Application of Artificial Neural Network in Electrical Power System

0.19 MB·English

by  P. Palanichamy

Indonesian Journal of Electrical Engineering and Computer Science
Vol. 9, No. 1, January 2018, pp. 77~80
ISSN: 2502-4752, DOI: 10.11591/ijeecs.v9.i1.pp77-80  77

Application of Artificial Neural Network in Electrical Power
System

P. Palanichamy
Department of Mathematics, AMET University, Chennai

Article Info ABSTRACT
Article history: The artificial neural network used to detect the fault in electrical machines
and can increase the function of new entry detection when compared to the
Received Jun 18, 2017
conventional method. In proposed artificial neural network has increased the
Revised Nov 27, 2017 precision and stability of system performance. The time-area vibration signs
Accepted Dec 19, 2017 of a pivoting machine with ordinary and flawed apparatuses are handled for
highlight extraction. The separated elements from unique and preprocessed
signs are utilized as contributions to both classifiers in view of ANNs and
Keywords: SVMs for two-class (typical or blame) acknowledgment. The quantity of
hubs in the concealed layer, if there should be an occurrence of ANNs, and
Electrical Machines
the extend basis work section parameter, in the event of SVMs, alongside the
Extraction and SVMs
choice of information components are enhanced utilizing genetic algorithm
Neural Network
(GAs).
Performance
Stability Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
P. Palanichamy,
Department of Mathematics,
AMET University, Chennai.

1. INTRODUCTION
In vector machines has two-class issues—in which the information are isolated by a hyper plane
characterized by various bolster vectors. A concise presentation of SVM is exhibited here for culmination.
Peruses are alluded to the instructional exercises on SVMs for points of interest. The SVM can be considered
to make a line or hyper plane between two arrangements of information for grouping. If there should arise an
occurrence of two-dimensional circumstance, the activity of the SVM can be clarified effectively with no loss
of simplification.

Problem Description
It is necessary to identify fault in electrical machines and power system lines during operation to
ensure continuous reliable operation of power systems. Conventional fault identification methodologies
based SVM’s and GA cannot identify problems or faults in large power systems and these methods are
complex.

Background
Condition viewing of machines is gaining more significance in industry due to the need to expand
dependability and to diminishing conceivable loss of creation because of machine breakdown [1]. The
utilization of vibration and acoustic outflow (AE) signs is very regular in the field of condition checking of
pivoting apparatus. By contrasting the signs of a machine running in typical and flawed conditions,
identification of deficiencies like mass unbalance, rotor rub, shaft misalignment, equip disappointments and
bearing imperfections is conceivable [2]. These signs can likewise be utilized to identify the beginning
disappointments of the machine parts, through the on-line checking framework, diminishing the likelihood of
disastrous harm and the down time [3].
Journal homepage: http://iaescore.com/journals/index.php/ijeecs
78  ISSN:2502-4752
Artificial neural networks (ANNs) have been connected in mechanized recognition and analysis of
machine conditions regarding these as grouping or speculation issues in view of taking in example from
illustrations or experimental information displaying [4]. In any case, the customary neural system approaches
have impediments on speculation offering ascend to models that can over fit to the preparation information
[5]. This lack is because of the streamlining calculations utilized as a part of ANNs for determination of
parameters and the factual measures used to choose the model. As of late, bolster vector machines (SVMs),
in view of factual learning hypothesis, are picking up applications in the ranges of machine learning, PC
vision and example acknowledgment due to high exactness and great speculation ability [6]. A recurrent the
functions of various vibrations based signals, attained under both normal and light loads and at low and high
sampling rates, are examined. The results explain the effectiveness of the extracted features from the attained
and preprocessed signals in analysis of the machine condition. Predictive Direct Power Control (PDPC) of
Grid-Connected Dual-Active Bridge Multilevel Inverter is explained in [7]. Proportional Integral Estimator
of the Stator Resistance for Direct Torque Control Induction Motor Drive is discussed in [8]. Comparison
Performances of Indirect Field Oriented Control for Three-Phase Induction Motor Drives is presented in [9].
Sensor less Control of BLDC Motor using Fuzzy logic controller for Solar power Generation is discussed in
[10].

Figure 1. Flow Chart of Diagnostic Procedure

1.1. Vector Machines
In vector machines has two-class issues—in which the information are isolated by a hyper plane
characterized by various bolster vectors. A concise presentation of SVM is exhibited here for culmination.
Peruses are alluded to the instructional exercises on SVMs for points of interest. The SVM can be considered
to make a line or hyper plane between two arrangements of information for grouping. If there should arise an
Indonesian J Elec Eng & Comp Sci, Vol. 9, No. 1, January 2018 : 77–80
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752  79
occurrence of two-dimensional circumstance, the activity of the SVM can be clarified effectively with no loss
of simplification. In Figure 4, a progression of focuses for two distinct classes of information are appeared,
circles (class An) and squares (class B). The SVM endeavors to put a direct limit (strong line) between the
two unique classes, and arrange it such that the edge (spoken to by specked lines) is augmented. The SVM
tries to arrange the limit with the end goal that the separation between the limit and the closest information
point in each class is maximal. The limit is then put amidst this edge between the two focuses. The
classification of vector machines is shown in figure 2.

Figure 2. Classification of Vector Machines

1.2. Proposed Artificial Neural Network
ANNs have been produced as parallel appropriated arrange models in light of organic learning
procedure of the human cerebrum. There are various uses of ANNs in information investigation, design
acknowledgment and control. Among various sorts of ANNs, multilayer perceptron (MLP) neural systems
are very mainstream and utilized for the present work. In this paper, the terms ANN and MLP have been
utilized reciprocally without different sorts of neural systems. MLPs comprise of an information layer of
source hubs, at least one concealed layers of calculation hubs or "neurons" and a yield layer. The quantity of
hubs in the information and the yield layers rely on upon the quantity of information and yield factors
separately. The quantity of shrouded layers and the quantity of hubs in each concealed layer influence the
speculation capacity of the system. For more modest number of concealed layers and neurons, the execution
may not be sufficient. The artificial neural network is shown in figure 3.

Figure 3. Artificial Neural Network

2. SIMULATION RESULTS
Table 1 shows the classification results for each of the sensor locations using the first 9 input
features (1–9). The test achievement was higher with SVMs in all cases aside from flag 3 where both were at
Application of Artificial Neural Network in Electrical Power System (P. Palanichamy)
80  ISSN:2502-4752
100%. For both classifiers, test achievement was unsuitable by and large. The calculation time (on a PC with
Pentium III processor of 533 MHz and 64MB RAM) for preparing the classifiers are additionally appeared,
however immediate examination is troublesome because of contrast in code productivity. In any case, it
ought to be specified that the distinction in calculation time caught be imperative if the preparation is done
disconnected.

Table 1. Comparison of Different Sensor Location
Straight ANN Straight vector machines
Data set Input features
Test success % Training time (S) Test success % Training time (s)
Signal 1 1-9 88 28 91 1.5
Signal 2 1-9 95 1 97 0.289
Signal 3 1-9 100 4 10 0.368
Signal 4 1-9 87 11 88 0.685
Signal 5 1-9 48 26 75 0.998
Signal 6 1-9 56 13 77 0.756
Signal 7 1-9 87 7.02 100 0.658

3. CONCLUSION
The parts of various vibration signals, gotten under both typical and light loads and at low and high
testing rates, have been researched. The arrangement exactness of SVMs was superior to of ANNs, without
GA. With GA-based choice, the exhibitions of both classifiers were similar at about 100%, even with various
load conditions and inspecting rates, utilizing just three components. For SVMs with six components, 100%
arrangement achievement was accomplished in all experiments even with a coarse range and step size of the
classifier parameter.

REFERENCES
[1] Kalogirou S A. Artificial neural networks in renewable energy systems applications: a review. Renewable and
sustainable energy reviews. 2001; 5. (4); 373-401.
[2] Kalogirou, S A. Applications of artificial neural networks in energy systems. Energy Conversion and
Management.1999; 40. (10); 1073-1087.
[3] Hsu Y Y, and Chen C R. Tuning of power system stabilizers using an artificial neural network. IEEE transactions
on energy conversion.1991; 6. (4); 612-619.
[4] Chemsi M, Agbossou K and Cardenas A. Neural network backpropagation algorithm control for PEM fuel cell in
residential applications. In Electrical Power and Energy Conference (EPEC). 2016; 1-6.
[5] Idris N H. Salim N A, Muhamad A F, Mahmud M N, Yasin Z M and Ab Wahab N. Application of Artificial Neural
Networks to the Identification of Power System Stability due to Transmission Line Outages. International Journal
of Simulation Systems, Science & Technology. 2016; 17. (41).
[6] Ibrahim M, Jemei S, Wimmer G, and Hissel D. Nonlinear autoregressive neural network in an energy management
strategy for battery/ultra-capacitor hybrid electrical vehicles. Electric Power Systems Research 2016; 136; 262-269.
[7] Goh, H. H., Aida, A., Lee, S. S., Sim, S. Y., & Goh, K. C. (2017). Predictive Direct Power Control (PDPC) of
Grid-Connected Dual-Active Bridge Multilevel Inverter (DABMI). International Journal of Power Electronics and
Drive Systems (IJPEDS), 8(4).
[8] Chaudhari, J. G., Bodkhe, S. B., & Aware, M. V. (2017). Proportional Integral Estimator of the Stator Resistance
for Direct Torque Control Induction Motor Drive. International Journal of Power Electronics and Drive Systems
(IJPEDS), 8(4).
[9] Aziri, H., Patakor, F. A., Sulaiman, M., & Salleh, Z. (2017). Comparison Performances of Indirect Field Oriented
Control for Three-Phase Induction Motor Drives. International Journal of Power Electronics and Drive Systems
(IJPEDS), 8(4).
[10] Sundaram, A., & Ramesh, G. P. Sensor less Control of BLDC Motor using Fuzzy logic controller for Solar power
Generation. International journal of MC square scientific research (IJMSR), 9(2).
Indonesian J Elec Eng & Comp Sci, Vol. 9, No. 1, January 2018 : 77–80