A fault cause identification methodology for transmission lines based on support vector machines

Abstract

This paper works on developing an algorithm based on support vector machines (SVM) which can automatically analyze, characterize, and classify a fault based on its root cause. Only single-phase grounding faults caused by external factors including lightning faults, wildfires faults, guano-caused flashovers, insulator contamination flashovers, object contacts and vehicle accidents are considered in this paper. From detailed analysis of the fault mechanisms and the waveforms of fault recorders, six influential factors are selected to characterize the six types of outages as follows: weather, season, time of day, DC component and high-frequency harmonic component in the zero sequence current and fault impedance magnitude. Discrete Fourier transform (DFT) is used for the analysis the frequency components of the fault phase voltage and current waveforms. The combination of these characteristics are used for training and testing the SVM architecture. In addition, genetic algorithm is applied to the SVM classifier to determine the optimal parametric settings which is proved that it can achieve higher classification accuracy. Successful testing of the proposed methodology proves its validity for identification of different fault reason types.