A fault classification and localization method in a power distribution network

Abstract

This paper presents a method of fault diagnosis in a power distribution network. A segment of a 88 kV power distribution network is modelled in Digsilent Power Factory. Various types of fault cases are obtained through an Electromagnetic Transient study on the model. Discrete wavelet transform (DWT) is used to extract features from transient fault currents measured at the source terminal of the network. The method uses two cycles of transient fault current measured at the source terminal after fault inception. The extracted features are subsequently fed into a support vector machine (SVM), Naive Bayes, support vector regression (SVR) and Gaussian process regression (GPR) schemes in order to diagnose various system faults. SVM is used to detect and classify various types of short circuit faults. The Naive Bayes is also used to classify faults. Furthermore, SVR and GPR schemes are used to estimate fault locations. A hybrid method using DWT, SVM and GPR is thus proposed. The feasibility of the proposed technique is tested on MATLAB. The results of the proposed method show that various types of faults can be classified with accuracy of up to 98.8% and a minimum estimation error for fault location.