An Intelligent MPPT Approach based on Neural-

Network Voltage Estimator and Fuzzy Controller,

Applied to a Stand-alone PV System

B.Bendib1,2 , F.Krim2 , Senior Member, IEEE, H. Belmili1, M. F. Almi1, S. Bolouma1

1
Unité de développement des équipements solaires (UDES)/EPST-CDER, Bou-Ismail, 42415, W. Tipaza, Algeria
2
University of Setif 1, power electronics and industrial control laboratory (LEPCI), route de Béjaïa, 19000, Sétif, Algeria
E-mail: [email protected], [email protected], [email protected]

Abstract—This paper presents an intelligent maximum power networks (ANN) are used to predict the maximum power point
point tracking (MPPT) method for a stand-alone photovoltaic (MPP) of PV module. This ANN model is a Multi-Layer
(PV) system using artificial neural networks (ANN) modelling Perceptron (MLP) network, its input vector contains the
and a fuzzy logic controller (FLC). The ANN is trained for temperature and the solar irradiation whereas its output vector
various conditions of solar irradiance and temperature to contains the corresponding MPP voltage, current and power of
estimate the MPP voltage. This voltage is then used by the FLC the PV module. Then, the fuzzy logic controller (FLC) uses the
as a reference voltage to generate the appropriate control signal MPP voltage, which is estimated by the trained ANN model, as
for the DC-DC converter. The proposed technique is a reference voltage to generate the desired control signal (duty
implemented in Matlab/Simulink and compared with the
ratio) for the DC-DC converter. The complete PV system is
conventional method of incremental conductance (IncCond).
Simulation results show a good performance of the ANN based
implemented in Matlab-Simulink software and the results are
fuzzy MPPT controller. compared with those obtained using IncCond methode.

Keywords—PV system; ; DC-DC converter MPPT; artificial II. STAND-ALONE PV SYSTEM MODELING
neural network (ANN); fuzzy logic controller (FLC); IncCond.
The stand-alone PV system under study is shown in Fig.1;
it consists of a PV generator, storage battery, DC-DC
I. INTRODUCTION converter and MPPT controller.
Recently, because of the soaring energy crisis and
environmental problems associated with the depletion of Ib
I DC-DC
natural resources, renewable energy and energy efficient Converter
G,T V
technologies have been attracting much attention [1]. Various =/=
Vb

renewable sources such as solar energy, wind energy,

geothermal etc. are harnessed for electric power generation. PV generator Batterie
Solar energy is a good alternative for electric power generation Duty ratio (D)
due to its availability and cleanliness. The solar energy is V
Fuzzy Logic
directly converted into electric energy by Photovoltaic (PV) Artificial
Controller
neural network
solar module. The PV modules have maximum operating (FLC)
(ANN) Vmpp
points corresponding to the surrounding conditions such as
MPPT
solar irradiance, and ambient temperature. DC-DC converters
may be used to regulate the voltage and current at the load, and
Fig. 1. Block diagram of the stand-alone PV system.
mainly to track the maximum power point (MPP) of the PV
module. In order to study these converters for stand-alone PV
systems, one first needs how to model the PV module that is A. PV generator
attached to the converter [2]. The mathematical model of the A PV solar cell equivalent electrical circuit can be
PV module may be useful in study of MPP tracking (MPPT) represented by a one-diode model as shown in Fig.2. The
algorithms, and mainly to simulate the PV system and its relation between the cell terminal current and voltage is the
components using circuit simulators. Therefore, for a proper following:
design of MPPT controller, an accurate simulation model of the
module is required. This is especially the case when the peak I = I ph − I o ⎡⎣exp( (V + I ⋅ RS ) a ⋅VT ) −1⎤⎦ − (V + I ⋅ RS ) RP (1)
power point changes continuously due to environmental
variations [3]. Several methods were used to predict and to Where: I and V are the output current and output voltage of
track the optimal point in which the output power of the PV the PV cell, Io is the diode's reverse saturation current, a: is the
module is maximum [4-11]. In this paper, the artificial neural diode ideality factor, and RS and RP are the series and parallel

978-1-4799-2399-1/14/$31.00 ©2014 IEEE 404

resistances respectively, VT (=k T/q) is the thermal voltage of TABLE I. ELECTRICAL CHARACTERISTICS DATA OF PV MODULE.
the PV cell, q is the electron charge (1,602·10-19 C), k is the Electrical characteristics Values
Boltzmann constant (1,380·10-23 J/K), and T is the temperature Open-circuit voltage (VOC) 22.3 V
of the p–n junction in Kelvin (K). Iph is the generated photo- Short-circuit current (ISC) 8.20 A
current, mainly depends on the insolation G and cell’s Optimum operating voltage (Vmpp) 17.5 V
temperature T, which is described as, Optimum operating current (Impp) 7.71 A
Maximum power at STC (Pmax) 135 W
Current temperature coefficient of ISC (0.055±0.01) %/K
I ph = ⎡⎣ I ph _ STC + K i (T −T STC ) ⎤⎦ (G G STC ) (2) Voltage temperature coefficient of VOC -(75±10) mV/K

where IPh_STC (in Ampere, A) is the photo-current generated Using the PV generator model given by the previous
under Standard Test Conditions (STC) (solar irradiance GSTC equations and the electrical specification listed in Table I, the
=1000 W/m2, spectral distribution of air mass AM =1.5, and I-V and P-V characteristic curves for different temperature and
cell temperature TSTC =25°C ). The constant Ki is the short insolation levels can be plotted. Fig.3 shows the I–V and P–V
circuit current coefficient, provided by the module characteristics of a PV module under STC conditions. There is
manufacturer. The cell’s saturation current Io depends on the a unique point on the P–V curve, known as the maximum
cell temperature according to the following formula [2], power point (MPP); at this point, the PV module is said to
I ph _ S T C + K i (T − T S T C ) (3) operate at maximum efficiency and produces its maximum
Io =
exp ⎡⎣ (V oc _ ST C + K v (T − T S T C ) ) a ⋅V T ⎤⎦ − 1 output power (Pmax) given by:
Pmax =V mpp ⋅ I mpp (5)
Where VOC_STC (in Volt, V) is the open circuit voltage under
STC; Kv is the open circuit voltage coefficient. The power Where: Impp and Vmpp are the optimal operating current and
output from single PV cells is relatively small. To produce the voltage of PV module respectively. This point changes with
required voltage and power, solar cells are connected in series and environmental conditions of temperature and solar radiation as
parallel. In this way, any desired current-voltage (I–V) and power- shown in fig.4. Fig. 4a shows that the PV module output
voltage (P-V) characteristic could be generated. Therefore, the I-V current (I) and power (P) are greatly influenced by the change
characteristic equation of a PV array (arranged in NP parallel and in irradiation G, whereas the output voltage V remains almost
NS series solar cells) can be described as, constant. On the other hand, for an increasing temperature one
can see that the voltage decreases significanty while the
⎛ ⎡ ⎛ N p ⋅V + N s ⋅ RS ⋅ I ⎞ ⎤ N p ⋅V + N s ⋅ RS ⎞ current remains almost unchanged as shown in fig. 4b. Finally,
I = N p ⎜ I ph − I o ⎢exp ⎜ ⎟⎟ −1⎥ − ⎟ (4) a set of three-dimensional (3-D) figures are proposed to cover
⎜
⎝ ⎢⎣ ⎝⎜ N p ⋅ N s ⋅ a ⋅VT ⎠ ⎥⎦ N p ⋅ N s ⋅ RP ⎟
⎠ the most probable situations at various solar irradiations and
In this paper, the Suntech STP135-12/Tb PV module is various temperatures with their corresponding optimal voltage
used for a Matlab simulation model. The module is made of 36 Vmpp, current Impp , and power Pmax, as shown in Fig.5.
multi-crystalline silicon solar cells in series, and its
specifications are listed in Table I [12].

a
Fig. 2. One-diode electrical model of PV solar cell.

b
Fig. 4. I–V , P–V characteristics of a PV module for : (a) various values of G,
at T= 25°C;(b) various values of T, at G =1000 W/m2.
Fig. 3. I-V, P-V characteristics of STP135 PV module at STC.

405
 The switch S operates at a high frequency to produce a
chopped output voltage. This converter is suitable for use
when the array voltage is high and the battery voltage is low.
The power flow is controlled by adjusting the on/off duty
cycle of the switching [14]. The average output voltage is
given by (5), [15].
V b = D ⋅V i (5)
The buck converter can be used to drive a low voltage load
from a high voltage PV array, and it can operate efficiently at
a
any insolation level. Therefore, the utilization of the buck
converter in PV applications with lower voltage batteries is
highly recommended [16]. The dynamic model of the used
buck converter (Fig.6) can be derived as:

⎡I ⎤
d ⎡V i ⎤ ⎡ 0 0 ⎤ ⎡V i ⎤ ⎡ C11 0 −D
C2 ⎤⎢ ⎥
(6)
⎢i ⎥ = ⎢ D +⎢
0 ⎥⎦ ⎢⎣ i L ⎥⎦ ⎢⎣0
⎥ Vb
dt ⎣ L ⎦ ⎣L −1
0 ⎥⎦ ⎢ ⎥
L
⎢⎣ I b ⎥⎦

III. MPP ESTIMATION USING NEURAL NETWORKS

b
A. Artificial Neural Network
In general, an artificial neuron (Fig.7) is a processing
element with n inputs x1, x2… xn (which are the external inputs
or the outputs of the other neurons) and only one output. Its
processing consists in assigning to its output yi the result of an
activation function,

⎣
n
(
y i = f ( ai ) = f ⎡ ∑ j =1w ij ⋅ x j ⎤ + bi
⎦ ) (7)

c Where ai: is the neuron activation state, f: is the neuron

activation function, bi: bias (representing the interne threshold
Fig. 5. (a) Current Impp, (b) Voltage Vmpp and (c) Power Pmax under irradiance of the neuron). The parameters wij connecting the inputs to the
variation from 50 to1200 W/m2 in a step of 50 W/m2 and temperature neurons are called synaptic weights or simply weights.
frome -10 to 70 oC in a step of 5°C.
Inputs Neuron i

x1 Wi ,1
B. DC-DC Converter x2 Wi ,2 ai yi
∑ f
In order to operate the PV module at its maximum power bi
point, or close to it, a specific circuit, called Maximum Power xn Wi,n
1
Point Tracker (MPPT), is employed. Usually, the MPPT is
achieved by interposing a power converter (DC-DC converter) Fig. 7. Artificial Neuron representation
between the PV generator and the load (battery), thus, acting
Hidden layer
on the converter duty cycle (D) it is possible to guarantee that Input layer Output layer
k=1, 2, 3
PV module is operating at its MPP [13]. Fig. 6 shows the j=1, 2
wi, j wk,i
y1= Voltage Vmpp (V)
circuit of the buck converter, whose output voltage (Vb) is less Irradiation G (W/m2) = x1
than or equal to the input voltage Vi (PV generator output y2 = Current Impp (A)
Temperature T (°C) = x2
voltage). i
y3= Power Pmax (W)

IL Ib 1
I bj
+ + bk
S 1
L

Vi C1 C2 Vb Fig. 8. MLP neural network model for MPP estimation.

An artificial neural network (ANN) is a set of information

- - processing elements (elementary neurons), with a specific
Fig. 6. The buck DC-DC converter circuit. topology of interconnections between these elements and a
training law to adapt the connection weights. A very important
number of models and network architectures exist [17, 18].

406
 ANN can be classified into two main categories based on IV. FUZZY LOGIC CONTROLLER
their connection structures: feed-forward (or non-recurrent)
A typical fuzzy logic controller includes three basic
and feedback (or recurrent). Feed-forward neural networks are
the most commonly used type, mainly because of the components, fuzzification module, inference engine, and
difficulty of training feedback networks, although the last- defuzzification module [23-24]; as shown in fig. 10.
mentioned are more suitable for representing dynamical
systems [19]. Multi-layer Perceptron (MLP) is the most known
feed-forward neural network [20].
In our case, to model the PV generator, we proposed a
MLP network with three layers (as shown in Fig.8) such that:
• The first layer: where the inputs to the network are the
solar irradiation G and the temperature T.
• The second layer: is the hidden layer, it constitutes the
heart of the neural network. Its activation functions are
the sigmoid functions (Tansig).
• The third layer: is the output layer which constituted by
three neurons (Vmpp, Impp, Pmax). Its activation functions Fig. 10. Structure of fuzzy logic controller.
are linear functions (Purelin).
B. Network Training The fuzzification module makes it possible to pass from the
real variables to fuzzy variables. The proposed fuzzy
The back-propagation is the supervised training method controller has two input variables, error E and change of error
more used to adjust the weights and bias of a MLP network
(CE). At a sampling instant k, these variables are expressed as
(so as to satisfy an optimization criterion). In this method, the
follows,
error between the two outputs (desired output and network
output) is calculated. It is used to update the weights wk,i (and ∗
E (k ) =V mpp (k ) −V ( k ) (8)
bias bk) of the output layer . Then by error back-propagation,
the intermediate errors, corresponding to the hidden layer are CE (k ) = E (k ) − E ( k − 1) (9)
thus calculated and allow the adjustment of the weights wij
Where V*mpp is the MPP voltage estimated by the ANN
(and bias bi ) of hidden layer. In our case, to train the neural
network, a set of 3-D figures is proposed to cover the most model, which is used here as a reference voltage to generate
probable situations at various irradiances, various the required control signal (duty ratio, D) for the DC-DC
temperatures with corresponding MPP current (Impp), voltage converter. The change in duty ratio (∆D) is used as the output
(Vmpp) and power (Pmax), as shown in Fig.5. It can be seen of proposed controller. Therefore, the control is done by
from this figure that the correlation between the optimum changing this duty ratio according to the variation in the input
voltage (current, power) and irradiance/temperature is non- variables (E and CE). As shown in fig.10, the input variables
linear. In this respect, estimation of the MPP using ANN can of the fuzzy controller (E, CE) are derived from the actual
give a good promising solution. Offline training was carried signals (e, ce) by multiplying with the corresponding scale
out using Matlab® software with Levenberg-Marquardt (L-M) gains (SE, SCE), and then converted to the linguistic variables
back-propagation algorithm. Although the L-M needs an
such as PB (positive big), PM (positive medium), PS (positive
important memory space in training stage, this algorithm is
preferred because it shows a faster convergence and a better small), Z0 (zero), NS (negative small), NM (negative
accuracy than other algorithms [21-22]. The training result of medium), NB (negative big) using basic fuzzy subset. Fig.11
our ANN is shown in Fig.9. shows the membership grades of seven basic fuzzy subsets for
input and output variables. In general, in fuzzy control, the
following inference techniques can be used: Max-Min, Max-
Prod and Somme-Prod. In this paper, the inference method of
Mamdani[25], which is the Max-Min fuzzy combination, is
used. Table.II shows the rule table of fuzzy controller, where
all the entries of the matrix are fuzzy sets of E, CE and ∆D.
Being given that converter requires a precise control signal D
at its entry it is necessary to envisage a transformation of this
fuzzy information into deterministic information; this
transformation is called defuzzification [23]. The most popular
defuzzification methods are the center of gravity (COG) and
the mean of maxima (MOM) methods [26]. In this paper, the
COG defuzzification method is used to compute the output of
fuzzy controller (∆D), which can be expressed as

Fig. 9. Training result of ANN block for MPP prediction.

ΔD = (∑ n
j =1
μ ( ΔD j ) ⋅ ΔD j ) ∑
n
j =1
μ ( ΔD j ) .
(10)

407
 μ(E) the IncCond technique experiences some oscillation around
μ ( CE ) the MPP.
μ (∆D. )

E, CE, ∆D

Fig. 11. Membership functions for input E, input CE and output ∆D.

TABLE II. FUZZY RULES (MATRIX’S INFERENCE)

CE NB NM NS ZO PS PM PB
E
NB NB NB NB NB NM NS ZO
NM NB NB NB NM NS ZO PS
NS NB NB NM NS ZO PS PM Fig. 12. Simulink implementation of ANN based fuzzy MPPT controller.
ZO NB NM NS ZO PS PM PB
PS NM NS ZO PS PM PB PB
PM NS ZO PS PM PB PB PB
PB ZO PS PM PB PB PB PB

V. SIMULATION RESULTS
Fig.12 shows the developed PV model system consisting
of PV array, DC-DC buck converter with an MPPT controller
connected to a 12-volt storage battery, as implemented in
Matlab/simulink environment. In this paper, the simulation
results obtained by ANN based fuzzy logic (ANN-FL) MPPT
technique, are compared with those obtained using
conventional Incremental Conductance (IncCond) technique.
The simulation configurations are under exactly the same
conditions. A 100-Hertz sampling frequency is used for both
MPPT algorithms. Fig.13 shows the simulation results of the
PV module output power, operating voltage, operating current,
and the duty ratio D using a buck converter under STC
conditions (25°C and 1000W/m2, AM.1.5). It can be seen that
the ANN-FL MPPT controller reduces the response time of
photovoltaic system. It is also seen that the system with
IncCond MPPT has a loss of energy. Fig.14 shows a zoom of
Fig.13. we see that there an oscillation around the operation Fig. 13. Comparison of power, voltage, current, duty ratio of IncCond and
point for the IncCond conventional technique. This is due to ANN-FL MPPT methods under STC conditions.
the continuous perturbation of the operating voltage in order to
reach the MPP. Whereas this phenomenon of oscillation is
avoided in ANN-FL MPPT technique, where signals of power,
voltage, current, and duty ratio remain almost constant.
In order to investigate the effect of rapidly changing
irradiation conditions, an irradiation ramp change was used
(Fig.15). A 2-second period for the increasing and decreasing
ramps was selected. The temperature was kept at a constant
value of 25°C. Under these operating conditions the ANN-FL
MPPT method is more performant. Fig.15 shows how the
power output P (W) of the ANN-FL MPPT increases linearly
when the irradiation increases, whereas the IncCond technique
experiences a deviation from the MPP. Likewise, during the
irradiation decrease, the power P (W) of the ANN-FL MPPT
decreases linearly, whereas the IncCond technique experiences
a deviation from the MPP.
Likewise, a temperature ramp change was used (Fig.16).
The irradiation was maintained constant at G = 1000 W/m2.
As shown in Fig.16, during the temperature increase, both
powers (ANN-FL and IncCond) decrease linearly, but the
IncCond technique takes a longer time to reach the MPP. On
Fig. 14. Steady-state comparison of IncCond and ANN-FL MPPT methods at
the other hand, during the decrease of the temperature, the a frequency of 100 Hz.
power P (W) of the ANN-FL MPPT increases linearly while

408
 [4] T. Hiyama and K. Kitabayashi, “Neural network based estimation of
maximum power generation from PV module using environment
information,” IEEE Trans Energy Conver, vol. 12, pp. 241-247, 1997.
[5] A. Al-Amoudi and L. Zhang, “Application of radial basis function
networks for solar-array modeling and maximum power-point
prediction,” IEE Proc-Gener.Transm.Distrib, vol.147, pp.310-316, 2000.
[6] MAG De Brito, L Galotto, LP Sampaio, G De Azevedo e Melo, and CA
Canesin,“Evaluation of the Main MPPT Techniques for Photovoltaic
Applications,” IEEE Trans. Ind. Electron, vol.60, pp.1156-1167, 2013.
[7] Z. Batushansky, and A. Kuperman, “Thevenin-based approach to PV
arrays maximum power prediction,” IEEE 26-th Convention of
Electrical and Electronics Engineers in Israel, pp. 598- 602, 2010.
[8] A. Safari and S. Mekhilef, “Simulation and hardware implementation of
incremental conductance MPPT with direct control method using cuk
converter,” IEEE Trans. Ind. Electron, Vol.58 (4), pp.1154-1161, 2011.
[9] K.H. Hussein and G. Zhao, “Maximum photovoltaic power tracking: an
algorithm for rapidly changing atmospheric conditions,” IEE Proc-
Gener. Transm. Distrib, vol. 142 (1) , pp. 59-64, January 1995.
[10] T. Esram and P.L. Chapman, “Comparison of photovoltaic array
maximum power point tracking technique,” IEEE Trans. Energy Conver,
Fig. 15. ANN-FL and IncCond MPPT under rapidly changing irradiation. vol. 22 (2), pp. 439-449, 2007.
[11] A.R. Reisi, M.H. Moradi, and S. Jamasb, “Classification and comparison
of maximum power point tracking techniques for photovoltaic system: a
review,” Renew. Sust. Energ. Rev., vol. 19, pp. 433–443, 2013.
[12] Suntech STP135-12/Tb –135W. Multi-crystalline photovoltaic module
datasheet;2001.
[13] R.F. Coelho, F.M. Concer, and D. Martins, “A simplified analysis of
DC-DC converters applied as maximum power point tracker in
photovoltaic systems,” 2nd IEEE Int. Sympos. Power Electr for Distribut
Gener Systems, pp.29-34, 2010.
[14] C. Hua, C. Shen, “Control of DC/DC converters for solar energy system
with maximum power tracking,” 23rd Int Conf on Indust Electr, Control
and Instrument (IECON 97), pp.827-329, November 1997.
[15] R.W. Erickson, D. Maksimovic, Fundamentals of power electronics. 2nd
edition, New York: Kluwer Academic Publishers; 2001.
[16] A.N. Abd El-Shafy, “An effective and safe charging algorithm for lead-
acid batteries in PV systems,” Int J Energy Res, vol.35, pp.733-40, 2011.
[17] S.H Huang, and H.C. Zhang, “Artificial neural networks in
manufacturing: concepts, applications, and perspectives,” IEEE Trans.
Comp. Pack. Man. Technology-Part A vol. 17(2), pp. 212-228, 1994 .
[18] M.R.G. Meireles, P.E.M. Almeida, and M.G. Simoes, “A comprehensive
Fig. 16. ANN-FL and IncCond MPPT under rapidly changing temperature. review for industrial applicability of artificial neural networks,” IEEE
Trans. Ind. Electron, vol. 50 (3), pp. 585- 601, 2003.
VI. CONCLUSION [19] D.A. Linkens, and H.O. Nyongesa, “Learning systems in intelligent
In this paper, an intelligent approach of MPPT using an artificial control: an appraisal of fuzzy, neural and genetic algorithm control
applications,”IEE Proc.-Control.Theory.Appl,vol.143,pp.367-386, 1996.
neural networks (ANN) estimator and a fuzzy logic controller (FLC)
is used for stand-alone PV systems. The PV module was modeled [20] M.H. Al-Shamisi , A.H. Assi and H.A.N. Hejase, ‘‘Artificial neural
using the single-diode electrical model and simulated on Matlab networks for predicting global solar radiation in Al Ain city UAE,” Int.
J. of Green Energy, vol. 10(5), pp.443-456, 2013.
software. In this work, we presented an ANN model which can
identify the MPP under any weather condition of solar irradiance and [21] M.T. Hagan, and M. Menhaj, “Training feedforward networks with the
temperature. A FLC uses the estimated MPP voltage as a reference Marquardt algorithm,”IEEE Trans. Neural. Net, vol.5, pp.989-993,1994.
voltage to generate the desired control signal for the dc-dc converter. [22] B.K. Bose, Power Electronics and Motor Drive – Advances and Trends.
The obtained results proved that the performances of the proposed MA: Elsevier, Burlington, 2006 .
ANN based fuzzy MPPT technique, in terms of precision and speed [23] C.Y. Won, D.H. Kim, S.C. Kim, W.S. Kim, and H.S. Kim, “A new
in the tracking of the MPP are much better than those of the IncCond maximum power point tracker of photovoltaic arrays using fuzzy
method. controller,” IEEE Power Elecronic. Specialists Conf, , pp.396-403, 1994
[24] S. Subiyanto, A. Mohamed, and M.A. Hannan, “Intelligent maximum
REFERENCES power point tracking for PV system using Hopfield neural network
optimized fuzzy logic controller,”EnergBuildings,vol.51,pp.29-38, 2012.
[1] K. Gopalakrishnan, S.K. Khaitan, and S. Kalogirou, “Soft computing in [25] E.H. Mamdani, and S. Assilian, “An experiment in linguistic synthesis
green and renewable energy systems,” Springer,Berlin Heidelberg, 2011. with a fuzzy logic controller,” Int. J. Hum-Comput St, vol. 51,pp. 135-
[2] M.G. Villalva and J.R. Gazoli, “Comprehensive approach to modeling 147, 1999.
and simulation of photovoltaic arrays,” IEEE Trans Power Electr. vol. [26] E. Van Broekhoven, B. De Baets, “Fast and accurate center of gravity
24 (5), pp.1198-1208, 2009. defuzzification of fuzzy system outputs defined on trapezoidal fuzzy
[3] K. Ishaque, Z. Salam, and Syafaruddin, “A comprehensive matlab partitions,” Fuzzy Sets and Systems, Vol 157 (7), pp. 904-918, 2006.
simulink PV system simulator with partial shading capability based on
two diode model,” Solar Energy,vol. 85, pp. 2217-2227, 2011.

409

Powered by TCPDF (www.tcpdf.org)