A Study of PRPD Statistics to Improve the Performance of PD Detection and Defect Type Identification

Insulation breakdowns in substations cause severe accidents. It is known that partial discharges (PD) tend to increase in advance of the insulation breakdowns. Monitoring the PD could be a clue for diagnosing the substation. In the diagnosis, phase-resolved partial discharge (PRPD) patterns are often used as a preprocessing. In this study, we examined which features of PRPD patterns were effective for PD detection and defect type identification, and applied these features to examine the effectiveness and the validity of features in the performance. As a result, the maximum phase difference of PD pulses and standard deviation of PD pulse amplitude were effective features for PD detection, and the dimensionality reduction by linear discriminant analysis (LDA) was the most effective for defect type identification. The future work involves applying the results to various situations in real substations.