Partial Discharge Pattern Recognition Based on Multiscale Features of Original Time Domain Signals

The detection of partial discharge (PD) is one of the most important methods for cable insulation condition diagnosis. When PD signal is detected in the cable, it often indicates that there are defects in the cable insulation. PD detection and pattern classification can provide information about the type and severity of cable insulation defect, which can guide cable maintenance and ensure the safe operation of cable. In this paper, four kinds of artificial defect model were prepared for PD simulation experiments of HV cable, including corona discharge, internal discharge, suspended discharge and surface discharge. Data were collected by high speed data acquisition card and stored in computer. After de-noising, the original time domain signal was decomposed by harmonic wavelet packet transform (HWPT), and the relative energy and sample entropy of different sub bands were calculated. Support vector machine (SVM) was applied to distinguish different type defects. Relative energy and sample entropy were used as features to train and test SVM. The results showed that each of the four kinds of defect recognition rate was over 90%, the average recognition rate was 94.5%. Therefore, the multiscale relative energy and sample entropy parameters based on HWPT can be used as characteristic quantities to realize the pattern recognition of PD signal of HV cable.