Energy Entropy Feature and Diagnosis of Partial Discharge Wavelet Packet in GIS Based on Support Vector Machine

摘要

Two physical models, Gas Insulated Switchgear (GIS) metal spike defect on high-voltage guide pole and metal spike defect on the surface of basin insulator, were designed to identify the GIS fault types in insulation faults and in turn locate the fault sources. The pulse current method was selected to measure the partial discharge signal. An on-line diagnosing method for wavelet packet energy entropy feature of PD in GIS, based on the support vector machine (SVM), was proposed. The 10-layer wavelet packet transform was conducted to refine the frequency band, followed by the first 130 frequency bands being divided into 13 sub-bands to calculate their energy entropy and analyze the data characteristics. The sub-band energy entropy was composed into the energy entropy vector and later brought into the SVM algorithm for fault discrimination, leading to an accuracy of 98.125%. This method shed some light on analyzing and diagnosing the GIS insulation fault.