Renewable and Sustainable Energy Reviews 59 (2016) 11491158

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

Optimal sizing for a hybrid power system with wind/energy storage

based in stochastic environment
Ahmad Mohamed Abd el Motaleb a,n, Sarah Kazim Bekdache a, Lzaro Alvarado Barrios b
a
b

Department of Electrical Engineering, University of Warwick, Coventry CV4 7AL, United Kingdom
Department of Electrical Engineering, University of Loyola Andaluca, Campus Sevilla 41014, Spain

art ic l e i nf o

a b s t r a c t

Article history:
Received 28 November 2014
Received in revised form
5 November 2015
Accepted 27 December 2015

For isolated power networks supplied by intermittent energy sources, several doubts have emerged
regarding the impact of the uncertainties on the networks reliability. This paper performs optimal sizing
for a hybrid power system with wind/energy storage sources based on stochastic modeling of historical
wind speed and load demand. The autoregressive moving average is used to stochastically model the
uncertainty of the load demand/wind speed and, the sequential Monte Carlo simulation is performed to
chronologically sample the system states. The contribution of the paper can be summarized as follows:
(1) an objective function based on self-adapted evolutionary strategy in combination with the Fischer
Burmeister algorithm is proposed to minimize the one-time investment and annual operational costs of
the wind/energy storage sources; and (2) the effect of the cycle efciency and charging/discharging rate
of different energy storage units on the system cost is investigated under different reliability and load
shifting levels. The computational performance of the proposed optimization solver is proven in order to
obtain the minimum possible investment cost. The presented case studies in this paper provide the
decision makers with the exibility to choose the suitable capacity installation at different values of
reliability and load shifting levels.
& 2016 Elsevier Ltd. All rights reserved.

Keywords:
Distributed generation (DG)
Load shifting
Stochastic modeling
Hybrid energy systems
Wind turbines
Energy storage sources

Contents
1.
2.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1149
Uncertainties modeling and probabilistic reliability evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1151
2.1.
ARMA models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1151
2.2.
Probabilistic reliability evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1151
3. Capacity optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1152
3.1.
Storage capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1152
3.2.
Storage constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1152
3.3.
Reliability constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1152
4. Optimal capacity algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1153
4.1.
Self-adapted evolutionary strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1153
4.2.
FischerBurmeister algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1153
5. Numerical simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1154
5.1.
Optimization with respect to reliability constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1156
5.2.
Effect of load shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1156
5.3.
Computational performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1157
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1157
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1157

Corresponding author.
E-mail address: A.Motalab@warwick.ac.uk (A. Mohamed Abd el Motaleb).

http://dx.doi.org/10.1016/j.rser.2015.12.267
1364-0321/& 2016 Elsevier Ltd. All rights reserved.

1150

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

1. Introduction
Deploying distributed generators (DGs) in power networks is
considered as one of the solutions to reduce reliance on fossil fuels
through exploitation of renewable energy sources. It also deferring
the investment for network upgrading as the load demand grows
[14]. In rural areas, it is very difcult and uneconomical to deliver
power over long distances through transmission lines to supply
such areas. The lack of an electrical network to supply remote
areas, high connection cost of grid extension and rough topography often leads to other options to supply energy. Stand-alone
hybrid systems which are dependent on renewable sources are
found to be a promising way to satisfy the energy supply
requirements for these areas [5]. The need for efcient electric
power sources in remote locations is a driving force for the
research in hybrid energy systems [6].
The wind power is considered as one of the possible energy
sources in rural areas. In general, wind energy avoids the full fuel
cost and a considerable portion of annual operation and maintenance costs of the displaced conventional units. However, the
level of avoided capital costs depends on the extent to which wind
power capacity can displace the conventional units, and so that
level is directly related to the capacity factor of the wind plants,
while the main drawback of the wind power is its uncertainty
nature [7,8]. With the market liberalization and the occurrence of
competitive markets, vital questions have arisen regarding the
optimum economical sizing of wind generators to contribute to
supplying the network with the constraints of power quality. In
the new power market model, the investors tend to favor dispatchable energy sources characterized by low capital costs, such
as gas turbines, despite their high operating costs. Consequently,
the wind energy, which is characterized by high capital cost, is
hurt by the shift to low capital cost technologies. Usually, batteries
are used to back intermittent energy sources, especially in the
small and medium-sized energy systems [9]. However, the problem of keeping the power balance is still more difcult for standalone networks supplied by intermittent generators. The characteristics of such grids require scheduling more reserve for
ensuring adequate security and reliability levels, but the higher
reserve requirements may substantially increase the investment
and operational costs of those systems.
Several studies have shown that the challenges and future
research topics to be addressed for DGs sizing/placement in the
hybrid power systems [1012]; some of these challenges include:
(1) modeling the network uncertainties; (2) investigating the
impact of DGs efciencies on the network planning and operational constraints; (3) investigating the impact of energy management strategies on the planning decisions; and (4) developing
more efcient computational algorithms for high-dimensional
space problems. In [13], an analytical technique was proposed to
optimize the DG capacity with respect to the network losses. In
[14], a heuristic curve-tted technique was employed to determine
the optimal location and size of distributed generators with
respect to the network losses. However, in [13,14], the network
uncertainties were not included.
In [15,16], Vrettos and Arabali performed optimal sizing of
stand-alone hybrid power systems; nevertheless, the reliability
analyses were not considered. In [17,18], Wang and Yang proposed
analytical methods to model the hybrid power systems and evaluate the reliability indices; but, these methods cannot properly
represent the random nature of the hybrid power systems [19]. In
[20,21], the capacity optimization model for hybrid power systems
was discussed using different heuristic optimization techniques. In
both references, a constraint was imposed to ensure that the
hybrid generation capacity always satises the load level; however, such a constraint leads to overestimation of the investment

cost, especially with the possibility of adopting load management

strategies. In [22,23], Different methodologies for allocating
energy storage systems in power networks were presented.
However in [22,23], the impact of the load sifting and energy
storage parameters on the system investment cost and reliability
were not investigated. Furthermore, in [24,25], optimal sizing
techniques of power systems were applied based on the genetic
algorithm (GA) to minimize the investment cost of the generation
system. Usually, the search capability of the GA is dependent on
the population size which could increase the computational burden and lead to non-optimal solution. The central idea of heuristic
optimization techniques is to nd the optimum solution for a
given problem, without the need for an exhaustive search. These
algorithms may fail in complex problems where the search space
is very large [26]. The computational performance of the proposed
optimization solver in this study, and its superiority over GA in
terms of accuracy and computational time will be claried in
Section 5.3.
In order to overcome the shortcomings of the previous
researches mentioned in the literature, this paper performs optimal sizing of wind/energy storage sources based on stochastic
modeling of historical wind speed and load demand. The ARMA
modeling is used to represent the uncertainty of load demand/
wind speed, and the SMCS is preformed to chronologically sample
the system states. Thereafter, an objective function is presented to
minimize the one-time investment and annual operational costs of
the wind/energy storage sources and, the reliability index, the
expected energy not served (EENS) is included as a constraint.
Several case studies are considered to illustrate the results under
different values of EENS/load shifting levels. Furthermore, the
effect of the cycle efciency and charging/discharging rate of different energy storage units on the system reliability will be claried. Two battery units (Nickel-cadmium and Lithium-ion) will be
separately simulated with the wind generation unit to clarify the
effect of batteries characteristics on the total hybrid system cost
and reliability. From the numerical simulations, it will be shown
that with increasing the reliability level at 0% load shifting, the
battery unit parameters are the dominant factors for determining
the total hybrid system cost. On the other hand, with increasing
the load shifting level at a constant reliability level, the battery
unit cost ($/kWh) is the dominant factor for determining the
hybrid system cost. The problem is solved using self-adapted
evolutionary strategy (SAES) in combination with the Fischer
Burmeister algorithm based on nonlinear complementarity problem (NCP) [27]. The competency of the proposed optimization
solver will be proven in order to obtain the minimum investment
cost with respect to the problem constraints. The contribution of
the study can be summarized as follows:
1) Proposing an optimization solver based on self-adapted evolutionary strategy in combination with the FischerBurmeister
algorithm, which depends on intelligent mutation of the offspring candidates without the need for exhaustive iterations
within a complex search space;
2) Investigating the impact of the energy storage units parameters
on the system investment cost under different values of reliability/load shifting levels, so as to provide the decision makers
with the exibility to choose the suitable capacity installation
without overestimation of the investment cost.
Three years historical data of load demand/wind speed are
used to process the ARMA models and provide the hourly samples
of 1 year scheduling period. The historical wind speed data is
based on the Warwickshire area, UK. Moreover, the historical load
data of one of the main feeders of Warwick University are obtained
from the energy tracker data base.

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

Nomenclature
St ; t ; t Hourly simulated, mean and standard deviation of
load/wind speed
APE
Absolute percentage error
wi ; wr ; wf Wind turbine cut-in speed, rated speed and
furling speed
Pr
Wind turbine rated power
wt
Wind generator converter efciency
bat ; wt Annual discount rate of the battery and wind
turbine units
mbat ; mwt Annual maintenance cost of the battery and wind
turbine units

1151

Q bat ; P bat Battery unit capacity and power rating

pbat ; pinv Unit price of the battery and inverter units
IC
Wind turbine unit and installation cost
P max
Battery maximum power rating
bat
PtL ; P tLi ; P tsh Hourly load demand, interrupted and shifted load
d ; c
Battery discharge/charge efciency
P tEd ; P tEc Hourly discharged/charged power
C
Battery stored energy
EENS
Expected energy not served
t
Time interval
IC
Wind turbine unit and installation cost

2. Uncertainties modeling and probabilistic reliability

evaluation
The historical load demand/wind speed record form a time
sequence, and therefore the ARMA technique can be used to predict a future load demand/wind speed models [28].
2.1. ARMA models
The ARMA(p, q) model is the combination of the autoregressive
(AR(p)) and the moving average (MA(q)):
yt

p
X
i1

i yt  i

q
X

j t  j

j0

where, i i 1; :::; p are the AR coefcients; whereas j j 1; :::; q

are the MA coefcients. The t  j is normally and independently
distributed random white noise process with zero mean and 2
variance. This model suggests that the value at any point in the
time series is a linear combination of both the p prior points, along
with the q prior white noise points.
Three years historical data are used to extract the ARMA coefcients and to provide the hourly samples of 1 year scheduling
period. For every month over every year of the historical data, the
data were divided into a number of segments; thus, there are several data points for the same hour over that month. Consequently,
for each hour of the 1 year scheduling period, the output of the
ARMA models were averaged and scaled using the mean and
standard deviation of the corresponding hour of the historical data.
The time series obtained using (1) is applied to calculate the
simulated hourly load/wind speed for the 1 year scheduling period
as follows [28]:
St t t yt

To ensure that the applied ARMA models for the planning study
are accurate, several ARMA models of different orders are compared against the actual wind speed/load demand data during the
year 2013 using the previous historical data, and the models which
show the least APE with respect to the actual historical data are
chosen to predict the wind speed/load demand for the 1 year
scheduling period.
The ARMA(5,4) and ARMA(4,3) models are used to predict the
wind speed and load demand, respectively. The ARMA models
demand were evaluated by the APE as follows:


y  St 
APEt t
3
yt
where yt is the actual daily wind speed/load demand at the hour t.
Fig. 1 claries the APE value for the predicted daily load demand
and wind speed samples from January 2013 until December 2013.

Fig. 1. Absolute percentage errors of daily wind speed and load demand samples.

The APE values are reasonable over the comparison period; however, on the holidays, the load demand show high APE values due
to unpredictable events, such as maintenance and random load
demand consumption.
The wind power can be subsequently obtained from the
simulated hourly wind speed [29], as follows:
8
w2  wi 2
>
>
< P r wr 2  wi 2 ; wi o w o wr
4
Pw P r ; wr ow o wf
>
>
: 0; Otherwise
where wt is set as at a predetermined value and multiplied by the
net wind power output as shown in (5). Similarly, the conversion
efciency is included for the energy storage unit. The cut-in speed,
rated speed and furling speed of the wind turbine are set as 2.5 m/
s, 13 m/s and 25 m/s, respectively. Moreover, the turbine rated
power is multiplied by a reasonably constrained parameter,wt to
balance the power mismatch and represent the total wind capacity
installation.
P wt Pw  wt

2.2. Probabilistic reliability evaluation

The uncertainties of the hybrid power system include the variations of load demand, wind speed and also the equipment failure
rate of the wind turbine unit. The two-state model is used to
generate the wind turbine updown sequence which is employed
with the wind speed to output the wind turbine generating

1152

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

capacity. The sequential state duration sampling is adopted to

provide the hourly system states. Based on [30], the following
steps are followed:
1) The wind turbine is initially assumed to be in the up-state;
2) The state duration of the wind turbine unit is sampled. The
sampled value of the state duration is shown by (6);
3) A load demand sample is generated by the ARMA load model;
4) Steps 23 are repeated for the scheduling period, T and, the
wind speed/load demand samples are arranged in a
chronological order;
5) The optimization solver is run to minimize the total hybrid
system cost for a specic samples number, N over the period,
T; and
6) The reliability index, EENS is calculated, as to be claried later.
D

1

ln R

The D and R are the state duration, and a uniformly distributed

random number between [0, 1]. is the outage/repair rate
depending on the present state. In this paper, the wind generating
unit can reside in one of two mutually exclusive states (up/down)
state. The forced outage rate of the wind model is assumed as 0.04.

3. Capacity optimization
An objective function is dened to minimize the one-time
investment and the annual variable costs of the wind/energy
storage sources over the whole life cycle of the hybrid system. In
this study, it is assumed that the whole life cycle is 15 years. The
total cost of the hybrid system is illustrated by (7).
X
min
Ci
7
i wt;ES

The subscripts wt and ES refer to the wind turbine and energy

storage system. The cost model for the wind turbine and energy
storage unit include the one-time investment cost, annual operational and maintenance cost, and annual discount rate. The cost
model components except the one-time investment are multiplied
by 15 to accumulate these costs over the whole life cycle.
According to [20], the total cost model of the energy storage unit
can be stated as follows:
C ES pbat  Q bat pinv  P bat bat  pbat mbat  15  Q bat

Two battery units (Nickel-cadmium and Lithium-ion) will be

separately simulated with the wind generation unit to clarify the
effect of batteries characteristics on the system reliability and cost.
The unit price of the Nickel-cadmium and Lithium-ion batteries
are set as 400 $/kWh and 600 $/kWh, respectively [31] and the
annual maintenance cost is set as 0.2 $/kWh. The inverter cost is
based on [32]. Similarly, the wind unit cost model can be appended to (7) as follows:
C wt IC  P r IC  wt mwt  15  P r

The wind unit and installation costs are set as 1500 $/kW and
450 $/kW, respectively and, the annual operation and maintenance
costs are set as 0.06 $/kW [22].
3.1. Storage capacity
The installed energy storage capacity must satisfy the maximum and minimum capacity constraints, (10). The minimum
capacity in this study is set to a null value. The maximum installed
capacity of the energy storage can be obtained according to the
size of area where the energy storage unit will be installed [21,,33].

Thus, the optimum energy storage capacity (with respect to the

cost) which can satisfy the reliability constraints must be equal or
less than the maximum capacity restricted by the size of the
installation area. In this study, a specic area size was assumed to
obtain the maximum installed capacity of the energy storage units.
This maximum capacity is used to indicate the maximum power
rating of the unit,P max
bat ; subsequently, the FischerBurmeister
function can be applied, as will be shown by (19).
Q min r Q bat r Q max

10

3.2. Storage constraints

The chargingdischarging battery process is based on [34] and
is shown in (11).
P tE

Ct 1 Ct  t d
d

for P twt  P tL P tLi P tsh o0

Ct 1 Ct t  P tEc  c
for P twt  P tL P tLi P tsh 4 0

11

The load management process applied in this study is a

sequential process: (1) If the wind energy is not sufcient to meet
the demand level and the battery stored energy is almost null,
then the load shifting strategy is applied; (2) the P tsh (restricted to a
specic percentage of PtL , as will be claried in Section 5) is
transferred to the next time interval; (3) if however, after applying
the load shifting, the power balance cannot be maintained, then a
specic level of the load demand is interrupted,P tLi ; (4) steps 13
are repeated and (11) is re-calculated until the power balance is
achieved.
In [35], it was claried that the charge/discharge rate is a vital
determinant for the battery lifetime. The lifetime of a battery
drops if the cell is charged or discharged at a rate higher than the
rated value due to detrimental chemical and structural changes
that occur during such a process. Therefore another constraint is
applied as shown in (12).
0 r P tEd r P max
Ed
0 r P tEc r P max
Ec

12

In [33], several values of discharge rate were provided for the

Lithium-ion and Nickel-cadmium batteries, and so the average
values of these batteries are applied in this study. The maximum
discharge at the rated capacity dened in (12) for the Lithium-ion
and Nickel-cadmium batteries are set as 2 hours and 3 hours,
respectively. The cycle efciencies of the Nickel-cadmium and
Lithium-ion batteries are set as 0.7 and 0.99, respectively [33].
3.3. Reliability constraints
The accumulation of the interrupted power demand over the
scheduling period is dened as the energy not served. On the
other hand, the EENS is a probabilistic index used in the reliability
analysis. The EENS is chosen for the reliability evaluation of the
proposed hybrid power system. The EENS is dened as shown in
(13), [30].
N
P

ENSi
EENS i 1
N

13

ENS is the energy not supplied, and N is the number of samples.

In order to satisfy a certain reliability level for the hybrid power
system, the following reliability constraint is applied.
EENS r EENSmax

14

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

4. Optimal capacity algorithm

Usually, the evolution strategies are based on stressing the
behavioral link between an individual and its offspring. Evolution
strategies employ real-coded variables, and depend on the mutation as a search operator.
4.1. Self-adapted evolutionary strategy
For SAES, a population of parents generatesoffspring by
randomly selecting one parent and mutating it, and this is repeated fortimes, provided that r . In this paper, the was chosen
to be 20, and the was chosen to be 100. The parents of the next
generation are selected as the best among the individuals generated by the mutation [36]. The SAES associates with each individual an extra variable which represents the variance of its
mutation distribution. The problem variables are mutated as follows:
X~ k X kg Z k ; k 1 to
1
Z k g
N0; 1; :::; N0; 1t
k

15

where, N0; 1; ; Z are the Gaussian distribution with zero mean

and unit variance, square root of the variance, and a random vector,
respectively. The letter g denotes the generation number, and the
symbol  denotes the mutated object variables. X is an individual
represented by a vector of n real-valued variables, along with an
extra variable related to the variance of the Gaussian distribution
commanding the mutations in its offspring, as shown in Fig. 2. For
the optimal progress rate, the k is mutated as follows:

~ k k ez0 zk

16

The mutation factors,z0 ; zk should be represented by Gaussian

distributions dependent on the learning parameter, [37].
z0 A N0; 20 ; 20 K 2

1
n

1
zk A N0; 2k ; 2k K 2 p
2 n
1

K  ln
2

droop control equation is restricted to the maximum power rating

of the energy storage,P max
bat :
From the previous text, it can be concluded that the battery
power rating can be determined through a piecewise function. The
problem of solving such a piecewise function lies in that the rating
of the energy storage unit that decides the state which will be
active is not known in advance, because it is dependent on the
rating of the wind energy unit owning different settings and
nominal values.
The proposed methodology employs FischerBurmeister algorithm based on nonlinear complementarity problem (NCP) [28],
through substituting the piecewise function by a scalar function.
The FischerBurmeister algorithm receives the generated individuals from the SAES and allows the exploration of feasible subspaces. Subsequently, the optimization solver is directed to the
promising feasible search areas and can be solved by less number
of iterations. Usually, when solutions neighboring a possible
optimum exist, the spread of the probability distribution that
regulates mutation should become narrower. The FischerBurmeister algorithm is used as a direct search tool to solve the nonsmooth problem. The following text explains the mathematical
formulation of the FischerBurmeister algorithm and thereafter,
its local search capability will be claried.
The droop formulation of active power depends on two possible states (with and without power limit reached). The problem
can be reformulated by (18).
( max
; P max
P bat  P bat
bat Z P bat
18
f
0
0
 mp P bat  P bat ; P bat o P max
bat
where,P 0bat ; 0 ; ; mp are the interface nominal power, nominal
frequency, equilibrium frequency, and droop regulation coefcient
(assumed as a null value in this study). The maximum power
rating of the energy storage unit,P max
bat is determined from the
maximum installed capacity, (10). The NCP reformed (18) to (19),
as follows:
0
0
0 r P max
bat  P bat ?  mp P bat  P bat  Z 0

17

Usually, the mutation mechanism does not take into account

the solution feasibility. Therefore, infeasible offspring frequently
appear. Furthermore, the computational burden is high because
the search capability of the SAES is dependent on the population
size. Thereby, the convergence rate will be detrimentally affected,
and the algorithm may get trapped in local optima. In order to
overcome this drawback, the FischerBurmeister algorithm is
appended to the SAES.
4.2. FischerBurmeister algorithm
Typically, the batteries are used to back intermittent energy
sources especially in the small and medium-sized energy systems
in order to satisfy the reliability requirements. For this purpose, it
is assumed that the energy storage interfaced via inverter unit is
regulated by the droop control, P bat  ; to achieve the reliability
constraints. The P bat and denote the battery power rating and
network frequency, respectively. As explained in Section 3, the

Fig. 2. Representation of an individual.

1153

19

The previous NCP equation can be formulated by means of the

FischerBurmeister function, as : R2 -R satisfying:

j; k 0 3 jZ 0; k Z 0; j:k 0

20

j 
Eq. (20)
where
can be reformulated by the FischerBurmeister as shown in (21).
q
j; k j2 k2  j k
21
k P max
bat  P bat and

mp P 0bat  P bat :

The effect of applying the FischerBurmeister lies in improving

the mutation capability of the SAES. At the beginning of the optimization process, high mutation rates are preferable because they
allow new individuals to jump away from parents and probe vast
regions of the search space (diversication). However, at a later
stage, large perturbations drive individuals away from the region of
the optimum solution. For each generated individuals, the individuals with the highest tness values are located over the Fischer
Burmeister constraints. The individual with the highest tness
values is chosen as the mean of the next parent. Thereafter, the
mutation rate, ~ k is multiplied by an adaption factor, pg :. This factor
is directly proportional to the standard deviation between the
individuals with the highest tness values. Normally, the factor will
decrease if the standard deviation between the individuals
decreases, and the reverse hold true, (22). In this way, the individuals are smoothly perturbed and will not be driven away from the
region of the optimum. Therefore, when solutions neighboring a
possible optimum exist, the spread of the probability distribution

1154

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

Fig. 3. Focusing on the optimum (a) and (b) The spread of the individuals over the FischerBurmeister constraints at the beginning and end of the optimization process,
respectively. (c) and (d) Gaussian distribution at the beginning and end of the optimization process, respectively.

that regulates the mutation will become narrower (intensication).

5. Numerical simulations

~ k pg  ~ k

In this section, the one-time investment and annual operational costs of the hybrid power system will be numerically discussed with respect to the reliability constraints through two
scenarios. In the rst scenario, the EENSmax will be varied through
ve steps, and the optimal hybrid power system cost will be
accordingly evaluated for two cases, the rst case includes the
Lithium-ion battery along with the wind energy unit, and the
second case includes the Nickel-cadmium battery along with the
wind energy unit. The purpose of this scenario is to clarify the
effect of battery unit parameters on the total hybrid power system
cost. In the second scenario, the load shifting effect on the total
hybrid power system cost will be investigated.
The planning model only provides the optimum values which
are shown in Tables 14 for each expected energy not served and
load shifting level. However, Figs. 58 represent the probability
density functions of the whole population used in the optimization
model. Fig. 4 shows the ow chart of the planning model. The
maximum cycle number is set to 100 as the termination criterion,
and for each cycle a population of 100 individuals is generated. The
purpose of plotting the whole population used in the optimization
model in Figs. 58 is to illustrate the behavioral tendency of the
population for each simulation scenario and that the conclusion
belong to the whole population, and not only to the optimum
solution. Thereby, the conclusion can be drawn and generalized.
Finally, the competency of the proposed optimization solver will
be claried.

22

Fig. 3 shows the spread of the individuals over the Fischer

Burmeister constraints at the beginning and the end of the optimization process, where the optimum rating of the energy storage
is identied.
unit, P opt
bat
The decision variables of the planning problem are included in
the following vector: X P bat ; wt . Where wt is a constrained
parameter multiplied by the wind turbine rated power, P r to balance the power mismatch. The Mechanism of evaluating the system installation capacity is shown in Fig. 4, and can be briefed as
follows:
1) The wind speed and load demand AMRA models are
established;
2) The wt parameter and P bat variable are initialized by the SAES;
3) The SMCS is run for N samples over the scheduling period, T. For
each sample, a large penalty coefcient is appended to the
objective function if the constraints are violated to assign a
higher cost to an infeasible solution;
4) The tness value for each individual is evaluated, and the
mutation rate is accordingly modied;
5) This process is repeated and the termination criterion is satised if the maximum cycle number is reached or, the difference
between two consecutive results is less than a pre-specied
tolerance;
6) The individuals are polled according to their tness values.

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

1155

Fig. 4. Mechanism of evaluating the hybrid power system installation capacity.

Table 1
Capacity installation costs at different reliability levels (Lithium-ion battery case).
EENS max MWh

0.25

Wind energy unit cost (M$)

Battery unit cost (M$)
Wind installation capacity
(MW)
Battery installation capacity
(kWh)
Total hybrid system cost (M$)

Table 2
Capacity installation
battery case).

costs

at

0.19

5.47
0.68
2.81

0.13

5.69
0.69
2.92

0.07

5.81
0.74
2.98

different

EENS max MWh

Wind energy unit cost (M$)
Battery unit cost (M$)
Wind installation capacity (MW)
Battery installation capacity
(kWh)
Total hybrid system cost (M$)

6.38

reliability

0.25

0.19

6.55

levels

0.13

6.27
0.82
3.22

6.8
0.89
3.49

7.09

7.69

(Nickel-cadmium

0.07

0.01

6.06
6.2
6.53
7.05
7.64
0.35
0.36
0.39
0.43
0.47
3.11
3.18
3.35
3.62
3.92
587.57 599.21 644.98 714.99 772.15
6.41

6.56

6.92

costs

at

different

load

shifting

levels

(Lithium-ion

0.01

838.96 852.84 920.97 1020.85 1101.6

6.15

Table 3
Capacity installation
battery case).

7.48

8.11

Level of load shifting (%)

20

25

Wind energy unit cost (M$)

Battery unit cost (M$)
Wind installation capacity (MW)
Battery installation capacity
(kWh)
Total hybrid system cost (M$)

5.53
5.3
5.05
4.83
4.66
0.72
0.62
0.56
0.48
0.46
2.84
2.72
2.59
2.48
2.39
888.91 773.85 691.23 592.76 576.18
6.25

5.92

30

35

5.61

40

5.31

5.12

Table 4
Capacity installation costs at different load shifting levels (Nickel-cadmium
battery case).
Level of load shifting (%)

20

Wind energy unit cost (M$)

Battery unit cost (M$)
Wind installation capacity (MW)
Battery installation capacity
(kWh)
Total hybrid system cost (M$)

5.83
5.51
5.14
4.99
4.73
0.38
0.33
0.29
0.25
0.24
2.99
2.83
2.64
2.56
2.43
621.86 541.75 483.74 414.62 402.98
6.21

25

5.84

30

35

5.43

5.24

40

4.97

1156

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

Fig. 5. Probability density functions of the installed wind power for Lithium-ion
and Nickel-cadmium cases, EENSmax r 0:13.

Fig. 8. Probability density functions of the total hybrid system cost for Lithium-ion
and Nickel-cadmium cases, 30% load shifting.

5.1. Optimization with respect to reliability constraints

Fig. 6. Probability density functions of the total hybrid system cost for Lithium-ion
and Nickel-cadmium cases, EENSmax r 0:13.

The EENSmax will be varied through ve steps from 0.25 MWh

to 0.01 MWh at 0% load shifting. Tables 1 and 2 illustrate the
optimal hybrid power system cost with respect to the reliability
constraint. For the case of the Nickel-cadmium battery, it is clear
that the investment cost of the wind energy unit is always higher
than its corresponding cost in the case of the Lithium-ion battery,
and subsequently, the total hybrid power system cost in the
Nickel-cadmium battery case is higher. The battery parameters
have signicant effect on the reliability constraints. The Nickelcadmium battery with its current parameters cannot satisfy the
strict reliability constraints, and consequently, the contribution of
the wind energy unit increases. Fig. 5 shows the probability density functions of the whole population for the installed wind
power at EENSmax r 0:13 and under 0% load shifting. The gure
claries that the population individuals in the Nickel-cadmium
case always record higher installed wind power values than their
corresponding in the Lithium-ion case, and accordingly, the total
hybrid system cost is higher in the Nickel-cadmium case, (Fig. 6).
5.2. Effect of load shifting

Fig. 7. Probability density functions of the installed wind power for Lithium-ion
and Nickel-cadmium cases, 30% load shifting.

In this section, the total hybrid system cost will be optimized

with respect to EENSmax r 0:07; and the load shifting strategy will
be applied for different percentage levels of the total system load.
Thus, if the load demand at time t exceeds the generation level,
then the modied load demand at time t 1, will include the
shifted load level. From Tables 3 and 4, it is clear that the total
hybrid system cost in the case of Nickel-cadmium battery signicantly drops with applying the load shifting. In this case, the
contribution of the wind energy unit reduces, and the battery unit
cost ($/kWh) is the dominant factor for determining the total
hybrid power system cost. Fig. 7 shows the individuals probability
density functions for the installed wind power under 30% load
shifting. The gure claries that the individuals for the Nickelcadmium case slightly record higher installed wind power values
than their corresponding individuals in the Lithium-ion case.
However, the total hybrid cost is always lower in the case of
Nickel-cadmium (Fig. 8). From this scenario, it can be stated that
with increasing the reliability level at 0% load shifting, the battery
unit parameters, cycle efciency and discharge rate are the
dominant factors for determining the hybrid system cost. On the
other hand, with increasing the load shifting level (if there is

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

1157

Table 5
Computational performance.
Item

Approximate run time for each simulation

Fig. 9. Total hybrid power system cost, EENSmax r 0:13 different algorithms.

Solver
MSAES

GA

SAES

0.24

0.67

0.32

gets trapped in local optima. On the other hand, the Fischer

Burmeister algorithm could be used as a direct search tool to solve
the non-smooth problem with high computational efciency and
could improve the mutation capability of the SAES. Thus, the
individuals were smoothly perturbed and were not driven away
from the region of the optimum. Also, it is clear that local search
capability of the GA solver is poor, because it records results with
high standard deviation between the consecutive solution candidates. Furthermore, it is observed from Table 5 that the proposed
solver is an efcient method to handle the optimization framework presented in this study in terms of the computational time.

Conclusion

Fig. 10. Computational performance, EENSmax r 0:13 different algorithms.

agreement between the power producers and consumers) at a

constant reliability level, the battery unit cost ($/kWh) is the
dominant factor for determining the hybrid system cost.
5.3. Computational performance
In this section, computational performance of the modied
self-adapted evolutionary solver (MSAES) based on the Fischer
Burmeister algorithm is compared against the conventional SAES;
additionally, the GA solver is included in this comparison using the
same population size. The analysis is established for 1000 simulations (Nickel-cadmium battery case) under 0% load shifting and
atEENSmax r 0:13 and the results are shown in Figs. 9 and 10. In
Fig. 9, the box plot is chosen to clarify the individuals total cost
obtained from the different solvers. On each box, the central mark
is the median, the edges of the box are the 25th and 75th percentiles, and the whiskers extend to the most extreme data points.
The Figures clarify the superiority of the proposed algorithm
over the typical solvers in terms of the result quality. In Fig. 10, the
results for each iteration are represented by the mean value of the
population.The MSAES records small standard deviation between
the consecutive solution candidates, and could reach to the steady
state faster than the other solvers. For the conventional SAES, the
computational burden is high because the search capability of the
SAES is dependent on the population size and therefore, the solver

This paper performs optimal sizing of hybrid power system

including wind/energy storage sources based on stochastic modeling for historical wind speed and load demand proles. The
purpose is to provide the decision makers with the exibility to
choose the suitable capacity installation under different values of
reliability/load shifting levels. The autoregressive moving average
is used to stochastically model the uncertainty of the load
demand/wind speed and, the sequential Monte Carlo simulation is
performed to chronologically sample the system states and the
reliability index, EENS is subsequently evaluated. Two contributions are presented in this study: (1) optimization solver based on
self-adapted evolutionary strategy in combination with the
FischerBurmeister algorithm, which depends on intelligent
mutation of the offspring candidates without the need for
exhaustive iterations within a complex search space; and
(2) investigating the impact of the cycle efciency and charging/
discharging rate of different energy storage units on the system
investment cost under different values of reliability/load shifting
levels. From the numerical simulations, it could be concluded that
with increasing the reliability level at 0% load shifting level, the
battery unit parameters are the dominant factors for determining
the total hybrid system cost. On the other hand, with increasing
the load shifting level at a constant reliability level, the battery
unit cost ($/kWh) is the dominant factor for determining the total
hybrid system cost. Furthermore, the competency of the proposed
optimization solver is claried in terms of the result quality and
computational time.

References
[1] Bollen MH, Hassan F. Integration of distributed generation in the power system. New York: Wiley-IEEE Press; 2011.
[2] Srivastava AK, Kumar AA, Schulz NN. Impact of distributed generations with
energy storage devices on the electric grid. IEEE Syst J 2012;vol. 6(1):1107.
[3] Mistry Khyati D, Roy Ranjit. Enhancement of loading capacity of distribution
system through distributed generator placement considering technoeconomic benets with load growth. Int J Electr Power Energy Syst 2014;
vol. 54:50515.
[4] Paliwal Priyanka, Patidar NP, Nema RK. Planning of grid integrated distributed
generators: a review of technology, objectives and techniques. Renew Sustain
Energy Rev 2014;vol. 40:55770.
[5] Lin Whei-Min, Hong Chih-Ming, Chen Chiung-Hsing. Neural-network-based
MPPT control of a stand-alone hybrid power generation system. IEEE Trans
Power Electron 2011;26(12):357181.

1158

A. Mohamed Abd el Motaleb et al. / Renewable and Sustainable Energy Reviews 59 (2016) 11491158

[6] Nehrir H, Wang Caisheng, Strunz K, Aki H, Ramakumar R, Bing J, Zhixhin Miao,
Salameh Z. Power and Energy Society General Meeting; 2012. p. 226.
[7] Lingfeng Wang, Singh C. Multicriteria design of hybrid power generation
systems based on a modied particle swarm optimization algorithm. IEEE
Trans Energy Convers 2009;24(1):16372.
[8] Tao Zhou, Francois B. Energy management and power control of a hybrid
active wind generator for distributed power generation and grid integration.
IEEE Trans Ind Electron 2011;58(1):95104.
[9] Li XJ, Hui D, Lai XK. Battery energy storage station (BESS)-based smoothing
control of photovoltaic (PV) and wind power generation uctuations. IEEE
Trans Sustain Energy 2013;4(2):46473.
[10] Keane A, Ochoa LF, Borges CLT, Ault GW, Alarcon-Rodriguez AD, Currie RAF,
Pilo F, Dent C, Harrison GP. State-of-the-art techniques and challenges ahead
for distributed generation planning and optimization. IEEE Trans Power Syst
2013;vol. 28(2):1493502.
[11] Georgilakis PS, Hatziargyriou ND. Optimal distributed generation placement in
power distribution networks: models, methods, and future research. PIEEE
Trans Power Syst 2013;vol.28(3):34208.
[12] Hemmati R, Hooshmand R-A, Khodabakhshian A. Comprehensive review of
generation and transmission expansion planning. Gener Transm Distrib IET
2013;vol. 7(9):95564.
[13] Hung Duong Quoc, Mithulananthan N, Bansal RC. Analytical expressions for
DG allocation in primary distribution networks. IEEE Trans Energy Convers
2010;vol. 25(3):81420.
[14] Abu-Mouti FS, El-hawary ME. Heuristic curve-tted technique for distributed
generation optimisation in radial distribution feeder systems. Gener Transm
Distrib IET 2011;vol. 5(2):17280.
[15] Vrettos EI, Papathanassiou SA. Operating policy and optimal sizing of a high
penetration RES-BESS system for small isolated grids. IEEE Trans Energy
Convers 2011;26(3):74456.
[16] Arabali A, Ghofrani M, Etezadi-Amoli M, Fadali MS, Baghzouz Y. Geneticalgorithm-based optimization approach for energy management. IEEE Trans
Power Deliv 2013;28(1):16270.
[17] Wang L, Singh C. Multicriteria design of hybrid power generation systems
based on a modied particle swarm optimization algorithm. IEEE Trans Energy
Convers 2009;24(1):16372.
[18] Yang H, Zhou W, Lu L, Fang Z. Optimal sizing method for standalone hybrid
solar-wind system with LPSL technology by using genetic algorithm. Sol
Energy 2008;82(4):35467.
[19] Kishore LN, Fernandez E. Reliability well-being assessment of PV wind hybrid
system using Monte Carlo simulation Proceedings of the international conference on emerging trends in electrical and computer technology (ICETECT),
Roorkee, India; 2011.
[20] Tianpei Zhou Wei Sun. Optimization of batterysupercapacitor hybrid energy
storage station in wind/solar generation system. IEEE Trans Sustain Energy
2014;5(2):40815.

[21] Wai Rong-Jong, Cheng Shan, Lin Yeou-Fu, Chen Yi-Chang. Installed capacity
selection of hybrid energy generation system via improved particle-swarmoptimisation. Gener Transm Distrib IET 2014;8(4):74252.
[22] Atwa YM, El-Saadany EF. Optimal allocation of ESS in distribution systems
with a high penetration of wind energy. IEEE Trans Power Syst 2010;vol.25
(4):181522.
[23] Xu Lin, Ruan Xinbo, Mao Chengxiong, Zhang Buhan, Luo Yi. An improved
optimal sizing method for wind-solar-battery hybrid power system. IEEE
Trans Sustain Energy 2013;vol. 4(3):77485.
[24] Ying-Yi Hong Ruo-Chen Lian. Optimal sizing of hybrid wind/pv/diesel
generation in a stand-alone power system using markov-based genetic algorithm. IEEE Trans Power Deliv 2012;27(2):6407.
[25] Evangelopoulos VA, Georgilakis PS. Optimal distributed generation placement
under uncertainties based on point estimate method embedded genetic
algorithm. Gener Transm Distrib IET 2014;vol.8(3):389400.
[26] Fernndez-Martnez JL, Mukerji T, Garca-Gonzalo E, Fernndez-Muiz Z.
Math Comput Model 2011;54:288999.
[27] Fischer A. Math Program Op Res 1992;24:26984.
[28] Karki R, Dhungana D, Billinton R. An appropriate wind model for wind integrated power systems reliability evaluation considering wind speed correlations. Appl Sci 2013;3(1):10721.
[29] Li Wang Tai-Her Yeh, We-Jen Lee Zhe Chen. Benet evaluation of wind turbine
generators in wind farms using capacity-factor analysis and economic-cost
methods. IEEE Trans Power Syst 2009;24(2):692704.
[30] Roy Billinton. Reliability evaluation of power systems. Pitman Advanced
Publishing Program; 1984.
[31] Chen Haisheng, Cong Thang Ngoc, Yang Wei, Tan Chunqing, Li Yongliang, Ding
Yulong. Progress in electrical energy storage system: a critical review. Prog
Natl Sci 2009;19:291312.
[32] Gonzalez Adolfo, . Gallachir Brian, McKeogh Eamon. Study of electricity
storage technologies and their potential to address wind energy intermittency
in Ireland. Sustainable Energy Ireland; 2004.
[33] Beaudin Marc, Zareipour Hamidreza, Schellenberglabe Anthony, Rosehart
William. Energy storage for mitigating the variability of renewable electricity
sources: an updated review. Energy Sustain Dev 2010;14:30214.
[34] Chen SX, Gooi HB, Wang MQ. Sizing of energy storage for microgrids. IEEE
Trans Smart Grid 2012;3(1):14251.
[35] Sharkh Suleiman Abu, Doerffel Dennis. A critical review of using the Peukert
equation for determining the remaining capacity of lead-acid and lithium-ion
batteries. J Power Sources 2006;155(2):395400.
[36] Beyer H-G. Toward a theory of evolution strategies: the (m, l) theory. Evol
Comput 1995;2(4):381407.
[37] Schwefel H-P, Rudolph G. Contemporary evolution strategies. In: Moran F,
Moreno A, Merelo JJ, Chacon P, editors. Advances in articial life. 3rd International conference on articial life. vol. 929 of Lecture Notes in Articial
Intelligence. Berlin: Springer; 1995. p. 893907.