Feature Parameters Extraction of GIS Partial Discharge Signals Based on Multiple Scale Higherorder Cumulants Matrix Singular Value Decomposition

Abstract

Due to the distortion of feature parameters, the Gaussian white noise will reduce the recognition accuracy of insulation defect types based on partial discharge (PD) feature parameters. PD UHF signals produced by defect models are analyzed by multiple scale decomposition with harmonic wavelet transform. The higher-order cumulants extracted from each scale PD UHF signal are composed as a trajectory matrix. Singular value sequence matrix of this trajectory matrix is obtained by singular value decomposition. The maximum value and singular entropy of singular value sequence matrix are selected as the feature parameters. These feature parameters were extracted from PD UHF signals and inputted to the support vector machine classifier for defect type recognition. Compared with another traditional method, the proposed feature parameters extraction method has higher recognition accuracy and better anti-interference performance.