Applied Energy 119 (2014) 118–130

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

A Maximum Power Point Tracking (MPPT) for PV system using Cuckoo

Search with partial shading capability
Jubaer Ahmed, Zainal Salam ⇑
Centre for Electrical Energy Systems, Faculty of Electrical Engineering, Universiti Teknologi Malaysia, 81310 Johor Bahru, Malaysia

h i g h l i g h t s g r a p h i c a l a b s t r a c t

 A soft computing MPPT method for

PV system, based on Cuckoo Search
(CS) is proposed.
 The method exhibits outstanding
Maximum Power Point Tracking
(MPPT) capability, good transient
performance and fast convergence.
 CS successfully handles the partial
shading condition.
 Compare to prominent MPPT
methods, P&O and PSO; CS shows
improved MPP tracking performance
and higher efficiency.

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

Article history: Objectives: This work proposes a Maximum Power Point Tracking (MPPT) for PV system using Cuckoo
Received 1 April 2013 Search (CS) method.
Received in revised form 13 November 2013 Methods: It is acknowledged that CS exhibits several advantages which include fast convergence, higher
Accepted 23 December 2013
efficiency using fewer tuning parameters. The paper outlines the concept of CS by highlighting the signif-
icance of the Lévy flight in influencing the algorithm’s convergence. The main equations that govern the
behavior of the search are also explained. To justify CS as a viable MPPT option, a comprehensive assess-
Keywords:
ment is carried out against two well established methods, namely Perturbed and Observed (P&O) and
Photovoltaic
MPPT
Particle Swarm Optimization (PSO). The evaluations include (1) gradual irradiance and temperature
Soft computing changes, (2) step change in irradiance and (3) rapid change in both irradiance and temperature. These
Cuckoo Search tests are carried out for both large and medium-sized PV systems. Furthermore, the ability of the algo-
Lévy flight rithm to handle the partial shading condition is demonstrated.
Partial shading Results: The results show that CS is capable of tracking MPP within 100–250 ms under various types of
environmental change. Besides, the power loss in steady state due to MPP mismatch is only
0.000008%. Furthermore, it can handle the partial shading condition very efficiently.
Conclusion: CS outperforms both P&O and PSO with respect to tracking capability, transient behavior and
convergence.
Practical implications: Due to these excellent features, it is envisaged that the CS can be suitably used as a
MPPT, particularly for large PV system.
Ó 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +60 7 5536187; fax: +60 7 5566272.

E-mail address: [email protected] (Z. Salam).

0306-2619/$ - see front matter Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2013.12.062
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 119

1. Introduction

With the continuous fall in the price of photovoltaic (PV) mod-

ules and the rising concern on the greenhouse gas emissions, solar
energy is rapidly becoming an important power source in the glo-
bal energy scenario. Technologically, PV system is relatively easy to
install, very safe, almost maintenance free and more importantly,
environment friendly. Large PV power systems are being installed
worldwide due to their medium and long term economic pros-
pects. Meanwhile, unused space, such as rooftops of homes, facto-
ries and large buildings can be effectively utilized to harvest solar
energy – as demonstrated by the success of the building integrated
PV (BIPV) initiatives in various countries [1]. Notwithstanding
these advantages, PV power systems, in general, still could not at-
tain the grid-parity due to the high initial investment cost. Despite
the continuous efforts to improve the efficiency of PV cells, the
manufacturing and fabrication processes, the inverter electronics
and others, one should not overlook the potential of enhancing
the system throughput by improving its Maximum Power Point
Tracker (MPPT) capability.
Due to the continuous variation in the environmental condi-
tion – primarily the temperature and solar irradiance, the P–V Fig. 1. The two-diode model of PV cell that is used throughout this paper (a) model
of a single cell (b) with series–parallel combination.
characteristics curve exhibits a non-linear, time-varying maxi-
mum power point (MPP) problem. To ensure that the maximum
power from PV arrays is always achieved, the MPPT is employed Recently, several works on an alternative SC method known as
in conjunction with the power converter (dc–dc converter and/or ‘‘Cuckoo Search’’ (CS) are attracting considerable interests [20].
inverter). To date, numerous MPPT algorithms are reported in Compared to other optimization techniques, CS is proven (in other
the literature; they are broadly classified into two categories, applications) to be more robust, has better convergence and exhib-
namely (1) the conventional and (2) soft computing methods. its higher efficiency [21,22]. Furthermore, it requires fewer tuning
Esram and Chapman [2] and Salam et al. [3] have reviewed var- parameters, which is an advantage when rapid design process is
ious techniques in both categories excellently. For conventional considered. After an exhaustive search, it is determined that the
MPPT, the widely used methods include Perturb and Observe use of CS for MPPT has not been reported in any technical publica-
(P&O) [4], Hill Climbing (HC) [5] and Incremental Conductance tion elsewhere. Hence based on this literature based gap, this work
(InCond) [6]. Besides, there are other simpler methods such as is carried out. The main contribution of this paper is to establish
the fractional short circuit current [7], fractional open circuit the feasibility of CS for MPPT and how it could be implemented
voltage [8], ripple correlation control [9], sliding control [10], in a practical PV system. Furthermore, to justify its viability, a com-
and mathematical-graphical approach [11]. In normal condition, prehensive evaluation is carried out against two well established
i.e. uniform irradiance, they are capable of tracking the MPP MPPT methods P&O and PSO. Several performance tests are carried
quite efficiently and exhibit very good convergence speed. De- out, namely (1) gradual environmental (irradiance and tempera-
spite these advantages, each of these methods inherit very seri- ture) change, (2) sudden step irradiance change, (3) rapid environ-
ous drawback, that is the continuous oscillation that occurs mental change and (4) the ability to handle the partial shading
around the MPP. The oscillatory behavior results in considerable condition. Finally, the paper ends with a discussion and recom-
loss of power during steady state. Numerous works are carried mendations on the prospects of using CS as the MPPT algorithm
out to minimize the oscillation, but it is achieved at the expense for a PV system.
of reduced tracking speed [12]. Additionally, none of these tech-
niques are capable of handling partial shading condition [13], i.e.
2. The two-diode PV model
a condition when a part or the whole module of the PV array
receives non-uniform irradiance.
In order to simulate a PV system, it is crucial to employ a reli-
To alleviate some of these problems, the MPPT techniques
able solar cell model. An accurate model is desirable, but it often
based on soft computing (SC) are proposed. Among them are the
results in computing speed trade-off. An excellent compromise is
Artificial Neural Network (ANN) [14], Fuzzy logic Controller (FLC)
presented by the two diode model [23], shown in Fig. 1(a). This
[15], Genetic algorithm (GA) [16], Differential Evolution (DE)
model requires less computational time and offers improved result
[17], Particle Swarm Optimization (PSO) [18] and Ant Colony Opti-
at low irradiances. For a PV system which comprises of Ns and Np
mization (ACO) [19]. Despite their flexibility, SC algorithms are
modules are connected in series–parallel (S–P) configuration, the
generally more complex and slower than the conventional meth-
equivalent circuit utilizing the two diode model is shown in
ods. For example, ANN requires very specific and prolong training
Fig. 1(b) [24]. The total current from these modules can be written
period to produce accurate results. Furthermore, due to its compu-
as:
tational intensive nature, ANN needs to be implemented using  
2 0 13
expensive microprocessor. On the other hand, FLC shows excellent Vþ Ns
IRs
Np
convergence speed, but its performance is subject to the program- I ¼ IPV Np  Id1 Np 4exp @ A5
mer’s experience and understanding of a specific PV module and a1 V T1 Ns
the environmental conditions in which the system is being in- 2 0   13  
stalled. Other algorithms such as GA and ACO are being used, but
V þ NNps IRs V þ NNps IRs
 Id2 Np 4exp @ A5    ð1Þ
mainly as an optimizer for the conventional MPPT; this approach a2 V T2 Ns Ns
Rp Np
is popularly known as the hybrid MPPT.
120 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

Table 1
The specifications of the PV module used throughout this paper.

Parameters Label Value Parameters Label Value

Short circuit current ISC 7.5 A Maximum power PMPP 110.22 W
Open circuit voltage VOC 20.7 V VOC coef. of temperature KV 0.084 V/°C
Current at Pmax IMPP 6.6 A ISC coef. of temperature KI 3.3e4 A/°C
Voltage at Pmax VMPP 16.7 V Cell in series per module 36

where I is the PV current and V is the PV voltage. The resistance Rs

and Rp are the series and parallel resistance, respectively, while VT is
the thermal voltage of the diodes. The light generated current (IPV)
is given by
G
IPV ¼ ðIPV STC þ K I ðT  T STC ÞÞ ð2Þ
GSTC
Note that IPV_STC is measured in the standard test condition (STC),
i.e. temperature T = 298 K and irradiance G = 1000 w/m2. Variable
KI is the short circuit current coefficient, which is usually provided
by the manufacturer. The diode saturation current is given by
ISC STC þ K I ðT  T STC Þ
Id1 ¼ Id2 ¼ ð3Þ
expððV OC STC þ K V ðT  T STC ÞÞ=V T Þ  1
In Eq. (3), ISC_STC and VOC_STC are the short circuit current and the
open circuit voltage in STC, respectively. Variable KV is the open
circuit voltage coefficient.
The specifications for the PV module used in this paper are
given in Table 1. These are typical for poly crystalline silicon cells.
Using these parameters, and by virtue of Eqs. (2 and 3), the corre-
sponding P–V and I–V curves are plotted in Fig. 2(a) and (b). These
are the static PV curves for various values of G and T. Each curve
presents its unique maximum power point (MPP) location.

3. Basics of P&O and PSO

In this work, the performance of the proposed CS will be evalu- Fig. 2. The I–V and P–V curves of the modules given in Table 1 (a) under varying
ated in comparison to Perturb and Observation (P&O) and Particle solar irradiance and (b) under varying temperature.
Swarm Optimization (PSO) MPPT. To facilitate discussion that shall
follow, brief overviews of both methods are presented here.
positions in the search-space, guided by the neighbor’s best posi-
tions. The general idea of particles movement in PSO is illustrated
3.1. P&O
by Fig. 3.
Based on Fig. 3, the individual particles position is defined by
The conventional P&O is the most widely used MPPT algorithm
[25,26]. Due to its effectiveness, P&O is considered as a standard xkþ1
i ¼ xki þ v kþ1
i ð5Þ
benchmark for any new MPPT algorithm to compare. The algo-
rithm first calculates the power (P) by sensing the voltage and cur-
where vi represent the velocity factor which is calculated by
   
rent. Then it provides a perturbation in the voltage based on the v kþ1
¼ wv ki þ c1 r 1 P besti  xki þ c2 r 2 Gbest  xki ð6Þ
i
change of power by following this basic rule:

V new ¼ V old þ / ðif P > Pold Þ

ð4Þ
V new ¼ V old  / ðif P < Pold Þ
In Eq. (4), / is known as the perturbed voltage. Clearly, the cru-
cial aspect of this algorithm is to determine the size of /. If / is
large, the convergence is fast but that causes large fluctuation in
the steady state power and vice versa. Whatever the case, the oscil-
lation causes undesirable losses of PV power during steady state.

3.2. PSO

PSO is an SC method that has been recently used for MPPT

[18,27]. It iteratively tries to improve a candidate solution (known
as particle) with regard to a given measure of quality. These parti-
cles move around in the search-space according to simple mathe-
matical formulae, exploiting their position and velocity. Each
particle is influenced by its own best position and the best known Fig. 3. Particles movement in PSO.
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 121

In Eq. (6), w is the inertia weight; c1 and c2 are the acceleration 4.3. Cuckoo Search algorithm
constants, while Pbest and Gbest are the personal and global best
positions, respectively. To start the optimization process, a solution Yang and Deb [28] have used three idealized rules for CS based
vector of voltage samples is initialized and the algorithm transmits on cuckoo’s brood parasitic behavior: (1) each cuckoo lays one egg
the calculated duty cycles to the power converter. These voltage at a time and places it in a randomly chosen nest, (2) the best nest
samples (represented by xi in Eq. (5)) serve as the initial particles with the highest quality of eggs will carry over to the next gener-
in the first iteration. Consequently all particles move towards their ation and (3) the number of available nests is fixed and the number
local best position Pbest. Among these particles, one of them is the of eggs (laid by a cuckoo) discovered by the host bird maintains a
global best Gbest. It gives the best fitness value. After calculating the probability Pa, where 0 < Pa < 1. If the cuckoo’s eggs are discovered,
velocity, which serves as a perturbation to the voltage, a new posi- the host bird can abandon its nest or destroy cuckoos’ eggs. Either
tion of the voltage is found. Through successive iteration all parti- way a new nest will be generated with a probability of Pa for a fixed
cles move towards global best position. As the particles approach number of nests. Based on these three rules, the CS algorithm can
the MPP, they get closer to the Gbest position. Correspondingly, be summarized in a pseudo code of Fig. 5.
the Pbest and Gbest factor in velocity term moves towards zero. When generating a new solution x(t+1) for a cuckoo, a Lévy flight
Eventually a zero velocity is achieved and the voltage position re- is performed as dictated by the following expression:
mains almost unchanged. Under this condition, the PV system ðtþ1Þ
reaches at MPP. xi ¼ xti þ a  Le
vyðkÞ ð8Þ
xti
where is samples/eggs, i is the sample number, t is the number of
iteration and a > 0 is the step size. Based on the constraints imposed
4. Cuckoo Search (CS)
by the optimization problem, it is important to tune the value of a
to get desired step size. In most cases a is used as in Eq. (9), i.e.
4.1. Cuckoos’ behavior  
a ¼ a0 xðtÞ ðtÞ
j  xi ð9Þ
CS is an optimization algorithm, inspired by the parasitic repro-
duction strategy of cuckoo birds [28]. It is observed that several where a0 is the initial step change. In Eq. (9), the difference between
species of cuckoos perform brood parasitism, i.e. by laying their two samples is used to determine the subsequent step size. The
eggs in other birds’ (host birds) nests [29]. Usually three types of product  in Eq. (8) indicates entry-wise multiplication i.e. a is mul-
brood parasitism are seen (1) intraspecific, (2) cooperative and tiplied with all samples individually. The value of Lévy (k) is found
(3) nest takeover. Some cuckoo species such as Tapera are intelli- from Lévy distribution given in Eq. (10), i.e.
gent enough to mimic the shape and color of the host bird to in- k
creases its reproduction probability. It is also presented in [29] vyðkÞ  u ¼ l ;
Le ð1 < k < 3Þ ð10Þ
that cuckoos lay their eggs at some specific time so that their eggs
hatch earlier than the host bird’s own. After the early hatching,
cuckoos destroy some of the host bird’s eggs to increase the chance
of their chicks getting more food. It is also a common phenomenon
that the host birds discover the cuckoo’s eggs and destroy these.
Sometimes they abandon their nest completely and go elsewhere
to build a new nest.

4.2. Lévy flight

Searching for a suitable host bird’s nest is an important part of

cuckoo’s reproduction strategy. Normally, the search for the nest is
similar to the search for food, which takes place in a random or in a
quasi-random form. In general, while searching for food, animals
choose directions or trajectories that can be modeled by certain
mathematical functions. One of most common model is the Lévy
flight. A recent study [30] shows that that fruit flies or Drosophila Fig. 4. A Lévy flight in 2-dimensional plane.
melanogaster, explore their landscape using a series of straight
flight paths punctuated by a sudden 90° turn, leading to a Lévy
flight style. Such behavior is adopted in the meta-heuristic search
algorithm for optimization problem. In CS, nest searching steps of
cuckoo is also characterized by Lévy flight. Mathematically, a Lévy
flight is a random walk where step sizes are extracted from Lévy
distribution according to a power law, i.e.

k
y¼l ð7Þ

where l is the flight length and k is the variance. Since l < k < 3, thus
y has an infinite variance. Fig. 4 depicts an example of Lévy flight in
a two dimensional plane. Due to the virtue of Lévy distribution, the
steps consist of many small steps and occasionally large-step, long-
distance jumps. Comparing to other meta-heuristic algorithms,
these long jumps may increase the search efficiency of cuckoo
search significantly in some cases, especial for multimodal, nonlin-
ear problems. Fig. 5. Pseudo code for CS algorithm [20].
122 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

In the context of MPPT algorithm, the structure of CS in Eq. (8) is where b = 1.5, j is the Lévy multiplying coefficient (chosen by the
similar to the HC/P&O method. This similarity, of course, does not designer), while u and v are determined from the normal distribu-
include the step size of the Lévy flight. However, there are several tion curves, i.e.
features that make CS much more robust than HC: (1) CS is a pop-
ulation based algorithm (like GA and PSO), but it exhibits elitism in
   
u  N 0; r2u v N 0; r2v ð13Þ
selection procedure (like harmony search), (2) in CS, the randomi-
zation is much more efficient; due to Lévy flight, the steps some- If C denotes the integral gamma function, then the variable ru
times get bigger, which provide faster convergence, (3) the and rv are defined as
number of parameters for tuning in CS is only two; GA and PSO
when requires three parameters or more and (4) unlike PSO, the 0 11b
performance of CS is not dependent on the samples initialization. Cð1 þ bÞ  sinðp  b=2ÞA
ru ¼ @   b1
and rv ¼ 1 ð14Þ
C 1þb
2
 b  ð2Þð 2 Þ
5. MPPT using CS
The respective power for the new voltage samples are measured
5.1. Governing variables and equations from the PV modules. By comparing the power values, the maxi-
mum power given by the voltage is selected as the new best sam-
To use CS for designing MPPT, appropriate variables have to be ple. Besides this best sample, others are randomly destroyed with a
selected for the search. First are the samples; in this case, they are probability of Pa-such process emulates the behavior of the host
defined as the values of the PV voltages, i.e. Vi (i = 1, 2. . .n). The to- bird discovering the cuckoo’s eggs and then destroying those. Then
tal number of samples is defined as n. Second is the step size, de- new random samples are generated to replace the destroyed ones.
noted by a. The fitness function (J) is the value of PV power at Consequently, the powers for all samples are measured again and
MPP. Since J is dependent on the PV voltage, thus J = f(V). the current best is selected by evaluating J. The iteration continues
Initially, the generated samples are applied to the PV modules until all the samples have reached the MPP.
and the power is set as the initial fitness value. The maximum
power provided by its corresponding voltage is considered as the
current best sample. Thereafter the Lévy flight is performed; conse- 5.2. MPPT searching mechanism
quently new voltage samples are generated based on the following
equation: To search the entire P–V curve, the initial samples need to be
distributed over the whole voltage span. The number of samples
ðtþ1Þ
Vi ¼ V ti þ a  Le
vyðkÞ: ð11Þ (n) is critical. A large n increases the search efficiency (i.e. improv-
ing the chances to converge to a correct value), but requires longer
where a = a0(vbestvi). A simplified scheme of the Lévy distribution
convergence time. In the case of MPPT problem, extensive simula-
is presented in [20] as:
! tions indicate that n = 3 is a good compromise and therefore used
u throughout this work. The CS searching mechanism under a uni-
s ¼ a0 ðv best  v i Þ  Le
vyðkÞ  j  1
ðv best  v i Þ ð12Þ form (normal) irradiance is depicted in Fig. 6(a). Here, three sam-
ðjv jÞb
ples, denoted by variables X (green1), Y (red) and Z (yellow) are
used. The upper subscript on the variable indicates the iteration
number. For the first iteration, Y0 is located nearest to the MPP; it
is therefore considered as the best value. Thus, X0 and Z0 are forced
to move towards the Y0 because the movements of the samples de-
pend on the path difference between the best sample and other sam-
ples. Furthermore, due to the Lévy flight, the step sizes for CS are
comparatively larger than PSO or GA; sometimes it is large enough
for a local sample to cross the best sample. This results in faster con-
vergence and guarantees the searches take place on both side of the
best position. On the other hand, the step size in PSO is based on the
vector summation of the global and local best samples, which always
lies at some point between this two. Nevertheless, it must be noted
that the random nature of CS causes the step sizes to vary in every
iteration. However, as the particle moves closer to MPP, the step size
gets smaller, as dictated by Eq. (12). Finally, after converging at MPP,
the step size reduces to zero. That explains why once convergence at
MPP is achieved; all samples continue to remain at the converged
point.
In the case of partial shading shown in Fig. 6(b), the samples are
initially distributed at various regions of the P–V curve. In the first
iteration, Y0 is in the best position; consequently, X0 and Z0 are
forced to leave their local positions and move towards Y0. However,
in the second iteration, Z2 reaches at a better position than the oth-
ers; therefore other samples move towards it. It is noticeable here
that true MPP lies slightly right to the Z2. Since the Lévy flight al-
lows for the local samples to cross the best sample, X and Y cross
Z2 and reach the MPP as well.

1
Fig. 6. The MPP searching mechanism by CS (a) normal condition and (b) partial For interpretation of color in Fig. 6, the reader is referred to the web version of
shading. this article.
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 123

6. Experimental set-up and implementation sample is generated. This iteration continues until all the samples
converge to the optimum point, i.e. MPP.
6.1. Experimental set-up One important point is worth noting here. When the all the
samples converge to MPP, their mutual distances approach zero;
Fig. 7 presents the set-up that is used to realize the MPPT algo- as a result, the power value of each sample are the same. However,
rithms. The buck-boost converter is selected due the several if a variation in the environmental condition (for example a change
advantages as highlighted in [18]. The converter is designed to in irradiance and temperature or partial shading) occurs, there will
operate in the continuous inductor current mode with the follow- be a change of power for the same value of voltage. When such
ing specifications: switching frequency (f) = 100 kHz, inductor occurrence is detected, the samples are again dispersed over the
(L) = 1 mH and capacitor C1 = C2 = 470 lF. At the input side, the PV curve; the algorithm recalculates new power values, thus
PV voltage and current is measured using voltage and current sen- causing unending ripple on the output power. This may not be
sor, respectively. Using these measured values, the MPPT algorithm desirable because the atmospheric change is a continuous phe-
generates VOUT. Then VREF is subtracted from the VOUT. The differ- nomenon and hence the above-mentioned condition can occur
ence is the error voltage VERROR, which is then fed to a PID control- continuously. To overcome the problem, a threshold value for the
ler. The output of the PID controller is compared to sawtooth normalized power is applied according to the formula below [27].
waveform to produce the duty cycle for the converter. This duty
cycle forces the converter to operate at desired voltage, i.e. at VMPP.
jPðsiþ1 Þ  P si j
> DP ð15Þ
The PID controller is preferable over the PI or P controller, as it Pðsi Þ
reduces the transient overshoot and improves the steady state sta-
where s is the iteration number and P is the value for power re-
bility [31]. However, the inclusion of the derivative term in the
corded for the samples. The variable DP is the normalized power
controller might results an undesirable noise at the input side,
tolerance. Its value is selected as 0.1, by trial and error process.
which to some extent, affects the output voltage. To overcome this
Thus, if the normalized power mismatch is bigger than 0.1, the sam-
problem, a filter is integrated in the derivative section of the PID.
ples will be dispersed on the PV curve, otherwise they remain on
The filter coefficient is set to 35; this value locates the pole location
the MPP.
of the filter in the derivative action. The parameters of PID control-
ler are tuned by trial and error method; the best values are found
as follows: P = 0.08, I = 15 and D = 0.01. 6.3. Initialization of the samples

It is a known characteristic of the PV module that its open cir-

cuit voltage changes with the variation in temperature. This phe-
6.2. Implementation of CS algorithm
nomenon is governed by the following equation:

The overall flow diagram of CS MPPT is shown in Fig. 8. The V oc ¼ V oc STC þ K v DT ð16Þ
algorithm is implemented in MATLAB. First, all constants and vari-
ables, namely the voltage, current, power, number of samples and where Voc_STC is the open circuit voltage at STC and KvDT is the var-
the value of b are initialized. The Initialization of the samples is de- iation of the open circuit voltage due to the temperature effect. For
scribed in detail in Section 6.3. Using the present value of voltage the simulation carried out in this paper, Kv is 0.084 V/°C, which is
and current, the power is calculated. The new value of voltage a typical value for a polycrystalline silicon cell [32]. If we consider
and power are stored in the voltage V ti and fitness J ti arrays, respec- that the temperature increases up to 50 °C (which is normal for a
tively. Furthermore, before the start of every iteration, a check is free air-flow PV system), DT will be (50–25) °C = 25 °C. Thus the
performed to determine if the samples have already achieved con- variation in Voc will be (25  (0.084)) = 2.1 V. Since the open
vergence or otherwise. If the samples have converged to MPP, they circuit voltage at STC is given as 20.7 V (in Table 1), then the varia-
 2:1 
will merge as a same value and so does the respective power. tion is 20:7  100 ¼ 10:14% of the Voc .From this calculation, it can
If the samples do not converge, all the power values of the cor- be concluded that if all the voltage samples are forced to be below
responding samples are measured and are stored in the J ti array. By 90% of the Voc, these will never be out of the range. That is the rea-
evaluating the array, the sample with highest power is chosen as son why the lowest (VL) and highest (VH) voltage level is set to be
the best sample. Thereafter, by virtue of Eq. (12) all other samples 0.1Voc and 0.9Voc, respectively. Using these values, the initial three
are forced to go towards this best value. The step sizes are calcu- samples are generated using
lated by performing the Lévy flight as described by Eqs. (13 and
14). Consequently, a new set of samples are found. Afterwards V ti ¼ V L þ Rand ð0; 1Þ  ðV H  V L Þ ð17Þ
the corresponding powers of these new samples are measured
Note that the exact values of the samples could not be fixed due to
from the PV panel. On the other hand, if any samples results in a
the random function. However they are guaranteed to remain inside
lesser power, then that particular sample is discarded and a new
the range of 0.1–0.9Voc.
There are three possibilities to initialize the samples. In the
case, where certain samples placed on the left and certain on the
right of MPP, it is easy for all the samples to converge to VMPP.
On the other hand, if all the three initial samples are placed to
the left or to the right of the VMPP, it will be very difficult for the
samples to converge. In fact for PSO, the samples could not con-
verge to VMPP because the MPP does not lie in between the global
best and local best position [27]. But in CS, when the local best
positions are moving towards the global best, two possible actions
will take place during the subsequent iteration: (1) the local best
will move closer to the global best and (2) local best will jump
across (i.e. going to the other side of) the global best. The latter
Fig. 7. The buck boost circuit used to realize the MPPT. phenomenon is due to the Lévy flight operation. It allows for the
124 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

Fig. 8. Flowchart for searching mechanism by CS.

algorithm to check if a better position than the current global best the 4th iteration, X4 jumps across Y4 to search for a better position.
exists. The movement of the samples can be verified by examining their
Three simulations runs are carried to illustrate the effect of positions after each iteration in Fig. 9.2(b). Eventually, all the
initial condition placement to the convergence. Searching in samples converge to VMPP within 250 ms. On the other extreme,
Fig. 6 depicts a case whereby the VMPP lies in between the initial three initial samples are generated on the right side of the VMPP,
samples, X0 and Z0. In Fig. 1(a), the X0, Y0 and Z0 are generated at i.e. 340, 408 and 381 V, as shown in Fig. 9.3(a). Initially, 340 V is
305, 250 and 350 V, while VMPP lies at 328 V. The movements of the best sample. After the 5th iteration the local best samples does
the samples are captured against the time waveform as shown in not move towards the global best; instead it crosses global best
Fig. 9.1(b). For clarity, the samples during the initial transients and moves to the other side of the MPP. From there, all the samples
are zoomed as the first inset of Fig. 9.1(b). Initially, Z0 is the best converge to VMPP within 260 ms. From these examples, it can be
sample; thus X0 and Y0 were supposed to move towards the Z0. concluded that the convergence of the samples MPP, to a large
However, in the first iteration Y0 performs a big jump due to extent, is not dependent on their initial positions. Rather, it is more
Lévy flight. It crosses Z0 and reaches 365 V. Similar phenomenon influenced by the random step sizes generated by the Lévy flight.
occurs at the 5th iteration, where X5 makes a jump to cross the best
sample Y5, as shown by the second inset of Fig. 9.1(b). All three
samples converge at VMPP within 220 ms. Fig. 9.2(a) presents the 7. Results
case in which all the samples are on the left side of the MPP. The
initial samples, i.e. X0, Y0 and Z0 are generated at 227, 280 and To evaluate the capability of CS, its performance is compared to
237 V, respectively. The VMPP is at 328 V. Initially, the global best two popular MPPT techniques, i.e. P&O (conventional) and PSO (soft
is Y0; consequently other samples are moving towards it. After computing). All three algorithms are subjected to the following
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 125

Fig. 9.1. (a) Initial position of the samples. (b) Voltage vs. time curve.

Fig. 9.2. (a) Initial position of the samples (b) Voltage vs. time curve.

Fig. 9.3. (a) Initial position of the samples. (b) Voltage vs. time curve.

tests: (1) continuous change in irradiance and temperature, (2) sud- irradiance and temperature requires at least a couple of seconds.
den step change in irradiance in, (3) rapid step changes in both irra- An important feature of CS is at steady state condition, the operat-
diance and temperature and (4) the ability to handle partial shading ing point remains firmly at MPP, i.e. the oscillation is practically
condition zero. This is in contrast to P&O (and other conventional methods),
in which the oscillation around MPP is prevalent.
7.1. Start-up test

This test is designed to determine the speed at start-up (from 7.2. Gradual change in irradiance and temperature
zero point). The PV array is built using the modules described in
Table 1, configured to a single string with 20 modules. The calcu- In a typical sunny day, both the irradiance and temperature
lated MPP at STC is 2205 W. The simulation is carried out for 1 s increase as the hour approaches midday and thereafter decreases
with a sampling rate of 10 ms. The CS parameters are set as the fol- towards the evening. To study the performance of the MPPT algo-
lowing: n = 3, b = 1.5 and Pa = 0.25. The Lévy multiplying coefficient rithms under these gradual changes, such environmental scenario
(j) is chosen to be 0.05. The resulting response of voltage, current is emulated over a period of 5.5 s. Fig. 11 presents the irradiance
and power is shown in Fig. 10(a–c), respectively. As can be ob- and temperature profile, in which the irradiance and temperature
served, the MPP is tracked within 150 ms. This is fast enough to are increased or decreased (in ramp) within one second rise or fall
comply with the real atmospheric variation, as, the changes in time, respectively.
126 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

Fig. 10. MPP tracking by CS under STC (a) voltage, (b) current and (c) power.

Fig. 11. Temperature and irradiance profile.

Fig. 12. Response of CS and P&O under gradual change in atmosphere.

The MPP tracking by CS and P&O when subjected to this profile

is illustrated in Fig. 12. A fixed step size of 5 V is chosen for P&O;
this value is selected to optimize the tradeoff between convergence 7.3. Step changes in irradiance with constant temperature
speed and steady state oscillation. For CS, the control parameters
remain the same as described in Section 7.1. Initially both CS and The step change in irradiance arises when a cloud passes over a
P&O require approximately 130 ms to reach the MPP. Once the PV array at a very fast rate. To evaluate the performance of the
steady state is reached, as expected, the P&O oscillates around algorithm under such condition, a set of irradiance steps are im-
the MPP. On the other hand, CS remains at MPP with almost zero posed on the array, as shown in Table 2. The step change is applied
power oscillation. These observations are clearly shown in inset at every one second interval. For this test, the modules in Table 1
image no. 2 (of Fig. 12). Moreover, during the gradual change in are connected in the S–P configuration of 100  100. The array is
irradiance and temperature (shown by the ramp of the profiles), deliberately made to be large to show the effect of steady state
P&O could not track the power ramp as perfectly as CS does, as de- oscillation by P&O. Throughout the experiment, the temperature
picted by the inset image no. 1. As a result, during these transitions, is maintained constant at 25 °C.
P&O loses power too. The step response for CS and P&O is shown in Fig. 14. To avoid
Fig. 13 compares the performance of CS and PSO. For PSO, vari- large oscillation around the MPP, a fixed step size of 5 V is selected
ables c1 and c2 is chosen as 0.2, while w is 0.4. These values are cho- for the P&O. This value is very conservative, considering the PV ar-
sen after extensive simulation trials and thus the PSO can be ray string voltage is 1670 V. Furthermore, it can be seen that even
regarded as well optimized. The random numbers are generated with a small step-size, P&O oscillate (around the MPP) with an
by MATLAB rand function. The result indicates that initially PSO re- average ripple of 200 W. On the other hand, as CS continuously
quires a much longer time, i.e. approximately 400 ms to settle at sticks to the MPP, the loss is almost zero. The comparison between
the first MPP. After convergence, both algorithm track the MPP per- CS and PSO is shown in Fig. 15. For the initial MPP tracking, PSO re-
fectly and remain almost ripple-free at steady state. However, dur- quires 600 ms. In the case of re-tracking (after each step change in
ing the power ramp, PSO sustains a prolong oscillation compared irradiance), PSO requires approximately 120 ms to settle to a new
to CS. This is due to the smaller step sizes, which forces the PSO MPP value, compared to CS which needs 100 ms. The most proba-
to utilize more samples in order to converge to MPP. This observa- ble reasons for the faster response of CS is the larger step size due
tion is consistent with discussions presented in Section 5.2. to the Lévy flight. Furthermore, PSO exhibits longer fluctuations in
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 127

irradiance and temperature. Such drastic change in irradiance is

quite a common occurrence-particularly in tropical countries dur-
ing pre and post monsoon season, where dark cloud rapidly covers
the sun. In case of temperature, rapid changes are unlikely; how-
ever, the step fluctuation in temperature is incorporated in order
to test the ability of the algorithm to perform under extreme envi-
ronmental conditions. The step change is set to occur every one
second, with the sequences shown in Table 3. The same module
as in Table 1 is used; in this case, 30 of them are connected in series
as a single string. As before, the sampling rate is 10 ms.
Fig. 16 shows the performance of P&O and CS. At the edge of the
step function (i.e. when the atmospheric change occurrence is de-
tected), CS disperses its samples all over the P–V curve and the
Fig. 13. Response of CS and PSO under gradual change in atmosphere.
search begins. Since the temperature step change is also included,
the power fluctuation is more prominent. Accordingly, the CS sam-
Table 2 ples are dispersed to larger area of the voltage span. That disper-
Simulated conditions for step change in irradiance. sion creates significant fluctuations for CS in transient state as
presented in the inset image. However, it has to be noted that
Time (s) 0 1 2 3 4 5
the power fluctuations only extend within a few hundreds of mil-
G (W/m2) 1000 950 850 700 775 900
lisecond from the edges of the steps. These transient behaviors
PMPP (MW) 1.104 (STC) 1.0456 0.9274 0.7485 0.8381 0.9866
may not significantly affect the overall yield of the PV system, com-
pared to the continuous oscillations of the P&O at steady state.
Fig. 17 compares the performance of CS and PSO. As can be seen,
both algorithms track the MPP perfectly in the steady state. How-
ever, in the transient performance of CS is better than the PSO. The
reasons for this have been explained earlier.

7.5. Handling partial shading

The MPP tracking becomes more complicated when the entire

PV array does not receive uniform irradiance-a condition known
as partial shading. Typically, it is caused by the clouds that strike
on certain spots of the array, while other parts are left uniformly
irradiated [33,34]. This problem is more prominent in array with
long string. Another source of partial shading-like characteristics
is exhibited by module irregularities; a common example would
be the presence of cracks on one or more modules of the array.
Fig. 14. Response of CS and P&O under step change of irradiance.
During partial shading, due to the action of the bypass diode, the
P–V curves are characterized by multiple peaks, i.e. several local
and one global peak. The conventional method (such as P&O, HC
and IC) – which relies on the identification of the P–V curve’s

Table 3
Simulated conditions for step change in irradiance and temperature.

Time (s) 0 1 2 3 4
T (°C) 25 35 30 20 15
G (W/m2) 1000 1400 1200 800 600
PMPP (W) 2205 2900 2587 1769 1295

Fig. 15. Response of CS and PSO under step change of irradiance.

the transient state because the algorithm has to perform more

checks while converging to MPP. However, in steady state both
methods track the MPP perfectly.

7.4. Step change in irradiance and temperature

This simulation is designed to determine the response of the

MPPT algorithms due to the simultaneous step changes in Fig. 16. Response of CS and P&O under rapid change of irradiance and temperature.
128 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

Fig. 18. P–V curve for MPP tracking under partial shading (a) P&O and (b) CS.

Fig. 17. Response of CS and PSO under rapid change of irradiance and temperature.

gradient, could not discriminate between the two. In most cases,

the algorithm is trapped at one of the local peak, with the conse-
quence of significant power losses.
CS is suitable to handle partial shading due to its natural ten-
dency to search for global peak. To illustrate such ability, a partial
shading scenario is set-up in simulation. Four PV modules, labeled
as A–D are connected in series. All the modules specifications are
the same as in Table 1. Initially each module receives a uniform
irradiance of 1000 W/m2. Consequently, there exists only one
Fig. 19. MPP tracking by CS and P&O under partial shading.
MPP at 440.9 W, as shown by Curve 1 of Fig. 18(a). After a lapse
of 1 s A–D are irradiated with 1000 W/m2, 800 W/m2, 500 W/m2
and 300 W/m2 respectively. As a result, multiple peaks are gener-
ated in the P–V curve, as shown by Curve 2 of the same figure. The
corresponding peaks (for the irradiances) occur at 115.0 W,
218.7 W, 280.1 W and 237.5 W. Since the modules are subjected
to the partial shading (after one second), the operating point has
to shift from Curve 1 to Curve 2. If P&O algorithm is utilized, the
operating will climb to the nearest peak, i.e. 237.5 as directed by
the arrows in Fig. 18 (a). Clearly, this peak is local, and conse-
quently power is lost. In the case of CS, when partial shading is de-
tected, new samples are dispersed over the entire voltage span of
Curve 2. The algorithm will begin the search for the global peak
using the procedures described in Section 6.2. As voltage samples
are located in different sections of the P–V curve, the algorithm
will not be trapped in any of the local peaks. After successive iter-
ations, the global MPP is tracked, as illustrated in Fig. 18(b). Fig. 20. MPP tracking by CS and PSO under partial shading.
The performance of P&O and CS under partial shading condition
can be summarized by observing output power timing diagram, 8. Discussion
shown in Fig. 19. The simulation is carried out for 2 s with a fixed
sampling rate of 10 ms. As it can be observed, at the point of partial 8.1. Tracking
shading occurrence (at 1 s), the P&O quickly gets trapped at the
local peak, i.e. 237.5 W. This result is consistent with the observa- For convenience, the tracking capability is summarized in
tion of the P–V curve in Fig. 18(a). On the other hand CS success- Table 4. Although the numbers are not absolute, i.e. given in
fully tracks the global MPP at 280.1 W, after going through the ranges, they can be considered as reasonable indicators on the
search for 300 ms (approximately after 30 sampling cycles). While relative performance of these methods. The variations arise due
CS is in the process of searching for the MPP, large fluctuations in to several factors such uncertain values of random numbers, the
power can be observed. This phenomenon is due to the random power level, location of the samples etc. For the initial power rise
nature of the algorithm and the trajectories of the Lévy flight. How- from zero to the steady state, CS appears to be the fastest; the
ever, in practice, this transient behavior is of little concern as its speed varies between 100 and 250 ms. The rapid convergence
duration is very insignificant compared to the steady state condi- can be attributed to the larger step sizes due to the Lévy flight.
tion. Once the steady state is reached, the CS algorithm continues The speed of P&O is on the same range as CS. This is to be expected
to maintain the MPP with almost zero fluctuation. because the algorithm only needs to incrementally climb from zero
Fig. 20 compares the performance CS and PSO subjected to same to MPP with a fixed step size. For faster convergence, the step size
partial shading patterns described above. As expected, both algo- can be increased, but at the expense of steady state oscillation.
rithms-being based on search mechanism, are able to handle the Alternatively, the P&O can be made more efficient if its adaptive
partial shading conditions well. However, CS is comparatively fas- version is implemented [35]. The PSO is the slowest among these
ter; after the occurrence of partial shading, CS tracks the MPP with- three; it requires 300–600 ms to reach steady state. The reason
in 300 ms, while PSO requires 520 ms. Besides, PSO presents bigger for this is the smaller step size as a result of the vector summation
fluctuations in transient state. between its global and local best particle.
 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130 129

Table 4
Summary of MPPT performance based on test carried out in Section 7.

MPPT Initial start-up Re-tracking speed after step Transient power Steady state oscillation No of tuning Ability to handle
algorithm speed (ms) atmospheric change (ms) fluctuation (% of power) para-meters partial shading
P&O 120–300 40–60 Low 0.2 1 No
PSO 300–600 100–500 High 0.0005 3 Yes
CS 100–250 50–300 Moderate 0.000008 2 Yes

For re-tracking of MPP (i.e. tracking to a new MPP after a step only in the algorithm. It is expected that the algorithm can be
change in irradiance and/or temperature, and P&O is the fastest) implemented easily using the standard (16 bit) modern microcon-
it takes in average, only about 50 ms to settle to a new MPP. This troller/microprocessor. The chosen sampling time, i.e. 10 ms, for
is because P&O only depends on the gradient of the PV curve; thus each MPP control updating is also realistic- considering the avail-
climbing to the direction towards MPP is much easier. On the other ability of a fast and low cost analogue to digital converter to
hand, both CS and PSO are search-based techniques. For every large sample the voltage and current. Furthermore, the values of the
change in irradiance or temperature, the algorithms will disperse circuit components (capacitor, inductor and switches) used for
their particles all over the voltage span; only then search can be simulation are carefully selected to ensure that these would be
initiated. Furthermore, the tracking formula is highly sensitive to compatible with the real hardware scenario. Based on these
the generated random numbers. As far as the transient perfor- observations, the next step of the work is to perform the hardware
mance is concerned, it is interesting to note that despite being validation of the proposed CS MPPT algorithm. It is envisaged that
faster, CS exhibits a lower power fluctuation compared to PSO. this work will attract considerable interest of PV community, both
researchers and practitioners.
8.2. Steady state oscillations
Acknowledgments
The main advantage of CS and PSO is absence of MPP oscilla-
tions which makes the energy loss at steady state almost zero.
The authors would like to thank Universiti Teknologi Malaysia
For P&O, the continuous oscillation causes considerable power loss,
and the Ministry of Higher Education Malaysia for providing the
particularly for large PV plants. For example, if the plant in Sec-
facilities and financial support (Research University Grant No.
tion 7.3 is considered, an exposure to the sunlight for an average
2423.00G40) to conduct this research.
of 8 h per day will result in an approximate energy loss of
584 kW h per year, i.e. 0.018% of the plant capacity. Although CS
and PSO exhibit power fluctuation during transient, this behavior References
may not significantly affect the overall yield of the PV system.
[1] Benemann J, Chehab O, Schaar-Gabriel E. Building-integrated PV modules. Sol
Energy Mater Sol Cells 2001;67:345–54.
8.3. Algorithm complexity [2] Esram T, Chapman PL. Comparison of photovoltaic array maximum power
point tracking techniques. Energy Convers, IEEE Trans 2007;22:439–49.
[3] Salam Z, Ahmed J, Merugu BS. The application of soft computing methods for
The P&O only requires one tuning parameter, namely the step MPPT of PV system: a technological and status review. Appl Energy
size. The algorithm is straightforward and easy to implement. For 2013;107:135–48.
PSO, the parameters that need to be tuned are three, i.e. w, c1 [4] Femia N, Petrone G, Spagnuolo G, Vitelli M. Optimization of perturb and
observe maximum power point tracking method. Power Electron, IEEE Trans
and c2. Through experience, it has been found that extensive trial
2005;20:963–73.
and error process is required to find the best combination for an [5] Koutroulis E, Kalaitzakis K, Voulgaris NC. Development of a microcontroller-
optimized performance. On the other hand, for CS only two param- based, photovoltaic maximum power point tracking control system. Power
Electron, IEEE Trans 2001;16:46–54.
eters need to be tuned. The programming effort is comparatively
[6] Lin C-H, Huang C-H, Du Y-C, Chen J-L. Maximum photovoltaic power tracking
easier. for the PV array using the fractional-order incremental conductance method.
Appl Energy 2011;88:4840–7.
[7] Masoum MA, Dehbonei H, Fuchs EF. Theoretical and experimental analyses of
8.4. Partial shadings photovoltaic systems with voltage and current-based maximum power-point
tracking. Energy Convers, IEEE Trans 2002;17:514–22.
Clearly, P&O is totally incapable of handling the partial shading [8] Ahmad J. A fractional open circuit voltage based maximum power point tracker
for photovoltaic arrays. In: Software technology and engineering (ICSTE), 2010
condition. This can be crucial in many situations, particularly for 2nd international conference on: IEEE; 2010. p. V1-247–V1-50.
building integrated PV system in densely populated areas. Both [9] Esram T, Kimball JW, Krein PT, Chapman PL, Midya P. Dynamic maximum
PSO and CS handle partial shading well. However, considering power point tracking of photovoltaic arrays using ripple correlation control.
Power Electron, IEEE Trans 2006;21:1282–91.
other aspects such as tracking capability, transient power fluctua-
[10] Kim I-S. Sliding mode controller for the single-phase grid-connected
tion and speed of convergence, CS can be considered a better photovoltaic system. Appl Energy 2006;83:1101–15.
option. [11] Papaioannou IT, Purvins A. Mathematical and graphical approach for
maximum power point modelling. Appl Energy 2012;91:59–66.
[12] Al-Amoudi A, Zhang L. Optimal control of a grid-connected PV system for
9. Conclusion maximum power point tracking and unity power factor. In: Power electronics
and variable speed drives, 1998 seventh international conference on (Conf
Publ No. 456)1998. p. 80–5.
In this paper, a novel MPPT method based on Cuckoo Search [13] Di Piazza MC, Vitale G. Photovoltaic field emulation including dynamic and
algorithm is proposed. The performance of CS is benchmarked with partial shadow conditions. Appl Energy 2010;87:814–23.
[14] Veerachary M, Yadaiah N. ANN based peak power tracking for PV supplied DC
two widely used MPPT algorithm P&O and PSO. The results demon-
motors. Sol Energy 2000;69:343–50.
strate that CS performs better than P&O and PSO in terms of con- [15] Kottas TL, Boutalis YS, Karlis AD. New maximum power point tracker for PV
vergence speed, transient fluctuations and steady state arrays using fuzzy controller in close cooperation with fuzzy cognitive
performance. The CS is also shown to be capable of tracking global networks. Energy Convers, IEEE Trans 2006;21:793–803.
[16] Messai A, Mellit A, Guessoum A, Kalogirou S. Maximum power point tracking
MPP under partial shading. Due to the simplicity of the CS it re- using a GA optimized fuzzy logic controller and its FPGA implementation. Sol
quires only 8 multiplications, 6 additions and tuning 2 parameters Energy 2011;85:265–77.
130 J. Ahmed, Z. Salam / Applied Energy 119 (2014) 118–130

[17] Taheri H, Salam Z, Ishaque K. A novel maximum power point tracking control [26] Fermia N, Granozio D, Petrone G, Vitelli M. Predictive & adaptive MPPT perturb
of photovoltaic system under partial and rapidly fluctuating shadow and observe method. Aerospace Electron Syst, IEEE Trans 2007;43:934–50.
conditions using differential evolution. Ind Electron Appl (ISIEA), IEEE Sym [27] Miyatake M, Veerachary M, Toriumi F, Fujii N, Ko H. Maximum power point
IEEE 2010:7. tracking of multiple photovoltaic arrays: a PSO approach. Aerospace Electron
[18] Ishaque K, Salam Z, Amjad M, Mekhilef S. An improved particle swarm Syst, IEEE Trans 2011;47:367–80.
optimization (PSO)–based MPPT for PV with reduced steady-state oscillation. [28] Yang X-S, Deb S. Cuckoo search via Lévy flights. Nature & Biologically Inspired
Power Electron, IEEE Trans 2012;27:3627–38. Computing. In: NaBIC 2009 World Congress on, IEEE, 2009. p. 210–4.
[19] Jiang LL, Maskell DL, Patra JC. A novel ant colony optimization-based [29] Teuschl Y, Taborsky B, Taborsky M. How do cuckoos find their hosts? the role
maximum power point tracking for photovoltaic systems under partially of habitat imprinting. Anim Behav 1998;56:1425–33.
shaded conditions. Energy Buildings 2012. [30] Reynolds AM, Frye MA. Free-flight odor tracking in Drosophila is consistent
[20] Yang X-S, Deb S. Multiobjective cuckoo search for design optimization. with an optimal intermittent scale-free search. PLoS ONE 2007;2:e354.
Comput Oper Res 2011. [31] Vaigundamoorthi M, Ramesh R. Experimental investigation of chaos in input
[21] Yang X-S, Deb S. Engineering optimisation by Cuckoo Search. Int J Math Model regulated solar PV powered Cuk converter. Int J Comput Appl 2012;43:11–6.
Numer Op 2010;1:330–43. [32] Ishaque K, Salam Z. An improved modeling method to determine the model
[22] Civicioglu P, Besdok E. A conceptual comparison of the Cuckoo-Search, particle parameters of photovoltaic (PV) modules using differential evolution (DE). Sol
swarm optimization, differential evolution and artificial bee colony Energy 2011;85:2349–59.
algorithms. Artif Int Rev 2011:1–32. [33] Ishaque K, Salam Z, Shamsudin A, Amjad M. A direct control based maximum
[23] Ishaque K, Salam Z, Taheri H. Simple, fast and accurate two-diode model for power point tracking method for photovoltaic system under partial shading
photovoltaic modules. Sol Energy Mater Sol Cells 2011;95:586–94. conditions using particle swarm optimization algorithm. Appl Energy
[24] Ishaque K, Salam Z, Taheri H, Syafaruddin. Modeling and simulation of 2012;99:414–22.
photovoltaic (PV) system during partial shading based on a two-diode model. [34] Sánchez Reinoso CR, Milone DH, Buitrago RH. Simulation of photovoltaic
Simul Model Pract Theory 2011;19:1613–26. centrals with dynamic shading. Appl Energy 2013;103:278–89.
[25] Chee Wei T, Green TC, Hernandez-Aramburo CA. Analysis of perturb and [35] Piegari L, Rizzo R. Adaptive perturb and observe algorithm for photovoltaic
observe maximum power point tracking algorithm for photovoltaic maximum power point tracking. Renew Power Gen, IET 2010;4:317–28.
applications. In: Power and energy conference, 2008 PECon 2008 IEEE 2nd
international, 2008. p. 237–42.