An IntelligentMPPT Method For PV Systems Operating Under Real Environmental Conditions

The International Journal Of Engineering And Science (IJES)
|| Volume || 5 || Issue || 10 || Pages || PP 01-10 || 2016 ||
ISSN (e): 2319 – 1813 ISSN (p): 2319 – 1805
www.theijes.com The IJES Page 1
An IntelligentMPPT Method For PV Systems Operating Under
Real Environmental Conditions
Ammar Hussein Mutlag1
, Hussain Shareef2
, Omar Nameer Mohammed Salim1
1
Electrical Engineering Technical College, Middle Technical University, Baghdad, Iraq
2
Department of Electrical Engineering, United Arab Emirates University, 15551Al-Ain, UAE
--------------------------------------------------------ABSTRACT-------------------------------------------------------------
The sun irradiance (G) and temperature (T) are the two main factors that affect the output power gained from
the photovoltaic (PV) DC–DC converter. Therefore, to enhance the performance of the overall system; a
mechanism to track the maximum power point (MPP) is required. Conventional maximum power point tracking
approaches, such as observation and perturbation technique, experience difficulty in identifying the true MPP.
Therefore, intelligent systems including fuzzy logic controllers (FLC) are introduced for the maximum power
point tracking system (MPPT). The selection of the membership functions (MFs) and the fuzzy sets (FSs)
numbers are crucial in the performance of the FLC based MPPT. Accordingly, this work presents numerous
adaptive neuro-fuzzy systems to automatically adjustthe fuzzy logic controller membership functions as an
alternative to the trial and error approach, which waste time and effort in MPPT design. For this purpose an
adaptive neuro-fuzzy system is developed in MATLAB/Simulink to determine suitable MFs and the FSs for the
fuzzy logic controller. The effects of different types of MFs and the FSs are deeply investigated using real data
collected from the rooftop PV system. The investigations show that the fuzzy logic controller with a triangular
membership function and seven fuzzy setsprovides the best results.
Keywords:Adaptive neuro-fuzzy system;MPPT; Converter; Photovoltaic
---------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: 17 May 2016 Date of Accepted: 29 October 2016
--------------------------------------------------------------------------------------------------------------------------------------
I. INTRODUCTION
Solar energy is inexhaustible, free and clean and it is considered as the core of renewable energy (RE) in the
recent times primarily because of running down of fossil fuels [1]. Among various RE resources, photovoltaic
(PV) system plays a very important role either in grid-connected or islanding configurations [2,3]. However, the
PV systems generate intermittent power under fluctuating weather which is the main issue that must be taken in
consideration[4,5]. The power-voltage and current–voltage characteristics are responsible for the power
generated from the PV cell [4]. Therefore, to work the PV generation at its peak; the MPPT mechanism is highly
significant in PV system [6].
Numerous MPPT mechanisms have been introduced by many scholars since year 1960. Some well-known
MMPT methods are incremental conductance (IC) method, perturb and observe (P&O) method and constant
voltage (CV) method [4,7–9]. The method of P&O was extensively used due to its simple control method as
well as the minimum number of its input parameters. However, the use of this algorithm leads to a loss in power
due to an enormous oscillation in the area of maximum power point (MPP). Others, like IC methods has been
proposed by some researchers [10,11], which somehow could eliminate the oscillations in the area of the MPP.
However, this kind of methods need good and accuracte sensor to measure either voltage or current.
Recently, the MPPT-based Artificial intelligence (AI) is widely used in PV converter with great dynamics and
high effectiveness. Various intelligent methods including fuzzy logic and artificial neural network (ANN) have
been mentioned in the literature. In [12], The MPPT-based ANN in a PV inverter power conditioning unit has
been suggested. In this study, the voltages and currents corresponding to a maximum power has been estimated
by the ANN. However, large training data are required before it can be used in the controller.
The fuzzy logic controller (FLC) is adaptable to complex systems [13];therefore, it is employed for the MMPT
[14]. The MFs and the rules are usually obtained in the conventional FLC using the trial and error process which
does not often lead to a convincing performance. Therefore, the FLC have been incorporates with the neural
network (NN) method to find the MFs together with its rules. The adaptive neuro fuzzy inference system
(ANFIS) is one of the quite popular techniques that combine the FLC with NN.
The ANFIS controller has been used in many applications such as induction motor drive [15]. ANFIS based
space vector modulation technique was used for a three-level inverter fed induction motor [16]. In [17], the
authors have presented an approach to design and implement a decentralized power system stabilizer which is
able to performing quite perfectly in a large range of variations of parameters of the system and conditions of

An IntelligentMPPT Method For PV Systems Operating Under Real Environmental Conditions
loading by using ANFIS. Chang et al. in [18] has developed a weighted evolving fuzzy neural network to
forecasting demand of electricity for each month inside country of Taiwan. In [19], the integrated diagnostic
system is used for detecting islanding condition.In the diagnostic system, neuro fuzzy controller for grid-
connected inverter-based distributed generation was investigated where the authors used the ANFIS for
islanding detection. Shareef et al. in [20] have proposed an adaptive neuro fuzzy inference system approach to
identify the real power transfer among generators where the 25-bus peninsula Malaysia equivalent system was
exploited as a test system to demonstrate the usefulness of the ANFIS result output with contrast to the method
of modified nodal equation. Thus from the literature, it is understood that the ANFIS has got excellent potential
for the MPPT.
In this study, the MPPT is developed using four different ANFIS controllers to assess their performance.The rest
of the work is organized as follows. The modeling of the PV system is mentioned in section 2. Adaptive neuro
fuzzy inference system is described in section 3; meanwhile the ANFIS based MPPT is discussed in section 4.
The results are discussed in section 5. Finally, the conclusion isrevealed in section 6.
II. MODELING OF THE PV SYSTEM
The G and T are in charge of the working point of PV system at the MPP [21,22]. Thecell current, I, which
represent the mathematical model of the PV cell can be express as [23]:
(1)
where is light-generated cell current (A), is cell reverse saturation current (A), is electronic charge, is
ideality factor, is Boltzmann’s constant, and is cell temperature (K).
Regarding to Eq. (1), the MPPT can be determine utilizing which varieswith G and T. The same relationship
can be utilized to determine the MMPT in a PV module if the number of PV cells is known. The mathematical
model can be used to determine the cell output current. However, this time consuming procedure and it is the
main obstacle of the mathematical model. The 25 PV modules types SolarTIFSTF are used in this work to
generate peak of 3 kW which are arrange in series-connected configuration. Table 1 shows the PV module
characteristics.
Table 1: PV module characteristics
PV module SolarTIFSTF
Maximum Power (PMPP) 120W
Open circuit voltage (Voc) 21.5 V
Short circuit current (Isc) 7.63 A
Voltage at maximum power (VMPP) 17.4 V
Current at maximum power (IMPP) 6.89 A
Nominal temperature 43.6º C
Nominal irradiance 1000 W/m2
III. ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM
Neural network (NN) with fuzzy logic (FL) makes an intelligent controller and it has been used in power
electronic converters to solve many problems of nonlinearity [24]. Foremost advantage of the FL is its capability
to represent the nonlinear input-output relationships by a place (if-then) rule to play that part. On the other hand,
the key advantage of the NN is its excellent learning capability.
The robustness of the FLC depends on the shape of the membership functions and the rules. These parameters
are usually obtained based on the trial and error process which is very difficult and time-consumption.
Therefore, the MFs are sometimes obtained by incorporating the FLC with NN. This popular hybridization is
known as adaptive neuro fuzzy inference system (ANFIS). ANFIS can effectively model the nonlinear systems
with fast learning and high accuracy.
3.1 Anfis Architecture
The ANFIS play the main part to arrange a best membership function [25,26]. Based on input and output data
set, ANFIS creates the fuzzy inference system (FIS).The membership function parameters can be attunedwith
either a back-propagation algorithm only or by a least square estimation (LSE) algorithm. In FL perspective,
only zero or first order Sugeno inference system proposed by Takagi-Sugeno can be used to systematically
generate fuzzy rules from an input and output data. Fig. 1 shows the ANFIS structure for two inputs, one output
with two rules.

Fig. 1. ANFIS structure for two inputs (x) and (y) and one output U.
In this structure, the input nodes represent training data while output node represents forecasted data. The inputs
for the ANFIS are (x,y), the output is (U), the circles symbolize fixed node and the squares symbolize adaptive
node. ANFIS exhibit in Fig. 1 contains two rules:
(2)
(3)
where are the nonlinear parameters and are the linear parameters.
ANFIS shown in Fig. 1 consists of five layers with different functions [26]. The first layer is the fuzzification
layer. Every node in this layer is an adaptive node:
, for i=1,2 (4)
where x, y are the input signals to this node, and are membership functions to the linguistic
variables Ai and Bi (positive big, positive medium, positive small.). The types of the membership functions used
in this study are gbell and triangular membership function as exhibited in Fig. 2. In the fuzzification process, Eq.
(5) is used for gbell membership function and Eq. (6) is used for triangular membership function.
(5)
(6)
where are parameter sets from the input membership function.
c+ac-a c
x
µ
Slop=-b/2a
ca b
x
µ

(a) (b)
Fig. 2. Membership functions types used in this study (a) gbell MF (b) triangular MF
The product of all input signals is the output of second layer (rules layer) which performs the inference process
as can be specific in Eq. (7):
(7)
The normalization layer is the third layer in the ANFIS system. Every node provides the ratio of the ith
rule’s
firing strength to the total firing level as shown in Eq. (8):
(8)
Each node of the layers computes the activation level of every rule, the fuzzy rules number are equivalent to the
number of layers. They calculate the weights (w) which are normalized. The forth layer is theconsequent layer
where the output of this layer can be expressed as in Eq. (9):
(9)
This layer output is generated from the inference of the rules. It establishes an adaptive correlation between
normalized firing value and result function.The final layer is the defuzzification layer where the output is
obtained by summing all incoming signals and transforming the fuzzy classification results into a crisp value, as
shown in the Eq. (10):
(10)
As exhibited in Fig. 1, there are two adaptive nodes which arethe first node and the fourth node. However, the
first node (fuzzification layer) there are three premise parameters named ( ), and three consequent
parameters named ( in the fourth node (consequent layer) [26].
3.2 ANFIS learning algorithm
The goal of the learning algorithm is to adjust the premise parameter ( ) and the consequent parameters
( so that the output of ANFIS is compatible with the data that has been trained. When the premise
parameters are settled to fix values the output can be represent in Eq. (11):
(11)
Next substituting Eq. (9) into Eq. (12):
(12)
Knowing and substituting it into Eq. (12) yields:
(13)
Finally, rearrange the Eq. (13), the output can be written as the Eq. (14):
(14)
Eq. (14) is a linear synthesis of consequent parameters ( ). Least square estimation (LSE)
algorithm can be adopted to determine the optimal consequent parameter's value. Oncethe premise parameters
are not fixed, a hybrid algorithm which it combined gradient descent back-propagation and LSE algorithms is
used to solve the problem. The hybrid algorithm includes two passes which are least square estimation (LSE)
and backward pass (gradient descent). The forward pass is mainly implemented to make consequent parameters
optimized when the premise parameters are fixed while the function of the backward pass is to optimize the
premise parameters. Therefore, it has been verified that the hybrid algorithm is extremely efficient to train the
ANFIS [27].
IV. ANFIS BASED MPPT
Two types of MFs have been considered in this work which are triangular and gbell membership functions. In
each type, five and seven FS have been used. Therefore, four ANFIS can be employed for MPPT which are:
 Triangular MF-five FS (5-tri) based MPPT
 Triangular MF-seven FS (7-tri) based MPPT
 gbell MF-five FS (5-gbell) based MPPT
 gbell MF-seven FS (7-gbell) based MPPT

The most important issue regarding the ANFIS is the training data to identify the MFs and the FSs. In this work,
the input training data are the G and T; meanwhile the output training data is cell output current, . The
inputtraining data (G and T) are generated using (15) and (16). These data are used to determine the output
training data ( ) through Table 1 and Eq. (1).
(15)
(16)
Regarding to (15) and (16), the limitation for the G and T should be define. Fig. 3 shows the historical data
fluctuation regarding G and T;hence, the following values have been adopted in this work:
 has been define as 1500 W/m2
 has been define as 0
 has been define as 45ºC
 has been define as 0
These maximum values have been chosen more than the historical data fluctuation and the minimum values
have been chosen less than historical data fluctuation to ensure that all the generated data using (15) and (16) are
covered the required data fluctuation.
Temperature(ͦC)Temperature(ͦC)
Irradiance(W/m²)
Time (hours)Time (hours)
Tdata
Tave
Gdata
Gave
Fig. 3. Measured hourly irradiation and temperature data for one year [28]
V. RESULTS AND DISCUSSION
The trained four ANFIS based MPPT have been evaluated using the input data (G and T) for five days which
are 22th
of March 2016, 6th
of April 2016, 21th
of April2016, 13th
of May 2016, and 28th
of May 2016 which are
collectedfrom the rooftop PV system. In this work, the aim is to investigate the effectiveness of four ANFIS
under real data condition using same meteorological data which are collected through the HOBOwarePro
software.
The ANFISeffectiveness-based MPPT has been evaluated using statistical analysis. For this analysis, three types
of indices have been used which are standard deviation (SDEV), mean square error (MSE), and extracted
energies. The SDEV evaluates the rate of variation in compare to the average value. High SDEVgive indication
that the data points are misled out over the values targeted meanwhile a minimum SDEVindicates that the data
points are near to the target value. Table 2 shows the effectiveness comparison of the ANFIS technique. The
ANFIS with 7-tri has achieved the better performance than the other ANFIS controllers.
Table 2: Standard deviation
Date 5-gbell 5-tri 7-gbell 7-tri
22th
of March 42.53635921 2.365545196 11.05373443 1.420979117
6th
of April 47.8999527 2.336575017 10.63208146 1.789202154
21th
of April 50.35506522 2.931421759 13.18874316 1.759782241
13th
of May 37.86859254 2.931387998 11.26272372 1.930723743
28th
of May 22.59633954 1.651998636 8.950492785 1.134269898

The second index is the mean square error (MSE). The quality of the signal is inversely proportional to the MSE
where the ANFIS output is move away from the accurate maximum powerwhen the MSE is increased;
meanwhile the decreasing in the MSE rate indicates that the output power is near to the accurate maximum
power. Table 3 shows the performance comparison using the MSE rates. Since the ANFIS with 7-tri achieved
the lower MSE; therefore, it achieved the robust performance compare with the other ANFIS controllers.
Table 3: Mean square error
22th
of March 2940.994578 6.667560774 227.9960801 2.377348134
6th
of April 3995.880805 6.910230339 246.436941 3.817990887
21th
ofApril 4679.006908 11.15435667 360.5945928 3.881629246
13th
of May 2173.950798 10.50674669 239.835553 4.268295513
28th
of May 778.2052292 2.954146174 139.5908919 1.346164582
Table 4 shows the extract energies based on four types of ANFIS. The extracted energies show that some
controllers produce power larger than others. However, the maximum of energy is extracted by (7-tri), as
boldfaced, and the minimum energy is extract by (5-gbell). The relative errors between (7-tri), and (5-gbell)
vary from 0.9% (on 28th
of May) to 6.01% (on 21th
ofApril). Getting 6% more energy is not ignorable in
measurement of performance of the energy. It is important to note that for very sunny days, as 22th
of March,
13th
of May, and 28th
of May, which correspond to the maximum extracted energy, the errors are significantly
small.
Table 4: Extracted energies (Wh)
22th
of March 1077.169328 1109.794271 1100.54038 1110.231903
6th
of April 791.6295322 831.69533 821.3478589 832.1146168
21th
of April 709.7169443 754.4337633 742.3689948 755.1490077
13th
of May 1071.569131 1097.406613 1088.158081 1098.055203
28th
of May 1668.904962 1684.79999 1677.560145 1685.030462
To further evaluate the performance of various ANFIS models, statistics analysis is conducted using Q-Q plots
for five days as shown in Figs. 4-8. Ordinarily, the essential idea is to calculate the theoretically predicted value
for every point of the data. In case of the data which go behind the assumed distribution, the points on the Q-Q
plot will fall around on a straight line. As can be seen from Figs. 4-8, the ANFIS based triangular MFs with five
and seven FS achieve better perform than the ANFIS based gbell MFs with five and seven FS.

Fig. 4.Q-Q plot for the 22th
of Marchusing (a) 5-gbell, (b) 5-tri, (c) 7-gbell, and (d) 7-tri
of Aprilusing (a) 5-gbell, (b) 5-tri, (c) 7-gbell, and (d) 7-tri

of Aprilusing (a) 5-gbell, (b) 5-tri, (c) 7-gbell, and (d) 7-tri
of Mayusing (a) 5-gbell, (b) 5-tri, (c) 7-gbell, and (d) 7-tri

of Mayusing (a) 5-gbell, (b) 5-tri, (c) 7-gbell, and (d) 7-tri
The Q-Q plots for five days are compared in Fig. 9. The figure indicates that the triangular membership function
achieved better performance than the gbell membership function. However, the 7-tri show robust and superior
performance compared with the 7-gbell.

Fig. 9. The Q-Q comparison for five days
VI. CONCLUSION
This paper introduced a new and effective maximum power point tracking method based on adaptive neuro-
fuzzy controller for 3 kW peak PV systems consisting of the 25 SolarTIFSTF PV modules. The proposed
method is simple in tuning. To select the best FLC-based MPPT, four ANFIS controllers have been developed
to track the MPP for PV system. Three types of indices have been used to evaluate the effectiveness of the
system which are SDEV, MSE and extracted energy. To evaluate the reliability and efficiency of the proposed
algorithm, the ANFIS model was tested using real data obtained from the rooftop PV system in 22th
of March
2016, 6th
of April 2016, 21th
of April 2016, 13th
of May 2016, and 28th
of May 2016. ANFIS with 7 triangular
MF has superior performance to track the MPP than the other ANFIS controllers where it achieved lowest
SDEV, MSE and highest extracted energy. Furthermore, it can be concluded that the choice of the MF type is
more important than the choice of the number of the FS in the design of the ANFIS controllers.
REFERENCES
[1]. A. Messai, A.Mellit, A.Guessoum, S. A. Kalogirou, Maximum power point tracking using a GA optimized fuzzy logic controller
and its FPGA implementation, Solar Energy85(2) (2011) 265–277.
[2]. Subiyanto, A. Mohamed,H. Shareef,Hopfield Neural Network Optimized Fuzzy Logic Controller for MaximumPower Point
Tracking in a Photovoltaic System, International Journal of Photoenergy2012 (2012) 1–13.

[3]. A. H. Mutlag, H. Shareef, A. Mohamed, M. A. Hannan, J. A. Ali, An Improved Fuzzy Logic Controller Design for PV Inverters
Utilizing Differential Search Optimization, International Journal of Photoenergy2014 (2014) 1–14.
[4]. T. Esram,P. L. Chapman, Comparison of photovoltaic array maximum power point tracking techniques, IEEE Transactions on
Energy Conversion22 (2) (2007) 439–449.
[5]. F. Dincer,M. E. Meral, Critical factors that affecting efficiency of solar cells, Smart Grid and Renewable Energy1 (1) (2010) 47–50.
[6]. G. Graditi, G. Adinolfi, G. M. Tina, Photovoltaic optimizer boost converters: Temperature influence and electro-thermal design,
Applied Energy 115 (c) (2014) 140–150.
[7]. E. Irmak,N. Güler, Application of a high efficient voltage regulation system with MPPT algorithm, Electrical Power and Energy
Systems44 (1) (2013) 703–712.
[8]. B. B. J. D. Retnam,A. Gounden, Power Electronic Interface with Maximum Power Point Tracking Using Line-commutated Inverter
for Grid-connected Permanent Magnet Synchronous Generator, Electric Power Components and Systems 43 (5) (2015) 543–555.
[9]. V. Salas, E. Olias, A. Barrado, A. Lazaro, Review of the maximum power point tracking algorithms for stand-alone photovoltaic
systems, Solar Energy Materials and Solar Cells 90 (11) (2006) 1555–1578.
[10]. D. Mi,Y. Jian, P. Ke, Zero average incremental conductance maximum power point tracking control for photovoltaic system, Proc
CSEE 30 (2010) 48–53.
[11]. F. Liu, S. Duan, F. Liu, B. Liu, Y. Kang, A Variable Step Size INC MPPT Method for PV Systems, IEEE Transactions on Industrial
Electronics 55 (7) (2008) 2622–2628.
[12]. A. K. Rai, N. D. Kaushika, B. Singh, N. Agarwal, Simulation model of ANN based maximum power point tracking controller for
solar PV system, Solar Energy Materials & Solar Cells 95 (2) (2011) 773–778.
[13]. T. Narongrit, K. Areerak, K. Areerak, A New Design Approach of Fuzzy Controller for Shunt Active Power Filter, Electric Power
Components and Systems 43 (6) (2015) 685–694.
[14]. I. Houssamo, F. Locment,Sechilariu, M. Experimental analysis of impact of MPPT methods on energy efficiency for photovoltaic
power systems, Electrical Power and Energy Systems46 (2013) 98–107.
[15]. Y. J. Mon, Airbag controller designed by adaptive-network-based fuzzy inference system (ANFIS). Fuzzy Sets and Systems 158
(24) (2007) 2706–2714.
[16]. D. Sukumar, J. Jithendranath,S. Saranu, Three-level Inverter-fed Induction Motor Drive Performance Improvement with Neuro-
fuzzy Space Vector Modulation, Electric Power Components and Systems 42 (15) (2014) 1633–1646.
[17]. H. E. A. Talaat,A. Abdennour,A. A. Al-Sulaiman, Design and experimental investigation of a decentralized GA-optimized neuro-
fuzzy power system stabilizer. Electrical Power and Energy Systems32 (7) (2010) 751–759.
[18]. P. C. Chang, C. Y. Fan, J. J. Lin, Monthly electricity demand forecasting based on a weighted evolving fuzzy neural network
approach, Electrical Power and Energy Systems33 (1) (2010) 17–27.
[19]. F. Hashemi, N.Ghadimi,B. Sobhani, Islanding detection for inverter-based DG coupled with using an adaptive neuro-fuzzy
inference system, Electrical Power and Energy Systems45 (1) (2013) 443–455.
[20]. H. Shareef, A. Khalid, M. W. Mustafa, A. Khairuddin, An ANFIS Approach for Real Power Transfer Allocation. Journal of
Applied Mathematics 2011 (2011) 1–14.
[21]. V. Quaschning, Renewable Energy and Climate Change. 1st ed., John Wiley & Sons, Chichester, West Sussex, United Kingdom,
2010.
[22]. F. Kininger, Photovoltaic Systems Technology. 1st ed., Universität Kassel, Wil-helmshöher Alle 73, 34121 Kassel, Germany, 2003.
[23]. S. J. E. Mineiro,S. Daher,F. L. M. Antunes, C. M. T. Cruz, Photovoltaic system for supply public illumination in electrical energy
demand peak, IEEE Conference on applied power electronics conference and exposition.Fortaleza, Brazil: IEEE Pres, 2004: 1501–
1506
[24]. M. M. Gupta, D. H. Rao, Neuro-control systems: theory and applications. Piscataway, NJ: IEEE press, USA, 1994.
[25]. B. Ozpineci, L. M. Tolbert,J. N. Chiasson, Harmonic optimization of multilevel converters using Genetic Algorithms, IEEE Power
Electronics Letters3 (3) (2005) 92–95.
[26]. E. D. Übeyli, Adaptive neuro-fuzzy inference system employing wavelet coefficients for detection of ophthalmic arterial disorders,
Expert Systems with Applications 34 (3) (2008) 2201–2209.
[27]. J. R. Jang, ANFIS: adaptive-network-based fuzzy inference system, IEEE Transactions on Systems, Man and Cybernetics23 (3)
(1993) 665–685.
[28]. M. Z. Daud, A. Mohamed,M. A. Hannan, An improved control method of battery energy storage system for hourly dispatch of
photovoltaic power sources, Energy Conversion and Management 73 (2013) 256–270.

An IntelligentMPPT Method For PV Systems Operating Under Real Environmental Conditions

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to An IntelligentMPPT Method For PV Systems Operating Under Real Environmental Conditions (20)

Recently uploaded (20)

An IntelligentMPPT Method For PV Systems Operating Under Real Environmental Conditions