Sie sind auf Seite 1von 24

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

A Modified Firefly Algorithm for Photovoltaic


Maximum Power Point Tracking Control
Under Partial Shading
D. F. Teshome, C. H. Lee, Y. W. Lin, and K. L. Lian Senior Member, IEEE

Abstract

Photovoltaic (PV) modules subjected to partial shading conditions (PSC) can drastically decrease their power
output. Hence, there have been various maximum power point tracking (MPPT) control algorithms developed to
reduce or counteract the shading effects. Recently, a new meta-heuristic algorithm known as firefly algorithm (FA)
was developed, which, under PSC, has been shown to successfully track the GMP. Nevertheless, the FA still has
some inherent problems, which may hinder the performance of the MPPT. This paper modifies the existing FA to
counteract these problems. As will be demonstrated in the paper, the proposed modified FA (MFA) method can
reduce the number of computation operations and the time for converging to the GMP that the existing FA requires.
Experimental results show that the proposed method can track the global point under various PSC, has a faster
convergence time, compared to the FA, and can effectively suppress the power and voltage fluctuations.

Index Terms

Maximum power point tracking (MPPT), photovoltaic (PV) array, partial shading, global optimization, firefly
algorithm

I. I NTRODUCTION

Partial shading on a photovoltaic (PV) string comprising multiple modules or substrings is known as a serious
problem that significantly decreases the energy utilization. Under partially shading conditions (PSC), the shaded
cells in a module become reverse biased and behave as a load, leading to the hot spot problem. To avoid this,
bypass diodes are used to conduct the current generated by the non-shaded cells within a module. However, the
connection of bypass diodes will change the uniform current-voltage (I-V) and power-voltage (P-V) characteristics
of the module, resulting in multiple peaks [1]. To maximize the efficiency of the module, it is necessary to track
the global maximum point (GMP).

D. F. Teshome, C. H. Lee, Y. W. Lin, and K. L. Lian are with Power and Energy Group of National Taiwan University of Science and
Technology, Taipei, Taiwan, 106, R.O.C. (e-mails: d10207804@mail.ntust.edu.tw; t40143031103@gmail.com; m10307213@mail.ntust.edu.tw ;
kllian@mouse.ee.ntust.edu.tw ).
This work was financially supported by the Ministry of Science and Technology under contract No.: 103-2221-E-011-099.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

The conventional MPPT methods such as perturb-and-observe (P&O) [2], [3], hill-climbing (HC) [4] and incre-
mental conductance (INC) [5] essentially rely on determining the gradient of the power with respect to the current,
voltage or duty cycle using the perturbation method for tracking movement [6]. Consequently, the main drawback
of these methods is that they cannot differentiate between a local maximum point (LMP) and the GMP, and the
attained point may be a LMP, leading to power losses. Note that in general, for a utility scale PV system, although
the site will be carefully selected to ensure good solar irradiance and to avoid partial shading (PS) caused by other
buildings or surrounding obstacles, the PS caused by clouds and the mutual shading occurred between adjacent PV
blocks are inevitable in practice [7], [8]. For residential PV installations, PS caused by buildings, trees, clouds, etc.
are very common.
Two approaches are generally used to reduce or counteract the shading effect. The first is based on hardware
fixtures such as dynamic reconfiguration of PV modules according to the shading patterns [9], [10]; distributed
architecture where each module has its own controller [11], [12], [13], [14]; multilevel converter system [15],
allowing each PV source to be controlled separately; and power electronic equalizers [16], [17] which ensure that
all the substrings operate at the same voltage under PSC. This approach is complex and costly [18]. The second
approach is to track the GMP by developing advanced control algorithms (ACA), and this will be the focus of this
paper.
The developed ACA in general can be classified into linear search (LS), artificial intelligence methods (AIM), and
meta-heuristic algorithms (MHAs). Many LS methods are essentially improvements of the conventional maximum
power point tracking (MPPT) methods. For example, Koutroulis and Blaabjerg [18] have proposed a two-stage
method. Firstly, a scanning process is utilized to detect the regions that contain the GMP. Then, in the second stage,
a P&O algorithm is used to find the GMP. However, the disadvantage of this method is that the searching step
(SS) greatly affects the probability of finding the GMP. When the SS is large, the system requires comparatively
less time to locate the GMP but may miss the GMP. On the other hand, a smaller SS can increase the probability
of locating the GMP but requires more time. Patel and Agarwal [19] have proposed a similar approach. First, 85%
of the open circuit voltage is used as the starting point to find and record a peak value. To check if there are
any subsequent peaks, the algorithm steps forward and conduct P&O to search and record the next peak value. If
the obtained value is lower than the previous peak, then the previous peak is the GMP; otherwise the algorithm
continues to move forward. Similar to [18], the problems with this method is that the size of the forward step
greatly affects the probability of tracking the GMP. In addition, as pointed out in [20] and [21], there are cases in
which this method fails to track the GMP. Boztepe et al. [22] have improved [18] by employing a power operating
triangle (POT) and voltage window to restrict the voltage range to be scanned. Nevertheless, the voltage step for
scanning still needs to be chosen with care to avoid missing the GMP. Other LS methods such as the DIRECT
(dividing rectangles) algorithm, proposed by Nguyen and Lehman [23], is based on a Lipschitz condition for finding
the maximum point. To ensure that the GMP is found, the search areas to be divided must also be selected wisely.
Lei et al. [24] proposed a sequential extremum seeking control (ESC) strategy for the GMP tracking. However, this
process requires the variation bound for the turning-point voltage to be found. This in turn requires knowledge of

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

the P-V or P-I characteristics under variable-shading circumstances, which may be difficult to obtain in practice.
In general, in order for LS methods to successfully track the GMP, one needs to choose the parameter carefully or
has a basic knowledge of the P-V or P-I characteristics under the PSC.
Researchers have also used AIM such as neural networks (NN) and fuzzy logic controllers (FLC) to develop
MPPT methods. In [25], a radial basis function and a three-layered feedforward NN are used to track the GMP.
However, the accuracy of this scheme depends on the volume of training data, and considerable computational effort
is needed to ensure reliability and accuracy under any shading condition. In [4], a FLC-based MPPT algorithm
with 16 rules defined in the logic rule table is presented. However, the FLC rule table is largely dependent on the
designers experience and a prior knowledge of how the specific PV system performs [21]. Hence, it is difficult to
generalize the system design, which is the major limitation of the FLC. For a more comprehensive review of FLC,
NN, and LS, [20], [21] are some of the good references.
Finally, a MPPT problem under PSC can be recast as an optimization problem without defined objective functions.
For these problems, MHAs are able to allocate the GMP by exploiting randomization to avoid being trapped in a
LMP and thereby enable the search globally. Chen et al. [26] demonstrated a MPPT method based on biological
swarm chasing behavior to increase the tracking performance. Nevertheless, only uniform insolation conditions were
considered and PSCs were not addressed. Miyatake et al. [27] have realized centralized MPPT control of modular
PV systems, and used the particle swarm optimization (PSO) to determine the individual module voltage. They
showed that the PSO can outperform conventional methods such as HC under PSCs. Ishaque et al. [28] and Liu et
al. [29] implemented PSO-based MPPT control algorithms in a PV system, consisting of a high-power single-stage
converter. However, as reported in [30], one problem of the PSO algorithm is the long GMP tracking time for large
search spaces. In [30], the authors proposed removing the random number from the PSO acceleration factors to
reduce the tracking time. Nevertheless, the value of the maximum change in particle velocity must be restricted,
otherwise LMPs may be obtained. Moreover, since the random numbers are removed, the algorithm lose the main
characteristic of a MHA and cannot be warranted to track the GMP in all PSC. Lian et al. [6] proposed a hybrid
method that combines PSO and P&O. Initially, the P&O is employed to identify the nearest LMP. Starting from
that point, the PSO method is used to search for the GMP. The advantage of this hybrid method is that the search
space of the PSO is reduced. Hence, the time required for convergence can be greatly decreased. Seyedmahmoudian
et al. [31] have proposed combining differential evolution (DE) and PSO. The DE and PSO are employed in an
alternative fashion where the PSO is applied every odd iteration, and the DE is run in every even iteration.
Researchers have also started to look for other MHAs for MPPT. Sundareswaran et al. [32] have proposed using
artificial bee colony (ABC) algorithm for MPPT tracking under PSC, and have shown that ABC performs better
than PSO in terms of convergence time. Nevertheless, as stated in [32], the number of bees plays a critical role in
deciding convergence to GMP. If the number is low, the algorithm may get trapped in a LMP. In [33], Jiang et al.
have proposed using the ant colony optimization method for MPPT. The algorithm is tested against both uniform
and shading pattern conditions. However, the performance is almost identical to that of PSO. In 2007, a new MHA
known as firefly algorithm (FA) was developed by Yang [34]. Sundareswaran et al. [35] implemented the FA for

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

MPPT under PSC and shown that it has better tracking performance than PSO and P&O. Nevertheless, the FA still
has some inherent problems, which may hinder the performance of the MPPT. Firstly, the amount of computation
required for each iteration is high. Secondly, the wandering motion of a fly can be excessive if the number of other
brighter flies is high. This may cause the tracking time for GMP to be excessively high. This paper improves the
existing FA to counteract the two above mentioned problems. As will be demonstrated in the paper, the proposed
modified FA (MFA) method can reduce the number of computation operations and the time for converging to GMP
that the existing FA requires. The rest of this paper is arranged as follows. The FA will be briefly reviewed in
Section II. Section III then describes the proposed MFA method. The experimental setup and a comparison of the
FA and the MFA are described in Section IV. Finally, the conclusions are presented in Section V.

II. F IREFLY A LGORITHM

In this paper, a voltage-based controller will be discussed. The PV voltage as the regulated variable is easier to
implement because the MPPT controller can quickly decide the initial points according to the percentage of the
open circuit voltage (Voc ) [36]. Duty-cycle as the regulated variable can also be done in a similar fashion.
The FA method proposed by Yang [34] has three fundamental assumptions. Firstly, all fireflies are unisex and
will move towards the brighter and more attractive ones until all of them have been compared (except for itself).
Secondly, the attractiveness of a firefly is related to its brightness, which depends on the distance between itself
and other flies. However, because of the light absorption of the air, the attractiveness decreases as the distance
increases. Finally, the brightness or light intensity of a firefly is determined by the value of the objective function
of a given problem. Mathematically, the FA algorithm can be expressed by three equations. The attractiveness,
can be quantitatively stated in (1).
(r) = 0 exp(rm ), m 1, (1)

where 0 is the initial attractiveness at r = 0, r is the distance between two fireflies, and is an absorption
coefficient controlling the reduction of the light intensity. m is an integer, and is set to 2. Equation (2) evaluates
the distance between two fireflies i and j, at positions xi and xj , respectively, and can be defined as Euclidean
distance. v
u d
uX
rij = kxi xj k = t (xi,k xj,k )2 (2)
k=1

where xi,k and xj,k are the k th component of the spatial coordinates of the ith and j th firefly, and d is the number
of the dimensions.
The movement of a firefly is determined by (3).
2 1
xi = xi + 0 erij (xj xi ) + (randn ), (3)
2
where [0, 1] and randn is a random perturbation value. Note that (3) clearly indicates that the movement of a
firefly is affected by attractiveness of a brighter fly and randomization. The randomization provides a good way to

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

move away from local search to the search on global scale. In general, a small value of tends to facilitate local
search while a large one promotes a global search [35].
The purpose of the MPPT block is to obtain Vref , which is sent to the PI controller. Therefore, the position (xi )
variable in (3) is actually the voltage references (Vref ) whereas the second and third terms of (3) can be regarded
as the correction terms for the voltage references. TABLE shows how the terminologies in FA match those of PV
systems.

TABLE I: Terminologies of FA algorithm and PV system

FA algorithm PV system
Firefly position Voltage reference (Vref )
Distance Voltage difference (Vref )
Attractiveness An exponential function of Vref
Brightness Power (Ppv )
Brightness of the brightest firefly Global maximum power (PGbest )

Since the converter can only respond to one command at a time, the flies are initialized and evaluated in a
successive manner. Fig. 1 summarizes the control action of the FA used for MPPT. Firstly, Vref is initialized, and
the number of flies is set to N, where N is a positive integer. Moreover, the global maximum power, PGbest is
initialized by ranking all the initialized flies according to their power levels. Note that the power is calculated by
multiplying the measured voltage (Vpv ) and current (Ipv ). Then, the program proceeds to a nested for-loop. In the
inner loop, the intensity of fly i, (i.e. the power output Ppv,i is compared against that of fly j (i.e. Ppv,j ), and note
that j 6= i. If Ppv,j > Ppv,i , the voltage reference is updated by (3); otherwise, it is set to a random value. This
nested loop is executed until all the flies have been compared against each other. Note that for the case of MPPT,
d = 1. Hence, rij in (3) should be as stated in (4).
q
rij = (xi xj )2 . (4)

Then, PGbest is updated by ranking the fireflies. Finally, the convergence criterion as defined in (5) is checked
to ensure that the GMP is reached.

|PGbest Ppv,i | < 1 , i = 1, . . . , N, (5)

where 1 is a tolerance value. Note that generally the tolerance values in the stopping criteria of the meta-heuristic
algorithms are iteratively optimized via multiple runs as reported in [35], [37], [38], and [39].

III. M ODIFIED F IREFLY A LGORITHM

The main problem of the original FA is that the position of each firefly is changed in a stepwise manner towards
the brighter fireflies. This is due to the fact that all the flies need to compare with each other, and each comparison
comes with a movement, which is governed by (3). This can be observed from Fig. 2. There are four fireflies

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

Begin

Initialize Vref and set the


number of fireflies, N

Obtain PGbest

Set i=1, and j=1

No Rank all the No


Is |PGbest-Ppv,i| 1 ? Is i N ?
i=1...N flies to
update PGbest
Yes
Yes
Calculate the power
(Ppv,i) for the firefly i i=i+1
by multiplying Vpv,i
and Ipv,i

No
Is j (N-1) ?

Yes
Calculate the power
(Ppv,j) for the firefly j
(j i)by multiplying
Vpv,j and Ipv,j
j=j+1
Evaluate (1) and (4) to
obtain the attractiveness
and distance

end No Set Vref


Is Ppv,j >Ppv,i ?
randomly

Yes

Update Vref
by (3)

Fig. 1: Flow chart for the Firefly algorithm

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

Initial position
for fly 1 The trajectory movement of
fly 1

Final position
for fly 1

fly 3
fly 4

fly 2

Fig. 2: Trajectory movements of firefly 1 in the original FA

in the space. Assume that flies 2, 3, and 4 are brighter than fly 1. Since it is difficult to describe the lightness
quantitatively, we use hue gradation to indicate the brightless level of the flies. Hence, fly 4 is the brightest, fly 3
is brighter than 2, and fly 1 has no coloration. As shown in the figure, fly 1 changes its position towards flies 2, 3
and 4, respectively, and the coloration (i.e. the brightness) of fly 1 also changes as its position changes. The zigzag
trajectories observed in Fig. 2 may cause the tracking time of the GMP to be excessively long.

Initial
position for
fly 1

fly 3
fly 4
xjavg

Final
position for
fly 2 fly 1

Fig. 3: Trajectory movements of firefly 1 in the proposed MFA

To overcome this problem, this paper proposes using the average of the coordinates of all the brighter fireflies
as the representative point, and the firefly will only move towards this point without wandering towards all the
brighter flies. Hence, the proposed modified FA (MFA) can drastically reduce the tracking time of the GMP.
This is illustrated in Fig. 3. The lightness levels of the four flies and the original position of fly 1 are the same
as those in Fig. 2. However, it only takes one step for fly 1 to move to the final position. The final position of fly
1 is the average of flies 2 to 4. Consequently, (3) is revised as follows:

2 1
xi = xi + 0 erijavg (xjavg xi ) + (randn ), (6)
2

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

where
q
rijavg = (xi xjavg )2 , (7)

and xjavg is the average coordinates of the brighter fireflies, and


L
1 X
xjavg = xj . (8)
L m=1
Note that L is the number of the brighter flies.
Fig. 4 summarizes the flow chart of the proposed method. Similar to the original FA, after initialization, the
program proceeds to a nested for-loop. The intensity of fly i, (i.e. the power output Ppv,i and that of a fly j (i.e.
Ppv,j ) are calculated. In the inner loop, Ppv,j is compared with Ppv,i to find out which flies are brighter than fly
i, and the number of these flies, which is L. Then, xjavg is calculated, and substituted into (6) and (7). Hence,
instead of updating xi with respect to each brighter fly, xi is updated with respect to the average of all the brighter
flies. This can drastically reduce the amount of computation.

IV. E XPERIMENTS AND VALIDATIONS

A. Configuration of the Experimental Setup

Fig. 5 shows the experimental setup of the PV system used to validate the proposed method. The PV array
in the figure is realized by a programmable PV emulator manufactured by AMETEK (Model ETS600X8C-PVF).
This PV emulator is able to mimic the output of a photovoltaic installation exposed to various shading conditions.
Moreover, in terms of performance evaluation of the two MPPT control algorithms, a PV emulator is a good choice
because it can ensure that the two algorithms are being tested in exact same conditions. The DC-DC converter is a
boost converter with an interleaved topology to reduce ripple currents, improve reliability, and increase efficiency
[40]. The controller is implemented in a 32-bit digital signal processor (DSP-TMS320F28035), which sends out
gating signals, P W M1 and P W M2 , to the gate drivers to control the MOSFET switches (SW1 and SW2 ) in
a complementary fashion. Vpv and Ipv are sent to the DSP via sensor circuits and A/D converters. The MPPT
controller takes Vpv and Ipv and determines the voltage reference, which is then sent to the PI controller. The PI
controllers proportional gain is 0.2, and its integral gain is 0.1. The load is a resistive load whose voltage is
regulated by a dc voltage source (Chroma programmable DC power supply 620120P-600-8). Thus, the load voltage
is held at 450V. Table II summarizes the parameter values of the converter circuit, and Fig. 6 shows the actual
setup. Depending on the PV system application, a dc/dc power converter is used to interface the PV array output
power to either a battery bank or a dc/ac inverter connected to the grid. Both of these types can be represented by
Fig. 5.
The parameter values of 0 and for both the FA and the proposed MFA are both set to 1 and 0.96, respectively.
Note that the parameters are iteratively optimized via multiple runs so as to reach GMP in each PSC.
To validate the proposed method and thoroughly compare it with the original FA, two types of case studies
static and dynamic are investigated. In the static case studies, V-I curves representative of different partial shadings,

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

Begin

Initialize Vref and set the


number of fireflies, N

Obtain PGbest

Set i=1 and j=1

Set L=0
No
No i=i+1
Rank all the
Is |PGbest-Ppv,i| 1 ? flies to Is i N ?
i=1...N update PGbest Update Vref by (6)
Yes
Yes Calculate the power
(Ppv,i) for the firefly i by Evaluate (7)
multiplying Vpv,i and Ipv,i

No Set xjavg
No Is By (8)
Is j (N-1) ?
L>0 ?

Yes
Yes
Calculate the power (Ppv,j)
for the firefly j (j i)by
multiplying Vpv,j and Ipv,j

end No
j=j+1 Is Ppv,j > Ppv,i ?

Yes

L=L+1

Fig. 4: Flow chart for the Proposed Modified Firefly algorithm

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

10

L1 D1
Ipv Rs

+ +
L2 D2
PV Vpv Cpv Cdc Vdc
Array T1 T2

- -

T1 T2

Gate driver Gate driver

PWM2 PWM3

DSP
TMS320F28035
Voltage
sensor A/D D/A
VPv,Ipv & Oscilloscope
Current converter converter
sensor

Fig. 5: Schematic diagram of the PV system.

AMETEK PV Emulator
(ETS600X8C-PVF)
Power Supply for Oscilloscope
DSP Control Board

DSP
Chroma Control
62012P- Board DC-DC
600-8 Converter
Loads

Fig. 6: Experimental setup of the PV system

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

11

TABLE II: Parameter Values of the Converter Circuit

Max. Input voltage 218 V


Max. Output voltage 450 V
Input current 0.6 5 A
Output current 0.6 2.8 A
Switching frequency 100 kHz
D1 , D2 IQBD30E60A1
MOSFET switches SPW35N60C3
Load 300
L1 , L2 2 mH
Cpv , Cdc 120 F, 470 F

shown in Fig. 7, are imported to the PV emulator to study their tracking times and MPPT capabilities. For the
dynamic cases, sudden and gradual changes of these P-V curves (and their variants) presented in Fig. 7 with respect
to time are investigated to compare the dynamic tracking performance under PSC.

B. Static Case Studies

To ensure reliable operation, the interval between each flies, Tint cannot be smaller than the response time of
the PV emulator, which is 4 ms. Hence, Tint is set to 100 ms.
1) Case 1: There are five LMPs, and the GMP occurs at the leftmost on the P-V curve, as shown in Fig. 7 (a),
and the MPPT controller needs to bypass four LMPs before reaching the GMP. Fig. 8 shows the voltage (upper
plot) and power (lower plot) waveforms when the FA is employed for the MPPT. Note that prior to the activation
of the controller, the dc voltage source at the load side has already been turned on, keeping the load voltage at
450 V, as shown in the figure. As can be seen from the voltage plot, Vpv closely tracks the voltage reference,
Vref , when the control algorithm is activated at t = 1.22 s 1 . This demonstrates that the PI controllers function
properly. However, it takes 2.5 s (i.e. 25 iterations or steps) for the FA to reach the GMP. The first five iterations
are simply outputting the initial values of the five flies. Then, each fly takes turns to move its position according
to the brightness of the other flies as presented in Fig. 1. Each fly moves in four steps, and GMP is reached at the
end of the fourth fly movement.
On the other hand, Fig. 9 shows the voltage (upper plot) and power (lower plot) waveforms when the proposed
method is employed. The control algorithm is activated at t = 1.28 s. It only takes 1.3 s (i.e. 13 iterations) to
reach the GMP. Similar to the FA, the first five steps are to output the initial values of the flies, and the rest of the
iterations are to update the movement of each fly. Since the proposed method uses the average of the coordinates of
the brighter flies as the representative point, each fly only needs to move one step to update its position.Therefore,
the proposed method drastically reduces the convergence time.

1 Note that prior to this is the soft start control for minimizing overcurrent.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

12

Ppv(W) Ipv(A)

277
2.52



110
110 Vpv(V) Vpv(V)
Ppv(W) Ipv(A)

303

2.07







146 Vpv(V) 146 Vpv(V)
Ppv(W) Ipv(A)

408


2.34




174 174
Vpv(V) Vpv(V)

Fig. 7: (a) Case 1: There are five LMPs, and the GMP occurs at the leftmost on the P-V curve ; (b) Case 2: There
are five LMPs, and the GMP occurs at the middle one on the P-V curve; (c) Case 3: There are five LMPs, and the
GMP occurs at the rightmost on the P-V curve.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

13

250
Initialization Stage #1 #2 #3 #4 #5

200
Voltage (V)

150
Controller
activated
100 at 1.22 sec
GMP is
reached Vpv
50
Vref

0
600 600

500 500

400 400
DC Bus(V)

Power(W)
300
277

200 Ppv 200


Pref = 277W
Vdc
100 100
Pref

0 0
0 1 2 3 4 5 6 7
Time (Sec)

Fig. 8: Voltage and power waveforms of the FA for Case 1

2) Case 2: There are five LMPs, and in this case, the GMP occurs at the middle one on the P-V curve as shown
in Fig. 7 (b). The GMP is 303 W. Fig. 10 shows the voltage and power waveforms when the FA is employed. As
seen from the figure, it takes 21 steps to reach the GMP. Fig. 11 shows the voltage and power waveforms of the
proposed method. It takes only 13 steps (i.e. 1.3 s) to reach the GMP.
For the waveforms of cases 3 (the GMP occurs at the rightmost on the P-V curve), they follow the similar trend
of the previous cases. Nevertheless, since both FA and MFA are both meta-heuristic methods, the convergence time
may be slightly different in each run. Therefore, we have run the above three cases for multiple runs and record
their best, worst and average convergence for reaching GMP in TABLE III. The amount of saving time in each
case is also shown in the table. For the best case, the MFA can save up to 67%, whereas it can save up to 24%for
the worst case. Moreover, the MPPT efficiency (s ) for static cases, is defined as (9) of the two algorithms.
GM Pact
s = 100%, (9)
GM Pth
where GM Pth is the theoretical maximum power that can be achieved, whereas GM Pact is the actual power that
is extracted using the MPPT algorithm. On average, the efficiencies are 99.30% and 98.8% for FA and MFA,
respectively.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

14

250
Initialization Stage
#2 #3 #4
#5
200 #1
Voltage (V)

150
Controller
activated
100 at 1.28 sec
GMP is
reached Vpv
50
Vref

0
600 600

500 500

400 400
DC Bus(V)

Power(W)
300
277

200 Ppv 200


Pref=277W Vdc
100 Pref 100

0 0
0 1 2 3 4 5 6
Time (Sec)

Fig. 9: Voltage and power waveforms of the proposed method for Case 1

TABLE III: Convergence Comparison Between FA and MFA

Case FA MFA Time Saved by MFA


1 2.6 s 1.4 s 46%
worst 2 2.5 s 1.9 s 24%
3 2.6 s 1.9 s 27%
1 2.3 s 1.2 s 48%
average 2 2.3 s 1.7 s 26%
3 2.5 s 1.5 s 40%
1 2.1 s 0.7 s 67%
best 2 2.0 s 1.2 s 40%
3 2.1 s 1.3 s 38%

3) Discussions: Although the static tracking efficiency for MFA is slightly lower than that of FA, the tracking
speed is almost twice (on average about 1.7 times faster) of that of FA. Thus, MFA can keep track of the GMP even
if the irradiance changes rapidly. For example, for the PV emulator we used, the tracking time for FA on average
is about 2.4 s for all three partial shading cases whereas that of MFA is 1.4 s. Therefore, for PV characteristics

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

15

250
Initialization Stage #1 #2 #3 #4 #5

200
Voltage (V)

150
Controller
activated
at 1.17 sec
100
Vpv

50 GMP is
Vref
reached

0
600 600

500 500

400 400
DC Bus(V)

Power(W)
300 303

Ppv
200 Pref = 303W 200
Vdc
100 Pref 100

0 0
0 1 2 3 4 5 6 7
Time (Sec)

Fig. 10: Voltage and power waveforms of the FA for Case 2

which change faster than 2 s, MFA can still keep track of the GMP while FA cannot. In addition, if one wants to
attain the efficiency higher than FA, methods such as the P &O can be utilized after MFA has converged to the
GMP. Fig. 12 shows that after 1.5 s, where MFA has converged, the algorithm then switches to P &O, and the
efficiency can be improved to 99.4%2 . The total tracking time is 1.87 s, which is still shorter than that of FA. Note
that other MPPT algorithms having similar efficiency as P &O, can also be used for improving the efficiency of
MFA. The trade-off between tracking time and efficiency should be carefully investigated for optimal performance.
Alternatively, one could also run P &O and other similar algorithms most of the time, and activate MFA only when
there is a significant change in the measured power. Nevertheless, such a topic is beyond the scope of this paper
and is left for future study.
Finally, it is worth noting that although the occurrence of the partial shading caused by the clouds usually do
not happen suddenly, the P &O method is unable to track the GMP, regardless of a ramp or step changes of PSCs,
as shown in [30], [35]. Therefore, it is important to test the performance of the proposed MFA for dynamically
tracking the GMP under PSC, which will be investigated in the next subsection.

2 As reported in [41], the tracking efficiency of P &O method under no partial shading is above 99.4%, and in some cases, it is over 99.5%.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

16

250
Initialization Stage
#2 #3 #4
#1 #5
200
Voltage (V)

150

Controller activated GMP is


100 at 1.30 sec reached
V
pv

50 Vref

0
600 600

500 500

400 400
DC Bus(V)

Power(W)
300 303

200 Ppv 200


Pref=303W
V
dc
100 100
Pref

0 0
0 1 2 3 4 5 6
Time (Sec)

Fig. 11: Voltage and power waveforms of the proposed method for Case 2

C. Dynamic Case Studies

Two scenarios are investigated for the dynamic case. The first one is to investigate the tracking performance
when the irradiance and shadings change in a way that the shapes of the P-V curves are significantly alternated.
The maximum point of the P-V pattern changes from 432 W (uniform P-V characteristic) to 277 W (case 1 in
Section IV B) within 1.3 sec. This is about 36% change of power level, a case to emulate insolation and partial
shading conditions are suddenly altered.
The second scenario is to investigate the continuous dynamic tracking capability. Similar to scenario 1, the starting
power level is 432 W, and the ending power level is 277 W. The changing sequences of the shaded patterns are,
however, as shown in Fig. 13. The shaded patterns and irradiance levels are changed every 5 s. This changing time
is selected based on the following reasons:
1) For the PV emulator we have used, the tracking time for MFA and FA under the worst cases are 1.9 s and
2.6 s, respectively. Hence, for dynamic tracking cases, the PSC change cannot be lower than 3 s.
2) Based on the European test standard EN50530 [42], a converter with MPPT capabilities need to be tested
under several changing irradiance with different time changes and ramp rates. For instance, with ramp change
from 30% to 100%, the time changes are 0.7 seconds (10 uniform P-V profile changes with a ramp time of

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

17

250
Initialization Stage #1 #2 #3 #4 #5

200
Voltage (V)

150 Controller activated


at 1.25 sec MFA
converged GMP is
100 reached
V
pv
50
V
ref

0
600 600

500 500

400 408
DC Bus(V)

Power(W)
300 300
Pref = 408W
200 P 200
pv
V
dc
100 100
P
ref
0 0
0 0.5 1 1.5 2 2.5 3 3.5
Time (Sec)

Fig. 12: Voltage and power waveforms of the MFA combined with P &O

7 seconds), 1.4 seconds, 2.3 seconds, 3.5 seconds, 5 seconds and 7 seconds.
However, the European test standard EN50530 only sets the standard for uniform PV changes (no bypass diodes),
and does not set the test standard for non-uniform PV changes (with bypass diodes). The PV changes set by EN50530
can be easily tracked by the P &O method because all the PV profiles have uniform PV characteristics. To test the
capabilities of the original and the proposed firefly algorithms for tracking the GMP even when the PV characteristics
are non-uniform, similar to [6], [29], [30], and [35], we set several sets of PSCs with different multiple peaks (with
GMP occurring in the left, middle, and right), having a ramp rate lower than 15%. The PV profiles in Fig. 13 of the
manuscript are more challenging and to some extent more practical (since all the profiles are gradually changed)
to test MPPT control algorithms.
Similar to [6], [29], and [31], (10) is used to detect if the GMP has been changed:
|Ppv Ppv,last |
P (%), (10)
Ppv,last
where Ppv,last represents the power at the GMP of the previous operating point, and P is set to 5%. In general,
(10) is needed for all the maximum point tracking algorithms, including the P &O method [43]. If it is satisfied,
the MPPT algorithms are reinitialized to search for the new maximum point.
Note that P in (10) needs to be carefully selected such that it is able to distinguish the fluctuations of the
output power caused by the random noises (or the voltage/current perturbations required in the MPPT algorithms)

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

18

Ppv(W) Ppv(W) Ppv(W)


432.49W
422.38W 407.62W

172V 172.46V 174.6V


Vpv(V) Vpv(V) Vpv(V)

Ppv(W) Ppv(W)
363.78W 387.82W

181.48V 177.97V Vpv(V)


Vpv(V)

Ppv(W)
Ppv(W) Ppv(W)
343.77W
302.97W 277

164.76V
Vpv(V) 146.45V Vpv(V) 110 Vpv(V)

Fig. 13: Partial shading change sequence.

from those ones caused by the irradiance variations [43]. Femia et al. [43] has shown how a theoretical optimized
value of P can be obtained for P &O method. Nevertheless, deriving a theoretical optimized value of P for a
FA or other meta-heuristic algorithms are beyond the scope of this paper. In practice, the value of P is iteratively
optimized by extensive simulations.
Figs. 14 and 15 show the resulted tracking responses for FA and MFA for scenario 1. First, both of the algorithms
try to track the GMP of the uniform P-V characteristic. Similar to the non-uniform cases presented in Section IV
B, the FA takes longer time to track the GMP. Then, the uniform P-V curve changes to the P-V curve of case 1,
and the GMP is suddenly changed to 277 W. According to [42], the tracking efficiency for the dynamic cases (d )
is calculated based on the following formula.

P
Vpv,i Ipv,i Ti
d = Pi 100%, (11)
j GM P th,j Tj

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

19

where Tj is period in which the GM P th,j is provided and Ti is period in which the Vpv,i and Ipv,i are sampled.
Hence, the tracking efficiency for FA and MFA are 99.12% and 99.84%, respectively. In terms of loss, the
proposed MFA (0.16%) is about 5 times smaller than that of FA (0.88%). This is due to the fact that MFAs
tracking time is much shorter than that of FA, which in turn can have a positive impact on the dynamic tracking
efficiency.

250

200
Voltage (V)

150

100
Vpv
50
Vref

0
600 600
432W
500 500

400 277W 400


DC Bus(V)

Power(W)
300 300

200 Ppv 200


Pref
100 100
Vdc
0 0
0 2 4 6 8 10 12 14
Time (Sec)

Fig. 14: (a) Top: voltage of the FA for dynamic variations of partial shading; (b) Bottom: power waveforms of the
FA for dynamic variations of partial shading

Fig. 16 and 17 show the tracking profiles of scenario 2 for the FA and MFA, respectively. As can be seen from
the figures, both methods can successfully track the GMP. However, the MFA exhibits much less power and voltage
fluctuations, as compared to the FA. The MFA takes less time to settle to the GMP whenever a power change
is detected. This case study clearly demonstrates that the MFA can effectively track the GMP with much shorter
convergence time and smoother profile.

V. C ONCLUSIONS

This paper described a new MPPT method that can be used to track the GMP when a PV panel is partially
shaded by cloud, snow, trees, and/or other buildings. The proposed approach essentially modifies the existing firefly
algorithm, and uses the average of all the brighter fireflies as the representative point so that the firefly will only
move towards this point without wandering towards all the brighter flies. Hence, the proposed method can drastically

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

20

250

200
Voltage (V)

150

100
Vpv

50 Vref

0
600 600
432W
500 500

400 277W 400


DC Bus(V)

Power(W)
300 300

200 Ppv 200

Pref
100 100
Vdc
0 0
0 2 4 6 8 10 12 14
Time (Sec)

Fig. 15: (a) Top: voltage of the MFA for dynamic variations of partial shading; (b) Bottom: power waveforms of
the MFA for dynamic variations of partial shading

decrease the tracking time and reduce the number of operations per iteration. Experimental results show that the
proposed MFA can save up to 67% of the tracking time, when compared to FA. Due to its short tracking time, the
dynamic efficiency of MFA in general is better than that of FA when fast irradiance change occurs.

VI. ACKNOWLEDGEMENT

The authors would like to sincerely thank the associated editor, anonymous reviewers, and Prof. Chung-Ming
Young from NTUST for their valuable comments and suggestions, which improved the quality of the paper.

R EFERENCES

[1] E. Romero-Cadaval, G. Spagnuolo, L. Garcia Franquelo, C. Ramos-Paja, T. Suntio, and W. Xiao, Grid-connected photovoltaic generation
plants: Components and operation, IEEE Industrial Electronics Magazine, vol. 7, no. 3, pp. 620, Sept. 2013.
[2] D. C. Jones and R. W. Erickson, Probabilistic analysis of a generalized perturb and observe algorithm featuring robust operation in the
presence of power curve traps, IEEE Transactions on Power Electronics, vol. 28, no. 6, pp. 29122926, June 2013.
[3] R. C. N. Pilawa-Podgurski, W. Li, I. Celanovic, and D. J. Perreault, Integrated cmos energy harvesting converter with digital maximum
power point tracking for a portable thermophotovoltaic power generator, IEEE Journal of Emerging and Selected Topics in Power
Electronics, vol. 3, no. 4, pp. 10211035, Dec 2015.
[4] B. Alajmi, K. Ahmed, S. Finney, and B. Williams, Fuzzy-logic-control approach of a modified hill-climbing method for maximum power
point in microgrid standalone photovoltaic system, IEEE Transactions on Power Electronics, vol. 26, no. 4, pp. 10221030, April 2011.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

21

250

200

150
Voltage (V)

100

V
pv
50
V
ref

0
600 600

432W 422W
500 408W 364W 500
388W
344W
400 303W 400
277W
DC Bus(V)

Power(W)
300 300

200 200
P
pv
P
100 ref 100
V
dc

0 0
5 10 15 20 25 30 35 40 45
Time (Sec)

Fig. 16: (a) Top: voltage of the FA for dynamic variations of partial shading (power gradually decreasing); (b)
Bottom:power waveforms of the FA for dynamic variations of partial shading (power gradually decreasing)

[5] G. C. Hsieh, H. I. Hsieh, C. Y. Tsai, and C. H. Wang, Photovoltaic power-increment-aided incremental-conductance mppt with two-phased
tracking, IEEE Transactions on Power Electronics, vol. 28, no. 6, pp. 28952911, June 2013.
[6] K. L. Lian, J. H. Jhang, and I. S. Tian, A maximum power point tracking method based on perturb-and-observe combined with particle
swarm optimization, IEEE Journal of Photovoltaics, vol. 4, no. 2, pp. 626633, 2014.
[7] T. Kerekes, E. Koutroulis, D. Sra, R. Teodorescu, and M. Katsanevakis, An optimization method for designing large pv plants, IEEE
Journal of Photovoltaics, vol. 3, no. 2, pp. 814822, April 2013.
[8] W. B. F. Corporation, Utility scale solar power plants: a guide for developers and investors. India: World Bank, 2012.
[9] G. Velasco-Quesada, F. Guinjoan-Gispert, R. Pique-Lopez, M. Roman-Lumbreras, and A. Conesa-Roca, Electrical pv array reconfiguration
strategy for energy extraction improvement in grid-connected pv systems, IEEE Transactions on Industrial Electronics, vol. 56, no. 11,
pp. 43194331, Nov 2009.
[10] D. Nguyen and B. Lehman, An adaptive solar photovoltaic array using model-based reconfiguration algorithm, IEEE Transactions on
Industrial Electronics, vol. 55, no. 7, pp. 26442654, July 2008.
[11] N. Femia, G. Lisi, G. Petrone, G. Spagnuolo, and M. Vitelli, Distributed maximum power point tracking of photovoltaic arrays: Novel
approach and system analysis, IEEE Transactions on Industrial Electronics, vol. 55, no. 7, pp. 26102621, July 2008.
[12] G. Petrone, G. Spagnuolo, and M. Vitelli, An analog technique for distributed mppt pv applications, IEEE Transactions on Industrial
Electronics, vol. 59, no. 12, pp. 47134722, Dec 2012.
[13] F. Wang, X. Wu, F. C. Lee, Z. Wang, P. Kong, and F. Zhuo, Analysis of unified output mppt control in subpanel pv converter system,
IEEE Transactions on Power Electronics, vol. 29, no. 3, pp. 12751284, March 2014.
[14] R. Bell and R. C. N. Pilawa-Podgurski, Decoupled and distributed maximum power point tracking of series-connected photovoltaic

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

22

250

200
Voltage (V)

150

100
V
pv
50
V
ref
0
600 600
432W 422W
500 408W 388W 364W 500
344W
400 303W 400
277W
DC Bus(V)

Power(W)
300 300
P
200 pv 200
P
ref
100 100
V
dc
0 0
5 10 15 20 25 30 35 40 45
Time (Sec)

Fig. 17: (a) Top: voltage of the MFA for dynamic variations of partial shading (power gradually decreasing); (b)
Bottom: power waveforms of the MFA for dynamic variations of partial shading (power gradually decreasing)

submodules using differential power processing, IEEE Journal of Emerging and Selected Topics in Power Electronics, vol. 3, no. 4, pp.
881891, Dec 2015.
[15] I. Abdalla, J. Corda, and L. Zhang, Multilevel dc-link inverter and control algorithm to overcome the pv partial shading, IEEE Transactions
on Power Electronics, vol. 28, no. 1, pp. 1418, 2013.
[16] L. Villa, T.-P. Ho, J.-C. Crebier, and B. Raison, A power electronics equalizer application for partially shaded photovoltaic modules,
IEEE Transactions on Industrial Electronics, vol. 60, no. 3, pp. 11791190, March 2013.
[17] M. Uno and A. Kukita, Single-switch voltage equalizer using multistacked buck-boost converters for partially shaded photovoltaic
modules, IEEE Transactions on Power Electronics, vol. 30, no. 6, pp. 30913105, June 2015.
[18] E. Koutroulis and F. Blaabjerg, A new technique for tracking the global maximum power point of pv arrays operating under partial-shading
conditions, IEEE Journal of Photovoltaics, vol. 2, no. 2, pp. 184 190, April 2012.
[19] H. Patel and V. Agarwal, Maximum power point tracking scheme for pv systems operating under partially shaded conditions, IEEE
Transactions on Industrial Electronics, vol. 55, no. 4, pp. 1689 1698, April 2008.
[20] Y.-H. Liu, J.-H. Chen, and J.-W. Huang, A review of maximum power point tracking techniques for use in partially shaded conditions,
Renewable and Sustainable Energy Reviews, vol. 41, pp. 436 453, 2015.
[21] K. Ishaque and Z. Salam, A review of maximum power point tracking techniques of pv system for uniform insolation and partial shading
condition, Renewable and Sustainable Energy Reviews, vol. 19, pp. 475 488, 2013.
[22] M. Boztepe, F. Guinjoan, G. Velasco-Quesada, S. Silvestre, A. Chouder, and E. Karatepe, Global mppt scheme for photovoltaic string
inverters based on restricted voltage window search algorithm, IEEE Transactions on Industrial Electronics, vol. 61, no. 7, pp. 3302
3312, 2014.
[23] D. Nguyen and B. Lehman, An adaptive solar photovoltaic array using model-based reconfiguration algorithm, IEEE Transactions on
Industrial Electronics, vol. 55, no. 7, pp. 2644 2654, July 2008.
[24] P. Lei, Y. Li, and J. Seem, Sequential esc-based global mppt control for photovoltaic array with variable shading, IEEE Transactions on
Sustainable Energy, vol. 2, no. 3, pp. 348 358, July 2011.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

23

[25] Syafaruddin, E. Karatepe, and T. Hiyama, Artificial neural network-polar coordinated fuzzy controller based maximum power point
tracking control under partially shaded conditions, IET Renewable Power Generation, vol. 3, no. 2, pp. 239 253, june 2009.
[26] L.-R. Chen, C.-H. Tsai, Y.-L. Lin, and Y.-S. Lai, A biological swarm chasing algorithm for tracking the pv maximum power point, IEEE
Transactions on Energy Conversion, vol. 25, no. 2, pp. 484 493, June 2010.
[27] M. Miyatake, M. Veerachary, F. Toriumi, N. Fujii, and H. Ko, Maximum power point tracking of multiple photovoltaic arrays: A pso
approach, IEEE Transactions on Aerospace and Electronic Systems, vol. 47, no. 1, pp. 367 380, Jan. 2011.
[28] K. Ishaque, Z. Salam, M. Amjad, and S. Mekhilef, An improved particle swarm optimization (pso) based mppt for pv with reduced
steady-state oscillation, IEEE Transactions on Power Electronics, vol. 27, no. 8, pp. 3627 3638, aug. 2012.
[29] Y.-H. Liu, S.-C. Huang, J.-W. Huang, and W.-C. Liang, A particle swarm optimization-based maximum power point tracking algorithm
for pv systems operating under partially shaded conditions, IEEE Transactions on Energy Conversion, vol. 27, no. 4, pp. 1027 1035,
dec. 2012.
[30] K. Ishaque and Z. Salam, A deterministic particle swarm optimization maximum power point tracker for photovoltaic system under partial
shading condition, IEEE Transactions on Industrial Electronics, vol. 60, no. 8, pp. 31953206, 2013.
[31] M. Seyedmahmoudian, R. Rahmani, S. Mekhilef, A. M. T. Oo, A. Stojcevski, T. K. Soon, and A. S. Ghandhari, Simulation and hardware
implementation of new maximum power point tracking technique for partially shaded pv system using hybrid depso method, IEEE
Transactions on Sustainable Energy, vol. 6, no. 3, pp. 850 1698, July 2015.
[32] K. Sundareswaran, P. Sankar, P. S. R. Nayak, S. P. Simon, and S. Palani, Ehanced energy output from a pv system under partial shaded
conditions through artificial bee colony, IEEE Transactions on Sustainable Energy, vol. 6, no. 1, pp. 198 209, July 2015.
[33] L. L. Jiang, D. L. Maskell, and J. C. Patra, A novel ant colony optimization-based maximum power point tracking for photovoltaic
systems under partial shaded condtions, Energy Buildings, vol. 58, no. 1, pp. 227236, March 2013.
[34] X. S. Yang, Nature-Inspired Metaheuristic Algorithms. UK: Luniver Press, 2008.
[35] K. Sundareswaran, S. Peddapati, and S. Palani, Mppt of pv systems under partial shaded conditions through a colony of flashing fireflies,
IEEE Transactions on Energy Conversion, vol. 29, no. 2, pp. 463472, June 2014.
[36] W. Xiao, N. Ozog, and W. Dunford, Topology study of photovoltaic interface for maximum power point tracking, IEEE Transactions
on Industrial Electronics, vol. 54, no. 3, pp. 1696 1704, june 2007.
[37] X.-S. Yang, S. Deb, M. Loomes, and M. Karamanoglu, A framework for self-tuning optimization algorithm, Neural Computing and
Applications, vol. 23, no. 7, pp. 20512057, 2013.
[38] A. Oi, C. Nakazawa, T. Matsui, H. Fujiwara, K. Matsumoto, and H. Nishida, Pid optimal tuning method by particle swarm optimization,
in SICE Annual Conference, 2008, Aug 2008, pp. 34703473.
[39] B. Akay and D. Karaboga, Computational Collective Intelligence. Semantic Web, Social Networks and Multiagent Systems: First
International Conference, ICCCI 2009, Wrocaw, Poland, October 5-7, 2009. Proceedings. Berlin, Heidelberg: Springer Berlin Heidelberg,
2009, ch. Parameter Tuning for the Artificial Bee Colony Algorithm, pp. 608619.
[40] P. W. Lee, Y. S. Lee, D. K. W. Cheng, and X. C. Liu, Steady-state analysis of an interleaved boost converter with coupled inductors,
IEEE Transactions on Industrial Electronics, vol. 47, no. 4, pp. 787 795, Aug. 2000.
[41] D. S. Morales, Maximum power point tracking algorithms for photovoltaic applications, Masters thesis, Aalto University, Dec. 2010.
[42] E. C. for Electrotechnical Standarization, Overall efficiency of grid connected photovoltaic inverters. Brussels: BSI, 2010.
[43] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, Optimization of perturb and observe maximum power point tracking method, IEEE
Transactions on Power Electronics, vol. 20, no. 4, pp. 963973, July 2005.

D. F. Teshome received the B.Sc. degree in electrical engineering from Jimma University, Jimma, Ethiopia in
2007, the M.Sc. degree in energy engineering and management from University of Lisbon, Lisbon, Portugal and the
M.Sc. degree in renewable energy from Universitat Politecnica de Catalunya (UPC), Barcelona, Spain in 2013. From
2007 to 2011, he was working as a Project Engineer in Ethio-Telecom at Addis Ababa, Ethiopia. He is currently
pursuing the Ph.D. degree in electrical engineering at National Taiwan University of Science and Technology, Taipei,

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2016.2581858, IEEE
Journal of Emerging and Selected Topics in Power Electronics

24

Taiwan.

C. H. Lee is currently pursuing his M. S. degree from the National Taiwan University of Science and Technology,
Taipei, Taiwan.

Y. W. Lin obtained his B. S. and M. S. degrees from the National Taiwan University of Science and Technology,
Taipei, Taiwan in 2014 and 2016, respectively. He is currently with Delta Products Corp., Chungli, Taiwan.

K. L. Lian received the B.A.Sc. (Hons.), M.A.Sc, and Ph.D. degrees in electrical engineering from the University
of Toronto, Toronto, ON, Canada, in 2001, 2003, and 2007, respectively. He was a Visiting Research Scientist at
the Central Research Institute of Electric Power Industry, Japan, from 2007 to 2009. Currently, he is an Associate
Professor at the National Taiwan University of Science and Technology, Taipei, Taiwan.

2168-6777 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.