Energy Education Science and Technology Part A: Energy Science and Research

2012 Volume (issues) Special Issue: 539-548

Predict and analysis of the full-wave rectifier

circuit with Fuzzy Modeling technique based on
performances of different membership functions
Remzi Tuntas*
Yuzuncu Yil University, Ercis Technical Vocational School of Higher Education Electronic and Communication
Technologies, Van, Turkey

Received: 31 August 2012; accepted: 18 November 2012

Abstract

In the present study, A neuro-fuzzy modelling technique was used to predict and simulation the full-wave rectifier circuit.
Structure of the Adaptive Neuro-Fuzzy Inference Systems (ANFIS) was improved and trained in MATLAB toolbox. A
hybrid learning algorithm consists of back-propagation and least-squares, and the real hardware data were used for training
the ANFIS network. Efficiency of the developed ANFIS modelling techniques with different membership functions (MFs)
was tested and a mean 99.999% recognition success was obtained. Furthermore, the full-wave rectifier circuit was simulated
with the HSPICE simulation program for testing the simulation speed of ANFIS and HSPICE. The comparison between
ANFIS and HSPICE indicates the feasibility and accuracy of the proposed neuro-fuzzy modelling technique. The results
showed that the proposed ANFIS simulation has much higher speed and accuracy in comparison with HSPICE simulation.
The neuro-fuzzy modelling technique can be simply used in software tools for designing and simulation of the full-wave
rectifier circuit and the other electronic circuit.

Keywords: ANFIS; Full-wave rectifier circuit; Predict and analysis; Different MFs.
©Sila Science. All rights reserved.

1. Introduction

AC/DC rectifiers are the first group circuits of the power switching applied in industrial fields. For DC power
supply devices Mercury-arc rectifiers were very favoured in the 1940s. Semiconductor material advancement
brought power devices, such as silicon controlled rectifier, power diode, Triac, insulated gate bipolar transistors
(IGBT) and metal oxide semiconductor field effected transistor (MOSFET), bipolar transistor (BT) and so on,
into the DC power supply in 1960s. The mathematical modelling of all AC/DC rectifiers is developed and these
rectifiers are widely used due to their durable and simplicity in worldwide. Several studies trying to full-wave
rectifier study in the literature. The majority of these studies present the simulations and experimentally
behaviour of the single-phase rectifier [1–4]. In previous studies a single-phase rectifier was experimentally
constructed in the laboratory. Experimental tests were performed to characterize its absorbed magnitude and
phase angle of harmonic currents [5]. A novel semi controlled switch mode rectifier (SMR) topology which is
proposed is analysed and showed some theoretical key operating wave forms of the SMR unit [6]. AC/DC
conversion is an obtained using a full-wave MOS rectifier with internal peak selection circuitry. A full-wave
rectifier has been fabricated in 0.35 m CMOS to rectify the low-frequency AC voltage from harvested vibration
energy [7]. A two-input vector circuit which is used the full-wave rectifier has been fabricated in an o, 8 µm
CMOS process [8]. The experimental and simulation results confirm the feasibility of the proposed full-wave
rectifier circuits. Proposed and realized a novel precision full-wave rectifier using an all-pass filter as a 90◦ phase
shifter [9]. The circuit gives a dc output voltage that is almost the same as the peak input voltage over a
frequency range of 50 Hz–1 MHz with a very low ripple voltage and low harmonic distortion.

____________
*
Corresponding author: Tel.: +90-432-388-2164 or +90 543 834 5049; fax: +90-432-388-2165.
E-mail address: [email protected] (R. Tuntas).
540 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548

Artificial neural networks (ANN) and fuzzy logic (FL) is used increasingly in various fields of engineering.
ANFIS is commonly used in complex system studies for modelling, control or parameter predicting. Fuzzy
modelling or fuzzy identification, first researched systematically by takagi and sugeno has found numerous
practical applications in control estimations and inference [10]. Functionality of the neural network of biological
neurons can be described as a simple processing elements based on an interconnection [11, 12]. There are simple
rules of fuzzy systems and are useful for simplifying the learning process [13]. The neuro-fuzzy architectures
developed with the combination of ANN and FL [14]. Hayati et al. have applied modelling and simulation of
combinational CMOS logic circuits by ANFIS [15]. An expert system for air flow control of HVAC system
based on ANFIS is presented [16]. Modelling of a 1MBI 150NH-060 Typed IGBT Device with Fuzzy and
Neural Network Approach is obtained by Remzi Tuntas [17]. Modelling and simulation of transistors and
circuits used the ANN [18, 19]. A fuzzy dynamic characteristic modelling and adaptive control method is
proposed by Li et al. [20].
In circuit design, the high speed simulation of circuit in the conditions in which the accuracy remains in
acceptable range is important. The purpose of simulators is to simulate the circuits quickly and accurately. Many
methods and parameters are used for modelling and simulation of rectifier circuits. The calculations of these
parameters in addition are used to other time-consuming calculations and the total increase in simulation time.
ANFIS uses a hybrid-learning algorithm to determine parameters of Sugeno-type fuzzy inference systems. And it
uses a combination of the least-squares method and the back propagation gradient descent method for training
membership function parameters to simulate a particular set of training data.
In this paper, we have proposed a simulation method that does not need these calculations directly. We have
used adaptive neuro-fuzzy inference system architecture based on the Takagi–Sugeno model for modelling of
rectifier circuits as a tool for designing and simulation of complex rectifier circuits in order to improve accuracy
and simulation time.
The rest of the paper is organized as follows. Section 2 provides a brief overview on full-wave rectifiers with
centre-tapped transformer; Section 3 presents the structure of ANFIS. The Root Mean-Square Error (RMSE)
which is a degree of distribution of the data and values of the coefficient of determination (R2) are described in
Section 4. The parameter values of electronic circuit established as experimental are given in section 5. The
simulation results of the proposed methodology are demonstrated in Section 6. Finally Section 7 reports
discussion and conclusion.

2. Full wave rectifiers

There are two types of full-wave rectifiers with centre-tapped transformer and bridge rectifiers. The single-
phase full-wave rectifier converts an alternative voltage source to the a dc source. A full wave rectifier converts
the all of the input signals to one of constant positive or negative value at its output by reversing the negative or
positive sections of the ac waveform. The outputs of the two half-wave rectifiers are combined to produce full-
wave rectification in the load. The Full-wave rectifier circuit is shown in Fig. 1(a). And the voltage waveforms
of the full-wave rectifier are shown in Fig. 1(b) [21]. The performance of the rectifiers is shown in terms of the
following parameters.

Fig. 1. (a) Full-wave rectifier with centre-tapped transformer (b) voltage waveforms of the full-wave rectifier the
average value of the load voltage VL is V and it is defined as.
dc
1T
V   V  t  dt (1)
dc T L
0
 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548 541

In the case of a full-wave rectifier, Fig. 1(b) indicate that VL(t )  V sin wt for both the positive and
m
negative half-cycles. Hence Eq. (1) can be re-written as
1
V   V sin wt d  wt  (2)
dc  m
0
Therefore full-wave,
2V
V  m  0.636 V (3)
dc  m
The root-mean-square (rms) value of load voltage VL is V , which is defined as
L
1
1 T 2  2
V    V  t  dt  (4)
L T L 
 0
In the case of a full-wave rectifier, VL(t )  V sin wt for both the positive and negative half-cycles.
m
Hence Eq. (4) can be re-written as

1
 
2
V   V sin wt d  wt  (5)
L 0 m
V
V  m  0.707 V (6)
L 2 m

Rectification process is commonly performed by semiconductor diodes. Full-wave rectification uses two
diodes with one for each half of the cycle, a centre-tapped transformer and a capacitor. Transformers are
machines making the desired change in the voltage and current values at specific power of electric energy. It is
an electrical circuit element that transfers energy from a primary winding to one or more coupled windings it
through the electromagnetic field. Another important part of an ac to dc converter is a diode. A diode is to permit
an electric current to pass in diode's forward direction, while blocking current in the reverse direction. Thus, the
diode is used to convert alternating current to direct current.

3. ANFIS architecture

ANFIS is an adaptive network which allows the use of fuzzy logic together with neural network [22, 23].
ANFIS uses the learning capacity of ANN to determine the input–output relationship and build the fuzzy rules
by defining the inputs structure. ANFIS networks have been implemented such as to rule-based process controls,
classification, pattern recognition problems and modelling. ANFIS is adaptive network that operates like
adaptive network simulator of fuzzy controllers of Takagi–Sugeno. The equivalent of this adaptive network is
fuzzy inference system (FIS). The non-linear and linear parameters are adjusted by ANFIS by using back
propagation gradient descent and least squares type of method for a given input output data set [24–27]. The
type-3 fuzzy inference system used in this study is proposed by Takagi and Sugeno [28].
The five-layered characteristic ANFIS structures have m inputs (x1, xm) each with n membership functions
(MFs), R rules and one output (f). If we suppose that fuzzy inference system has two inputs (x, y) and one output
(f), a first order Sugeno model can be described with two rules as follows:
Rule 1: If x is A1 and y is B1, then f1 =p1x+q1y+r1,
Rule 2: If x is A2 and y is B2, then f2 =p2x+q2y+r2.
The associated inference method is shown in Fig. 2. The output of each rule is a linear combination obtained
by adding a constant term to the input variables. The final output of this inference system is the average of
weighted of each rule output. The structure of ANFIS is shown in Fig. 3.
The individual layers of this ANFIS structure are summarized as follows:
Layer 1: Every node i in this layer is adaptive with a node function. Outputs of this layer is given by,

O1   ( x) (7)
i Ai
542 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548

where O1 is the output of the node i in the first layer, x is the input to node i, Ai the linguistic term associated
i
with this node function and µAi is the membership function for Ai fuzzy sets, Usually μAi (x) is chosen as a bell-
shaped, such as

Fig. 2. The inference method of the Sugeno model.

Fig. 3. The ANFIS architecture based on Takagi–Sugeno.

1
 ( x)  (8)
Ai
2 bi


1  x - c / a 

i i 
Or
  
2
 x  ci 
 ( x)  exp     (9)
  ai 
Ai
 
 
Where x is the input and {ai, bi, ci} is the parameter set. Parameters in this layer are referred to as premise
parameters.
Layer 2: Every node in this layer is a fixed node that calculates the firing strength wi of a rule. The output of
each node is the product of all the incoming signals to it and is given by,
O2  w    x  x  y  , i  1, 2 (10)
i i Ai Bi
Layer 3: Each node in this layer is a fixed node. Every i the node calculates the ratio of the i the rule’s firing
strength to the sum of firing strengths of all the rules. The output from the i the node is the normalized firing
strength given by,
w
O3  w  i , i  1, 2 (11)
i i w w
1 2
Layer 4: Every node in this layer is an adaptive node with a node function given by
 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548 543

i i i i i i 
O4  w f  w p x  q y  r , i  1, 2
i  (12)
Where wi is the output of Layer 3 and {pi, qi, ri} is the consequent parameter set.
Layer 5: This layer contains of only one fixed node that calculates the overall output as the summation of all
incoming signals, i.e.
i w f
O5  overall output   w f  i i (13)
i
i
i i i w
i

Terms of the proposed ANFIS structure the values of premise parameters are given. The final output can be
phrased as a linear combination of the consequent parameters. The output f in Fig. 3 can be written as

w w
f 1 f  2 f w f w f
w w 1 w w 2 11 2 2 (14)
1 2 1 2
  1 1      
 wx p  wy q  w r  w x p  w y q  w r
1 1 1 1 2 2 2 2 2 2    
where f is linear in the result parameters (p1, q1, r1, p2, q2, r2). Result parameters are identified by the least
squares estimate, in the forward pass of the learning algorithm. In the backward pass, the error signals are
propagated backward from the output layer to the input layer. In this backward pass, the premise parameters are
updated by the gradient descent algorithm [29, 30].

4. Performance measurements

The Root Mean-Square Error (RMSE) which is a degree of distribution of the data and values of the
coefficient of determination (R2) are used to compare the predicted and actual values for model validation. The
RMS and R2 are formulated by Eq. (15) and Eq. (16), respectively [16].

n
m  1( y t )
pre, m mea, m 2 (15)
RMS 
n
n
m  1( y t )2 (16)
pre, m mea, m
R2  1 
n
 m  1(t )2
mea, m

where n is the number of data patterns in the independent data set, ypre,m indicates the predicted, tmea,m is the
measured value of one data point and m is the mean value of all measured data points.

5. Experimental

We have used the real hardware of the full-wave rectifier circuit in the Fig. 4 for generating the training and
testing data. It consists of two diode (1N4001), a sinusoidal AC voltage source, a three windings linear
transformer, an inductance, a capacitance and a resistance. The parameter values of the circuit of Fig. 4 were Vi
= 50 V, Vf = 0.7 V, f = 50 Hz, L = 0.2 H, C = 0.0005 F and R = 10 Ω.

Fig. 4. The full-wave rectifier circuit.

544 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548

6. Modelling and simulation of full-wave rectifier circuit

A hybrid learning algorithm [31] used for ANFIS training consists the learning of the premise parameters by
back-propagation and the learning of the consequence parameters by least-squares estimation. In the forward
transition, an input is given to the ANFIS and the consequent parameters are updated by the least-squares
estimation, and in the backward transition the premise parameters are updated by the gradient descent. The loop
is repeated until a desired error is reached.
In this study, three separate ANFIS model (ANFIS model I, ANFIS model II and ANFIS model III) was
performed to predict the model of the full-wave rectifier circuit in Fig. 1. Each ANFIS model has three inputs
and one output. The sinusoidal input signal, a past time value of input signal and a past time value of the output
signal were used as the inputs {x (t), x (t-1), y (t-1)} of the each ANFIS model. Furthermore, respectively the
diode current ID1 for ANFIS model I, the diode current ID2 for ANFIS model II and the load output voltage Vload
for ANFIS model III were used as the output signal of the ANFIS models. The training and testing data for the
ANFIS models were obtained by using the real hardware experimental data. The block diagram of developed
modelling structure is showed on Fig. 5. After training processes were successfully completed, the predicting
output dynamic curves of this full-wave rectifier circuit were obtained. For this purpose, these ANFIS models
were trained in MATLAB 7.10.0.499 program. The existing experimental data set consists of two sets, one to be
used for training of the models (75% of the data), and the rest for testing the performance. The total 400 input–
output data pair was used for the training and testing of the ANFIS models. The 300 pairs of these data were
used for stages of training and the remaining 100 pairs were used for stages of test.

Fig. 5. The block diagram of developed modelling structure.

For each input, it was used four membership function. The each ANFIS model was trained for 200 epochs
using different MFs, which was given in Table 1. Their performances were compared in terms of the RMSE and
R2 values for the training and testing data sets of the three separate ANFIS model in Table 2, 3 and 4,
respectively.

Table 1 Membership functions

Type of MF Descriptions
GAUSSMF Gaussian curve membership function.
GAUSS2MF Two-sided Gaussian membership function Pi-shaped curve membership function.
PIMF Triangular membership function.
TRIMF Generalized bell curve membership function.
GBELLMF

Table 2. Predicting performance of ANFIS model I with different MFs for the current of diod1 (ID1) of the
full-wave rectifier circuit
Membership functions For the training data sets For the testing data sets
RMS R2 RMS R2
Gaussmf 6.0689e-005 0.999999 0.000118 0.999999
Gauss2mf 7.7141e-005 0.999999 0.003241 0.999971
Pimf 2.2934e-005 0.999999 0.000706 0.999991
Trimf 5.3739e-005 0.999999 0.000146 0.999998
Gbellmf 3.3396e-005 0.999999 0.000327 0.999995
Table 3 Predicting performance of ANFIS model II with different MFs for the current of diod1 (ID2) of the full-wave rectifier circuit
Membership functions For the training data sets For the testing data sets
RMS R2 RMS R2
Gaussmf 1.8065e-005 0.999999 0.000793 0.999992
Gauss2mf 2.6343e-005 0.999999 0.007012 0.999794
Pimf 6.7720e-006 0.999999 0.003603 0.999976
Trimf 2.7525e-005 0.999999 0.004412 0.999983
Gbellmf 3.0904e-005 0.999999 0.002204 0.999945
 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548 545
Table 4. Predicting performance of ANFIS model III with different MFs for the output voltage (Vload) of
the full-wave rectifier circuit
Membership functions For the training data sets For the testing data sets
RMS R2 RMS R2
Gaussmf 2.9686e-005 0.999999 0.001210 0.999988
Gauss2mf 3.4140e-005 0.999999 0.003204 0.999973
Pimf 3.5474e-005 0.999999 0.013023 0.999856
Trimf 2.0892e-005 0.999999 0.002506 0.999981
Gbellmf 2.2036e-005 0.999999 0.002112 0.999984

The best performance was obtained by using Gaussian curve membership function. The RMSE values is
0.000118, 0.000793 and 0.001210 and the R2 values is 0.999999, 0.999992 and 0.999988 for ANFIS model I,
ANFIS model II and ANFIS model III, respectively. The structure, architecture and training parameters of
ANFIS models are presented in Table 5 for Gaussian curve membership function.

Table 5. the structure, architecture and training parameters of ANFIS models for Gaussian curve
membership function (gaussmf).
Architecture and training Values ANFIS parameters Values
parameters
The number of layers 5 Number of nodes 158
The number of input 3 Number of linear parameters 256
The number of output 1 Number of nonlinear parameters 24
The number of MFs 4 Total number of parameters 280
Type of input MFs Gaussian Number of training data pairs 300
Activation functions Log-sigmoid Number of fuzzy rules 64
Learning rule Hybrid learning algorithm (back-propagation for nonlinear
parameters and Least square errors for linear parameters)

Then, HSPICE were used as a general simulator for comparing and testing the ability of proposed simulation
method. Four various data sets that were shown in Table 6 were used for testing the simulation speed of ANFIS
model III and HSPICE.

Table 6. the comparison between HSPICE and ANFIS model III simulation time for ALU
Number of data Simulation time (s) Speed improvement
HSPICE ANFIS
800 43 1.02 42.6 times
2500 139 2.11 65.9 times
4000 198 2.95 67.1 times
5000 235 3.37 69.7 times

The obtained results show that the proposed simulation method is much faster than HSPICE simulation. The
better and faster results are obtained with increasing number of data as shown in Table 6.
Furthermore the mean percentage error comparisons of HSPICE, ANFIS, and real hardware data have been
done for 75 random data sets. The results of this comparisons show that the mean percentage error between
HSPICE and real hardware is 6.13 % and the mean percentage error between ANFIS and real hardware is 2.4 %.
In this case, the harmony between ANFIS and real hardware is much better than the harmony between HSPICE
and real hardware as shown on the same graph in Figs. 6-8.
The current wave curve of the D1 diod
0.8
Actual Output
0.7 ANFIS Predicted Output
HSPICE Simulation Output
0.6

0.5
ID1 (Amper)

0.4

0.3

0.2

0.1

-0.1
0 20 40 60 80 100
Sample Number

Fig. 6. The ANFIS predicted output for testing data, actual output and HSPICE simulation output for the ID1
diode output current of the full-wave rectifier circuit.
546 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548

The current wave curve of the D2 diod

0.8
Actual Output
0.7 ANFIS Predicted Output
HSPICE Simulation Output
0.6

0.5

ID2 (Amper)
0.4

0.3

0.2

0.1

-0.1
0 20 40 60 80 100
Sample Number

Fig. 7. The ANFIS predicted output for testing data, actual output and HSPICE simulation output for the ID2
diode output current of the full-wave rectifier circuit.

The output voltage of the full-wave rectifier circuit

8

5
Vload (Volt)

1
Actual Output
0 ANFIS Predicted Output
HSPICE Simulation Output
-1
0 20 40 60 80 100
Sample Number

Fig. 8. The ANFIS predicted output for testing data, actual output and HSPICE simulation output for the V load
output voltage of the full-wave rectifier circuit.

From these results, it is clear that the behaviour of the full-wave rectifier circuit is approximated very
accurately by the obtained ANFIS models, so we can apply this model as a simulator in design and simulation of
rectifier circuits. The results are directly dependent on computer processing capability. For calculations and
simulations, the characteristics of the processor used in this study are given as Intel Core i5, 3100 MHz, 4 GB,
DDR3, HDD, 500 GB, windows XP PRO SP3.

7. Conclusion

In this paper, an ANFIS approach for the predicting and simulation of full-wave rectifier circuit was
presented. The developed ANFIS models were applied to full-wave rectifier circuit. For this purpose a computer
program was developed by using MATLAB programming language. The experimental data were obtained from
real hardware in order to train the ANFIS models. The input–output data sets of this circuit were first stored and
then these data sets were used to obtain its intelligent model based on ANFIS. The ANFIS models were trained
and tested for different MFs. Then, the best ANFIS models were obtained by comparing the performances of
MFs. Efficiency of the developed neuro-fuzzy modelling technique was tested and a mean 99.999% recognition
success was obtained. Then, for comparing and testing the ability of our proposed simulation method, we used
HSPICE as a general simulator. The proposed ANFIS model was tested through rectifier circuits, to compare
with experimental data and HSPICE simulation. Results of obtained showed that the ANFIS model simulation is
 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548 547
much faster than, HSPICE simulation. This study showed that the values of predicted with the ANFIS can be
used to predict full-wave rectifier circuit quite accurately. Therefore, faster and simpler solutions can be
achieved with ANFIS.

References

[1] Gift S J G, Maundy B. Versatile precision full-wave rectifiers for instrumentation and measurement. IEEE
Trans Instrum Meas 2007;56:1703–1710.
[2] Djukic SR. Full-wave current conveyor precision rectifier. Serbian J Elect Eng 2008;5:263–271.
[3] Toumazou C, Lidgey FJ, Chattong S. High frequency current conveyor precision full-wave rectifier.
Electronics Leiters 1994;30:745–746.
[4] Song TT, Nianci H. A novel zero voltage and zero current switching full bridge PWM converter. IEEE
APEC 2003;2:1088–1092.
[5] Sainz L, Pedra J, Mesas J J. Single-phase full-wave rectifier study with experimental measurements.
Electric Power Systems Research 2007;77:339–351.
[6] Manias S. Novel full bridge semicontrolled switch mode rectifier. IEE Proceedings-b 1991;138:252–256.
[7] Guilar NJ, Amirtharajah R, Hurst P J. A Full-Wave Rectifier With Integrated Peak Selection for Multiple
Electrode Piezoelectric Energy Harvesters. IEEE J Solid-State Circuit 2009;44:240–246.
[8] Chang CC, Liu SI. Current-mode full-wave rectifier and vector summation circuit. Electronics Letters
2000;36:1599–1600.
[9] Sahu PP, Singh M, Baishya A. A Novel Versatile Precision Full-Wave Rectifier. IEEE Transact Instrument
Measurement 2010;59:2742–2746.
[10] Tagaki T, Sugeno M. Fuzzy identification of systems and ıts applications to modelling and control. IEEE
Trans Systems 1985;15:116–132.
[11] Haykin S. in: Neural Networks, A Comprehensive Foundation, Macmillan College Publishing Company,
New York, 1994.
[12] Hagan MT, Beale HB. in: Neural Network Design, PWS Publishing Company, Boston, 1996.
[13] Takagi T, Sugeno M. Fuzzy identification of systems and its applications to modeling and control. IEEET
Transact Syst Man Cybern 1985;15:116–132.
[14] Jang J S R, Sun C T. Neuro-fuzzy modeling and control. Proceedings of the IEEE SpecialIs sueon Fuzzy
Logic in Engineering Applications 1995;83:378–406.
[15] Hayati M, Rezaei A, Seifi M, Naderi A. Modeling and simulation of combinational CMOS logic circuits by
ANFIS. Microelectron J 2010;41:381–387.
[16] Soyguder S, Alli H. An expert system for the humidity and temperature control in HVAC systems using
ANFIS and optimization with Fuzzy Modeling Approach. Energy Build 2009;41:814–822.
[17] Tuntas R. Modeling of a 1MBI 150NH-060 Typed IGBT Device with Fuzzy and Neural Network
Approach, International Symposium on Innovations in Intelligent Systems and Applications, Yildiz
Technical University in Istanbul, Turkey, 2005, 116–119.
[18] Andrejevic M, Litovski V. Electronic circuit modeling using artificial neural network. J Autom Control
2003;13: 31–37.
[19] Djeffal F, Chahdi M. An approach based on neural computation to simulate the nanoscale CMOS circuits:
application to the simulation of CMOS inverter. Solid-State Electron 2007;51:48–56.
[20] Li H B, Sun Z Q, Min H B, et al. Fuzzy dynamic characteristic modeling and adaptive control of nonlinear
systems and its application to hypersonic vehicles. Sci China Inform Sci 2011;54:460–468.
[21] Lee YS, Chow MHL. Diode Rectifiers. Power Electronics Handbook (Third Edition), Department of
Electronic and Information Engineering, The Hong Kong Polytechnic, University Hung Hom, Hong Kong,
pp. 149-181, 2011.
[22] Turkmen I, Guney K. Computation of association probabilities for single target tracking with the use of
adaptive neuro-fuzzy inference system. Turkish J Electrical Eng Comput Sci 2005;13:105–118.
[23] Culliere T, Titli A, Corrieu J M. Neuro-fuzzy modelling of nonlinear systems for control purposes, in:
Proceedings of the IEEE International Conference on Fuzzy Systems 1995;4:2009–2016.
[24] Jang J R. ANFIS: adaptive-network-based fuzzy inference system. IEEE Trans Syst Man Cybern
1993; 23: 665–685.
[25] Mashrei M A, Abdulrazzaq N, Abdalla T Y, et al. Neural networks model and adaptive neuro-fuzzy
inference system for predicting the moment capacity of ferrocement members. Eng Struct
2010;32:1723–1734.
[26] Rojas I, Bernier J L, Alvarez R R, et al. What are the main functional blocks involved in the design of
adaptive neuro-fuzzy inference systems? in: IEEE-INNS-ENNS Conf Neural Netw 2000;6:551–556.
548 R. Tuntas / EEST Part A: Energy Science and Research Special Issue (2012) 539-548

[27] Abdelrahim E M, Yahagi T. A new transformed input-domain ANFIS for highly nonlinear system modeling
and prediction, in: Electrical and Computer Engineering, Canadian Conf 2001; l: 655–660.
[28] Jang J S R. ANFIS: Adaptive-Network-based Fuzzy Inference Systems. IEEE Transact Syst Man Cybern
1993;3:665–685.
[29] Qian N. On the momentum term in gradient descent learning algorithms. Neural Network 1999;12:145–151.
[30] Shoorehdeli MA, Teshnehlab M, Sedigh AK, Khanesar MA. Identification using ANFIS with intelligent
hybrid stablelearning algorithm approaches and stability analysis of training methods. Appl Soft Comput
2009;9:833–850.
[31] Chiu S. Fuzzy model identification based on cluster estimation. J Intelligent Fuzzy Syst 1994;2:762–767.