COMPARATIVE STUDY BETWEEN FUZZY LOGIC AND ARTIFICIAL NEURAL NETWORK (ANN) ALGORITHMS FOR SINGLE PHASE SHUNT ACTIVE POWER FILTERS (SAPFs

Vol 63, No. 11;Nov 2013

377 Jokull Journal

COMPARATIVE STUDY BETWEEN FUZZY LOGIC AND ARTIFICIAL

NEURAL NETWORK (ANN) ALGORITHMS FOR SINGLE PHASE SHUNT

ACTIVE POWER FILTERS (SAPFs)

ALIYU SABO, NOOR IZZRI ABDULWAHAB, HAMISU USMAN, MOHD AMRAN MOHD RADZI,

NASHIREN FARZILAH MAILAH

Abstract-The excessive use of power electronics devices in industrial, commercial and residential

purposes have lead to the deterioration of supply current and voltage wave forms, injecting harmonic pollution

on to the supply system, which no doubt generates a major concern to power system engineers. Restricted

standards set by IEEE-519 or IEC-61000-3-2 necessitates that, those unwanted harmonic injected current (into

the utility networks) to fall below a specified range. Present day technological development, particularly in the

field of power converters, introduces the application of active power filters as a modern weapon for harmonic

current mitigation and reactive power compensation leading us to IEEE norms realization. This paper depicts

comprehensively, comparison in terms of total harmonic distortion (THD) and simulation results, the most

effectively fast response in harmonic mitigation between fuzzy logic based active power filter (APF) control and

its artificial neural network (ANN) controller counterpart. In both cases, results were developed via simulation

studies under MATLAB/SIMULINK environment.

Keywords: Shunt active power filter, harmonics, total harmonic distortion, artificial neural network,

fuzzy logic controller and power factor.

1. INTRODUCTION

Present day technological development, particularly in the field of power electronic offers remarkable achievements

in terms of its fast switching and high quality capability, less heavy in weight, operational flexibility, robustness,

size minimization etc. On the other hand, they also dispense harmonics in networks due to their nonlinearity nature.

Harmonics level minimization in the grid caused due to these development becomes a serious issue threatened

power system engineers particularly, the utility companies. Classical passive filters techniques are adopted during

the 20th

age, which are now seen as an armature technology as it adversely depends on networks impedance resulting

in an unwanted resonance issues. Active power filters which is adopted as the latest matured development in

handling effectively current harmonics mitigation, power factor optimization and reactive power compensation due

to its flexibility and robustness in handling many types of harmonic pollution agent’s devices (power electronic

devices) also providing quick response to load variations. This new trend finally abolishes the armature passive

filters usage and its drawbacks.

Various configurations of APF have been exploited together with its distinct control strategies in ensuring optimal

operating state [1][2][3]. Several controller techniques for shunt APF’s have been developed in the past [4][5]. In

[4][5][6], harmonic current extraction strategies can be realize via time-based, heterodyne, frequency-based,

instantaneous power compensation and pattern recognition and learning (where mainly neural network technique).

Similarly, previous work of the shunt APF presents an optimization algorithm in extracting current harmonics [4][7].

In general, shunt APF works on the principle of current harmonic injection. Its main job is to deliver compensated

current to the supply network at the common coupling point, as such, to cancel the harmonics current and reactive

power drawn by the nonlinear loads. SAPF configuration could be either voltage source inverter (VSI) or current

source inverter (CSI) types, depending on applications to be used. But (VSI) type is the most widely employed in

active filtering due to its simplicity and popular in recognition topology. Basically, shunt APF comprises the power

circuit and control circuit. The power circuit incorporates the IGBT or MOSFET semiconductors switching devices

with an interfacing inductor for taking care of the harmonics ripples after compensation by the active filter, DC

Vol 63, No. 11;Nov 2013

378 Jokull Journal

capacitor for maintaining DC voltage by the inverter. While the control circuit is the main brain of the filter, which

control the semiconductor switching gating signal for realization by the shunt APF. The depiction of the shunt APF

configuration is shown (fig 1).

Fig. 1 Single phase shunt APF configuration

In this paper, fuzzy logic and neural network algorithms in shunt APF control is been compared with results

obtained discussions to verify the most effective algorithm in shunt APF control for current harmonics mitigation.

Section 2 of this paper entails the control system development (2.1 for neural network and 2.2 for fuzzy logic

algorithms). Section 3 depicts the simulation outputs and lastly, section 4 dispenses the conclusion.

2. CONTROL TECHNIQUE

2.1 Neural network algorithm

This section explains the operation of the SAPF with adaptive/artificial neural network (ANN) control in harmonic

extraction. The algorithm is based on the concept of sum sine and cosine parts, attached each to an appropriate

coefficient which represents a signal periodically, as such estimates the polluted component. Load current

presented by fundamental and harmonic components as below (for each sample k in a digital operation) with △t as

the sampling time and w as the fundamental frequency.

With and as the amplitudes of the sine and cosine components of the measured nonlinear current, where n,

is the number of harmonics, up to N maximum number. In vectorial form,

(3)

With the weight matrix = and the sine and cosine vector

.

Vol 63, No. 11;Nov 2013

379 Jokull Journal

Fig. 2 conventional ANN extraction topology

Fig. 2 shows an ANN algorithm with a sole purpose of generating the right value of by training accurately .

The delta rule block is the brain of this extraction circuit (algorithm). The W-H minimizes the average square error

between the actual measured signal and the estimated signal written as

Where is the square of the vector and

The harmonic orders N, determines the weight matrix dimension presents a drawback of excess computation

and simulation time (to complete a single iteration of the computation). In the comparison aspect for the ANN, a

modified ANN algorithm is proposed with an advantage of mitigating the drawback presents by the above classical

algorithm.

Where, ,

, is the sampling time and alpha ( ) is the learning rate.

Fig. 3 A modified ANN extraction topology

Equation (5) above gives the algorithm (modified W-H algorithm) the liberty irrespective of N Present to alone

update the weights of the fundamental part. Mathematically, the elements are orthogonal for all, as such, the

modification is so much feasible and suitable, resulting in fast estimation and computation (enhancing the speed of

the iteration). A factor range 0 to 1 is incorporated. Fig. 3 above presents the configuration of the modified

proposed ANN algorithm which is far limited in complexity to its classical algorithm.

Vol 63, No. 11;Nov 2013

380 Jokull Journal

Challenge has been recorded in the implementation of this method as in a selection of a good alpha value ( ).

Previous works have been carried out which reports some suitable range of the factor ( ) [8] but, this paper skips the

learning rate value selection as appropriate and good simulation results with a preferred alpha value are recorded.

Recommendations for further study/work on a good range of alpha value would be beneficial. In the classical

version of the ANN algorithm (4), would follow harmonic order N if decided. However, in (5), only

contains the fundamental component as such, the alpha value is taken below 1/N which realizes an optimum

performance by the algorithm.

The nonlinear load currents fundamental sine component deducted from the nonlinear load current produces the

harmonic current .

This work further proposes a preferred value based on this two considerations: how quick the algorithm can

generates the current harmonics that will be treated as the reference signal by the pulse-width modulation (PWM)

in switching the gates of the VSI which produces the compensating current and how feasible the scheme can

generates and produce the harmonic current ) to minimizes total harmonic distortion (THD) . The complete

model of the simulation in MATLAB/SIMULINK environment using the modified ANN is shown in fig 4 below.

Fig. 4 Complete simulation model of the system

2.2 Fuzzy logic algorithm

2.2.1 Harmonic extraction technique

Basically, harmonic extractions can be achieved either in frequency domain or time domain approach. In this work,

a synchronous reference frame (SRF) a time domain was used for the extraction of the harmonic reference current to

the APF. The SRF is only applicable for three-phase system, which we modeled it as its three-phase under balanced

system condition and only one-phase was selected for the reference current harmonic extraction. In synchronous

reference frame, the distorted currents due to non-linear loads are transformed from a-b-c stationary reference frame

to its d-q reference frame. The SRF for the extraction of the harmonic signal is depicted in fig 5.

pwm

Discrete,Ts = 5e-005 s.

powergui

Vc

VS

v+-

VM3

v+-

VM2

v+-

VM1

VL

g

A

B

+

-

UB2

A

B

+

-

UB1

Subtract

Conn2Conn1

Subsystem1

In1

In2

Out1

Subsystem

RL R1

Uref Pulses

PWMG

0.003

LRL

ISIL

IH1

-1

Gain1

-1

Gaini

+-

CM2

i+

-

CM1

C2

C1

ACVS

Vol 63, No. 11;Nov 2013

381 Jokull Journal

Fig 5 SRF for extraction of harmonic

2.2.2 Fuzzy Logic Controller

Fuzzy logic controller (FLC) was first introduced by Professor Zadeh Lotfi of University California Berkeley in

early 1960s. He proposed a technique of how to process an imprecise data with complex input. The idea of Zadeh’s

was fully utilized after the introduction and availability of modern computers and controllers applications. Fuzzy

logic controllers gain interest by many researchers and system engineers in the application of control system analyst,

as well as in the control algorithm for shunt APF applications. FLC has advantages of simplicity in design

procedures that does not require any accurate mathematical modeling, can work with an imprecise input of the

system and it can also work with non-linearity. Fuzzy controller is very robust than classical controllers such as PI

and PID controllers [5]. In our study, a Mamdani fuzzy controller was chosen and designed with linguistic term “if

then” rules. The linguistic variables for the rule base was selected as, positive large (PL), positive medium (PM),

positive small (PS), negative large (NL), negative medium (NM), negative small (NS) and zero (Z). However, Fig. 2

depicts the general structure of FLC.

The design of the FLCis characterize as follows.

1. Five memberships function for each two inputs error (e) and its derivative (Δe) are used.

2. Seven memberships function for the output.

3. Mamdani implication in the design was also used.

4. Centre of area (COA) was used for the Deffuzzification process.

5. Implication using “min-max” approach

6. twenty five rules for both error and change of error are used.

Triangular membership functions are used due to its simplicity, completeness and easy to implement.

Fig. 6 Structure of FLC

Vol 63, No. 11;Nov 2013

382 Jokull Journal

As it was defined for the linguistic rule variables, table 1 below shows the “if then” rules for the five membership

functions selected for each of the input error (e) and the change of error (Δe). For the two inputs, only twenty five

possible rules are possible based on (5*5) =25 “if then” combinations rules.

Table 1 Linguistic variable rules for the FLC

Complete simulink model of the single-phase shunt APF is shown in fig 7. Parallel RC load with full bridge rectifier

are employed for the non-linear load which are the main source of harmonic producing load due to their non-

linearity in nature.

Fig. 7 Complete simulink model of single-phase shunt active power filter

3. SIMULATION RESULTS

3.1 ANN results and discussions

The performance and effectiveness of the modified ANN algorithm is discussed below, with the single phase shunt

APF system was tested and observed by simulation studies in MATLAB SIMULINK. The simulation’s sole aim

was to be accurately chosen at some ranges which will be implemented experimentally. The nonlinear load was

built by a series resistor (RS = 50Ω) and inductor (LS = 3mH) with a diode rectifier connected to a capacitor (CL =

470μF) and a resistive load (RL = 15Ω). The closed loop system has the following parameters (L = 1mH), r = 0.1Ω,

C = 1000μF, VS = 100V, 50HZ, Vdc = 150V, fswitching = 5 kHz. Discrete solver was used which performed the

simulation digitally at time sample Ts= 50μs.

Initially, the circuit was simulated without connecting the SAPF which records a THD of 16.59%. Chosen alpha

values 0.0009 to 0.001 at the SAPF presence is considered.THD measurements at distinct values of alpha are

reported as in table I above. The learning rate’s good chosen value was 0.004 (as shown in the table below) with a

THD measured within an acceptable range. Smaller value could reduce THD but, dispenses less fast increase in

for each k sample. These selection was made as to help the scheme produce a good and stable quicker and also

make it obeyed the IEEE-519-1992 THD% requirements.

Vol 63, No. 11;Nov 2013

383 Jokull Journal

TABLE I. THD SIMULATION RESULT FOR SPECIFIED ALPHA VALUES ( )

LEARNING RATE (α) THD MEASUREMENT (%)

0 (without SAPF) 33.59%

3.84%

3.87%

3.92%

3.83%

3.94%

4.06%

4.06%

4.06%0.0008

0.0009

0.001

0.002

0.003

0.004

0.005

0.006

The waveform of the current harmonics extracted by the ANN algorithm via a good chosen alpha value (0.004) is

presented below (fig. 8). Similarly, fig. 9 presents the simulated waveform for the supply voltage Vs (a), the supply

current IS (b) and the nonlinear load current IL (c) respectively all at 0.004. From the outputs (Table I and Fig. 5), the

THD was 3.92% with a stable , resulting in compensated to nearly in phase with VS and as such achieving a

nearly maximum PF.

Fig. 8 Harmonic extraction by ANN algorithm at 0.004 learning rate

(a)

(b)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

time in second (S)

Curr

ent

in (

A)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-100

-80

-60

-40

-20

0

20

40

60

80

100

time in second (S)

Voltage in (

V)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

time in second (S)

Cur

rent

in (

A)

Vol 63, No. 11;Nov 2013

384 Jokull Journal

(c)

Fig. 9 Simulated waveforms for the source voltage VS (a), the source current IS (b) and the load current IL (c)

respectively, all at 0.004 learning rate.

Fig. 10 Harmonic spectrum of the source current IS at 0.004 learning rate

Fig. 10 presents the graph of harmonic distributions of compensated source current via chosen alpha value

(0.004). The FFT clearly indicates the real simulation results (THD %) conforming to IEEE-519 harmonic

recommendations.

3.2 Fuzzy logic results and discussions

The parameters of the simulation model are as follow: VS= 25V peak amplitude, F= 50HZ, RS= 40Ω, CL= 500µF,

RL= 150Ω, RF= 40Ω, LF= 1mH. In the simulation, RC parallel load were fed from the diode rectifier as a nonlinear

load. In the same vein, the simulation results are also presented in this section. AS shown in Fig. 11, the result of the

nonlinear load due the harmonic produced by the rectifier shows a waveform which is opposite to the source

voltage. Fig. 12 depicts the filter current which also distorted due to the same nonlinear load, while Figure 13 and

Figure 14 are the corresponding source voltage and source current respectively and they seems to be purely

sinusoidal and in phase to each other. In the same analysis, Figure 15 and Figure 16 indicate the FFT analysis of the

THD which are also found to be 39.82% and 3.91% respectively. These THD have drastically reduced and is within

the recommended IEEE 519-1992 harmonic limits.

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-1.5

-1

-0.5

0

0.5

1

1.5

time in second (S)

Curr

ent

in (

A)

0 0.02 0.04 0.06 0.08 0.1

-1

0

1

Selected signal: 5 cycles. FFT window (in red): 1 cycles

Time (s)

0 1000 2000 3000 4000 50000

0.5

1

1.5

2

2.5

3

Frequency (Hz)

Fundamental (50Hz) = 1.871 , THD= 3.83%

Mag (

% o

f F

undam

enta

l)

Vol 63, No. 11;Nov 2013

385 Jokull Journal

Fig. 11 Load current before applying shunt active power filter

Fig. 12 Filter current

Fig. 13 Source Voltage

Fig. 14 Source Current after compensation

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

time in Second (S)

Curr

ent

in A

mps (

A)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

time in Second (S)

Curr

ent

in A

mps (

A)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-25

-20

-15

-10

-5

0

5

10

15

20

25

time in Second (S)

Voltage in V

olt (

V)

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

time in Second (S)

Curr

ent

in A

mps (

A)

Vol 63, No. 11;Nov 2013

386 Jokull Journal

Fig. 15 FFT spectrum of the load current before compensation

Fig. 16 FFT spectrum of the load current after compensation

4. CONCLUSION

This paper depicts comprehensively, comparison in terms of total harmonic distortion (THD) and simulation results,

the most effectively fast response in harmonic mitigation between fuzzy logic based active power filter (APF)

control and its artificial neural network (ANN) controller counterpart. In both cases, results were developed via

simulation studies under MATLAB/SIMULINK environment as shown and discussed above. At first, the shunt APF

(as an alternative and sophisticated tool in mitigating current harmonics) optimal performance is justified. A

modified ANN was effectively developed in detecting harmonic components and a PWM was also implemented in

generating switching strategies for the filter. A novelty was introduced (a learning rate factor) which is chosen

between 0 and 1 (in this case, good and suitable alpha value was realized) producing a perfect result as the

simulation outputs reports. For the fuzzy logic algorithm, THD for the source current with and without shunt active

power filter obtained to be 3.91% and 39.82% respectively. However, the corresponding total harmonic distortion

for the load voltage and source voltage found to be 18.77% and 0% respectively. This shows that, the fuzzy logic

controller have shown its capability in mitigating harmonic current and voltage produced by non-linear loads, with

the reactive power of the system. Results obtained for THD found to be within the required IEEE 519-1992

harmonic standard limit for both algorithms. It is clearly seen from the results that, the ANN algorithm reports less

Vol 63, No. 11;Nov 2013

387 Jokull Journal

significant THD than the fuzzy logic due to the modification of learning rate (which can be set to the rate at which

delivers the lowest THD) and also, owing to its quick adaptation to load variations with less time consumption and

simplicity in operation.

REFERENCE

[1] H. Komurcugil, “Hysteresis current-control strategy for single-phase half-bridge shunt active power filters,”

in 2013 12th International Conference on Environment and Electrical Engineering, 2013, pp. 35–40.

[2] E. B. Majid Yavari, “A New Strategy to Control Three-Phase Shunt Active Filters under Balanced and

Unbalanced Conditions by Controlling One Phase Current,” in 4th Power Electronics, Drive Systems &

Technologies Conference (PEDSTC2013), Feb 13-14,2013, Tehran, Iran, 2013, no. 1, pp. 529–534.

[3] B. Pal, Y. ; Swarup, A. ; Singh, “Comparison of Three Control Algorithms for Single-Phase UPQC,” in

Energy, Automation, and Signal (ICEAS), 2011 International Conference on, 2011, pp. 7–11.

[4] M. Amran, M. Radzi, N. A. Rahim, and S. Member, “Neural Network and Bandless Hysteresis Approach to

Control Switched Capacitor Active Power Filter for Reduction of Harmonics,” IEEE TRANSACTIONS ON

INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY 2009, vol. 56, no. 5, pp. 1477–1484, 2009.

[5] I. Sahu and D. A. Gadanayak, “Comparison between Two Types of Current Control Techniques Applied to

Shunt Active Power Filters and Development of a Novel Fuzzy Logic Controller to Improve SAPF

Performance,” in International Journal of Engineering Research and Development, 2012, vol. 2, no. 4, pp.

1–10.

[6] S. K. Ram and B. B. Das, “Comparison of different control strategy of Conventional and Digital Controller

for Active Power Line Conditioner (APLC) For Harmonic Compensation,” in 2013 12th International

Conference on Environment and Electrical Engineering, 2013, pp. 209–214.

[7] S. Golestan, M. Monfared, J. M. Guerrero, and S. Member, “A Novel Method for Extraction of the

Reference Compensating Current for Single-Phase Shunt Active Power Filters,” in 2011 International

Conference on Electrical Engineering and Informatics 17-19 July 2011, Bandung, Indonesia, 2011, no. July,

pp. 0–4.