2nd International Conference on Advances in Computer Science and Engineering (CSE 2013)

Power Quality Improvement by Shunt Active Performance Filters Emulated by

Artificial Intelligence Techniques
Radek Martinek, Jakub Manas, Jan Zidek and Petr Bilik
Department of Cybernetics and Biomedical Engineering
Faculty of Electrical Engineering and Computer Science, VSB Technical University
Ostrava, Czech Republic
e-mail: {radek.martinek; jakub.manas;jan.zidek; petr.bilik}@vsb.cz
abilities are used by the authors of this article to predict the
changes of current in supply networks which are caused by
connecting of non-linear loads. The article is primarily
focused on systems which use the combination of fuzzy
system techniques [4] and the artificial intelligence [3]. The
authors deal with use of Adaptive Neuro-fuzzy Interference
system based on ANFIS control of three-phase power filter
[6] for harmonic mitigation [4].

AbstractThis article is focused on progressive strategies of

shunt active performance filters with intension to compensate
higher harmonic currents of non-linear load. The authors of
this article deal with usage of a combination of fuzzy system
techniques and artificial intelligence, which is also known as
Adaptive Neuro Fuzzy Interference Systems- ANFIS for
improve in terms of quality of electrical energy in supply
network. Within the executed experiments there was created a
complex adaptive system based on the ANFIS structure.
Authors have been examining different structures of the
ANFIS network (structure). Experimental results based on real
signals are showing up, that mentioned types of control with
the use of ANFIS reach very good compensation attributes of
the harmonic currents of non-linear load. These control
strategies of the active shunt filter are compared on a base of
total THD harmonic distortion of current networks after the
compensation.

II.

Currently, the quality of electrical energy [7] is still more

and more discussed question. Harmonically [8] generated
non-linear appliances [8] are some kind of pollution of the
supply networks [7].To these non-linear loads belong:
controlled actuations, motor starters
electric lights
PCs, UPS and other electric devices
All of these devices are generating undesirable harmonic
currents to the supply network. Caused by this, there is a
voltage distortion in the supply network and it also causes an
aggravation in quality of electrical energy. In this article, the
most important marker of electrical energy quality will be
harmonic distortion of current [9]. Distortion of the
supplynetwork can cause some critical damage, for example:
overheating of transformers and motors (even if they
are correctly dimensioned)
possible oscillation of motors (vibrations can appear)
in case of magnetic switchers, device can be turned
off
flicker, PC failures (loss of data).
All effects caused by bigger distortion of the supply
network lead to economic losses [7]. Therefore, it is very
important to mitigate dominant harmonics to the
specifications outline in IEEE 519-1992 harmonic standard
[10]. Today, there are 2 basic types of higher harmonic
elimination- passive [11] and active [12] filter.

Keywords-Adaptive Neuro Fuzzy Inference Systems ANFIS;Total Harmonic Distortion - THD;Power Quality;Shunt
Active Power Filter- SAPF.

I.

INTRODUCTION

Currently, there is a big attention given to a problem of

adaptive systems, because many of modern applications need
this kind of a progressive accession of adaptation. Topicality
of examined problem corresponds to improvement of a
microprocessor technique performance:multi-core processors
[2], Field Programmable Gate Array - FPGA [2]. In fact, low
performance of processors was the main limiting factor in
usage of the sophisticated complex adaptive filters. In 2013,
we can say that the age of changes is coming, which is
evocated with several trends in area of information
technologies.
The idea of the adaptation comes from attributes of the
bioplasm, its ability of living organisms to adapt their
behavior to changes in an environment they live in,
especially, when these changes are adverse. This
phenomenon is commonly called learning. To this group of
systems, which are capable of adaptation we can also count
mechanical systems beside of those biological ones, then we
are talking about so called machine learning.[3] The ability
to learn is sometimes considered as a definition for
intelligence, see [3]. So its genuine that there is a big effort
to acquire such an attribute also to technical systems. These
technical adaptive systems are characterized by the ability to
adapt their parameters according to anactual information
about controlled system or executed signal. Exactly these

2013. The authors - Published by Atlantis Press

CURRENT STATUS OF EXAMINED PROBLEM

III.

SHUNT ACTIVE PERFORMANCE FILTERS WITH ANFIS

The active shunt filter represent additional performance

electric switcher [15] plugged to a non-linear load, see fig. 1.
The adaptive system is here to watch and predict the changes
of harmonics, which are created by plugging the non-linear
loads, or let us say to adapt to network current changes in
time. Incoming switcher current is driven to produce same
levels of the harmonics as the non-linear load with the help
of ANFIS, but in opposite phase. These 2 levels of

157

harmonics eliminate each other in a point where they meet.

After this, the final current is without undesired harmonics.
With such a method, current taken from the network is
filtered and this way also voltage deformations caused by the
load can be eliminated and the effectiveness of the network
will be improved too. The principle of this action is already
known for several decades, but the realization was possible
just after the arrival of modern and fully controlled
components, like are modern IGBT [16], transistors, tyristor
modules IGCT [16] supplemented by anti-shunt diodes [17].
Better description of the shunt active filters activity can be
found in [11,14,17].

In fig. 1 is shown simplified principal scheme of the

shunt active filter for three-phase system, which is controlled
by ANFIS. The filter consists of the controlled current
generator (rectifier with PWM [17]) plugged shunt to a load.
The active filter doesnt need external source of power
supply, capacitor is charged by controlling periphery. For
basic processing are used well known practices such a Clarke
transformation [14], Fourier transform [18], linear filtration
[19], we can mark them as a conventional processing (see fig.
1). Better description of these DSP methods is beyond area
of this articles topic, you can find more information in
[14,18,20].

Figure 1. Simplified principal scheme of the shunt active filter for three-phase.

IV.

the design parameters which are determined during the

training process.
The ANFIS architecture to implement these two rules is
shown in fig. 2 [21], inwhich a circle indicates a fixed node
whereas a square indicates an adaptive node. As figure
illustrates, ANFIS architecture consists of five layers [3].

ADAPTIVE NEURO FUZZY INFERENCE SYSTEMS

ANFIS (Adaptive Neuro-Fuzzy Inference System) [23]

represents a forward, adaptive neural network that is
functionally equivalent to the fuzzy inference system of the
Sugeno type (Takagi-Sugeno) [22]. A Two Rule Sugeno
ANFIS has rules of the form:

If x is A1 and y is B1
THEN f 1 = p1 x + q1 y + r1

If x is A2 and y is B 2
THEN f 2 = p 2 x + q 2 y + r2

(1)

(2)

where x and y are the inputs, Ai and Bi are the fuzzy sets,
fi, i =1,2 are the output of fuzzy system, and pi, qi and ri are

Figure 2. Simple T-S FIS system.

158

Mathematical derivation and description of function of

single layers (fig. 2) is well-known in common and good
described in a number of specialized publications, e.g. in [1,
4, 21, 22]. The authors of paper address to the detailed
description of ANFIS in their own publication [20].
V.

The programmable source HP6834B allows togenerate

three-phase sinus waveforms up to 300V/5A amplitude and
frequency up to 5kHz (100 harmonic for 50Hz).
VI.

DESCRIPTION OF EXPERIMENTAL WORKSTATION

To verify functionality of the designed ANFIS system,

three-phase voltage and current waveforms have been
measured. For this has been used the experimental
workstation with programmable source HP6834B [25] (see
fig. 3) and with developed IED (intelligent electronics device)
which allow us to measure the parameters of the power
energy quality. To source have been connected different nonlinear loads.
For automatic tests was the source controlled remotely
with developed IED using GPIB interface. The developed
IED allows to measure up to three voltages and currents with
sampling frequency up to 40kS/s for a channel. Input voltage
levels are adapted to measure on voltage transducers
(57,7/100 V) and also to a direct measuring in low voltage
networks (230/400 V). Current levels are adapted to measure
on current transducers (1/5A) or for the use of current clamps
(20/200A).
Every input has its own anti-aliasing filter to keep a
sampling theorem. All measuring ranges were calibrated
with NI 9225 [30] and NI9227 [31] modules.The developed
application in IED allows to measure and evaluate the power
quality parameters according to actual standards IEC 610004-30, IEC 61000-4-7, IEC 61000-4-15 and EN 50160 [33].
To application were implemented the techniques of the
ANFIS system what has allowed us to watch its results in
real-time.

RESULTS OF EXECUTED EXPERIMENTS

Within the executed experiments have been examined

different networks (structures) of ANFIS. The review of used
network models is in tab. 1. For the ANFIS system
buildinghavebeen used ANFIS functions [26], for training
(estimation) EVALFIS function [27]. Better description of
work with functions ANFIS and EVALFIS in MATLAB [28]
environment can be found in [26,27,28].
TABLE I.

THE REVIEW OF USED ANFIS MODELS.

ANFISModel

numberofnodes
numberoflinear
parameters
numberofnonlinear
parameters
totalnumberof
parameters
numberoffuzzyrules

21

35

53

75

12

27

48

75

12

18

24

30

24

45

72

105

16

25

In fig.4 are reference current waveforms through the

single phases.For a better readability of executed
experiments was presented length of time window 0,05s. The
waveforms of single phases are marked by different colors:
phase A is blue,phase B is red and phase C is black.

Figure 4. Reference voltage process.

Infig.5 are current waveforms through single phases after

the connection of non-linear load. The authors plugged to the
experimental supply network real non-linear loads (for
example PC, regulated actuations, e.t.c.) see chapter 2.
This way were captured the real hardly distorted
waveforms of the current which we can, from a view of used
ANFIS system, consider as a primary waveforms. The THDI
distortions are for phase a=97,6%, b=99,5% a c=96,8%.

Figure 3. Experimental workplace.

159

models are very equal to each other, see tab. 2. More difficult
models bring us just a bit of an improvement. This is,
however, paid by huge increase of requirements for
calculations (see tab. 1) so from the view of own
implementation they appear as inconvenient.
The frequency spectra of the source current before and
after the compensation are compared in fig. 8 where the
amplitude of each harmonic is represented in a percentage to
the fundamental.

Figure 5. Primary current waveforms from non-linear load.

Infigures from 6 to 7 are displayed final waveforms of

reconstructed currents for ANFIS models A and ANFIS
models B.

Figure 8. Comparision of amplitude current spectras from non-linear load

and reconstructed current waveforms using the ANFIS model A.

With the other ANFIS models the THD values have been
really low, so their presence in fig.8 would have just a small
relegating value. The results of single executed ANFIS
models are nicely presented in tab. 2.
TABLE II.

RESULTS OF EXECUTED EXPERIMENTS

ImprovementoftheTHDValue
Phasea
Phaseb
Phasec

Figure 6. Reconstructed current waveforms for ANFIS model A.

Without
ANFIS
ANFIS
modelA
ANFIS
modelB
ANFIS
modelC
ANFIS
modelD

97,6%

99,5%

96,8%

10,05%

6,80%

10,93%

1,21%

1,39%

2,78%

0,64%

0,75%

1,21%

0,31%

0,62%

0,86%

VII. SUMMARY
This article was dedicated to a new control strategy of
shunt active performance filters. The effort of the authors
was to design system, which would effectively compensate
higher harmonic currents of non-linear load in modern power
networks, where is impossible to predict frequencies of
disturbing signals and these frequencies also change in time.
The executed experiments tested the functionality of
designed ANFIS system. The system was reaching satisfying
results. The THD of the source current after compensation is

Figure 7. Reconstructed current waveforms for ANFIS model B.

If we compare the results of single ANFIS models, we

can declare that the worst results have reached the ANFIS
model A, see fig.6. This is caused by simplicity of used
model (see tab.1), however this model has the lowest
requirements for calculations. If we compare the results of
other ANFIS models B, C and D we can allege that those

160

for ANFIS model A approx. 10%. For ANFIS models B, C a

D is THD approx. 1 % which is less than 5 %, the harmonic
limit imposed by the IEEE-519 [10] and IEC-61000-3 [29]
standards.
The used method provides really better response than p-q
method [13]in some cases because it is easier to implement
on a microprocessor.
Also when there are variable
frequencies because of the supply network instability, used
method will follow basic frequency in a moment when the
frequency loop will be closed.
The adaptive Neuro-Fuzzy controller training and
fundamental wave assessment was done by using offline
training method. Next step will lead to investigation of the
possibilities in applying online training method to track
harmonics which vary in time and are located in power
systems.

[13]
[14]

[15]
[16]
[17]
[18]

ACKNOWLEDGMENT
This researchwas supported in part by The Ministry of
Education, Youth and Sports of Czech Republic under the
project KONTAKT II registration numberLH12183
and project SGS no. SP2013/203 VSB-TU Ostrava, FEECS.

[19]
[20]

REFERENCES
[1] J. Patel, R. Gianchandani, "ANFIS Control for Robotic Manipulators:

[2]
[3]

[4]

[5]

[6]
[7]
[8]

[9]
[10]
[11]
[12]

Adaptive Neuro Fuzzy Inference Systems for Intelligent Control."

LAP LAMBERT Academic Publishing (25 Nov 2011), 88 pages,
ISBN-13: 978-3846591710
Korf, Richard E. Artificial intelligence search algorithms. Chapman
& Hall/CRC, 2010.
Hands-on Experience with a Sorting ApplicationN. K., Bett, J. N.
Nderu, and P. K. Hinga, "Adaptive Neuro-fuzzy Inference system
based control of three-phase hybrid power filter for harmonic
mitigation." In Journal: Emerging Technology and Advanced
Engineering, pp. 75-81, ISSN 2250-2459, Volume 2, Issue 8, August
2012. Article (CrossRef Link)
L. WenNa, W. PeiFeng, L. Yang, "A Method of Adaptive Colored
Noise Cancellation Based on ANFIS."International Conference on
Image Analysis and Signal Processing, 2009. IASP 2009, Taizhou, pp.
386 388, ISBN: 978-1-4244-3987-4. Article (CrossRef Link)
Zhang, Yu, Yupeng Tang. "Active Power Filter Based on Adaptive
Detecting Approach of Harmonic Currents." Journal of
Electromagnetic Analysis and Applications 1.4 (2009): 240-244.
Article (CrossRef Link)
B. Dixit , Amit Yadav, Electrical Power Quality , First, 2010, p. 172,
ISBN : 978-93-80386-74-4.
D. Chapman, "Power Quality Application Guide, Harmonics Causes
and Effects." March 2001, Copper Development Association. Article
(CrossRef Link)
S. Fukuda, K. Muraoka, T. Kanayama, "Adaptive learning based
current control of active filters needless to detect current
harmonics."In Applied Power Electronics Conference and Exposition,
2004. APEC'04. Nineteenth Annual IEEE (Vol. 1, pp. 210-216).
IEEE. Article (CrossRef Link)
IEEE Recommended Pratices and Requirements for Haronic Control
in Electrical Power Systems. IEEE std 519-1992 (CrossRef Link)
P.Brandstetter, Active power filter, VSB-TU Ostrava,(CrossRef Link)
Z.l Salam, "Hybrid Active Power Filter." 184 pages, LAP Lambert
Academic Publishing (17 Aug 2009), ISBN-13: 978-3838307510.
R. Martinek, J. Zidek, "Use of Adaptive Filtering for Noise Reduction
in Communication systems," In Conference Proceeding: The
International Conference Applied Electronics (AE), pp. 1-6, ISBN

[21]
[22]

[23]
[24]
[25]

[26]

[27]
[28]

[29]
[30]
[31]
[32]

161

978-80-7043-865-7, ISSN 1803-7332, INSPEC Accession

Number: 11579482, Pilsen, Czech Republic, 8-9 September 2010.
Article (CrossRef Link)
Z. Paclt, "Control strategy of a shunt active power filter", PhD thesis,
CVUT, 2011.
S. Kalaschnikow, S. Hansen, L. Asiminoaei, H. G. Moos, "Active
compensation of harmonics in industrial applications". In Electrical
Power Quality and Utilisation, 2007. EPQU 2007. 9th International
Conference on (pp. 1-6). IEEE. Article (CrossRef Link)
L. A. Morn, J. W. Dixon, J. R., Espinoza, R. R. Wallace, "Using
active power filters to improve power quality."In 5th Brazilian Power
Electronics Conference (pp. 501-511). Article (CrossRef Link)
M. Rashid, "Power Electronics Handbook." Hardbound, 1362 Pages,
Published: December 2010, Butterworth
M. J. Roberts, "Signals and Systems: Analysis Using Transform
Methods and MATLAB." USA, The McGraw-Hill Companies, 2008,
1026 p. ISBN 0-07-293044-6.
R. Martinek, J. Zidek, "A System for Improving the Diagnostic
Quality of Fetal Electrocardiogram, "In Journal: Przeglad
Elektrotchniczny (Electrical Review), R. 88 NR 5b/2012, pp. 164173, ISSN 0033-2097, Warszawa, Poland, May 2012. Article
(CrossRef Link)
J. G. Proakis, D..K. Manolakis, "Digital Signal Processing: Principles,
Algorithms, and Applications." Pearson; 4 edition (15 Feb 2006), 948
pages, ISBN-13: 978-0132287319.
R. Martinek, J. Zidek, "Refining the diagnostic quality of the
abdominal fetal electrocardiogram using the techniques of artificial
intelligence,"In Journal: Przeglad
Elektrotchniczny (Electrical
Review), R. 88 NR 12b/2012, pp. 155-160, ISSN 0033-2097,
Warszawa, Poland, December 2012. Article (CrossRef Link)
O. Castillo, P. Melin, "Type -2 Fuzzy Logic: Theory and
Applications." 2008, XIV, 244 p. 188 illus, ISBN 978-3-540-76283-6.
M. Hammer, O. Janda, J. Ertl. "Selected Soft-Computing Methods in
Power
Oil
Transformer
Diagnostics

part
1.
"
In Journal: Elektrorevue. pp. 1-13, ISSN 1213-1539, 2012. Article
(CrossRef Link)
A. Shukla, R. Tiwari, R. Kala, "Towards Hybrid and Adaptive
Computing." 1st Edition, 2010, 450 p., 138 illus, ISBN 978-3-64214343-4.
HP - Agilent 6834B, Agilent 6834B Specifications. (CrossRef Link)
MathWorks, Matlab R2012b - Adaptive Noise Cancellation,
demonstration of adaptive nonlinear noise cancellation using the
Fuzzy Logic Toolbox functions ANFIS and GENFIS1. (CrossRef
Link)
MathWorks, Matlab R2012b - Fuzzy Logic Toolbox, web: (CrossRef
Link), MathWorks, Matlab R2012b - genfis1 and anfis - Generate
Fuzzy Inference System structure from data using grid partition, web:
(CrossRef Link)
International Statndard IEX 61000-3-3, 2002, (CrossRef Link)
R. Martinek, M. Al-Wohaishi, J. Zidek, Software Based Flexible
Measuring Systems for Analysis of Digitally Modulated Systems, In
Conference Proceedings: The 9th Roedunet International Conference,
RoEduNet, pp. 397-402, ISBN 978-1-4244-7335-9, ISSN 2068-1038,
Sibiu, Romania, 24-26 June 2010. Article (CrossRef Link)
NI 9225, 3-Channel, 300 Vrms Analog Input Module (CrossRef Link)
NI 9227, 4-Channel Current Input C Series Module (CrossRef Link)
Electromagnetic
compatibility
(EMC),
INTERNATIONAL
STANDARD IEC, (CrossRef Link)
Czech version of the European Standard EN 50160:1999. (CrossRef
Link)