Partial Discharge Identification Using Multiple Features Fusion Based on Evidence Theory and Support Vector Machine

Abstract

The time resolved partial discharge (TRPD) signals collected by Ultra High Frequency (UHF) sensors of transformer contain important informations of the properties of faults. Wealth of information can be utilized for fault identification by different signal analyzing techniques, such as time domain analysis, frequency domain analysis and the wavelet analysis. An integrated characteristic information fusion system based on the Platt posterior probability model and support vector machine (SVM) one-against-one method is proposed. From the recognition result of the four typical insulation defects, the recognition system shows strong advantages in accuracy and reliability.