Hindawi Publishing Corporation

ISRN Renewable Energy

Volume 2014, Article ID 673840, 13 pages
http://dx.doi.org/10.1155/2014/673840

Research Article
Efficient Maximum Power Point Tracking Algorithm for
PV Application under Rapid Changing Weather Condition

Khaled M. Bataineh and Amr Hamzeh

Department of Mechanical Engineering, Jordan University of Science and Technology, Irbid, Jordan

Correspondence should be addressed to Khaled M. Bataineh; [email protected]

Received 10 October 2013; Accepted 28 October 2013; Published 27 March 2014

Academic Editors: A. Bosio and S. Dai

Copyright © 2014 K. M. Bataineh and A. Hamzeh. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

This study presents a novel search algorithm of maximum power point tracking for photovoltaic power generation systems. The I-V
characteristics and the P-V power output under specific irradiation and temperature conditions are simulated. The performance
of the algorithm under fully shaded and sudden partially shaded conditions as well as variable insulations levels is investigated.
The developed algorithm performs a wide-range search in order to detect rapidly changing weather conditions, and keeps the
simulated stand-alone or grid-connected systems continuously operating close to the maximum power point. The performance of
the developed algorithm, under extremely changing environmental conditions, is found to be superior compared to that of other
conventional algorithms. The results of this study show that, under uniform radiations conditions, the developed algorithm takes
only half of the time required by the Perturbation and Observe algorithms to reach maximum power point MMP. Furthermore,
when PV is subjected to sudden partial shading conditions, the algorithm rapidly detects these changes and reaches the new MMP
in less than a second.

1. Introduction the voltage of the PV cell, there is a unique maximum power

point at particular weather conditions, and this maximum
It is now widely accepted that the nonrenewable sources in power point keeps changing with the irradiance levels and
the world are finite and it is only a matter of time before ambient temperature. Therefore, a maximum power point
reserves will essentially be consumed [1, 2]. It has been proven tracking (MPPT) algorithm is commonly used, to obtain the
that the use of nonrenewable energy sources has severe effect maximum possible power under varying weather conditions
on the environment. Due to environmental awareness and and loads. Because of the nonlinear I-V relation, the power
technological advancement, high oil price, and government versus voltage P-V relation has more complicated behavior
support, the renewable electricity generation capacity has especially when the weather conditions change. This complex
reached an estimated 240 gigawatts (GW) worldwide in 2007 behavior makes analytical solution very difficult and force
while it was 160 GW in 2004 [3]. The solar photovoltaic (PV) researchers to develop numerous techniques for finding
power system is attractive renewable energy source due to its MPP. MPPT algorithms are implemented using digital signal
availability and economic feasibility [4, 5]. Stand-alone PV processors (DSP) or microcontrollers. They control DC/DC
systems are found suitable for powering remote areas [4]. converters to drive the PV panel to operate at their MPP.
The power produced by a PV module depends on the The main methods used to achieve MPPT for PV cells
operating temperature, the amount of falling solar irradiance are Estimation Methods, Heuristic Method, and Search Algo-
over the PV Cells array, and the load connected [5, 6]. rithms. System modeling method [7–9], curve-fitting method
The power output of PV cells depends on the nonlinear [10, 11], open circuit method [12, 13], and short circuit method
current voltage (I-V) characteristics relationship. Because [14, 15] are examples of Estimation Methods. They run PV
of this nonlinear relationship between the current and cells to track an Estimated Maximum Power Point rather than
2 ISRN Renewable Energy

the Actual Maximum Power Point. In general, Estimation IL

I
Methods depend on an approximated mathematical model
to calculate an estimated MPP. PV cell current-voltage data, Rs
irradiance, and temperature levels are the required inputs for G
these methods. The main advantages of Estimation Methods
are their simple implementation and fast response. On the V
other hand, they are expensive and inaccurate, require the use T
of many sensors, demand large computational power, and fail
under rapidly changing atmospheric conditions.
Heuristic methods are recently developed to overcome Figure 1: Simple PV equivalent circuit model.
the problem associated with the inaccuracy of the PV cell
mathematical model. The succesful development of these
methods is attributed to the recent advances in nonlinear Practical PV device
control method. Fuzzy Logic Control methods [16, 17], Ideal PV cell I
Neural Network methods [18–22] and Genetics Algorithm
methods are examples of Heuristic Methods developed to RL
achieve MPPT of PV cells. The Fuzzy Logic Control method Ipv Id Rp
V
effectively tracks the MPP under various weather conditions.
However, the performance highly depends on the expertise of
the rule-based system designer which might lead to the failure
of the controller in tracking the MPP under partial shading
condition. The outcomes of MPPT using Neural Network Figure 2: Single-diode model of the theoretical PV with additional
methods are highly related to the accuracy and efficiency parameters for the improved PV model.
of the designed algorithm, the size of the training database,
and the network training quality. Furthermore, they require
gathered data for various conditions and multiple locations
adaptation to different solar panels and different environ-
to guarantee a better performance.
mental conditions is considered in developing the proposed
Unlike estimation methods, Search Algorithms track search algorithm.
the actual MPP rather than an estimated MPP. However,
they continuously search for the MPP by increasing or
decreasing the PV cell output voltage. Many methods are 2. PV Cells Model
developed implementing the search algorithms concept.
Some of those methods are the differentiation method [23, The modern solar cells are fabricated from a p-n junction
24], the Perturbation and Observe methods, or the modified and prepared in a small thickness semiconductor layer. These
Perturbation and Observe methods [25–30], the Incremental cells create electric current when subjected to sunlight. The
Conductance method [31], the Parasitic Capacitance method PV cells act as a form of diode in which the parameters of
[32], the Fibonacci Search method [33, 34], the Slide Control this diode define the circuit model. Walker [38] proposed
method [35], and the Particle Swamp Optimization method a simple approximate circuit for the PV cells as shown in
[36, 37]. The main advantage of these methods can be Figure 1. He modeled the PV cells as a constant current source
summarized as follows. No previous knowledge about the connected in parallel with a diode. The model includes the
PV cell characteristics is required, no or minimal use of necessary series resistance but neglects the shunt resistance.
sensors is required to show acceptable response to rapidly The characteristics equation of the current 𝐼 produced by the
changing conditions. On the other hand, search algorithms PV cell is proportional to the intensity of the radiation fallen
waste energy as they continuously oscillate around the MPP. on the PV cells [38] as follows:
Unfortunately, solving the oscillation problem requires com-
plex computational power. Furthermore, the performance of 𝐼 = 𝐼pv − 𝐼0 [𝑒𝑞(𝑉+𝐼𝑅𝑠 )/𝑎𝑘𝑇 − 1] , (1)
these methods degrades significantly under partial shading
conditions and fails under sudden partial shading conditions. where 𝐼pv is the current (A) generated by the photovoltaic cell
Unfortunately, none of the developed search algorithms can due the incident light, 𝑘 = 1.3806503 × 10−23 (J/K) is the
successfully deal with all extreme weather conditions, namely, Boltzmann’s constant, 𝑇 is the temperature in Kelvins (K) at
rapidly changing, partial shading, and sudden partial shading the p-n junction, 𝑞 = 1.60217646 × 10−23 (C) is the electron
conditions. charge, 𝐼0 is the diode saturation or leakage current (A), 𝑉 is
All the developed methods found in the literature still the PV cell output voltage (Volt), and 𝑎 is the diode ideality
suffer from weak performance under extreme weather con- factor.
ditions. It is the objective of this study to develop an efficient Villalva et al. suggested adding several parameters in
novel search algorithm that deals with rapidly changing, order to capture the behavior of actual panels which consist
partial shading and sudden partial shading conditions to of a number of connected photovoltaic cells [39]. Figure 2
track the MPP of PV cells. Minimal user interface for wider demonstrates the equivalent circuit of a photovoltaic cell
ISRN Renewable Energy 3

I to the experimental maximum power point 𝑃max,𝑒 [39]. The

+ i 1 method yields the following equations:
−
Rs Current measurement
𝑞 𝑉𝑚𝑝 + 𝑅𝑠 𝐼𝑚𝑝
𝑃max,𝑚 = 𝑉𝑚𝑝 {𝐼pv − 𝐼0 [exp ( )]
+ Ipv V + 𝐾𝑇 𝑎𝑁𝑠
+ 2 V
−v
s − Rs1 Voltage measurement s − 𝑉𝑚𝑝 + 𝑅𝑠 𝐼𝑚𝑝
− },
1
𝑅𝑝
Im
𝑅𝑝 = 𝑉𝑚𝑝 (𝑉𝑚𝑝 + 𝐼𝑚𝑝 𝑅𝑠 )
Ramp
𝑞 𝑉𝑚𝑝 + 𝑅𝑠 𝐼𝑚𝑝
Figure 3: MATLAB/Simulink model of improved PV panel. × ({𝑉𝑚𝑝 𝐼pv − 𝑉𝑚𝑝 𝐼0 [exp ( )]
𝐾𝑇 𝑎𝑁𝑠
−1
+𝑉𝑚𝑝 𝐼0 − 𝑃max,𝑒 }) .
with the additional parameters for the improved model. The
improved I-V characteristic of a PV array is given as [39] (5)
From the above equations, it is obvious that output power
𝑞(𝑉+𝐼𝑅𝑠 )/𝑎𝑉𝑡 𝑉 + 𝑅𝑠 𝐼 of a PV module depends on the solar irradiance values and
𝐼 = 𝐼pv − 𝐼0 [𝑒 − 1] − , (2)
𝑅𝑝 ambient temperature.

where 𝑉𝑡 = (𝑁𝑠 𝑘𝑇/𝑞) is the thermal voltage of the array with 3. Partial Shading Problem
𝑁𝑠 cells connected in series and 𝑅𝑠 and 𝑅𝑝 are the equivalent
series and shunt resistances (Ω) of the array, respectively. The The temperature, the irradiation levels, and the shading
series resistance is the sum of several internal and structural of the system affect the performance of photovoltaic cells.
resistances within the device. The parallel resistance models Partial shading problems arise due to the existence of clouds
the current leakage through the p-n junction. or building shadows. This problem makes the photovoltaic
Figure 3 shows a MATLAB/Simulink model for the power characteristics more complicated with multiple peaks
improved PV panel mathematical model. Figure 4 shows a in power. This reduces the efficiency of most MPPT tech-
MATLAB/Simulink model of the PV circuit built to obtain niques. The effect of partial shading problem appears signifi-
the I-V characteristics according to (2). The photogenerated cantly for large arrangement of panels. Under partially shaded
current is function of solar radiation and temperature of the condition, it has been found that the I-V curves have multiple
p-n junction is given by [39] as follows: stairs while the P-V curves have multiple peaks [40].

𝐺 4. Boost Converter Design

𝐼pv = (𝐼pv,𝑛 + 𝑘𝑖 Δ𝑇) , (3)
𝐺𝑛
In this paper, we utilized a boost convertor to change the PV
panel operating point to its MPP. The operation of the DC-
where 𝐼pv,𝑛 is the photogenerated current at the nominal DC converters is controlled by the MPPT algorithm making
conditions (25∘ C, 1000 W/m2 ), Δ𝑇 = 𝑇 − 𝑇𝑛 , 𝑇𝑛 is the the power output of the panel operate at its the maximum
nominal temperature in kelvin, 𝐺 is the radiation fallen on level. The MPPT algorithms are usually implemented using
the device, and 𝐺𝑛 is the nominal radiation. Figure 5 shows either digital signal processors (DSP) or a microcontroller.
the Simulink model for calculating 𝐼pv . The saturation current DC-DC boost converter shown in Figure 7 transforms
equation derived by [39] is given as an unstable 𝑉𝐼 voltage source into a higher-level stable
output voltage 𝑉𝑂. Over the past years, these converters
𝐼sc,𝑛 + 𝑘𝑖 Δ𝑇 have shifted from the conventional analog control to pulse
𝐼0 = , (4) width modulated PWM digital control. These converters
(𝑒 oc,𝑛 +𝐾V Δ𝑇)/𝑎𝑉𝑡 )
((𝑉 − 1) use solid-state components including MOSFETS, transistors,
and diodes to operate a digital switch (on-off switching)
where 𝐼sc,𝑛 is the short circuit current at nominal conditions to control the resultant output of these DC-DC converters.
and 𝑘𝑖 and 𝑘V are the current and voltage coefficients, These converters employ capacitive and inductance elements
respectively. 𝑉oc, 𝑛 is the open circuit current at nominal to store and transfer energy and eliminate the noise (lowpass
conditions. Figure 6 shows the Simulink model used to filter). Unlike analog control methods, the digital control
calculate saturation current 𝐼0 . The values of 𝐼sc,𝑛 , 𝑘𝑖 , 𝑘V , and methods of DC-DC converters are quite resilient and flexible
𝑉oc,𝑛 are supplied by the PV panel manufacturer. Villalva et al. in changing the software. Also, they offer the advantage of
suggested a method to adjust the values of 𝑅𝑠 and 𝑅𝑝 based implementing more complex control algorithms.
on the fact that there is only pair {𝑅𝑠 , 𝑅𝑝 }, which makes the Boost converters have two modes of operation. The
simulated maximum power point 𝑃max,𝑚 = 𝑉𝑚𝑝 × 𝐼𝑚𝑝 , equal closed-switch mode starts the diode reverse bias mode.
4 ISRN Renewable Energy

1 +
+ ×
V eu
Add
2 Product 1 Math function
×
I
Product 5 +
0.15 + 2
1 − Ipv −
Rs Subtract Subtract 1 Im
Constant
1.25 ×
÷ ×
q/(a ∗ K ∗ Ns ) 4
Divide
3 I0 Product 2
T 1
Vt a

Figure 4: MATLAB/Simulink model for calculating the I-V characteristics.

+
0.065
× +
ki Add
Product 3
3.56
1 + Ipv,n
T −
× 2
Subtract 1 2 ×
298 1 Product 1 Ipv
G ÷
Tn dT
Divide
1000

Gn

Figure 5: PV cell MATLAB/Simulink model for calculating 𝐼pv .

It causes the input power source to store energy in the Table 1

inductor as well as the capacitor to discharge into the storage Specification Min Max Units
battery, while the open-switch mode starts the diode forward
Input voltage 0 30 V
bias mode causing higher output energy supply from both the
power source and the inductor to the capacitor and the load. Output voltage 50 51 V
The equation defining the ratio between the input voltage and Output power 0 150 W
the output voltage is given as Operation frequency 1 MHz
Voltage allowed output ripple 50 mV
𝑉in 1
= , (6)
𝑉out 1 − 𝐷
(iii) The boost inductor value plays a key role in deter-
where 𝐷 = 𝑇on /(𝑇on +𝑇off ) is the duty cycle of the input PWM mining the operational mode of the system. In order
signal. The microcontroller sends the proper PWM signal to to have continuous operation mode, the inductance
the boost converter switch to reach the MPP advised by the should satisfy the following equation:
MPPT algorithm.
In order to design an appropriate boost converter, the 𝑅 × 𝑡𝑠 (𝑀 − 1)
𝐿≥ × , (7)
following steps are carried out. 2 𝑀3

(i) The boost converter specifications listed in Table 1 where 𝑅 is the load resistance, 𝐿 is the inductance,
are selected to satisfy continuous operation under all 𝑀 is the maximum voltage gain which is equal
conditions. to (𝑉out /𝑉in ), and 𝑡𝑠 is the switching period. We chose
the closest available conductor value of 68 𝜇H.
(ii) A 200 kHz switching frequency is chosen to reduce
the size of the boost converter components and (iv) The maximum duty cycle is calculated according to
decrease the power loss. (6) and is found to be 0.846.
ISRN Renewable Energy 5

42.1 + 0.065 × +
+ ki +
Voc,n
Add Product 2
Add 1
×
0.5 3.87 ×
÷ 1
Product 1
kv Isc,n
Divide I0
× eu +
2
1 Vt a −
Product 3 Math function
dT Subtract 1
1

kv1

Figure 6: MATLAB/Simulink model for calculating 𝐼0 .

L calculated and chosen specifications, the Simulink model of

c p
Diode the PV panel system with the boost converter is shown in
RL iL = iC
Figure 8.
d VO
Q1 C
g
s R
5. Theory of the Proposed MPPT Algorithm
VI +
− There are several factors to consider when developing and
a Rc choosing the techniques for performing MPPT, such as the
Drive circuit ability of an algorithm to detect multiple maxima, costs,
ia and convergence speed. MPPT is naturally a maxima-finding
process. The proposed Maximum Power Point tracking algo-
rithm implements the search algorithms concept. The reasons
Figure 7: Circuit diagram of boost converter. behind this choice as mentioned previously are no previous
knowledge about the PV cell characteristics is required, sim-
ple implementation, and guaranteed convergence. The main
(v) The capacitance value is given by disadvantage of search algorithms are that; they waste energy
as they continuously oscillate around the MPP and they show
𝐼𝑝𝑘 2 × 𝐿 inadequate response under partial shading conditions and fail
𝐶= , (8)
2 × Δ𝑉 × (𝑉0 − 𝑉𝑖 ) under sudden partial shading conditions.
Most search algorithms model the data as a 1-D function
where 𝐼𝑝𝑘 is given by and go about a Brute-Force method of finding the maxima
of the function. These kinds of algorithms require a large
𝑉𝑖 amount of processing time. Other algorithms like the Shubert
𝐼𝑝𝑘 = × 𝐷 × 𝑡𝑠
𝐿 algorithm [40–42] rely on Lipschitz continuity. They use
(9)
8 1 weighing parameters that emphasize on the local search
= × 0.846 × = 0.5 A. versus the global search for the optima. However, these
66 × 10−6 2 × 10−6
constants may not exist or could not be easily computed
The allowed ripple in the output voltage is required to be especially for optimizing nonlinear control system which is
50 mV. Hence, the desired capacitance value obtained from the case in this study. Also, these constants are required
(8) is as to be large enough to exceed the rate of change of I-V
curve. This might lead to a large number of iterations as the
0.52 × 66 rate of convergence towards the optimal point slows down.
𝐶= = 153 𝜇F. (10)
2 × 0.05 × (13 − 2) Regardless of these draw backs, the Lipschitz method remains
highly attractive due to the ability to bound the rate of change
The equivalent series resistance needed to limit the output of the function; thus, searching algorithms can be easily
ripple to 5 mV is calculated by implemented and one parameter is required to be specified,
Δ𝑉0 0.05 that is, Lipschitz constant [41].
ESR = = = 10 mΩ. (11) Eliminating the need to specify the constant and making
Δ𝐼0 0.5
the algorithm consider both local and global search are the
To sooth the signal, almost twice of the calculated criteria for developing the new algorithm. In this study,
capacitor value is used, that is, 330 𝜇F. Using all the design we followed Jones et al. methodology [41] by utilizing
6 ISRN Renewable Energy

IL
+ i To workspace 1
−
Ipv
+ −i
2 𝜇H Diode 1 IL
1 G
2
G R1
Connection Switch
2 T port 1 g 1
+ C
T −v
Connection Vpv
port
PV

Pulse generator

Vpv Search Change pulse

Ipv duty width
Embedded Level-2 M-file
MATLAB function S-function

Figure 8: Full simulator model.

the advantages of Shubert algorithm, mainly the bounding of

the rate of change. Jones et al. presented an algorithm called
DIRECT algorithm, that is, a modification of the standard
Lipschitzian to overcome the problems of normal Shubert’s f(x)
algorithm [40] (Figure 11). In order to clarify the logical
development and the features of the proposed algorithm,
Slope-𝛼
we begin by reviewing Shubert’s method and discussing its
main drawbacks; then we will present Jones’s algorithm that
eliminates the need to specify a Lipschitz constant. Finally, we f1
present the newly developed algorithm. Slope 𝛼

5.1. Lipschitzian Optimization. The goal is to find the maxi-

mum functional value of 𝑓(𝑥). The normal Shubert method
a x1 b
can be summarized as follows.
Lipschitz continuity states that a function 𝑓(𝑥) defined Figure 9: An initial lower bound for 𝑓 using the Lipschitz constant.
on the closed interval [𝑙, 𝑢] is called Lipschitz continuous on
[𝑙, 𝑢] if there exists a positive constant, the Lipschitz constant,
such that
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑓 (𝑥) − 𝑓 (𝑥󸀠 )󵄨󵄨󵄨 < 𝛼 󵄨󵄨󵄨𝑥 − 𝑥󸀠 󵄨󵄨󵄨 , ∀𝑥, 𝑥󸀠 ∈ [𝑙, 𝑢] . (12) The initial lower bound of 𝑓 is denoted by 𝑓1 and given
󵄨 󵄨 󵄨 󵄨
Let us take a hypothetical function 𝑓(𝑥) defined on [𝑎, 𝑏]. by
If we substitute 𝑎 and 𝑏 for 𝑥󸀠 into the definition of Lipschitz-
continuity, we get the following two inequalities for 𝑓(𝑥),
where 𝑥 ∈ [𝑎, 𝑏], (𝑓 (𝑎) − 𝑓 (𝑏))
𝑓1 = − 𝛼 (𝑏 − 𝑎) . (15)
𝑓 (𝑥) > 𝑓 (𝑎) − 𝛼 (𝑥 − 𝑎) , 2𝛼
(13)
𝑓 (𝑥) > 𝑓 (𝑏) + 𝛼 (𝑥 − 𝑏) .
The Shubert algorithm uses this straightforward idea in
The inequalities (13) form V-shaped formed from the two finding the minimum or the maximum of 𝑓(𝑥). Shubert’s
lines with slopes −𝛼 and +𝛼 with the intersection occur below algorithm is an iterative algorithm that continues to perform
𝑓(𝑥) as shown in Figure 9. the same operation on the regions [𝑎, 𝑥1 ] and [𝑥1 , 𝑏], elimi-
The point of intersection for the two lines 𝑥1 is easy to
nating the higher 𝑓1 value intervals to finally reach the global
calculate and is given as
minimum value as demonstrated by Figure 10.
(𝑎 + 𝑏) (𝑓 (𝑎) − 𝑓 (𝑏)) Evaluating the function at the center of any interval rather
𝑥1 = + . (14)
2 2𝛼 than the bounds of the interval is the main idea of DIRECT
ISRN Renewable Energy 7

f(x)
f(x)

a x1 b a x1 x2 b
(a) (b)

f(x)

a x1 x2 x3 b
(c)

Figure 10: Iterations of the Shubert algorithm in dividing the intervals of minimum 𝑓1 [41].

f(c)
a c1 b

fmax + 𝜀|fmax |
a1 c2 b1 = a2 c1 b2 = a3 c3 b3 fmax

Figure 11: Dividing strategy of DIRECT algorithm.

algorithm developed by Jones et al. [41]. Mathematically, this

can be expressed as

𝑓 (𝑥) > 𝑓 (𝑐) + 𝐾 (𝑥 − 𝑐) , for 𝑥 ≤ 𝑐,

(16)
𝑓 (𝑥) > 𝑓 (𝑐) − 𝐾 (𝑥 − 𝑐) , for 𝑥 ≥ 𝑐,
(bj − aj )/2 (b − a)/2
where 𝑐 = (𝑎 + 𝑏)/2. Thus, the lower bound equation has
Potentially optimal
to take into account the function value at the center of the
Nonoptimal
interval:
𝛼 (𝑏 − 𝑎) Figure 12: Set of potentially optimal intervals.
Lower bound = 𝑓 (𝑐) − . (17)
2
Figure 13 shows the interval-dividing strategy of the
DIRECT algorithm when a sampling interval [𝑎, 𝑏] has been interval is then divided into three intervals [𝑎1 , 𝑏1 ], [𝑎2 , 𝑏2 ],
specified. Assume that the algorithm has already taken the and [𝑎3 , 𝑏3 ] resulting in two new center points to be evaluated
sample 𝑐 at the center of [𝑎, 𝑏] in the previous step. This 𝑐1 , 𝑐2 . The sample 𝑐 simply becomes the center of the new
8 ISRN Renewable Energy

P
I
POP

OP DOP
󳰀
POP

V VMMP V
(a) (b)

Figure 13: Change in power under partially shaded condition identification in (a) I-V and (b) P-V.

middle interval. The algorithm then evaluates the three (2) Calculate the angles in radians that each of the points
samples to decide the next sampling interval. It is clear that makes with 𝑃0 ; then sort them in increasing order, and
only two new samples in each dividing iterations are required push them onto a stack.
for evaluation. Further subdivision for the potential interval (3) If 𝑃0 forms a left turn with the last two points in the
containing optima is carried out until the optimal point is stack, we push 𝑃0 onto the stack; else, we discard and
found. Theoretical details are found in reference [41]. The fact make the next point in the stack 𝑃0 and repeat.
that in convex hull functions local optima are global optima
is used to select potentially optimal interval. Suppose that we (4) Repeat step number (3) until you encounter 𝑃0 again.
have partitioned the interval [𝑙, 𝑢] into intervals [𝑎𝑖 , 𝑏𝑖 ] with A simplified approach is to calculate the direction cross
midpoints 𝑐𝑖 % for 𝑖 = 1, . . . , 𝑚. Let 𝜀 > 0 be a positive constant product of the two vectors formed from three points 𝑃0 -𝑃1
and let 𝑓max be the current best function value. Interval 𝑗 and 𝑃1 -𝑃2 . If the value is positive, it is a left turn and thus
is said to be potentially optimal if there exists some rate-of- we keep the point and the interval. If it is negative, then we
change constant 𝐾 ̃ > 0 such that [41] discard the interval all together.

̃ (𝑏𝑗 − 𝑎𝑗 )
𝐾 ̃ (𝑏𝑖 − 𝑎𝑖 )
𝐾 5.2. Golden Section Search (for Rapidly Changing Conditions).
𝑓 (𝑐𝑗 ) + ≥ 𝑓 (𝑐𝑗 ) + , ∀𝑖 = 1, 𝑚, The golden section algorithm is used to detect the envi-
2 2 ronmental change by continuously oscillating around the
(18)
maximum power point. The Golden Section Search method
̃ (𝑏𝑗 − 𝑎𝑗 )
𝐾 is used to find the maximum or the minimum of a unimodal
󵄨 󵄨
𝑓 (𝑐𝑗 ) + ≥ 𝑓min + 𝜀 󵄨󵄨󵄨𝑓min 󵄨󵄨󵄨 . (19) function by calculating the function at three different points.
2 In this study, Golden Section Search (GSS) MPPT algorithm
uses the voltage as the search variable. The main advantage
The inequality (18) selects intervals that would improve of GSS algorithm is its fast convergence compared to many
the current function value. For intervals with the same other MPPT algorithms. The MPPT algorithm is developed
length, the interval with the highest function value at its with the limiting parameters for fast convergence. The main
center point is chosen to be the potentially optimal interval steps in GSS algorithm are as follows.
(POI). The inequality (19) ensures that the POIs exceed the
current best solution by a nontrivial amount 𝜀|𝑓max |. Figure 12 Initialization
demonstrates how convex hull sets help choose POIs that
satisfy both (18) and (19) [41]. If we construct convex hull
(1) Determine 𝑥𝑙 and 𝑥𝑢 which is known to contain the
from the function values at the center points, the intervals
maximum of the function 𝑓(𝑥).
that make up convex hull are considered POIs. Graham’s Scan
is efficient algorithm used to create a convex hull out of the set (2) Determine two intermediate points 𝑥1 and 𝑥2 such
of center points [43]. Graham’s scan is a phase algorithm that that
can be summarized as follows, given a set of points 𝑆𝑝 . √5 − 1
𝑥1 = 𝑥𝑙 + (𝑥𝑢 − 𝑥𝑙 ) ,
2
(1) Find the point in 𝑆𝑝 with the maximum value. If two (20)
or more have the same value, use one with the lowest √5 − 1
𝑥2 = 𝑥2 − (𝑥𝑢 − 𝑥𝑙 ) .
𝑥 coordinate. Call it 𝑃0 . 2
ISRN Renewable Energy 9

Set m = 1
[a1 , b1 ] = [a, b]
c1 = (a1 + b1 )/2
Evaluate f(c1 )
Set fmax = f(c1 )
Set iteration counter p = 1

Identify the set S of potentially

Environment change happens. One needs to find global OP
optimal intervals using Graham’s
scan
No Continue
Yes No environment oscillating
Ct < 4
Select any interval j in set S change around MPP

𝜎 = (bj − aj )/3
cm+1 = cj − 𝜎 xu = mpp , mpp = xL , Ct = Ct + 1
cm+2 = cj + 𝜎
xL = mpp + (xu − mpp )/0.62
Evaluate f(cm+1 ) and evaluate f(cm+2 )
Update fmax
Yes xu = mpp , mpp = xl
No
In the partition, add the left and right subintervals Ct = Ct + 1
P( mpp ) > P(xl )
xL = mpp + (xu − mpp )/0.62
[am+1 , bm+1 ] = [aj , aj + 𝜎], center point cm+1
[am+2 , bm+2 ] = [aj + 2𝜎, bj ], center point cm+2
Yes Yes Ct < 4 No

[aj , bj ] = [aj + 𝜎, aj + 2𝜎]

No
P( mpp ) > P(xu )
m=m+2
xL = mpp , mpp = xu
S = S − {j} Ct = Ct + 1
xu = mpp + 0.62( mpp − xL )
No
Is S empty?

Yes GSS: xu = mpp + 0.5

xL = mpp − (0.5/0.62), Ct = 0 Yes No
OP interval is identified and reached Ct < 4

No Yes
Is MPP found?

Figure 14: Flowchart of the proposed search algorithm.

(3) If 𝑓(𝑥1 ) > 𝑓(𝑥2 ), then new points 𝑥𝑙 , 𝑥1 , 𝑥2 , and 𝑥𝑢 (5) If 𝑥𝑢 − 𝑥𝐿 < 𝜀 (a predefined condition), then the
are updated as maximum occurs at (𝑥𝑢 + 𝑥𝐿 )/2 stop iteration; else,
go to step 2.
𝑥𝑙 = 𝑥2 , 𝑥2 = 𝑥1 , 𝑥𝑢 = 𝑥𝑢 ,
The intermediate points 𝑥1 and 𝑥2 are chosen such that
√5 − 1 (21)
the ratio of the distance from these points to the boundaries of
𝑥1 = 𝑥𝑙 + (𝑥𝑢 − 𝑥𝑙 ) .
2 the search region is equal to the golden ratio. The golden ratio
which is equal to 1.61803398 makes the algorithm converge at
(4) If 𝑓(𝑥1 ) < 𝑓(𝑥2 ), then the new points 𝑥𝑙 , 𝑥1 , 𝑥2 , and a constant speed.
𝑥𝑢 are updated as
𝑥𝑙 = 𝑥𝑙 , 𝑥𝑢 = 𝑥1 , 𝑥1 = 𝑥2 , 5.3. Implementation for PV System. Theoretically, 𝑉MMP falls
(22) between 0 V and 𝑉oc max . In the present study, a DC/DC
√5 − 1 converter is used to vary the optimal point OP of the PV
𝑥2 = 𝑥𝑢 + (𝑥𝑢 − 𝑥𝑙 ) .
2 system. Hence, the duty cycle would be in the range of (0, 1).
10 ISRN Renewable Energy

− 6
2

Current (A)
1 4
G G
2 +
Connection
T T port 1 1 2
Connection
port
0
Subsystem 0 5 10 15 20 25 30 35
Figure 15: Masked Simulink model of PV cells. Voltage (V)

1000 W/m2 , 25∘ C 1000 W/m2 , 40∘ C

1000 W/m2 , 60∘ C 1500 W/m2 , 25∘ C
However, this range can be controlled to be much smaller
Figure 16: I-V characteristic curves of a practical photovoltaic
in practice. The smaller range will increase the convergence
device under different weather conditions.
speed. The maximum duty cycle is calculated according to (6)
and found to be 0.846.
As mentioned previously, the I-V curves have multiple 120
stairs while the P-V curves have multiple peaks under 100

Power (W)
80
partially shaded condition as shown in Figure 15. To explain
60
the main idea of the new algorithm, assume that DIRECT has 40
successfully found the maximum power point as shown in 20
Figure 13(a). When weather condition changes happen, the 0
OP will move to a different point due to the change of the I-V 0 5 10 15 20 25 30
curve. Since the duty cycle is not changed, the power of the PV Voltage (V)
󸀠
is decreased from 𝑃OP to 𝑃OP as explained in Figure 13(b). To 1000 W/m2 , 25∘ C
detect environment changes on the PV arrays, GSS algorithm 2000 W/m2 , 25∘ C
continuously oscillates around the current 𝑉MMP . The GSS
is chosen due to its rapid local searching for optimal point. Figure 17: P-V curves plotted for different weather conditions.
When executing the GSS algorithm within small interval
around the current 𝑉MMP , the GSS has the ability within
four iterations to decide whether environment changes have different solar irradiation to allow producing different values
occurred or not. If OP is not found when GSS iterated of photovoltaic current 𝐼pv .
four times, DIRECT algorithm is called to search for global In order to clarify the complexity associated with partial
OP. This will increase the response time of the proposed shading, sample simulations are carried out and their results
algorithm. Figure 14 shows the simplified flowchart of the are shown in Figures 20 and 21.It can be seen in Figures 20
DIRECT search algorithm incorporated with GSS. and 21 that there are two local maxima in the power-voltage
curves. As mentioned previously, many MPPT algorithms
6. Results are incapable of dealing with the effects of partial shading
and might mistakenly drive the system to its local maximum
The simulation results are carried out using MATLAB/ instead of the desired global maximum.
Simulink to validate the performance of the proposed MPPT
algorithm. 6.2. Performance under Uniform Weather Condition. The
Performance and operation of the proposed search algorithm
6.1. Photovoltaic Model Simulation Results. The developed have been evaluated using MATLAB/Simulink. The sampling
photovoltaic cell Simulink model built and shown in Figure 15 time is chosen to be 0.05 s. For the implemented proposed
is used to simulate the performance of PV cells under MPPT algorithms, the simulation results have been obtained
different temperatures and irradiance levels. Figure 16 shows during starting up of the system. The results have been
the I-V characteristic curve of a practical photovoltaic device obtained for a solar irradiance value of the proposed system
under different weather conditions. It can be seen that the I-V that is tested under two uniform radiation levels: 1000 W/m2
characteristics are dependent on the levels of irradiance and and 2000 W/m2 . As shown in Figures 22 and 23, the proposed
the temperature of PV cell. Figure 17 shows the power-voltage MPPT algorithm found the global maximum in a relatively
variations under different weather condition. It is clear that short time, that is, in less than 0.8 seconds, with small oscil-
the P-V curve has single peak that could be easily found by lation in steady state. IT is also observed that the power loss
conventional searching method. from oscillation is insignificant. To evaluate the effectiveness
To investigate partial shading conditions, first, a masked of the proposed algorithm, its performance is compared with
model of a single PV cell is built as shown in Figure 18. Then, that of the Perturbation and Observe algorithm [25–31]. The
a model for a panel consisting of three PV cells connected in result in Figure 24 shows that Perturbation and Observe
parallel is built as shown in Figure 19. Each cell is subjected to needed 1.9 seconds to reach the MPP.
ISRN Renewable Energy 11

T
1
T Out 1
G 1
2 Ipv
G Subsystem

Figure 18: Masked Simulink model to calculate 𝐼pv .

1 T
T Out 1 +
2 G Ipv1 +
G Add
Subsystem

+ 1
T +
Out 1 Add 1 Ipv
3 G Ipv2
G1 Subsystem 1

T
Out 1
4 G Ipv3
G2 Subsystem 2

Figure 19: Simulink model for partial shading.

3.5 60
3 50
2.5
Power (W)

40
Power (W)

2
30
1.5
20
1
10
0.5
0 0
0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Voltage (V) Time (s)

Figure 20: The variation of the I-V under varying irradiance (600, Figure 22: The simulated power curves for the 1000 W/m2 , 25∘ C
200, and 100 W/m2 ). fully shaded designed algorithm.

60 120
50 100
Power (W)

80
40
Power (W)

60
30 40
20
20 0
0 0.2 0.4 0.6 0.8 1 1.2
10
Time (s)
0
0 5 10 15 20 25 Figure 23: The simulated power curves for the 2000 W/m2 , 25∘ C.
Voltage (V)

Figure 21: The variation of the P-V under partially shaded condition
(600, 200, and 100 W/m2 ).
the PV panels are subjected to uniform insolation condition.
This condition is maintained for 0.4 s before it is changed
to partially shaded condition. The cells temperatures are
6.3. Proposed MPPT Algorithm under Partially Shaded. The kept constant at 25∘ C. Figure 25 shows that the MMPT
proposed scheme for MPPT algorithm is tested under algorithm maintains the MPP until the radiation level varies
partially shaded conditions. The simulations were con- at 0.4 s. It can be seen from Figure 25 that the proposed
ducted with two consecutive scenarios. In the first scenario, algorithm immediately detects the weather changes and starts
12 ISRN Renewable Energy

100
References
Power (W)

[1] K. Bataineh and N. Fayez, “Thermal performance of building

50 attached sunspace in Jordan climate,” in Proceedings of the
1st International Nuclear and Renewable Energy Conference
0 (INREC ’10), Amman, Jordan, March 2010.
0 0.5 1 1.5 2 2.5
[2] K. Bataineh, “Numerical simulations for average temperature
Time (s) differential stirling engine,” Journal of Technology Innovations in
Figure 24: The simulated power curves for the 2000 W/m2 , 25∘ C Renewable Energy, vol. 2, no. 3, 2013.
Perturbation and Observe [25–31]. [3] K. Bataineh and D. Dalalah, “Assessment of wind energy
potential for selected areas in Jordan,” Journal of Renewable
Energy, vol. 59, pp. 75–81, 2013.
[4] K. M. Bataineh and D. Dalalah, “Optimal configuration for
design of stand-alone PV system,” Smart Grid and Renewable
120
100 Energy, vol. 3, no. 2, 2012.
Power (W)

80 [5] K. Bataineh and A. Hamzeh, “Efficient maximum power point

60 tracking algorithm for photovoltaic cells,” in Proceedings of
40 the 1st WSEAS International Conference on Industrial and
20
Manufacturing Technologies, Athens, Greece, 2013.
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 [6] M. Adel Hamdy, “A new model for the current-voltage out-
Time (s) put characteristics of photovoltaic modules,” Journal of Power
Sources, vol. 50, no. 1-2, pp. 11–20, 1994.
Figure 25: MMP tracking under partial shading (600, 200, and [7] T. Takashima, T. Tanaka, M. Amano, and Y. Ando, “Maximum
100 W/m2 ). output control of photovoltaic (PV) array,” in Proceedings of the
35th Intersociety Energy Conversion Engineering Conference and
Exhibit (IECEC ’00), pp. 380–383, Las Vegas, Nev, USA, July
2000.
[8] N. Takehara and S. Kurokami, “Power control apparatus and
immediately searching for the new MPP. As it can be seen, in method and power generating system using them,” Patent US
Figure 25, the proposed MPPT found the new MPP after 0.7 s. 5,654,883, 1997.
[9] K. Nishioka, N. Sakitani, K.-I. Kurobe et al., “Analysis of the
temperature characteristics in polycrystalline Si solar cells using
7. Conclusion modified equivalent circuit model,” Japanese Journal of Applied
Physics, vol. 42, no. 12, pp. 7175–7179, 2003.
A novel algorithm of maximum power point tracking for [10] J. C. H. Phang, D. S. H. Chan, and J. R. Phillips, “Accurate analyt-
photovoltaic power generation system is presented. A math- ical method for the extraction of solar cell model parameters,”
ematical model of the PV panel is presented based on the Electronics Letters, vol. 20, no. 10, pp. 406–408, 1984.
theory of photovoltaic. The V-I characteristics and the P- [11] D. Lafferty, “Coupling network for improving conversion effi-
V power output under several irradiation levels and tem- ciency of photovoltaic power source,” US 4,873,480, 1989.
perature conditions are simulated. The proposed algorithm [12] P. Chetty, “Maximum power transfer system for a solar cell
is implemented in a PV panel connected to DC-DC boost array,” US 4,604,567, 1986.
converter with resistive loading. A full Simulink MATLAB [13] M. A. S. Masoum and H. Dehbonei, “Optimal power point
model is built to simulate the performance of the proposed tracking of photovoltaic system under all operating conditions,”
algorithm. The proposed MPPT algorithm is evaluated under in Proceedings of the 17th Congress of the World Energy Council,
extreme weather conditions. The results show that various Houston, Tex, USA, 1998.
advantages are gained with the proposed scheme compared [14] S. M. Alghuwainem, “Matching of a dc motor to a photovoltaic
to perturbation and observe algorithms. The response time is generator using a step-up converter with a current-locked loop,”
smaller and the oscillations around the MPP were reduced IEEE Transactions on Energy Conversion, vol. 9, no. 1, pp. 192–
198, 1994.
to obtain steady state maximum power output. It is found
that the proposed algorithm quickly identifies the MPP of the [15] T. Noguchi, S. Togashi, and R. Nakamoto, “Short-current
pulse-based adaptive maximum-power-point tracking for a
solar panels under extreme weather conditions. Furthermore,
photovoltaic power generation system,” Electrical Engineering in
the MMPT algorithm sustains its performance when sub- Japan, vol. 139, no. 1, pp. 65–72, 2002.
jected to sudden changes in the insulation levels. In addition,
[16] P. Takun, S. Kaitwanidvilai, and C. Jettanasen, “Maximum
it is capable of finding the global maximum points under both power point tracking using fuzzy logic control for photovoltaic
fully and partially shaded conditions. systems,” in Proceedings of the International MultiConference of
Engineers and Computer Scientists (IMECS ’11), pp. 986–990,
Hong Kong, March 2011.
Conflict of Interests [17] M. S. A. Cheikh, C. Larbes, G. F. T. Kebir, and A. Zerguerras,
“Maximum power point tracking using a fuzzy logic control
The authors declare that there is no conflict of interests scheme,” Revue des Energies Renouvelables, vol. 10, no. 32, pp.
regarding the publication of this paper. 387–395, 2007.
ISRN Renewable Energy 13

[18] T. Hiyama, S. Kouzuma, and T. Imakubo, “Identification of opti- [32] A. Brambilla, M. Gambarara, A. Garutti, and F. Ronchi, “New
mal operating point of PV modules using neural network for approach to photovoltaic arrays maximum power point track-
real time maximum power tracking control,” IEEE Transactions ing,” in Proceedings of the 30th Annual IEEE Power Electronics
on Energy Conversion, vol. 10, no. 2, pp. 360–367, 1995. Specialists Conference (PESC ’99), pp. 632–637, July 1999.
[19] K. Ro and S. Rahman, “Two-loop controller for maximizing [33] M. Miyatake, T. Kouno, and M. Nakano, “Maximum power
performance of a grid-connected photovoltaic-fuel cell hybrid point tracking control employing fibonacci search algorithm
power plant,” IEEE Transactions on Energy Conversion, vol. 13, for photovoltaic power generation system,” in Proceedings of
no. 3, pp. 276–281, 1998. the International Conference of Power Electronics (ICPE ’01), pp.
622–625, Seoul, Republic of Korea, October 2001.
[20] A. Hussein, K. Hirasawa, J. Hu, and J. Murata, “The dynamic
performance of photovoltaic supplied DC motor fed from DC- [34] N. A. Ahmed and M. Miyatake, “A novel maximum power point
DC converter and controlled by neural networks,” in Proceed- tracking for photovoltaic applications under partially shaded
ings of the International Joint Conference on Neural Networks insolation conditions,” Electric Power Systems Research, vol. 78,
(IJCNN ’02), pp. 607–612, May 2002. no. 5, pp. 777–784, 2008.
[35] M. Zhang, J. Wu, and H. Zhao, “The application of slide
[21] X. Sun, W. Wu, X. Li, and Q. Zhao, “A research on photovoltaic technology in PV maximum power point tracking system,” in
energy controlling system with maximum power point track- Proceedings of the 5th World Congress on Intelligent Control and
ing,” in Proceedings of the Power Conversion Conference, pp. Automation (WCICA ’04), pp. 5591–5594, June 2004.
822–826, 2002.
[36] M. Miyatake, F. Toriumi, T. Endo, and N. Fujii, “A novel
[22] L. Zhang, Y. Bai, and A. Al-Amoudi, “GA-RBF neural network maximum power point tracker controlling several converters
based maximum power point tracking for grid-connected pho- connected to photovoltaic arrays with particle swarm optimiza-
tovoltaic systems,” in Proceedings of the International Conference tion technique,” in Proceedings of the European Conference on
on Power Electronics, Machines and Drives, pp. 18–23, April Power Electronics and Applications (EPE ’07), September 2007.
2002. [37] S. R. Chowdhury and H. Saha, “Maximum power point tracking
[23] L. T. W. Bavaro, “Power regulation utilizing only battery current of partially shaded solar photovoltaic arrays,” Renewable Energy,
monitoring,” Patent US 4,794,272, 1988. vol. 34, no. 10, pp. 2093–2100, 2009.
[24] C. Hua and J. R. Lin, “DSP-based controller application in bat- [38] G. Walker, “Evaluating MPPT converter topologies using a mat-
tery storage of photovoltaic system,” in Proceedings of the IEEE lab PV model,” Journal of Electrical and Electronics Engineering,
22nd International Conference on Industrial Electronics, Control, Australia, vol. 21, no. 1, pp. 49–55, 2001.
and Instrumentation (IECON ’96), pp. 1705–1710, August 1996. [39] M. G. Villalva, J. R. Gazoli, and E. Ruppert Filho, “Modeling and
[25] J. H. R. Enslin, M. S. Wolf, D. B. Snyman, and W. Swiegers, “Inte- circuit-based simulation of photovoltaic arrays,” in Proceedings
grated photovoltaic maximum power point tracking converter,” of the Brazilian Power Electronics Conference (COBEP ’09), pp.
IEEE Transactions on Industrial Electronics, vol. 44, no. 6, pp. 1244–1254, Mato Grosso do Sul, Brazil, October 2009.
769–773, 1997. [40] H. Patel and V. Agarwal, “MATLAB-based modeling to study
the effects of partial shading on PV array characteristics,” IEEE
[26] A. Al-Amoudi and L. Zhang, “Optimal control of a grid- Transactions on Energy Conversion, vol. 23, no. 1, pp. 302–310,
connected PV system for maximum power point tracking and 2008.
unity power factor,” in Proceedings of the 7th International
Conference on Power Electronics and Variable Speed Drives, pp. [41] B. O. Shubert, “A sequential method seeking the global maxi-
80–84, September 1998. mum of a function,” SIAM Journal on Numerical Analysis, vol.
9, no. 3, pp. 379–388, 1972.
[27] N. Kasa, T. Iida, and H. Iwamoto, “Maximum power point [42] D. R. Jones, C. D. Perttunen, and B. E. Stuckman, “Lipschitzian
tracking with capacitor identificator for photovoltaic power optimization without the Lipschitz constant,” Journal of Opti-
system,” in Proceedings of the 8th International Conference mization Theory and Applications, vol. 79, no. 1, pp. 157–181, 1993.
on Power Electronics and Variable Speed Drives, pp. 130–135,
September 2000. [43] E. A. Galperin, “The cubic algorithm,” Journal of Mathematical
Analysis and Applications, vol. 112, no. 2, pp. 635–640, 1985.
[28] L. Zhang, A. Al-Amoudi, and Y. Bai, “Real-time maximum
power point tracking for grid-connected photovoltaic systems,”
in Proceedings of the 8th International Conference on Power
Electronics and Variable Speed Drives, pp. 124–129, September
2000.
[29] W. Xiao, W. G. Dunford, P. R. Palmer, and A. Capel, “Application
of centered differentiation and steepest descent to maximum
power point tracking,” IEEE Transactions on Industrial Electron-
ics, vol. 54, no. 5, pp. 2539–2549, 2007.
[30] J. M. Enrique, J. M. Andújar, and M. A. Bohórquez, “A
reliable, fast and low cost maximum power point tracker for
photovoltaic applications,” Solar Energy, vol. 84, no. 1, pp. 79–
89, 2010.
[31] K. H. Hussein, I. Muta, T. Hoshino, and M. Osakada, “Max-
imum photovoltaic power tracking: an algorithm for rapidly
changing atmospheric conditions,” IEE Proceedings, vol. 142, no.
1, pp. 59–64, 1995.