Solar Energy 80 (2006) 1072–1088

www.elsevier.com/locate/solener

Methodology for optimal sizing of stand-alone

photovoltaic/wind-generator systems using genetic algorithms
Eftichios Koutroulis a,*, Dionissia Kolokotsa b,
Antonis Potirakis a, Kostas Kalaitzakis a
a
Department of Electronic and Computer Engineering, Technical University of Crete, GR-73100 Chania, Greece
b
Department of Natural Resources and Environment, Technological Educational Institute of Crete, GR-73133 Chania, Greece

Received 25 May 2005; received in revised form 4 November 2005; accepted 17 November 2005
Available online 27 December 2005

Communicated by: Associate Editor Mukund Patel

Abstract

A methodology for optimal sizing of stand-alone PV/WG systems is presented. The purpose of the proposed method-
ology is to suggest, among a list of commercially available system devices, the optimal number and type of units ensuring
that the 20-year round total system cost is minimized subject to the constraint that the load energy requirements are com-
pletely covered, resulting in zero load rejection. The 20-year round total system cost is equal to the sum of the respective
components capital and maintenance costs. The cost (objective) function minimization is implemented using genetic algo-
rithms, which, compared to conventional optimization methods such as dynamic programming and gradient techniques,
have the ability to attain the global optimum with relative computational simplicity. The proposed method has been
applied for the design of a power generation system which supplies a residential household. The simulation results verify
that hybrid PV/WG systems feature lower system cost compared to the cases where either exclusively WG or exclusively
PV sources are used.
 2005 Elsevier Ltd. All rights reserved.

Keywords: Photovoltaic power systems; Wind power generation; Genetic algorithms

1. Introduction istics, they are usually used in hybrid system config-

urations. The block diagram of a stand-alone hybrid
Photovoltaic (PV) and Wind Generator (WG) PV/WG system is shown in Fig. 1. Battery chargers,
power sources are widely used in order to supply connected to a common DC bus, are used to charge
power to consumers in remote areas. Due to their the battery bank from the respective PV and WG
almost complementary power production character- input power sources, which are usually configured
in multiple power generation blocks according to
*
Corresponding author. Tel.: +30 28210 37233; fax: +30 28210
the devices nominal power ratings and the redun-
37542. dancy requirements. Depending on the battery char-
E-mail address: [email protected] (E. Koutroulis). ger technology, the maximum available power can

0038-092X/$ - see front matter  2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.solener.2005.11.002
 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1073

Nomenclature

aw WG size (m2) n2 conversion factor (%)

as PV size (m2) N PV
ch number of PV battery chargers
cw WG cost per unit area ($/m2) NPV total number of PV modules
cs PV cost per unit area ($/m2) Pm
ch power rating of the selected battery char-
d(t) average daily demand at day t (kW h) ger (W)
W(t) WG energy production at day t per unit Pm PV maximum power of one PV module un-
area (kW h/m2) der STC (W)
S(t) PV energy production at day t per unit h WG installation height (m)
area (kW h/m2) href reference WG height (m)
N number of PV modules P iWG ðt; hÞ power transferred to the battery bank
Nb number of batteries at hour t of day i, from a WG installed
a PV module cost ($) at height h (W)
b battery cost ($) vi(t, h) wind speed at height h (m/s)
NP number of PV modules connected in par- viref ðtÞ reference wind speed measured at height
allel href (m/s)
NS number of PV modules connected in series a power law exponent
P iM ðt; bÞ maximum output power of the PV Cn battery bank total nominal capacity
array on day i (1 6 i 6 365) and at hour (A h)
t (1 6 t 6 24) (W) DOD maximum permissible depth of discharge
I iSC ðt; bÞ PV module short-circuit current (A) (%)
ISC,STC PV module short-circuit current under Cmin minimum permissible battery capacity
STC (A) during discharging (A h)
b PV module tilt angle () Ci(t) available battery capacity at hour t of
Gi(t, b) global irradiance incident on the PV day i (A h)
module placed at tilt angle b (W/m2) nB battery efficiency (%)
KI short-circuit current temperature coeffi- VBUS DC bus voltage (V)
cient (A/C) P iB ðtÞ battery input/output power (W)
V iOC ðtÞ PV module open-circuit voltage (V) Dt simulation time step (h)
VOC,STC open-circuit voltage under STC (V) nSB number of batteries connected in series
KV open-circuit voltage temperature coeffi- VB nominal voltage of each individual bat-
cient (V/C) tery (V)
T iA ðtÞ ambient temperature (C) NBAT total number of batteries
NCOT Nominal Cell Operating Temperature CB nominal capacity of each battery (A h)
(C) P ire ðtÞ total power transferred to the battery
FFi(t) Fill Factor bank during day i and hour t (W)
b1 PV module tilt angle from January until NWG total number of WGs
April and September until December () P iL ðtÞ DC/AC inverter input power (W)
b2 PV module tilt angle for the rest of the P iload ðtÞ power consumed by the load at hour t of
year () day i (W)
Vm DC battery charger maximum input voltage ni DC/AC converter efficiency (%)
(V) J(x) 20-year round total system cost function
Vm OC PV modules maximum open-circuit volt- (€)
age (V) Cc(x) total capital cost function (€)
P iPV ðt;bÞ PV power transferred to the battery Cm(x) maintenance cost function (€)
bank on day i (W) x vector of the decision variables
ns battery charger conversion factor hlow WG tower lower height limit (m)
n1 battery charger power electronic inter- hhigh WG tower upper height limit (m)
face efficiency (%) CPV capital cost of one PV module (€)
1074 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

CWG capital cost of one WG (€) CINV capital cost of the DC/AC inverter (€)
CBAT capital cost of one battery (€) yBAT expected number of battery replacements
MPV maintenance cost per year of one PV during the 20-year system operation
module (€/year) M PV
ch maintenance cost per year of one PV bat-
MWG maintenance cost per year of one WG (€/ tery charger (€/year)
year) MINV maintenance cost per year of one DC/AC
MBAT maintenance cost per year of one battery inverter (€/year)
(€/year) Cmax absolute minimum of function J(x) cal-
Ch WG tower capital cost per meter (€/m) culated at each generation (€)
Chm WG tower maintenance cost per meter psc initial probability of Simple Crossover
and year (€/m/year) psac initial probability of Simple Arithmetical
C PV
ch capital cost of one PV battery charger (€) Crossover
pwac initial probability of Whole Arithmetical
y PV
ch expected number of PV battery charger
Crossover
replacements during the 20-year system
pum uniform mutation probability
lifetime
pbm boundary mutation probability
yINV expected number of DC/AC inverter
pnum non-uniform mutation probability
replacements during the 20-year system
lifetime

be extracted from the PV and WG power sources terminals are connected in parallel. The energy pro-
(Maximum Power Point Tracking, MPPT), (Lore- duced from each PV or WG source is transferred to
nzo, 1994; De Broe et al., 1999). The battery bank, the consumer load through the battery charger and
which is usually of lead-acid type, is used to store the DC/AC inverter, while the energy surplus is
the energy surplus and to supply the load in case used to charge the battery bank.
of low wind speed and/or irradiation conditions. Because of the intermittent solar irradiation and
A DC/AC converter (inverter) is used to interface wind speed characteristics, which highly influence
the DC battery voltage to the consumer load AC the resulting energy production, the major aspects
requirements. The outputs of all battery chargers, in the design of PV and WG power generation sys-
the battery bank and the DC/AC converter input tems are the reliable power supply of the consumer

Fig. 1. Block diagram of a hybrid PV/WG system.

 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1075

under varying atmospheric conditions and the cor- Using the values of d(t), W(t) and S(t) during the
responding total system cost. Considering the wide year, a plot of aw versus as is obtained and the opti-
range of commercially available device types, it is mal configuration is defined on this curve as the
essential to select the number and type of PV mod- point where
ules, WGs and batteries, and their installation das cw
details such that power is uninterruptedly supplied ¼ ð3Þ
daw cs
to the load and simultaneously the minimum system
cost is achieved. However, using this method the battery size is not
A sizing method of stand-alone PV systems has included in the optimization process.
been presented by Shrestha and Goel (1998), which A method for the selection of the optimum com-
is based on energy generation simulation for various bination of battery and PV array in a hybrid PV/
numbers of PVs and batteries using suitable models WG system has been presented by Borowy and Sal-
for the system devices (PVs, batteries, etc.). The ameh (1996). The system operation is simulated for
selection of the numbers of PVs and batteries various combinations of PV array and battery sizes
ensures that reliability indices such as the Loss of and the Loss of Power Supply Probability (LPSP) is
Load Hours (LOLH), the lost energy and the sys- calculated for each combination. Then, for the
tem cost are satisfied. In a similar method, Markov desired LPSP, the PV array versus battery size are
chain modeling is used for the solar radiation. In plotted and the optimal solution, which minimizes
this case, the number of PVs and batteries are the total system cost, is defined as the point on the
selected according to the desired System Perfor- sizing curve where
mance Level (SPL) requirement, which is defined dN b
as the number of days that the load cannot be satis- ¼ ð4Þ
dN b a
fied and it is expressed in terms of probability
(Maghraby et al., 2002). where N, Nb are the number of PV modules and bat-
A design method for hybrid PV/WG systems, teries and a, b are the costs ($) of a PV module and a
based on energy balance, has been proposed by Kel- battery, respectively.
logg et al. (1998). Using the hourly average data of In a similar method (Bagul et al., 1996), the N
wind speed, solar radiation and consumer power versus Nb curve is plotted using a probabilistic anal-
demand, the difference of generated and demanded ysis of the daily energy surplus. In both methods,
power (DP) is calculated over a 24-h period. The the WG size is not included in the optimization
numbers of PV modules and WGs are finally process.
selected, using an iterative procedure where the sys- According to the methods proposed by Chedid
tem operation is simulated for various numbers of and Rahman (1997) and Yokoyama et al. (1994)
PVs and WGs, such that DP has an average value the optimal sizes of the PV and WG power sources
of zero. The total annual cost for each configuration and the batteries are determined by minimizing the
is calculated and the combination with the lowest system total cost function using linear programming
cost is selected to represent the optimal mixture. techniques. The total system cost consists of both
The seasonal variation of PV and WG power the initial cost and yearly operation and mainte-
generation is taken into account in the method pro- nance costs.
posed by Markvart (1996), where the problem of Common disadvantage of the methods described
optimal sizing the PV and WG power sources in a above is that the proposed sizing methodologies do
hybrid generation system is formulated as follows: not take into account system design characteristics
such as the number of battery chargers, the PV
minimize : system cost ¼ cs  as þ cw  aw ð1Þ modules tilt angle and the WG installation height,
subject to : dðtÞ 6 W ðtÞ  aw þ SðtÞ  as ð2Þ which highly affect both, the resulting energy pro-
duction and the installation and maintenance costs.
where aw and as are the WG and PV sizes (m2), cw Also, the minimization of the system cost function
and cs represent the corresponding costs per unit has been implemented either by linearly changing
area ($/m2), d(t) is the average daily demand at the values of the corresponding decision variables
day t (kW h) and W(t), S(t) are the WG and PV en- or employing linear programming techniques,
ergy productions at day t per unit area (kW h/m2), resulting in suboptimal solutions and increased
respectively. computational effort requirements.
1076 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

In this paper, an alternative methodology for the which compared to conventional optimization
optimal sizing of stand-alone PV/WG systems is methods, such as dynamic programming and gradi-
proposed. The purpose of the proposed methodol- ent techniques, has the ability to attain the global
ogy is to suggest, among a list of commercially optimum with relative computational simplicity.
available system devices, the optimal number and GAs have been applied to the design of large power
type of units ensuring that the 20-year round total distribution systems (Ramirez-Rosado and Bernal-
system cost is minimized subject to the constraint Agustin, 1998) and the solution of power economic
that the load energy requirements are completely dispatch problems (Li, 1998) because of their ability
covered, resulting in zero load rejection. The 20- to handle complex problems with linear or non-lin-
year round total system cost is equal to the sum of ear cost functions both, accurately and efficiently. In
the respective components capital and maintenance the proposed method, GAs are selected because
costs. The decision variables included in the optimi- they have shown to be highly applicable to cases
zation process are the number and type of PV mod- of large non-linear systems, where the location of
ules, WGs and battery chargers, the PV modules tilt the global optimum is a difficult task. Due to the
angle, the installation height of the WGs and the probabilistic development of solutions, GAs are
battery type and nominal capacity. The minimiza- not restricted by local optima. Thus, given the num-
tion of the cost (objective) function is implemented ber of non-linearities that exist in similar problems,
employing a genetic algorithms (GA) approach, GAs appear to be a useful approach.

Fig. 2. Flowchart of the proposed optimization methodology.

 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1077

This paper is organized as follows: the proposed The current–voltage and power–voltage charac-
methodology is outlined in Section 2, the hybrid sys- teristics of a PV array in each power generation
tem modeling and operation simulation is described block shown in Fig. 1, which consists of NP modules
in Section 3, the system cost minimization algorithm connected in parallel, and NS modules connected in
using GAs is analyzed in Section 4 and the simula- series, are shown in Fig. 3(a). The maximum output
tion results are presented in Section 5. power of the PV array on day i (1 6 i 6 365) and at
hour t (1 6 t 6 24), P iM ðt; bÞ (W), is calculated using
2. The proposed methodology the specifications of the PV module under Standard
Test Conditions (STC, cell temperature = 25 C and
A general block diagram outlining the proposed solar irradiance = 1 kW/m2), provided by the man-
methodology is shown in Fig. 2. The optimization ufacturer, as well as the ambient temperature and
algorithm input is fed by a database containing irradiation conditions, according to the following
the technical characteristics of commercially avail- equations:
able system devices along with their associated per
P iM ðt; bÞ ¼ N S  N P  V iOC ðtÞ  I iSC ðt; bÞ  FFi ðtÞ ð5Þ
unit capital and maintenance costs. Various types
i
of PV modules and WGs, batteries with different G ðt; bÞ
I iSC ðt; bÞ ¼ fI SC;STC þ K I ½T iC ðtÞ  25 g ð6Þ
nominal capacities, etc., are stored in the input data- 1000
base, which is implemented in the form of text files V iOC ðtÞ ¼ V OC;STC  K V  T iC ðtÞ ð7Þ
for easy maintenance. NCOT  20  C i
The first step of the optimal sizing methodology T iC ðtÞ ¼ T iA ðtÞ þ G ðt; bÞ ð8Þ
800
consists of a system simulation procedure in order
to examine whether a system configuration, com-
prising a certain number of system devices and
installation details, fulfills the load power supply
requirements during the year. The data used in this
case are the daily solar irradiation on horizontal
plane, the hourly mean values of ambient tempera-
ture and wind speed and the consumer power
requirements for an one year time period. The sec-
ond step of the optimal sizing procedure consists
of a method employing GAs, which dynamically
searches for the system configuration, which subject
to the criterion set in the first step, minimizes the
system total cost. For each combination of system
device types, the optimal sizing procedure is per-
formed computing the corresponding optimal total
system cost and devices configuration. After all
device type combinations have been optimally sized
as described above, the combination with the lowest
cost and the corresponding devices mixture are dis-
played as the overall optimal system configuration.

3. The PV/WG system modeling and operation

simulation

In the proposed method, the system operation is

simulated for one year with a time step of 1 h. The
power produced by the PV and WG power sources
is assumed to be constant during that time period.
Therefore, the power generated by the renewable Fig. 3. PV and WG output power characteristics: (a) PV module
energy sources is numerically equal with the energy current–voltage and power–voltage characteristics and (b) WG
generated within this time step. power versus wind speed characteristic.
1078 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

where I iSC ðt; bÞ is the PV module short-circuit cur-

rent (A), ISC,STC is the short-circuit current under
STC (A), Gi(t, b) is the global irradiance (W/m2)
incident on the PV module placed at tilt angle b
(), KI is the short-circuit current temperature coef-
ficient (A/C), V iOC ðtÞ is the open-circuit voltage (V),
VOC,STC is the open-circuit voltage under STC (V),
KV is the open-circuit voltage temperature coeffi-
cient (V/C), T iA ðtÞ is the ambient temperature
(C), NCOT is the Nominal Cell Operating Temper-
ature (C), provided by the manufacturer and FFi(t)
is the Fill Factor (Markvart, 1994). Fig. 4. Deviation of the measured PV array output power from
The value of Gi(t, b) is calculated using the daily the corresponding maximum power using a PV charger without
MPPT function.
solar irradiation on the horizontal plane as analyzed
by Lorenzo (1994). The PV modules tilt angle can
be selected by the system designer to be either con-
stant during the year, b, or variable with angle b1 to the battery bank. The available solar irradiance
corresponding to the months from January until during the measurements ranges from 0 to 900 W/
April (day numbers 1–104) and September until m2 and the resulting average value of the conversion
December (day numbers 290–365) and b2 corre- factor n2 is approximately 70%.
sponding to the rest of the year. The number of PV battery chargers, N PV ch , which
The number of PV modules connected in series in is also equal with the total number of PV power
the PV array, depends on the battery charger max- generation blocks, depends on the total number of
imum input voltage, V m DC (V), and the PV modules PV modules, NPV
maximum open-circuit voltage level, V m OC (V)
N PV  P m
m N PV
ch ¼
PV
ð11Þ
V DC Pm
NS ¼ m ð9Þ ch
V OC where P mch is the power rating of the selected battery
The PV power actually transferred to the battery charger (W), and P m PV is the maximum power of one
bank on day i, P iPV ðt; bÞ (W), is related to the max- PV module under STC (W), both specified by their
imum output power of the PV array, P iM ðt; bÞ, manufacturers.
through the battery charger conversion factor, ns, The diagram of the WG output power versus
which is defined as wind speed is shown in Fig. 3(b). Such a diagram
is provided by the manufacturer and it usually indi-
P iPV ðt; bÞ cates the actual power transferred to the battery
ns  ¼ n1  n2 ð10Þ
P iM ðt; bÞ bank from the WG source, taking into account
the effects of both the charger power conversion effi-
where n1 is the battery charger power electronic ciency and the MPPT operation, if available. Thus,
interface efficiency, specified by the manufacturer modeling of the WG battery charger characteristics
and n2 is a conversion factor, which depends on has not been considered in the proposed method.
the battery charging algorithm and indicates the This diagram is input to the optimization algorithm
deviation of the actual PV power generated, from in the form of a lookup table, storing the corre-
the corresponding maximum power. sponding output power and wind speed points. Dur-
In case that the battery charger operates accord- ing the optimization process, the power transferred
ing to the MPPT principle, n2 is approximately to the battery bank, P iWG ðt; hÞ (W), at hour t of
equal to 1, otherwise it differs significantly. An day i, from a WG installed at height h (m), is calcu-
example of the PV array measured output power lated using the following linear relation:
and the corresponding maximum available power
P2  P1
during the bulk-charging phase of a commercial P iWG ðt; hÞ ¼ P 1 þ ½vi ðt; hÞ  v1  ð12Þ
battery charger without MPPT capability, is v2  v1
depicted in Fig. 4. During the bulk-charging phase where vi(t, h) is the wind speed (m/s) at height h and
of this charger, the PV array is directly connected (P1, v1), (P2, v2) are WG power and wind speed pairs
 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1079

stored in the lookup table, such that The modeling equations described above form the
v1 < vi(t, h) < v2. simulation algorithm, which is used to verify
In order to calculate the wind speed, vi(t, h), at whether a solution derived by the GA-based cost
the desired WG installation height, h, which is usu- function minimization procedure fulfills the load
ally different from the height corresponding to the power requirements during the whole year. The cor-
wind speed input data, the exponential law is used responding flowchart is depicted in Fig. 5. The algo-
 a rithm input data set consists of the daily solar
i i h
v ðt; hÞ ¼ vref ðtÞ  ð13Þ irradiation on horizontal plane, the hourly mean
href
values of ambient temperature and wind speed,
where viref ðtÞ is the reference (input) wind speed the load power requirements during the year and
(m/s) measured at height href (m) and a is the power the specifications of the system devices, while it is
law exponent, ranging from 1/7 to 1/4. executed with a time step of 1 h. Initially, the total
The battery bank, with total nominal capacity Cn power, P ire ðtÞ (W), transferred to the battery bank
(A h), is permitted to discharge up to a limit defined from the PV and WG power sources during day i
by the maximum permissible depth of discharge (1 6 i 6 365) and hour t (1 6 t 6 24) is calculated
DOD (%), which is specified by the system designer as follows:
at the beginning of the optimal sizing process, i.e.
P ire ðtÞ ¼ N PV  P iPV ðt; bÞ þ N WG  P iWG ðt; hÞ ð19Þ
C min ¼ DOD  C n ð14Þ
where Cmin is the minimum permissible battery where h is the WG installation height, NPV is the
capacity during discharging (A h). total number of PV modules and NWG is total the
Depending on the PV and WG energy produc- number of WGs. The PV modules tilt angle b is
tion and the load power requirements, the battery either constant during the year, or takes the values
state of charge is accumulated during the simulation b1, b2 as described above.
period as follows: Then, the DC/AC inverter input power, P iL ðtÞ
P iB ðtÞ (W), is calculated using the corresponding load
C i ðtÞ ¼ C i ðt  1Þ þ nB Dt ð15Þ power requirements, as follows:
V BUS
C i ð24Þ ¼ C iþ1 ð0Þ ð16Þ P iload ðtÞ
P iL ðtÞ ¼ ð20Þ
where Ci(t), Ci(t  1) is the available battery capac- ni
ity (A h) at hour t and t  1, respectively, of day i,
nB = 80% is the battery round-trip efficiency during where P iload ðtÞ (W) is the power consumed by the
charging and nB = 100% during discharging load at hour t of day i, defined at the beginning of
(Borowy and Salameh, 1996), VBUS is the DC bus the optimal sizing process and ni is the DC/AC con-
voltage (V), P iB ðtÞ is the battery input/output power verter efficiency (%), specified by the manufacturer.
(W) [P iB ðtÞ < 0 during discharging and P iB ðtÞ > 0 According to the above power production and
during charging] and Dt is the simulation time step, load consumption calculations, the resulting battery
set to Dt = 1 h. capacity is calculated
The number of batteries connected in series, nSB ,
depends on the nominal DC bus voltage and the • If P ire ðtÞ ¼ P iL ðtÞ then the battery capacity
nominal voltage of each individual battery, VB remains unchanged.
(V), and it is calculated as follows: • If P ire ðtÞ > P iL ðtÞ then the power surplus
P iB ðtÞ ¼ P ire ðtÞ  P iL ðtÞ is used to charge the bat-
V BUS
nSB ¼ ð17Þ tery bank and the new battery capacity is calcu-
VB lated using Eq. (15). In case that the battery
The battery bank nominal capacity is related SOC reaches the 100% SOC limit then the
with the total number of batteries, NBAT, the num- remainder of the available power is not used.
ber of series connected batteries and the nominal • If P ire ðtÞ < P iL ðtÞ then the power deficit
capacity of each battery, CB (A h), as follows: P iB ðtÞ ¼ P ire ðtÞ  P iL ðtÞ required to cover the load
energy requirements is supplied by the battery
N BAT bank and the new battery capacity is calculated
Cn ¼ CB ð18Þ
nSB using Eq. (15).
1080 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

Fig. 5. Flowchart of the algorithm simulating the system operation.

The above steps are repeated until either the simula- ciples of natural genetics. GAs are very different
tion time period finishes, at day number 365 and the from traditional search and optimization methods
24th hour, indicating successful system operation, used in engineering design problems. Fundamental
or the battery bank is discharged below the lowest ideas of genetics in biology are borrowed and used
permissible limit defined by Eq. (14). In such case, artificially to construct search algorithms that are
the system operation is considered to fail and the robust and require minimal problem information.
corresponding configuration is rejected, since it does A typical constrained, single variable optimiza-
not guarantee reliable power supply of the load. tion problem can be outlined as follows:
Maximisex f ðxÞ
4. System cost minimization algorithm using GAs ð21Þ
subject to the constraint : xmin 6 x 6 xmax
Genetic algorithms are adaptive search and opti- For the solution of such a problem with GAs the
mization approaches that work mimicking the prin- variable x is typically coded in some string struc-
 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1081

tures. Binary-coded or floating point strings can F ðN PV ; N WG ; N BAT ; N PV

ch ; h; bÞ
also be used, while the length of the string is usually 8 PV
determined according to the accuracy of the solu- < C max  J ðN PV ; N WG ; N BAT ; N ch ; h; bÞ;
>
tion desired. The GA, as any evolution procedure ¼ if C max  J ðN PV ; N WG ; N BAT ; N PV
ch ; h; bÞ > 0
>
:
for a particular problem, must have the following 0; otherwise
components (Michalewicz, 1994):
ð24Þ
• A generic representation for potential solutions subject to the constraints
to the problem, similar to the system modeling
presented in the previous section. N PV P 0 ð25Þ
• A way to create an initial population of potential N WG P 0 ð26Þ
solutions. N BAT P 0 ð27Þ
• An evaluation function that plays the role of the
N PV
ch P 0 ð28Þ
environment, rating solutions in terms of their
‘‘fitness’’ and hlow 6 h 6 hhigh ð29Þ
" #
• Genetic operators (such as crossover and muta- 0 6 b 6 90 or
tion) that alter the composition of children. ð30Þ
0 6 b1 6 90 and 0 6 b2 6 90
" #
In the proposed method, the GA optimal sizing Simulation ðN PV ; N WG ; N BAT ; N PV
ch ; h; bÞ ¼ Successful or

methodology outputs the optimum number of Simulation ðN PV ; N WG ; N BAT ; N PV

ch ; h; b1 ; b2 Þ ¼ Successful
WGs, PV modules, batteries and PV battery charg- ð31Þ
ers, along with the optimum PV modules tilt angle
and WGs installation height, comprising the set of where hlow, hhigh are the WG tower lower and upper
decision variables, such that the 20-year round total height limits (m), respectively, specified by the WG
system cost function, J(x) (€), is minimized. The manufacturer, CPV, CWG, and CBAT are the capital
total system cost function is equal to the sum of costs (€) of one PV module, WG and battery,
the total capital, Cc(x) (€), and maintenance cost, respectively, MPV, MWG, and MBAT are the mainte-
Cm(x) (€), functions nance costs per year (€/year) of one PV module,
WG and battery, respectively, Ch is the WG tower
minfJ ðxÞg ¼ minfC c ðxÞ þ C m ðxÞg ð22Þ capital cost per meter (€/m), Chm is the WG tower
x x
maintenance cost per meter and year (€/m/year),
where x is the vector of the decision variables listed C PV
ch is the capital cost of one PV battery charger

above. (€), y PV
ch , yINV are the expected numbers of PV bat-

Thus, multi-objective optimization is achieved by tery charger and DC/AC inverter replacements dur-
minimizing the total cost function consisting of the ing the 20-year system lifetime and it is equal to the
sum of the individual system devices capital and system lifetime (20 years) divided by the Mean Time
20-year round maintenance costs Between Failures (MTBF) of power electronic con-
verters (Holtz et al., 1994), CINV is the capital cost
J ðN PV ; N WG ; N BAT ; N PV of the DC/AC inverter, (€), yBAT is the expected
ch ; h; bÞ
number of battery replacements during the 20-year
¼ N PV  ðC PV þ 20  M PV Þ þ N WG  ðC WG system operation, because of limited battery life-
time, M PV ch , MINV maintenance costs per year (€/
þ 20  M WG þ h  C h þ 20  h  C hm Þ
year) of one PV battery charger and DC/AC inver-
þ N BAT  ½C BAT þ y BAT  C BAT ter, respectively and Cmax is the absolute minimum
PV
of function J(x) calculated at each generation, (€).
þ ð20  y BAT  1Þ  M BAT  þ N PV PV
ch  C ch  ðy ch þ 1Þ The Simulation(Æ) function performs the system
þ N PV PV PV simulation described in Section 3, in order to verify
ch  M ch  ð20  y ch  1Þ þ C INV  ðy INV þ 1Þ
that the examined system configuration fulfils the
þ M INV  ð20  y INV  1Þ ð23Þ uninterrupted power supply requirement of the load
during the whole year.
or, equivalently, by maximizing the GA fitness Real-Coded GA techniques are introduced for the
function solution of the optimization problem based on the
1082 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

mechanism of natural selection and natural genetics. The first step of the algorithm iteration is the fit-
Each chromosome consists of five genes in the form: ness function evaluation for each chromosome of
[NPVjNWGjNBATjhjb]. In case that the PV modules the corresponding population. If any of the result-
inclination changes seasonally, then six genes are ing fitness function values is lower than the lowest
used in the optimization algorithm and each chromo- value obtained at the previous iterations then this
some is of the form: [NPVjNWGjNBATjhjb1jb2]. The value is considered to be the optimal solution of
number of PV battery chargers has not been incorpo- the minimization problem and the corresponding
rated into the chromosome configuration in order to chromosome consists of the hybrid system optimal
reduce the algorithm complexity, since, as expressed operational parameter values. This optimal solution
in Eq. (11), it depends on the PV array power rating, is replaced by better solutions, if any, produced in
which in turn depends on the number of PV modules, subsequent GA generations during the program
NPV. However, the number of PV battery chargers evolution.
affects the total system cost, thus the associated cost In order to select the chromosomes, which will be
has been included in the GA fitness function, as subject to the crossover and mutation operations in
expressed in Eq. (24). order to produce the next generation population, a
The flowchart of the GA optimization process is selection operation is applied based on the roulette
depicted in Fig. 6. An initial population of 30 chro- wheel method (Michalewicz, 1994).
mosomes, comprising the 1st generation, is gener- The crossover mechanism uses the following
ated randomly and the constraints described by three operators:
inequalities (25)–(31) are evaluated for each chro-
mosome. If any of the initial population chromo- Simple Crossover, (SC) with initial probability
somes violates the problem constraints then it is psc = 10%.
replaced by a new chromosome, which is generated Simple Arithmetical Crossover, (SAC) with ini-
randomly and fulfils these constraints. tial probability psac = 10%.

Fig. 6. The GA optimization process.

 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1083

Whole Arithmetical Crossover (WAC) with ini- to enhance the fine local tuning capability during
tial probability pwac = 10%. If the chromosomes the optimization process.
c1 = [u1 . . . um] and c2 = [w1 . . . wm] are to be crossed For each chromosome a random number, r,
then the resulting offspring consists of (0 6 r 6 1) is generated
c01 ¼ awac  c1 þ ð1  awac Þ  c2 and c02 ¼ ð1  awac Þ
c1 þ awac  c2 . The parameter awac is set equal to • r < pum then this chromosome is selected for UM,
0.75. • pum 6 r < (pbm + pum) then the BM operator is
For each chromosome selected for crossover, a applied,
random number r (0 6 r 6 1) is generated. Depend- • (pbm + pum) 6 r < (pbm + pum + pnum) then the
ing on the value of r, one of the previous crossover chromosome is selected for NUM and finally,
operators is applied • r P (pbm + pum + pnum) then none of the three
mutation operators is applied.
• r < psc then this chromosome is selected for SC,
• psc 6 r < (psc + psac) then the SAC operator is In case that the application of any of the cross-
applied, over or mutation operators described above, results
• (psc + psac) 6 r < (psc + psac + pwac) then the in a chromosome which does not satisfy the optimi-
chromosome is selected for WAC and finally, zation problem constraints, then a ‘‘repair’’ proce-
• r P (psc + psac + pwac) then none of the three dure is performed and that chromosome is
crossover operators is applied. replaced by the corresponding parent. In case of
SC operation, where each new chromosome is gen-
Next, the selected chromosomes are subject to erated by two parents, then the chromosome is
the mutation mechanism, which is performed using replaced by the parent with the best fitness function
the following three operators: value.
Uniform Mutation, (UM). During UM a gene is The GA optimization process described above is
randomly selected and it is assigned a new value, repeated until a predefined number of population
randomly selected from the corresponding range generations have been evaluated.
of values which fulfill the optimization problem
constraints. This range of values is calculated for 5. Simulation results
the selected gene, considering the values of the other
genes within the chromosome constant. The muta- The proposed method has been applied to the
tion probability, pum, is 10%. design of a stand-alone hybrid PV/WG system in
Boundary mutation, (BM). The boundary muta- order to power supply a residential household
tion probability, pbm, is 3%. located in the area of the Technical University of
Non-uniform mutation, (NUM). The non-uni- Crete (TUC) with geographical coordinates defined
form mutation probability, pnum, is 35%, in order as

Fig. 7. Distribution of the consumer power requirements during the day.

1084 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

Latitude: 35.53 (3531 0 4800 N), year 2003, were recorded using a properly developed
Longitude: 24.06 (2403 0 3500 E) and data-acquisition system installed at the TUC
Altitude: 150 m (approx.) above sea level. campus.
The technical characteristics and the related cap-
The distribution of the consumer power require- ital and maintenance costs of the hybrid system
ments during a day (Lazou and Papatsoris, 2000) is devices, which are input to the optimal sizing proce-
shown in Fig. 7. It is assumed that the user appli- dure, are shown in Tables 1–5. The installation cost
ances are energy efficient and energy saving load has been included in the capital cost of the devices,
management strategies are applied, which are com- while the maintenance cost of each unit per year has
mon practices in renewable energy applications. been set at 1% of the corresponding capital cost.
The daily solar irradiation on horizontal plane, as The MTBF of the battery chargers and DC/AC
well as the hourly mean values of ambient tempera- inverter has been set equal to 40,000 h. The expected
ture and wind speed, plotted in Fig. 8 during the battery lifetime has been set at 3 years resulting in
yBAT = 6.
The optimal sizing results, consisting of both the
devices type and their number, for each combina-
tion of system devices included in Tables 1–5, are
shown in Table 6. The overall optimal solution is
combination #5, resulting in a 20-year round system
cost of 37,524 €. It is noteworthy that a PV charger
with MPPT capability has been selected in the opti-
mal solution, although of higher per unit cost com-
pared to the other PV charger option as shown in
Table 5, since it achieves better exploitation of the
available PV power. The corresponding battery
bank has an optimal total nominal capacity of
920 A h and the available capacity variation during
the year is depicted in Fig. 9. The maximum depth
of discharge is approximately 63.7%, thus indicating
the system capability to supply reliably the con-
sumer load. Also, the variation of the system total
cost (fitness function) during the GA optimization
procedure of Combination #5 is shown in Fig. 10.
It can be noted that a near optimal solution was
derived during the early stages of the GA genera-
tions evolution. The CPU time required for the evo-
lution of the 721 generations in the diagram of
Fig. 10 is approximately 3 min (Pentium IV,
2.6 GHz).
The optimal sizing results, for each combination
of system devices included in Tables 1–5, in case
that the power source consists either only of
WGs or only of PV modules are tabulated in
Tables 7 and 8, respectively. It is observed that
in both cases the overall optimal solutions, i.e.
combinations #1 and #5, respectively, result in
substantially higher total system cost compared to
the hybrid PV/WG system design. Also the PV-
only system results in approximately double total
Fig. 8. Hourly mean values during the year 2003 of meteorolog-
cost compared to the WG-only system because of
ical conditions: (a) ambient temperature, (b) wind speed and (c) the much higher per unit capital cost of PV
daily irradiation on horizontal plane. modules.
 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1085

Table 1
Batteries specifications
Type Nominal capacity (A h) Voltage (V) DOD (%) Capital cost (€) Maintenance cost per year (€/year)
1 230 12 80 264 2.64
2 100 12 80 126 1.26

Table 2
WGs specifications
Type Power hlow (m) hhigh (m) Capital Maintenance cost Tower capital Tower maintenance cost per year
rating (W) cost (€) per year (€/year) cost per meter (€/m) and per meter (€/year/m)
1 1000 8 15 1681 16.81 55 0.55
2 400 8 15 512 11.9 11 0.11

Table 3
PV modules specifications
Type Voc (V) Isc (A) Vmax (V) Imax (A) Pmax (W) NCOT (C) Capital cost (€) Maintenance cost per year (€/year)
1 21.6 3.48 17.3 3.18 55 43 265.81 2.66
2 21 7.22 17 6.47 110 43 519.14 5.19

Table 4
DC/AC inverter specifications
Type Efficiency (%) Power rating (W) Capital cost (€) Maintenance cost per year (€/year)
1 80 1500 1942.0 19.42

Table 5
PV battery chargers specifications
Type n1 (%) n2 (%) Power rating (W) Capital cost (€) Maintenance cost per year (€/year)
1 95 100 300 200.0 2.0
2 95 70 240 94.0 0.94

A remark concerning the tilt angles b1 and b2 installation procedures impose tilt angles less than
tabulated in Tables 6 and 8, is that the tilt angle val- 60 for the site under consideration. These b2 tilt
ues corresponding to the winter period, b1, are angle values, which minimize the system cost, result
sometimes lower than the relevant ones for the sum- in case that the system components optimal mixture
mer period, b2. This is due to the solar irradiation is such that PV energy surplus is produced during
profile of the site under consideration, exhibiting the summer period, thus the zero load rejection con-
prolonged cloudy intervals during the winter, result- straint is satisfied without requiring further reduc-
ing in low optimal b1 tilt angle values, since the tion of the b2 angle value.
main component of the global irradiation under In all preceding cases, the proposed algorithm
cloudy conditions is the diffused radiation, which convergence to the global optimum solution has
is maximized at low tilt angles. Also, the tilt angle been verified, using a properly developed simulation
b2 values are sometimes higher compared to the typ- program, by linearly changing the values of all deci-
ical angle values calculated using the latitude of the sion variables included in the optimization process
installation site. For example, b2 = 73 in the com- and calculating the corresponding 20-year round
bination of row #5 in Table 6 and b2 = 79 in the system cost. This procedure is repeated for each
combination of row #3 in Table 8, etc., while typical combination of system components types. The
1086 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

Table 6
Hybrid system optimal sizing results
Combination Device type Number Number Number WG Number PV tilt PV tilt Total
of of of height of PV b1 () b2 () cost (€)
WG PV Charger Inverter Battery
PVs WGs batteries (m) Chargers
1 1 1 1 1 1 19 3 6 15 4 50 44 40497.29
2 1 1 1 1 2 22 3 10 15 4 53 12 39144.08
3 1 1 2 1 1 17 3 8 15 4 54 28 41440.38
4 1 1 2 1 2 20 3 14 15 5 54 54 40400.16
5 1 2 1 1 1 11 3 4 15 4 42 73 37524.83
6 1 2 1 1 2 11 3 10 15 4 46 60 38979.35
7 1 2 2 1 1 9 3 8 14 5 45 55 41910.67
8 1 2 2 1 2 12 3 12 15 6 52 21 40183.68
9 2 1 1 1 1 22 16 9 15 4 50 32 53247.56
10 2 1 1 1 2 29 14 17 15 6 51 17 53975.95
11 2 1 2 1 1 30 16 9 15 7 52 72 55068.04
12 2 1 2 1 2 29 16 20 15 7 51 78 55775.79
13 2 2 1 1 1 13 14 9 15 5 58 26 53462.76
14 2 2 1 1 2 14 16 16 15 6 50 19 54444.93
15 2 2 2 1 1 15 16 9 15 7 53 49 54843.40
16 2 2 2 1 2 15 17 19 15 7 52 59 55919.74

Fig. 9. Simulated available battery capacity corresponding to the optimal solution, during the year.

Fig. 10. The system total cost during the GA optimization.

 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088 1087

Table 7
Optimal sizing results for WG-only power source
Combination Device type Number of WGs Number of batteries WG height (m) Total cost (€)
WG Inverter Battery
1 1 1 1 5 10 15 43860.50
2 1 1 2 5 24 15 46598.42
3 2 1 1 31 21 15 142753.22
4 2 1 2 31 48 15 146346.74

Table 8
Optimal sizing results for PV-only power source
Combination Device type Number Number Number PV tilt PV tilt Total
of of of b1 () b2 () cost (€)
PV Chargers Inverter Battery
PVs batteries PV chargers
1 1 1 1 1 57 26 11 15 11 88453.02
2 1 1 1 2 58 59 11 12 20 92836.10
3 1 2 1 1 73 28 17 31 79 94220.92
4 1 2 1 2 75 62 18 24 49 98337.56
5 2 1 1 1 29 26 11 12 54 88337.69
6 2 1 1 2 31 57 12 6 64 92880.97
7 2 2 1 1 36 28 17 30 47 93362.81
8 2 2 1 2 39 61 18 19 21 97812.03

resulting optimal solution, for each combination of variables or employing linear programming tech-
system components types, is equal to the solution niques, resulting in suboptimal solutions and
derived using the proposed GA optimization increased computational effort requirements.
method, in terms of both the total system cost and In this paper, a methodology for the optimal siz-
the values of the decision variables, thus proving ing of hybrid, stand-alone PV/WG systems, has
the GA optimization procedure convergence to the been presented. The purpose of the proposed meth-
global optimum solution. However, the CPU time odology is to support the selection, among a list of
required by the simulation program to derive the commercially available system devices, the optimal
optimal solution for each combination of system number and type of PV modules, WGs and PV bat-
components types is approximately 20 h, while tery chargers, the PV modules tilt angle, the instal-
using the proposed method the corresponding time lation height of the WGs and the battery type and
required is approximately 3 min. nominal capacity. The optimal number and type
of each system component is calculated such that
6. Conclusions the 20-year round total system cost is minimized
subject to the constraint that the load power
The major aspects in the design of PV and/or requirements are completely covered, thus resulting
WG power generation systems are (a) the reliable in zero load rejection. The 20-year round total sys-
power supply of the consumer under varying atmo- tem cost is equal to the sum of the respective com-
spheric conditions and (b) the corresponding total ponents capital and maintenance costs. The cost
system cost. Past proposed PV/WG system sizing (objective) function minimization is implemented
methods suffer the disadvantage of not taking into using genetic algorithms, which compared to con-
account system design characteristics such as the ventional optimization methods, such as dynamic
number of battery chargers, the PV modules tilt programming and gradient techniques, have the
angle and the WG installation height, which highly ability to attain the global optimum with relative
affect both the resulting energy production and the computational simplicity. The proposed method
installation and maintenance costs. Also, in the has been applied to the design of a power generation
existing literature, the minimization of the system system in order to supply a residential household.
cost function is implemented either by linearly The simulation results verify that hybrid PV/WG
changing the number of the corresponding decision systems result in lower system cost compared to
1088 E. Koutroulis et al. / Solar Energy 80 (2006) 1072–1088

cases where either exclusively WG or exclusively PV Lazou, A., Papatsoris, A., 2000. The economics of photovoltaic
sources are used. stand-alone residential households: a case study for various
European and Mediterranean locations. Solar Energy Mate-
rials & Solar Cells 62, 411–427.
References Li, F., 1998. A comparison of genetic algorithms with conven-
tional techniques on a spectrum of power economic dispatch
Bagul, A.D., Salameh, Z.M., Borowy, B., 1996. Sizing of a stand- problems. Expert Systems with Applications 15, 133–142.
alone hybrid wind-photovoltaic system using a three-event Lorenzo, E., 1994. Solar Electricity: Engineering of Photovoltaic
probability density approximation. Solar Energy 56 (4), 323– Systems, first ed. Progensa.
335. Maghraby, H.A.M., Shwehdi, M.H., Al-Bassam, G.K., 2002.
Borowy, B.S., Salameh, Z.M., 1996. Methodology for optimally Probabilistic assessment of photovoltaic (PV) generation
sizing the combination of a battery bank and PV array in a systems. IEEE Transactions on Power Systems 17 (1), 205–
wind/PV hybrid system. IEEE Transactions on Energy 208.
Conversion 11 (2), 367–373. Markvart, T., 1994. Solar Electricity, first ed. Wiley.
Chedid, R., Rahman, S., 1997. Unit sizing and control of hybrid Markvart, T., 1996. Sizing of hybrid photovoltaic-wind energy
wind-solar power systems. IEEE Transactions on Energy systems. Solar Energy 57 (4), 277–281.
Conversion 12 (1), 79–85. Michalewicz, Z., 1994. Genetic algorithms + data struc-
De Broe, A.M., Drouilhet, S., Gevorgian, V., 1999. A peak power tures = evolution programs, second ed. Springer-Verlag.
tracker for small wind turbines in battery charging applica- Ramirez-Rosado, I.J., Bernal-Agustin, J.L., 1998. Genetic algo-
tions. IEEE Transactions on Energy Conversion 14 (4), 1630– rithms applied to the design of large power distribution
1635. systems. IEEE Transactions on Power Systems 13 (2), 696–
Holtz, J., Lotzkat, W., Stadtfeld, S., 1994. Controlled AC 703.
drives with ride-through capability at power interruption. Shrestha, G.B., Goel, L., 1998. A study on optimal sizing of
IEEE Transactions on Industry Applications 30 (5), 1275– stand-alone photovoltaic stations. IEEE Transactions on
1283. Energy Conversion 13 (4), 373–378.
Kellogg, W.D., Nehrir, M.H., Venkataramanan, G., Gerez, V., Yokoyama, R., Ito, K., Yuasa, Y., 1994. Multiobjective optimal
1998. Generation unit sizing and cost analysis for stand-alone unit sizing hybrid power generation systems utilizing photo-
wind, photovoltaic and hybrid wind/PV systems. IEEE voltaic and wind energy. Transactions of the ASME: Journal
Transactions on Energy Conversion 13 (1), 70–75. of Solar Energy Engineering 116, 167–173.