Partial Discharge Pattern Recognition Based on Kolmogorov-Smirnov Cluster Analysis of Discharge Pulse Sequences

Partial discharge (PD) detection is an important way to evaluate the insulation state of power equipment. The phase-resolved partial discharge (PRPD) pattern has too many fingerprint features and the spectrum will be blurred with noise under field conditions. This article starts from the perspective that can directly describe the microscopic nature of the random characteristics of PD and selects the statistical properties of PD pulse sequences, such as pulse time interval and pulse amplitude probability distribution. In this paper, a novel PD pattern recognition method is developed based on Kolmogorov-Smirnov (K-S) cluster analysis. First, various typical defect types such as void, creeping, and corona discharge pulse sequences are obtained through the corresponding simulation model. Then, the optimal characteristic vector is formed with the discharge sequence's pulse interval and pulse amplitude. Finally, based on the two-dimensional K-S distance (K-S2D), the cluster analysis and identification of different discharge types are realized. In addition, this paper discussed the robustness of the proposed method to outliers. The numerical simulation test results show that the proposed method can effectively distinguish PD types, with high recognition accuracy and strong anti-noise ability.