the optimization module.
In this step, the trial vector Ui,G+1 is developed from the elements Therefore, a valid control vector is obtained by assigning val-
of the target vector Xi,G and the elements of the mutant vector Vi,G+1 . ues to the control variables that lie within allowed bounds of each
 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587 1585

Fig. 6. LOEE for (a) PMS (a) and (b) PMS (b).

individual. However, some modifications to the original optimiza- 5. Case study and results
tion algorithm are made to match the specific characteristics of the
present problem, as follows: 5.1. System data

The hybrid energy system under study has been described in

- Some of the control variables, including number and type of each Section 2. In practical system planning, the designer has to choose
component, must be non-negative integers. Therefore, the opti- the system components from a set of commercially available ones.
mization procedure should handle the discrete variables in all As a case study, various models and types of components with dif-
initialization, crossover and mutation steps. ferent rated powers, costs and other specification [15], are the input
- The upper limit, associated with CSi and DSi (i = 2,. . .,n) float in to the design procedure. The system location is a region in Arde-
accordance with the values of CSk and DSk (k = 1,. . .,i −
1), because bil city (38◦ 14 57 N/48◦ 18 5 ’E), in the north west of Iran. For this
n
CSi and DSi must satisfy the equality constraint CSk = region, the hourly data for solar irradiance, wind speed and ambient
m k=1
temperature all provided within one year period [2]. The consid-
k=1
DSk = 1. This constraint modification is applied to the value
that the optimization procedure assign to CSi and DSi , from device ered load profile is the IEEE reliability test system [33] with 30 kW
i = 1 through n, one by one, i.e.: peak power.

5.2. Results

i−1
CS1max = 1, CSimax = 1 − CSk (i = 2, . . . , n) (35)
The data are analyzed to determine the optimum design and
k=1
PMS parameters for the case study. The results of the used opti-
mization algorithms, i.e., GA, LDI-PSO and DEA, are presented in
A similar equation is used by replacing CS by DS and n by m. Table 1. It can be seen in this table that DEA revealed the best per-
- The integration of the fuzzy model with the optimization pro- formance with respect to optimization of objective functions and
cedure for handling the multi-objective problem consists of the resultant fuzzy index. Hence, DEA is nominated to demonstrate
developing membership functions of objectives as well as con- the results of this problem in more detail. Since, this problem should
straints imposed on the operation of the system. By solving be solved in off-line mode, the computation time and convergence
a single objective optimization for each objective function, its speed are not of great concern.
minimum and maximum values are obtained and the corre- In order to well present the applicability and efficacy of the pro-
sponding membership function is achieved. By the way, using posed PMS, which is the significant idea of this paper, a simulation
maximum and minimum power ramp rates of each compo- is also carried out by using another PMS, referred to as PMS (b).
nent (ramp ratei,min and ramp ratei,max ), rated limitations on SOCi Based on the PMS (b) which is analogous with the PMS (dispatch
(SOCi,min and SOCi,max ) and power boundaries of each component strategy) employed in the HOMER software [8], Prem (t) in time step
(rated poweri,min and rated poweri,max ), all as the operation con- t is dispatched among components by the priorities defined by
straints, their corresponding membership functions are obtained designer, unless the marginal cost of energy storage or production
(as shown in Fig. 4). of the dispatched power by the components is in contrast with
- To treat the uncertain parameters, several scenarios are generated
and simulated. Hence, all the three objective functions introduced
in Section 3.1 are calculated for all the scenarios. The algorithm Table 1
Results of three evolutionary algorithms.
utilizes the concept of the “expected objective function”, which
aggregates all the objective function values corresponding to PMS (a) PMS (b)
the simulated scenarios representing the uncertain parameters,
GA LDI-PSO DEA DEA
in the form of an average value [12]. Then, using the expected
Cost (D) 1,042,242 967,375 912,560 918,337
objective function values, the fuzzy index is calculated, and the
Fuel emissions (kg) 2086.2 2632.8 1827.7 2888.6
algorithm continues with the same algorithmic operations found LPSP (W/h) 0.0067 0.0083 0.0072 0.0075
in DEA. In this way, the results represent the solution with an Number of iterations to converge 29550 196 227 153
overall consideration of all scenarios associated with the uncer- Computation time (s) 173,520 121,311 134,548 72,549
tain parameters in the system. Fuzzy index 0.9980 0.9973 0.9996 0.9915
1586 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587

Table 2
Optimized values for design parameters of the case study.

Design data DEA

Equipments data
Number 9
PV panel
Type 10
Number 6
Wind turbine
Type 6
Number 28
Battery
Type 5
Number 6
Fuel cell
Type 4
Number 8
Electrolyzer
Type 4
Number 7
Diesel generator
Type 3
Hydrogen tank Number 6 Fig. 8. Monthly dispatch shares storage devices for saving the excess power.
Inverter nominal power (kVA) Type 35
Installation data
Table 3
Wind tower height (m) 32
Optimized values of limitations of storage devices.
67
58 Storage device limitations DEA
63
46 Minimum state of charge of batteries (SOCmin %) 21
38 Maximum state of charge of batteries (SOCmax %) 100
PV panels monthly 31 Minimum level of Hydrogen in tanks (%) 33
tilt angle ␤1 − ␤12 (◦ ) 22
30
26 values. In fact, CS and DS parameters are interdependent for compo-
43 nents that utilize the same energy storage device in both charging
59
53
and discharging states. For instance, electrolyzers and fuel cells use
the same energy container namely hydrogen tanks to store energy
and consume it, respectively. Table 3 contains optimized values of
operating limitations of storage devices.
the predefined priorities. In this case, the priority of components
The attained values for parameters of PMS by represent nearly
to store or produce Prem (t) is sorted ascendingly according to
the overall optimum operating condition of the system considering
their corresponding cost of energy storage or production. The
a compromise among all the considered objectives. The parameters
optimization results using PMS (b) are included in Table 1, too. It
are separately calculated for each month, representing the influ-
can be observed that PMS (a) have better results than PMS (b). The
ence of monthly and seasonal changes in load demand and climatic
fuel emission is comparably less than that of PMS (b), showing
patterns on the utilization of the hybrid system. Hence, the results
that the utilization of back-up and storage devices other than the
are adapted with any circumstances that alter the operation of the
diesel generator is more effective when PMS (a) is employed. Fig. 6
system resulting in more accuracy and optimality.
presents the LOEE in both PMS cases, confirming the better LPSP of
Table 1 for the case of PMS (a).
6. Conclusions
Table 2 presents the result of design parameters, including the
type and number of each component. The installation data con-
This study has been dedicated to review optimal design and
sisting of the monthly PV panels tilt angle and wind turbines
operation strategy selection of hybrid RES-based stand-alone
installation height are also listed in the table.
energy systems, including various generators and storage media.
The monthly values of charge and discharge shares are pre-
Considering resource uncertainties, associated with wind speed
sented in Figs. 7 and 8, respectively. Fig. 7 presents that the DS
and solar irradiation, a novel method has been proposed, to deter-
of fuel cells is more significant in winter’s months than summer’s
mine the power management strategy of the system along with
months and the opposite is true for diesel generators. The values
sizing parameters of system components so that the overall cost
obtained for DS parameters also have great influence on the CS
of the system during its useful lifetime, unmet load and pollutant
emissions have been minimized. The PMS parameters are monthly
charge shares (CS) of storage devices and monthly discharge shares
(DS) of generator devices, limitations on their start-up/shut-down
power thresholds (HBR) and on their level of available energy (SOC).
Based on the proposed method, the values of PMS parameters,
the sizing and design parameters have been determined by the
iterative optimization algorithm. The PMS parameters have been
separately determined for each month to adapt monthly variations
in the load and climate patterns. Then in each iteration, these values
have been applied to the hourly simulation of the system opera-
tion and evaluated in each single time step to meet the operational
constraints of the components, consisting of the nominal power,
SOCmin and SOCmax , HBR, and power ramp rate of each device. Oth-
erwise, these values have been modified by either of two operators,
namely CSC or DSC, in that time step. These modifications have
Fig. 7. Monthly dispatch shares of backup devices for providing the deficient
ensured that while the optimum operation strategy and commit-
demand. ment of components has been attained, all constraints associated
 S. Abedi et al. / Renewable and Sustainable Energy Reviews 16 (2012) 1577–1587 1587

with operational characteristics of the components have been sat- [9] Mohamed FA, Koivo HN. System modelling and online optimal management of
isfied. However, if after completion of all steps of simulation in a MicroGrid using mesh adaptive direct search. International Journal of Electrical
Power & Energy Systems 2010;32:398–407.
time step, the constraints were still unsatisfied, some discarded or [10] Garcia RS, Weisser D. A wind–diesel system with hydrogen storage: joint opti-
unsupplied energy remains. The summation of these values during misation of design and dispatch. Renewable Energy 2006;31:2296–320.
all time steps of the simulation has been involved in the objective [11] Bernal-Agustín JL, Dufo-Lopez R. Multi-objective design and control of hybrid
systems minimizing costs and unmet load. Electric Power Systems Research
functions to be minimized. 2009;79:170–80.
The proposed method has tested on a hybrid [12] Giannakoudis G, Papadopoulos AI, Seferlis P, Voutetakis S. Optimum design
PV–wind–diesel–hydrogen–battery system. To efficiently handle and operation under uncertainty of power systems using renewable energy
sources and hydrogen storage. International Journal of Hydrogen Energy
the nonlinear mixed-integer multi-objective optimization prob-
2010;35:872–91.
lem, a modified version of DEA accompanied by fuzzy technique [13] Ierapetritou M, Acevedo J, Pistikopoulos E. An optimization approach for
has implemented. The system has been optimized for summations process engineering problems under uncertainty. Computers & Chemical Engi-
neering 1996;20:703–9.
of all objective values of each function calculated for all scenarios
[14] Dufo-López R, Bernal-Agustín JL, Contreras J. Optimization of control strategies
generated by applied uncertainty models. The presented results of for stand-alone renewable energy systems with hydrogen storage. Renewable
the solution can be categorized into three groups: Energy 2007;32:1102–26.
[15] Dufo-Lopez R, Bernal-Agustin JL. Multi-objective design of PV–wind–diesel–
hydrogen–battery systems. Renewable Energy 2008;33:2559–72.
- Size (design) data: The optimum number and type of each com- [16] Mondol JD, Yohanis YG, Norton B. The impact of array inclination and orienta-
ponent of the system to be installed. tion on the performance of a grid-connected photovoltaic system. Renewable
- Installation data: The optimum values for monthly inclination Energy 2007;32:118–40.
[17] Chang YP. Optimal the tilt angles for photovoltaic modules in Taiwan. Interna-
angles of solar panels and the optimum tower height for wind tional Journal of Electrical Power & Energy Systems 2010;32:956–64.
turbines installation to maximize the exploitable power from RES. [18] Zghal W, Kantchev G, Kchaou H. Optimization and management of the energy
- Operation strategy (PMS): The monthly CS and DS, and boundaries produced by a wind energizing system. Renewable and Sustainable Energy
Reviews 2011;15:1080–8.
associated with SOC. [19] Hybrid optimization model for electric renewables (HOMER).
wwwnrelgov/international/homer.
By comparing the results of design parameters and objective [20] Mostafaeipour A, Sedaghat A, Dehghan-Niri A, Kalantar V. Wind energy feasi-
bility study for city of Shahrbabak in Iran. Renewable and Sustainable Energy
functions values (accompanied by that of the system operation
Reviews 2011;15:2545–56.
and unsupplied energy), with the results of applying a user- [21] Atwa Y, El-Saadany E, Salama M, Seethapathy R. Optimal renewable resources
defined PMS, it has been concluded that the design decisions and mix for distribution system energy loss minimization. IEEE Transactions on
Power Systems 2010;25:360–70.
operation of such systems are highly affected by the employed
[22] Bagul A, Salameh Z, Borowy B. Sizing of a stand-alone hybrid wind–photovoltaic
PMS; and the superiority of the proposed PMS has been con- system using a three-event probability density approximation. Solar Energy
firmed. It should be mentioned that the proposed method can 1996;56:323–35.
be effectively employed for any composition of hybrid energy [23] Auld AE, Samuelsen S, Brouwer J, Smedley K. Load-following strategies for evo-
lution of solid oxide fuel cells into model citizens of the grid. IEEE Transactions
systems. on Energy Conversion 2009;24:617–25.
[24] Zhang W, Liu Y. Multi-objective reactive power and voltage control based on
References fuzzy optimization strategy and fuzzy adaptive particle swarm. International
Journal of Electrical Power & Energy Systems 2008;30:525–32.
[25] Bellman RE, Zadeh LA. Decision-making in a fuzzy environment. Management
[1] Shahnia F, Majumder R, Ghosh A, Ledwich G, Zare F. Operation and control of Science 1970:B17–41.
a hybrid microgrid containing unbalanced and nonlinear loads. Electric Power [26] Baer B, Butler K, Hiyama T, Kubokawa J, Lee K, Luh P, et al. Tutorial on fuzzy
Systems Research 2010;80:954–65. logic applications in power systems; 2000.
[2] Kashefi Kaviani A, Riahy G, Kouhsari S. Optimal design of a reliable [27] Storn R, Price K. Differential evolution – a simple and efficient heuristic for
hydrogen-based stand-alone wind/PV generating system, considering compo- global optimization over continuous spaces. Journal of Global Optimization
nent outages. Renewable Energy 2009;34:2380–90. 1997;11:341–59.
[3] Nelson D, Nehrir M, Wang C. Unit sizing and cost analysis of stand- [28] Varadarajan M, Swarup K. Differential evolutionary algorithm for optimal reac-
alone hybrid wind/PV/fuel cell power generation systems. Renewable Energy tive power dispatch. International Journal of Electrical Power & Energy Systems
2006;31:1641–56. 2008;30:435–41.
[4] Seeling-Hochmuth G. A combined optimisation concet for the design and [29] Ba os R, Manzano-Agugliaro F, Montoya F, Gil C, Alcayde A, Gómez J. Opti-
operation strategy of hybrid-PV energy systems* 1. Solar Energy 1997;61: mization methods applied to renewable and sustainable energy: a review.
77–87. Renewable and Sustainable Energy Reviews 2011;15:1753–66.
[5] Muselli M, Notton G, Louche A. Design of hybrid-photovoltaic power gen- [30] Elbeltagi E, Hegazy T, Grierson D. Comparison among five evolutionary-
erator, with optimization of energy management. Solar Energy 1999;65: based optimization algorithms. Advanced Engineering Informatics 2005;19:
143–57. 43–53.
[6] Koutroulis E, Kolokotsa D, Potirakis A, Kalaitzakis K. Methodology for opti- [31] del Valle Y, Venayagamoorthy GK, Mohagheghi S, Hernandez JC, Harley RG.
mal sizing of stand-alone photovoltaic/wind-generator systems using genetic Particle swarm optimization: basic concepts, variants and applications in
algorithms. Solar Energy 2006;80:1072–88. power systems. IEEE Transactions on Evolutionary Computation 2008;12:
[7] Maghraby H, Shwehdi M, Al-Bassam GK. Probabilistic assessment of pho- 171–95.
tovoltaic (PV) generation systems. IEEE Transactions on Power Systems [32] Shi Y, Eberhart R. A modified particle swarm optimizer. IEEE 1998:69–73.
2002;17:205–8. [33] IEEE Reliability Test System. http://wwweewashingtonedu/research/pstca/rts/
[8] Dufo-López R, Bernal-Agustín JL. Design and control strategies of PV-Diesel pg tcartshtm.
systems using genetic algorithms. Solar Energy 2005;79:33–46.