Partial Discharge Source Identification Using Phase-Resolved Patterns and a Weighted K-Nearest Neighbors Method

This paper introduces a methodology for identifying partial discharge (PD) sources through phase-resolved patterns and the application of a weighted k-nearest neighbors (WKNN) algorithm. The discrete wavelet transform (DWT) is employed to mitigate noise in signals acquired from a high-frequency current transformer (HFCT). The four-quadrant model of partial discharge patterns is utilized. Initially, three distinct types of partial discharge (PD) simulators are evaluated: corona discharge, surface discharge, and internal discharge, representing the three prevalent forms of PD in high-voltage equipment. Moreover, phase-resolved partial discharge (PRPD) and phase-resolved pulse sequence (PRPS) methodologies are adopted to diagnose various PD sources. The four-quadrant model of PRPS patterns is subsequently analyzed in relation to these PD types. Additionally, the correlation between the maximum and average values of PD signals suggests that Quadrants I and III serve as practical bases for WKNN, providing a reference for future evaluations of the three PD sources. The findings confirm that this approach accurately classifies different PD sources during field measurements.