Adaptive Separation Method for Interference Pulses in Partial Discharge Signals of Converter Transformers Based on IFCM Clustering

Wideband partial discharge (PD) detection has been widely applied due to its ability to retain rich discharge information. However, the presence of significant noise at converter stations severely interferes with on-site PD detection. Effectively separating PD from noise interference is key to improving the accuracy of wideband PD detection. Traditional denoising algorithms based on signal decomposition techniques can suppress noise, but they also lead to significant attenuation of PD signals. Signal separation techniques based on waveform features can retain the complete waveform and reduce PD signal attenuation, emerging as a new trend in denoising. However, existing algorithms based on clustering are affected by data distribution, making it difficult to effectively separate PD signals from interference pulses, causing noticeable noise residue. Moreover, these algorithms struggle to directly handle long-duration signals collected on-site, as the signals first require extracting each pulse individually. To solve the problems, this article proposes a denoising algorithm based on adaptive separation of interference pulses. Initially, pulse waveforms are extracted from the raw signal based on the instantaneous rate of change of energy in the time series, thereby removing nonpulse noise. Then, using the −20-dB bandwidth, kurtosis, and peak factor (PF) of the pulse waveforms as features, an iterative fuzzy C-means (IFCM) clustering is applied to progressively separate interference pulses, ultimately preserving PD signals. The denoising results for PD signals from the laboratory, superimposed with on-site noise, and measured signals from on-site discharge calibration and PD testing demonstrate that the algorithm can adaptively remove complex interference while fully preserving the PD signals.