Segmented Spectrum Energy Characteristics-based Density Clustering of Partial Discharge Signals

Recent research results and on-site operation experience have shown that there may be small and sporadic partial discharge (PD) before some insulation flashover, but the existing online monitoring methods cannot effectively monitor them, which may cause non-early-warning failures. Pulse excitation measurement method can be used to effectively improve the data storage efficiency, and can realize the high-precision and long-term measurement of sporadic pulses. However, further analysis of discharge characteristics and fault location need to screen and identify discharge pulses from a large number of signals, including interference signals. In this process, small and sporadic pulses are very likely to be ignored, which will affect subsequent analysis and judgment. Therefore, in this paper, a feature extraction method based on the signal spectrum was proposed, and the DBSCAN density clustering algorithm was used to process the discharge signals, which can classify the discharge signal automatically and effectively so that a large number of pulse signals can be screened and identified efficiently and quickly, to solve the identification problem of small and sporadic pulses, and to avoid the non-early-warning faults of the equipment to a large extent. At the same time, it also lays a good foundation for the subsequent pattern recognition and fault location, which makes greatly improve the efficiency of the overall processing of the PD signal.