SVM‐based method for high‐impedance faults detection in distribution networks

Abstract

Purpose The purpose of this paper is to present a new pattern recognition‐based algorithm to detect high‐impedance faults (HIFs), including only with broken conductor and arcs, in distribution networks. Design/methodology/approach In the proposed method, using discrete wavelet transform, the time‐frequency‐based features of the current waveform are calculated. Then, to extract the best feature set of the generated time‐frequency features, principle components analysis (PCA) is applied and finally support vector machines (SVM) is used as a classifier to distinguish between the HIFs, including only with broken conductor and arcs, and other similar phenomena such as capacitor banks switching, no load transformer switching, load switching, insulator leakage current and harmonic loads. Findings The experimental results have shown that using SVM with PCA as the feature extraction method and radial basis function (RBF) as the kernel function has acceptable security and dependability performances in distinguishing HIFs, including only with broken conductor and arcs, from other similar phenomena and is superior to the Bayes and multi‐layer perceptron neural network classifiers. Originality/value Using new combination of time‐frequency‐based features with SVM provides a new algorithm to detect HIFs, including only with broken conductor and arcs, that has acceptable security and dependability.