International Journal of Innovative Technology and Exploring Engineering (IJITEE)

ISSN: 2278-3075, Volume-8 Issue-2S December, 2018

Artificial Neural Network Controlled Shunt Active

Power Filter for Minimization of Current
Harmonics in Industrial Drives
P. Thirumoorthi, Raheni T D

Abstract: In this paper development of three phase voltage Because of its effective operation, VCSAPF is examined as
controlled shunt active power filter is designed to compensate the viable solution to minimize the current harmonics produced
harmonic current present in nonlinear load. The designed system by nonlinear load. To extract reference compensating
overcomes the limitations of passive filter because of its
current several control techniques such as synchronous
resonance and bulky size. Voltage controlled shunt active power
filter is the effective method for compensating harmonic elements reference frame D-Q theory, instantaneous power of P-Q
caused by rectifier with RLE (nonlinear) load. In the proposed theory, synchronous detection algorithm and wavelet
system classical PI controller is implemented and minimizes the transform are applicable in the present scenario. P-Q theory
ripple voltage of the DC capacitor voltage. The techniques used is one of the most simple andconvenienttechnique to
to control algorithm deals with the concept of instantaneous implement. Due to the complexity of the conventional
power of P-Q theory and a combination of neural network based
technique the proposed intelligent controller behaves for
intelligent technique to calculate three phase reference
compensating current. The results ofPI based instantaneous leading patterns to compensate the reference current [6-7].
power of P-Q theory and intelligent technique such as artificial
neural network based back propagation algorithm is II. INSTANTANEOUS P-Q THEORY BASED
implemented and simulation are carried out in MATLAB/ VOLTAGE CONTROLLED SHUNT ACTIVE
Simulink environment. POWER FILTER
The proposed system consist of three phase power supply,
Key Words: PI controller, Voltage Controlled Shunt Active
Power Filter (VCSAPF), Artificial Neural Network (ANN), Total front end rectifier, PI controller, P-Q theory, and hysteresis
Harmonic Distortion (THD) band current controller and voltage source inverter. The
principle of VCSAPFis mainly used to minimize the
I. INTRODUCTION harmonics fed by rectifier with RLE load. The system is
During recent years the development of power electronic implemented in classical P-Q theory. The configuration
based systems creates a major issue in acting electrical scheme of voltage controlled shunt active power filter is
system applications. The main source of the production and shown in Fig. 1.
Front End Rectifier
distribution of the electricity creates an issue and degrades Ra La ILa
the quality of the system. Because of the rapid increase in a
RL
power electronic based loads such as thyristors, front end Rb Lb ILb

rectifier, uninterruptible power supply (UPS), fluorescent Rc Lc

ILc
b
LL

lamps and electric arc furnace draws a non-sinusoidal c

current that creates harmonics in power system [1-3]. Three Phase

AC Mains
Therefore these types of loads create a series problem
mainly harmonic interference in industrial applications. The Filter
Current
VSI

above problem is mitigated by passive filters and provides a

Vs IL
distortion less current in the distribution network. But
passive filters are used for low voltage systems and they are Controller Part -
Cdc

suitable to work at the low impedance path, in order to Vdc

PI and ANN
controller
adjust the harmonic frequencies that are simple and Gate Pulses

inexpensive. In spite passive filter draws fixed Vdc Ref

compensation, resonance with the system impedance and
lower bandwidth is a majorproblem. However in industrial Fig. 1 Configuration scheme of voltage controlled shunt
applications voltage controlled active power filters are active power filter
extensively used for harmonic compensation[4-5]. Moreover
Three phase AC source is fed to rectifier that converts ac to
various types ofpower filters are feasible to overcome the dc power and connected to RLE load. Source current
above problem such as series active power filter and shunt harmonics is not minimized due to presence of non linear
active power filter. Series active power filter is expensive, condition. The control circuit receives input from source
complex and minimize the voltage harmonics. voltage and load current and control circuit produces gate
pulses by implementing P-Q theory.
Revised Manuscript Received on December 28, 2018.
P. Thirumoorthi, Department of Electrical and Electronics Engineering
Kumaraguru College of Technology,Coimbatore, Tamilnadu, India.
Raheni T D Department of Electrical and Electronics Engineering
Kumaraguru College of Technology,Coimbatore, Tamilnadu, India.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 150 & Sciences Publication
 Artificial Neural Network Controlled Shunt Active Power Filter for Minimization of Current Harmonics
in Industrial Drives

The corresponding pulses are given as input to the VCSAPF The net input is calculated in the following equation given
to generate compensating filter current andconnected at a below:
point of common coupling. binawa w
P-Q theory explains about the transformation from three Whereas a1 and a2 are the activations of the input
phase abcof current and voltage coordinates to two phase αβ neurons A1 and A2 that is fed to the output of input signals.
voltage and current coordinate by Clarke’s transformation The output b of output neuron B can be obtained by using
and the corresponding αβ is transformed to active and the activations over the net input.
reactive power component p-q. b f bin
The corresponding power components are converted to two Output neuron = function (calculated net input)
phase αβ current by low pass filter. Then the αβ current is
Topology of ANN consists of tan sigmoid function is
converted to abc reference compensating current by using
depicted in Fig. 4.
inverse Clarke transform and the output of control circuit
produces gate pulses. The load current produced by rectifier
with RLE load and filter current produced by P-Q theory ∑
have equal magnitude but opposite phase to that of load
current cancel each other and produce distortion less current
in supply side. Block diagram of PI controller based
Vdc
instantaneous power of P-Q theory is shown in Fig. 2. ∑

∑ Ploss

∑
Vdc Ref

30
Tansig

Fig. 4.Topology of ANN

Through many interconnection and number of neurons are

involved, the power of the human mind appears across it. In
order to improve the characteristic performance of voltage
controlled shunt active power filter, multilayer neural
network is chosen[11-12]. The design parameters consist of
Fig. 2.Block diagram of P-Q theory input layer, hidden layer with 30 neurons and output layer
with a neuron. The values are chosen from conventional
III. INTELLIGENT CONTROLLER - ANN proportional integral controller and these data’s are used to
In recent years artificial neural network plays a vital role train the network by selecting back propagation algorithm
that has the capacity to learn by specimen that makes them depending on users need. The activation function is selected
more extensible and robust. While implementing neural as tan sigmoid for the input layer and positive linear
network, it does not require any internal structure to perform activation function for target layer.For training the
that task. The most primary element of the human brain is algorithm, Marquardt Levenberg Back Propagation (MLBP)
the special type of cell is neuron and it provides the abilities is used. Each training session requires 265 values with 10
to realize and executes the previous existence to every iterations and the error occurrence is calculated. Back
action[8-10]. A1 and A2 are the input neuron that imparts the propagation algorithm is mainly used for artificial neural
signals and B is the output neuron that receives the signals. network to estimate the error improvement of every neuron.
The input neurons namely A1 and A2 are associated with the The network is based on first iterative optimization
output neuron B over interdependent weights (W1 and W2). algorithm to vary the weight of the neuron. Because of the
The simple neural network architecture is shown in Fig. 3. estimated error at the output and delivered back to the
network it is termed as backward propagation of errors. For
better internal representation of the network, back
propagation algorithm is the best one.

Fig. 3.Neural network architecture

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 151 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8 Issue-2S December, 2018

The internal representation of the network is mainly used to Step 1: Start the process and initialize weights of the
get accurate results. Because of its efficiency in the network, network and level the input/output data.
back propagation algorithm is more significant to estimate Step 2: The network of the structure is chosen and
the delta rule.The back propagation algorithm is the correspondingly the hidden layers for each neuron is
improvement of least mean square algorithm in which it selected.
changes the network weights to minimize the mean squared
Step 3: The activation function is selected and these
error between the actual and desired outputs of the network.
activation functions can be uniform and may be vary for
different layers.
IV. IMPLEMENTATION OF P-Q THEORY BASED
ANN CONTROLLER Step 4: Training pair for each training set is determined. To
determine the output of the network, it is based on input
The neural network based instantaneous power P-Q vector set andinitial weights.
theory modeling is shown in Fig.5. Step 5: The network output is determined by taking the error
output and the corresponding error is propagated back that
varies the weights in such a way it minimizes the error.
Itagain start from the output layer and fed backward to input
layer.
Step 6: Repeat steps from 4 to 6 and for each vector the
training set is analyzed unless the error for the given vector
set is lower than the required minimum error.
Neural network model and best validation performance is
shown in Fig. 6 and Fig. 7.

Fig. 5. Neural Network of P-Q theory modeling

Vdc and Vdc reference are the inputs fed to hidden layer
and compensating current is the output that is observed from
the P-Q theory. In neural network adaption of the weights
(w) and bias (b) arebased on the computation of mean square
error between its outputs. The process for the calculation of
two phase αβ coordinate is given below.
Iα = 2/3 Ia1/3IbIc

Iβ1/√3 IbIc Fig. 6. Neural network model

Therefore the compensating current obtained from two phase In the back propagation network, testing of data’s are
αβ coordinate to three phase abc frameis given by, implemented only in the feed forward path and during
ICa√2/3IC1 implementation of the network, even though network has
the ability to operate in more hidden layers, it is more
ICb√2/30.5 IC1) √3/2 IC2
sufficient to operate under one hidden layer. During training
ICc√2/3-0.5IC1√3/2  IC2 process it requires more time to train the network but it
produces an output in a clear manner when compare to other
From the above equations compensating currents are
networks.
calculated from α-β to abc frame. To calculate the derivative
The best validation performance is obtained by training and
function for the least square error with respect to the weights
testing the network by selecting number of neurons and
and bias of the network, first order iterative
epochs in a clear manner.
optimizationalgorithm is used. The squared error function is
defined by the following equation:
E  ½ TC2
E is the squared error, T is the Target output for a given
sample and C is the actual output of the neuron. The
constant ½ is included while differentiating the constant is
cancelled. Training steps for back propagation algorithm is
given below:

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 152 & Sciences Publication
 Artificial Neural Network Controlled Shunt Active Power Filter for Minimization of Current Harmonics
in Industrial Drives

integral controller. The results are described for rectifier

with RLE load, PI controller and ANN tuned back
propagation algorithm.The parameters aided in the proposed
system are supply voltage = 440V, R=300Ω, L=1mH,
E=1V. The output of source voltage THD without filter is
2.24%.

Fig. 7. Best validation performance

The algorithm consists of input and output layer to vary
the weights in a network to analyze the arrangements of Fig. 8.Source voltagewaveform -Rectifier with RLE load
input in a correct manner. Sourcecurrent total harmonic distortion - rectifier with
Back propagation algorithm is different from other RLE load (non linear) is 12.33%.
techniques because during learning period itself the weights
are calculated. The error signals are determined by obtaining
the difference between the calculated and targeted output
that measures in the output layer. Training set for neural
network provides a good performance while taking
proportional and integral control in the range of 1.25
and6.5.The configuration parameter for the voltage
controlled shunt active power filter is given in Table I:

Configuration Parameters
Source voltage 440 V, 50 Hz, Three phase
Source inductance 0.5mH
Fig. 9. Source current waveform –Rectifier with RLE
Load resistance 300Ω load
Load inductance 1mH
Filter inductance 0.1017H
Filter capacitance 100µF
Proportional gain(Kp) 1.25
Integral gain(Ki) 6.5
Vdc 620V
Cdc 3600µF

Fig. 10.Source current THDof rectifier with RLE load-

Table I. Configuration Parameter FFT analysis
V. RESULTS AND DISCUSSIONS
It is observed from the simulation result the waveform of
The proposed voltage controlled shunt active power filter source current is non sinusoidal nature due to the occurrence
is designed and rectifier with RLE load is a nonlinear. It is of nonlinear load and the above result isrectified by
viewed that rectifier with RLE load results in harmonic implementing conventional PI controller.
current is not minimized to IEEE standard due to presence The source voltagetotal harmonic distortion with IRP theory
of nonlinearity. With the addition of VCSAPF the current based on PI controller is 0.55%.
harmonics is minimized by instantaneous power of P-Q
theory and ANN tuned proportional integral controller. To
train the network, values obtained from PI controller is
chosen to train the neurons in order to get a distortion less
current in supply side when compared to proportional

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 153 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8 Issue-2S December, 2018

The source currenttotal harmonic distortion with ANN

based back propagation algorithm is 2.73 %.

Fig. 11. Source voltage waveform –IRP theory based on

PI controller
Source current total harmonic distortionwith IRP theory Fig. 15. Source current waveform– ANN tuned back
based on PI Controller is 4.62%. propagation algorithm

Fig. 12.Source current waveform- IRP theory based on

PI controller

Fig. 16.Source Current THD of ANN controller –FFT

analysis

The system is designed and the performance is verified for

the proposed system with RLE load. By implementing PI
controller based IRP theory source current harmonics is
minimized in which filter current gets injected that are equal
in magnitude and opposite in phaseopposite to that of load
current cancel each other and produce distortion less current
Fig. 13.Source Current THD ofPI controller – FFT in supply side as per IEEE standard. In a similar manner, by
analysis implementing intelligent technique it minimize the current
harmonics to an acceptable limit and obtains better
performance. Source current THD is described for different
Itis observed from the simulation result with the addition of controller is shown in Table II:
VCSAPF generate the compensating filter current which is
equal to the harmonic current. Thus it minimizes the source
current harmonics and produce distortion less current in the Controller Source current Source
mainsupply. – THD voltage- THD
The source voltagetotal harmonic distortion with ANN
based back propagation algorithm is 0.21%. Rectifier 12.33% 2.24%
with RLE
load
With PI 4.62% 0.55%
With PI - 2.73% 0.21%
ANN

Table II.Comparison Table

Fig. 14.Source voltage waveform – ANN tuned back

propagation algorithm

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 154 & Sciences Publication
 Artificial Neural Network Controlled Shunt Active Power Filter for Minimization of Current Harmonics
in Industrial Drives

VI. CONCLUSION
Voltage controlled shunt active power filter with ANN
controller has been implemented for minimization of source
current harmonics. The present system is designed to
mitigate the harmonics produced by nonlinear load.
Compared to conventional controller, neural network
minimizes the source current harmonics that has higher
attenuation against higher order harmonics and improves the
power quality of the system. Conventional controller
regulates average level of the DC capacitor voltage,whereas
neural network avoids mathematical operation and
simplifies the control configurations. Compensation is
employed for power quality improvement by minimizing the
current harmonics in various fields such as variable speed
drives, electric arc furnace and UPS application.

REFERENCES
1. Tey,L.H,Chu,Y.C “Improvement of power quality using adaptive
shunt active filter” ,IEEE Transactions on Power Delivery,vol. 20,
April 2005.
2. Akagi .H, “Modern active filters and traditional passive filters”,
Bulletin of the polish academy of technical sciences, vol. 54, no. 3,
pp.255-269, 2006.
3. Ahmed M. Mohammed, “Analysis and simulation of shunt active
power filter for harmonic cancellation of nonlinear loads”, Eng&
Tech. Journal, vol.28, no.16, 2010.
4. Thirumoorthi P, N.Shanmugam, Yadaiah N, “Power quality
improvement in induction furnance through harmonic current
compensation”, International Conference in Green Computing
Communication and Electrical Engineering(ICGCCEE),IEEE,DOI-
10.1109/ICGCCEE.2014.6922383, 6-8 March 2014.
5. Anju Jacob, Babitha T Abraham, NishaPrakash and Riya Philip, “ A
review of active power filters in power system applications”,
International Journal of Advanced Research in Electrical, Electronics
and Instrumentation Engineering(IJAEREEIE), vol.3, issue:6, June
2014.
6. S. Rahmani, A. Hamadi, K. Al-Haddad, and L.A. Dessaint, “A
combination of shunt hybrid power filter and thyristor-controlled
reactor for power quality”, IEEE Trans. Ind. Electron., vol. 61, no. 5,
pp. 2152–2164, May 2014.
7. PapanDey and SaadMekhlief, “Current harmonics compensation with
three phase four – wire shunt hybrid active power filter based on
modified D-Q theory”, Institute of Engineering and Technology,
Power Electronics ,vol.8, issue:11, pp. 2265-2280, 2015.
8. Jarupulasomlal, VenuGopalaRao.Mannam,Narishma
Rao.vutlapalli,“Power quality improvement in distribution system
using ANN based shunt active power filter”, IEEE Trans. Ind.
Electron., vol. 61, no. 5, pp. 2152–2164, Apr 2015.
9. B.Mangaiyarkarasi, P M Balasubramaniam, “Implementation of
artificial neural network controlled shunt active power filter for
current harmonics compensation”, International Journal of Advanced
Research in Computer and Communication Engineering ,vol . 4,
issue: 5, May 2015.
10. AbderrahmenBenyamina, Samir Moulahoum, IlhamiColak and
RamazanBayindir, “Hybrid fuzzy logic artificial neural network
controller for shunt active power filters”, International Conference on
Renewable Energy Research and Applications, Nov 2016.
11. Badoni, Bhim Singh and Alka Singh, “Implementation of echo – state
network based control for power quality improvement”, IEEE
Transaction on Power Electronics, 2016.
12. PrakashCh.Tah, AnupK.Panda and BibhuP.Panigrahi, “ Shunt active
power filter based on radial basis function neural network and p-q
power theory”, International Journal of Power Electronics and Drive
System (IJPEDS) , vol. 8, no.2, pp – 667-676, June 2017.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: BS2693128218/19©BEIESP 155 & Sciences Publication