Classification of Shunt Type Faults using Wavelet Transform and Convolutional Neural Network

Abstract

Power system fault detection and its classification play an important role in protection of AC systems. In this study, non-stationary fault patterns those include three phase voltages and current waveforms are obtained for various types of shunt type of faults on a long transmission model using Simulink. Fault location, fault resistance, and distance are considered as the key parameters to study their influence on fault patterns. There are three methods of developing mathematical models for fault classification problems; one: quantitative method; two: qualitative method and three: data-driven method. As the dimensions of the system increases, it is getting more complex to develop mathematical model that can capture the dynamic behaviour of system. Artificial neural network using Levenberg Marquardt has been explored to solve the problem of recognition of faults. The pattern recognition by Levenberg Marquardt somewhat lacks generality and the selection of topology is quite tedious task. On the other hand, it is important to differentiate the fault signal from the disturbances in voltage and current waveforms due to transient disturbances. The other data driven method use feature extraction techniques. Owing to the advantage of the wavelet transform (WT) technique selection of variable size window proportion to frequency of signal processing is proposed to differentiate the fault signal from transient disturbance signal. The noise in the experimental result gives rise to non-zero wavelet coefficient during the steady-state. This has been improved by removing the unwanted noise by selecting proper filter such that fault-induced transient remain retained. The fault signal data has been transformed into informative data using WT. Convolutional neural network (CNN) has been explored with the training set of informative data to solve the problem of fault classification. The results obtained from ANN and CNN are compared to illustrate the capability of CNN.