Detection and Identification Techniques for Multi-PD Source in GIS

SI Wen-rong,LI Jun-hao,YUAN Peng,YANG Jing-gang,LI Da-jian,LI Yan-ming

DOI: https://doi.org/10.3321/j.issn:0258-8013.2009.16.019

2009-01-01

Abstract:A novel detection and identification approach of multi-partial discharge (PD) in gas insulated switchgear (GIS) based on wideband detection technique is presented in this paper. The realization of the approach involves the classification of PD pulse sequence and identification of each sub-group with phase resolved partial discharge (PRPD) based on least squares support vector machine (LS-SVM). The classification module is composed of the time-frequency feature extraction for PD pulses and the competitive learning network (CLN) unsupervisory clustering, which divides PD pulses into sub-groups, in each group PD pulses show homogeneous stochastic features. Then the fingerprints of each sub-group are deduced from PRPD, and the identification is performed with LS-SVM discriminant function trained with the PD fingerprints produced by single insulation defect of GIS. Simulation test results show that the technique is feasible and practical, which provides a good approach for developing the multi-PD detection and identification system for GIS.