Study on Time-Frequency Entropy Method to Make Feature Extraction for DC PD Pulse Waveshapes

How to develop a dc partial discharge (PD) detection and pattern recognition system for multi-PDs and noise suppression used for HVDC test is introduced in this paper. The fast grouping technique for dc PD pulsed current sequence (PCS) is used to develop the system, which is composed of parameters extraction and unsupervised clustering analysis methods for dc PD pulsed current waveforms in time domain. Focus is made on the time-frequency entropy method (TFEM) used to make parameters extraction for dc PD PCS measured by wideband detection technique with a sampling rate of 100 MS/s, which gives a 2D parameters distribution. After the parameters extraction made for the noised simulation PD signals, the results are analyzed using an unsupervised clustering method, the competitive learning network (CLN), which shows that TFEM does a good separation for dc PD pulsed current waveforms with the same characteristics and the noise pulses. This provides a good method to make parameters extraction for dc PD pulsed current waveforms to develop the multi-PDs detection and pattern recognition system.