20120140502021 2

International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 2, February (2014), pp. 173-185, © IAEME
173
ANALYSIS OF GENERATED HARMONICS DUE TO SINGLE PHASE PWM
AC DRIVES LOAD ON POWER SYSTEM USING ARTIFICIAL NEURAL
NETWORK
Dharmendra Kumar singh, Ekta Singh Thakur, Smriti Kesharwani, Dr. A.S.Zadgaonkar
Dr. C.V. Raman University Kargi Road Kota Bilaspur (C.G), INDIA
ABSTRACT
Recently harmonic distortion in power systems is attracting significant attention. Traditional
methods for harmonic distortion analysis using either FFT or DFT are, however, susceptible to the
presence of noise or sub-harmonics in the distorted signals. Harmonic detection by using Fourier
transformation requires input data for one cycle of the current waveform and also requires time for
the analysis in next coming cycle. In this paper, an alternative method using neural network
algorithm has achieved satisfactory results for fast and precise harmonic detection in noisy
environments.
In this paper we are identifying the harmonics component in power system generated by a
control scheme of single-phase to three-phase PWM converters for low power three-phase induction
motor drives. Currently there are number of methods are available for identifying the harmonic
components in power system but we use the intelligence system (ANN) for this work.
Keyword: Power system, Harmonics, Artificial Neural Network, VFD.
1 INTRODUCTION
The increasing application of power electronic facilities in the industrial environment has led
to serious concerns about source line pollution and the resulting impacts on system equipment and
power distribution systems. Power systems, in the presence of electronic equipment, can produce
harmonics in the power signal waveforms. Power converters, specifically, are responsible for a
disproportionate amount of the harmonics troubling power systems today [1]. Converters are used in
variable-speed drives, power supplies, and UPS (uninterruptible power supply) systems; the term
converter can refer to rectifiers, inverters, and cycloconverters. Arc furnaces are another significant
source of harmonics. Harmonics in power systems can be the source of a variety of undesirable
effects.
INTERNATIONAL JOURNAL OF ADVANCED RESEARCH IN ENGINEERING
AND TECHNOLOGY (IJARET)
ISSN 0976 - 6480 (Print)
ISSN 0976 - 6499 (Online)
Volume 5, Issue 2, February (2014), pp. 173-185
© IAEME: www.iaeme.com/ijaret.asp
Journal Impact Factor (2014): 7.8273 (Calculated by GISI)
www.jifactor.com
IJARET
© I A E M E

174
In recent years, neural network has got special attention to the researchers because of its
simplicity, learning and generalization ability and it has been applied in the field of engineering such
as in harmonic detection [2-6]. In this paper we identify the harmonics component in power system
generated by a control scheme of single-phase to three-phase PWM converters for low power three-
phase induction motor drives. In low power residential and industrial applications, where only a
single-phase utility is available, a single-phase to three-phase power converter system is required to
feed the three-phase induction motor drives. Conventionally, a full–bridge diode rectifier plus three-
leg PWM inverter has been used. However, the diode rectifier produces harmonic currents to flow
into the supply [7]. Traditionally and currently there are different methods are available and in
use for identify the harmonic components in power system but we use the intelligence system
(ANN) for this work.
2 VARIABLE FREQUENCY DRIVES
INDUCTION motor for many years has been regarded as the workhorse in industrial
applications. In the last few decades, the induction motor has evolved from being a constant speed
motor to a variable speed, variable torque machine. Its evolution was challenged by the easiness of
controlling a DC motor at low power applications. When applications required large amounts of
power and torque, the induction motor became more efficient to use. With the invention of variable
voltage, variable frequency drives (VVVF), the use of an induction motor has increased. Variable
frequency Voltage Source Inverters (VSI's) are widely used to control the speed of 3-phase squirrel
cage Induction Motors (IM) over a wide range by varying the stator frequency. In particular the
VSI's are widely preferred in industries for individual medium to high power variable speed drive
systems, driving a group of motors connected in parallel at economic costs. Most modern variable
frequency drives operate by converting a three-phase voltage source to DC using rectifier. After the
power flows through the rectifiers it is stored on a dc bus. The dc bus contains capacitors to accept
power from the rectifier, stores it, and later deliver that power through the inverter section. The
inverter contains transistors that deliver power to the motor. The “Insulated Gate Bipolar Transistor”
(IGBT) is a common choice in modern VFDs. The IGBT can switch on and off several thousand
times per second and precisely control the power delivered to the motor. The IGBT uses “pulse
width modulation” (PWM) technique to simulate a sine wave current at the desired frequency to the
motor [8-10].
Fig(1): Block diagram of variable frequency drive with 3 phase induction motor load input and
output signal also show

175
In this paper we use a single-phase to three-phase PWM converters for low power three-
phase induction motor drives which is shown in fig(1). In low power residential and industrial
applications, where only a single-phase utility is available, a single-phase to three-phase power
converter system is required to feed the three-phase induction motor drives. Conventionally, a full–
bridge diode rectifier plus three-leg PWM inverter has been used. However, the diode rectifier
produces harmonic currents to flow into the supply.
Fig(2): Circuit diagram of variable frequency drive
Fig(3): single-phase to three-phase PWM converters for low power three-phase induction motor
drives
3 ARTIFICIAL NEURAL NETWORK
The most popular Artificial Neural Network (ANN) architecture is multilayer Feedforward
Network with backpropagation (BP) learning algorithm.This network, as its name indicates is made
up of multilayer Thus architecture of this class besides processing on input, an output layer also have
one or more intermediary layers called hidden layers. The computational units of the hidden layer are
known as the hidden neurons or hidden units. The hidden layer aids in performing useful
intermediary computations before directing the input to the output layer. The input layer neurons are

176
links are referred to as input hidden layer weights. Again the hidden layer neurons are liked to the
output layer neuron and the corresponding weights are referred to as hidden output layer weights.
A two layers network in which one is input layer and other is output layers called single layer
feed forward network. In this architecture the input layer receive the input signals and after
processing it forwarded to output layer for output the data. The synoptic links carrying the weights
connected every input neuron to the output neuron but not vice-versa. Such a network is said to be
feed forward in type or acyclic in nature. Despite the two layers, the network is termed single layer
since it is the output layer, alone which performs computation [11-13].
Fig (4): Multilayer Feed-forward Network
4. BACKPROPAGATION LEARNING RULE
There are several different training algorithms for feed-forward networks. All these
algorithms use the gradient of the performance function to determine how to adjust the weights to
minimize performance. The gradient is determined using a technique called back-propagation.
Back-propagation is a systematic method of training multilayer Artificial Neural Networks. It
is built on high mathematical foundation and has very good application potential. Even though it has
its own limitations, it is applied to a wide range of practical problems and has successfully
demonstrated its power.
The Back-propagation learning algorithm approach to be followed is basically a gradient
descent along the error surface to arrive at the optimum set of weights. The error is defined as the
squared difference between the desired output and the actual output obtained at the output layer of
the network due to application of an input pattern from the given input-output pattern pair. The
output is calculated using the current setting of the weights in all the layers. The optimum weight
may be obtained if the weight are adjusted in such a way that the gradient descent is made along the
total error surface [13].
5 ANN DESIGNING PROCESS
ANN designing process involves five steps: gathering input data, normalizing the data,
selecting the ANN architecture, and Training the Network, Validation-testing the network [14].

177
5.1 Gathering Input Data
The configuration of the experimental system and experimental system block diagram is
shown below in
Experimental –setup
Fig(5): Image of experimental set-up
Fig(6): Block diagram of experimental set-up
In the above block diagram set-up, a transformer is connected with power supply. A linear or
non-linear load is connected with this transformer. Due to transformer and other loads are generated
harmonics in power system. Due to this power supply waveform is distorted. A data acquisition card
is connected at power common connection to collect the distorted current/voltage waveform or data.
These collected waveform/data transmitted to PC through RS-485 for ANN input which is designed
in MATLAB. Collected data is shown in waveform in fig (7).

178
Fig(7): Supply current waveform when Variable frequency drive loaded with three phase induction
motor
5.2 Normalization of input and output data sets
Normalization of data is a process of scaling the numbers in a data set to improve the
accuracy of the subsequent numeric computation and is an important stage for training of the ANN.
Normalization also helps in shaping the activation function. For this reason [-1, 1] normalization
function has been used.
Fig(8): Normalised current waveform of collecting waveform/data

179
Fig(9): Normalization and scalling of current waveform for ANN Input
5.3 Selecting the ANN Architecture
The numbers of layers and the number of processing element per layer are essential decision
for selecting the ANN architecture. Choosing these parameters to a feed forward backpropagation
topology is the art of the ANN designer. In this paper the ANN configuration has 32 iput neurons
receiving 32 sampled points of the distorted waveform and 32 output neurons producing the
magnitude of harmonic components up to t e 33th odd harmonics. The hidden layer has 65 neurons
to bridge input layer with output layer. For a set of input there is a corresponding set up of output
“target” values already stored in a data array . ANN Toolbox in MATLAB is used for this work. The
designed network is shown in fig(10)
Fig(10): Designed ANN for harmonics component identification

180
5.4 Training of the ANN Model
The ANN model used is executed by a structured computer program that can update neurons
almost simultaneously .Before the start of training, the initial weight were randomized to value
between -0.5 and +0.5. These input and target outputs were “shown” to the ANN in a sequential
manner so that the weights were updated step by step according to the backpropagation learning
algorithm. The error between the actual output and the target was evaluated after every update. The
backpropagation learning algorithm employed [15] works toward reduction of the RMS error, and
the training ceases as the total sum of square error reaches just below the error critia initially set. The
weights are then supposed to have converged enough that they should represent the non-linear
transfer functions between inputs and outputs of the ANN model accurately
Fig(11): Training of designed ANN
It was observed that during the initial stage of training the rate of convergence in weights
update was fast at a learning rate of 0.05. This was seen in the rapid steady drop in total sum of
square. Subsequently training yielded a slower convergence rate. The learning rate ή was constantly
reduced whenever the total sum of square value changed too slowly. It was also reduced when the
total sum of square value oscillated for a prolonged period of training epochs due to entrapment in
local minima.
5.5 Testing
To test the generalizing capabilities of the magnitude networks the distorted waveforms that
contained harmonics up to the 33rd odd harmonic with no noise added were considered for the
training process.
After the training and testing, the ANN used for unfamiliar input which is collected from
experimental set-up for the identification of the harmonic component.
6 RESULT AND DISCUSSION
Fig (12) and fig (13) shows the output of ANN for input voltage and current which is
collected from the experimental set-up. From the graph of ANN output we observe that odd
harmonics generated in power system due to the single-phase to three-phase PWM converters for
low power three-phase induction motor drives load.

181
Fig(12): ANN Output for odd harmonicst
Fig(13): 1st
node ANN Output for dc component
Fig(14): 2nd
node ANN Output for fundamental frequency

182
Fig(15): 3rd node ANN Output for 3rd
harmonics
Fig(16): 4th
node ANN output for 5th
harmonics
Fig(17): 5th
harmonics

183
Fig(18): 7th
harmonics
Fig(19): 8th
harmonics
Fig(20): ANN output for phase angle for harmonics

184
7 CONCLUSION
An artificial neural network model is developed and implemented for measuring harmonics
component in power system. This model is tested offline under different condition. the result
outcome from offline test indicate that the ANN model has providing very high accuracy in
harmonic component measurement , the proposed ANN model is implemented on pc with MATLAB
software using a data acquisition card. It was tested off-line under different conditions. The result of
the off-line test indicates that the ANN model has very high power system harmonics component
measurement accuracy. The developed ANN model was implemented on a PC with MATLAB
Software (with ANN Toolbox) using a data acquisition card. The ANN model was able to measure
the harmonic components of voltage and current at various levels. The data is collected at Machine
lab in Dr.C.V.Raman University where the system is available. The output of the ANN show that
due to the single-phase to three-phase PWM converters for low power three-phase induction motor
drives odd current harmonics that is 3rd
,5th
,7th
,9th
,11th
,13th
,15th
,17th
etc are generated in power
system .So proper filter is required for elimination harmonics from load to supply.
8 ACKNOWLEDGEMENTS
I would like to express my sincerest gratitude to all staff of EEE Department Dr C.V. Raman
University who has contributed, directly or indirectly, in accomplishing this paper. Special thanks to
extend Miss Pallavee Jaiswal for her suport in completing this Paper.
9 REFERENCES
[1] J.F. Chicharo, and T.S. Ng, “Gradient based adaptive IIR notch filtering for frequency
estimation”, IEEE Trans. Acoust. Speech & signal Process., 38, (5), 1990, pp.769-777.
[2] N. Pecharanin, M. Sone, and H. Mitsui, “An application of neural network for harmonic
detection in active filter” , 1994 IEEE International Conference on Neural Networks, Vol. 6,
27 June - 2 July 1994, USA, pp.3756-3760.
[3] W W L Keerthipala, T C Low, and C L Tham, “An Application of A Back -Propagation Type
Neural Network for Harmonic Distortion Analysis”, International Power Engineering
Conference 1995, 27 Feb. - 1 Mar. 1995, Singapore, pp. 501-506.
[4] N. Pecharanin, H. Mitsui, and M. Sone, “Harmonic detection by using neural network”, 1995
IEEE International Conference on Neural Networks, Vol. 2, 27 Nov. - 1 Dec. 1995 Australia,
pp. 923-926.
[5] A.A.M. Zin, M. Rukonuzzaman, H. Shaibon, and K.I. Lo, “Neural network approach of
harmonics detection” , 1998 International Conference on Energy Management and Power
Delivery (EMPD ’98), Vol. 2, 3-5 March 1998 Singapore, pp. 467-472.
[6] M. Rukonuzzaman, A.A.M. Zin, H. Shaibon, and K.I. Lo, “An application of neural network
in power system harmonic detection”, The 1998 IEEE World Congress on Computational
Intelligence, Vol. 1, 4-9 May 1998 USA, pp. 74-78.
[7] P. Enjeti and A. Rahman, "A new single phase to three phase converter with active input
current shaping for low cost ac motor drives," IEEE Trans. On IA, vol. 29, no. 4,
pp. 806-813, 1993.
[8] R. Hoadley, “Line and Load Considerations for AC Drives: Harmonics”, Rockwell
Automation Allen radley 2002 Automation Fair, Anaheim, Ca.
[9] D. Paice, “Optimized 18-pulse type ac/dc or dc/ac Converter System”, U.S Patent
#5,124,904, Jun 1992.

185
[10] M.F. McGranaghan and D.R. Mueller, "Designing Harmonic Filters for Adjustable-Speed
Drives to Comply with IEEE-519 Harmonic Limits," IEEE Transactions on Industry
Applications, Vol. 35, No. 2, March/April, 1999, pp. 312-318.
[11] K. Toyama, T. Takeshita, and N. Matsui, "Stability and initial estimation of power source
voltage sensorless single-phase PWM ac/dc converter," Trans. of JIEE, vol. 116-D, no. 3,
pp. 354-360, 1996.
[12] P. Enjeti and A. Rahman, "A new single phase to three phase converter with active input
current shaping for low cost ac motor drives," IEEE Trans. On IA, vol. 29, no. 4,
pp. 806-813, 1993.
[13] Jaker.M. ZArada “Introduction to Artificial Neural Systems” Jaico Publishing houre ISBN :-
81-7224-650-1.
[14] Dharmendra kumar singh, Dr. Moushmi Kar and Dr. A.S.Zadgaonkar “Analysis of Generated
Harmonics Due to Transformer Load on Power System using Artificial Neural Network”,
International Journal of Electrical Engineering & Technology (IJEET), Volume 4, Issue 1,
2013, ISSN Print : 0976-6545, ISSN Online: 0976-6553, pp. 81 - 90.
[15] Dharmendra Kumar Singh, A. S. Zadgaonkar, “Power System harmonics Analysis Using
Multi-Layer Feed Forward Artificial Neural Network Model” International Journal of
Electronics and computers”, vol 4 no1 2012 ISSN: 0975-3796.
[16] M. Depenbrock, “Pulse width control of a three-phase inverter with non sinusoidal phase
voltage of a three-phase PWM inverter”, Proc. IEEE Int. semiconductor Power Conversion
Conf., Orlando, Florida, USA, pp. 399-403, 1977.
BIOGRAPHIES
Dharmendra kumar obtained M. Tech. Degree in Electronics Design and Technology from Tezpur
University, Tezpur, Assam in the year 2003. Currently he is pursuing research work in the area of
Power Quality under the guidance of Prof A. S. Zadgaonkar.
Ekta Singh Thakur has obtained B. E. degree in Electrical Engineering from Chhattisgarh Swami
Vivekananda Technical University.She pursuing M.Tech. from Dr C.V. Raman University.
Smriti Kesharwani has obtained B. E. degree in Electrical Engineering from Chhattisgarh Swami
Vivekananda Technical University. She pursuing M.Tech. from Dr C.V. Raman University.
Dr. A. S. Zadgaonkar has obtained B. E. degree in Electrical Engineering from Pt. Ravishankar
Shukla University, studying at Govt. Engineering College, Raipur in 1965. He obtained M. E. in
1978 from Nagpur University. His research paper for M. E. was awarded “best paper” by the
Institution of Engineers (India) in the year 1976 & 1977 respectively. The testing technique for
quality of wood developed by him was included in ISI in 1979. He was awarded Ph. D. in 1985 by
Indira Gandhi Kala & Sangeet University, Khairagah for his work on “Acoustical and Mechanical
Properties of Wood for Contemporary Indian Musical Instrument Making.” He obtained another Ph.
D. in 1986 by Pt. Ravishankar Shukla University on “Investigation of Dynamic Properties of Non-
Conducting Materials Using Electrical Analogy.” He has 47 years of teaching experience. He is
currently adding glory to the post of Vice Chancellor of Dr. C. V. Raman University. He has
published more than 500 technical papers for journals, national and international conferences. He
was the Joint Director, Technical Education, Govt. of Chhattisgarh in 2004 & the Principal of NIT,
Raipur in 2005. He is life member of Acoustical Society of India, Biomedical Society of India,
Linguistic Society of India, Indian Society for Technical Education and many social bodies.

20120140502021 2

Recommended

Recommended

More Related Content

What's hot

What's hot (19)

Viewers also liked

Viewers also liked (7)

Similar to 20120140502021 2

Similar to 20120140502021 2 (20)

More from IAEME Publication

More from IAEME Publication (20)

Recently uploaded

Recently uploaded (20)

20120140502021 2