Wavelet-PCA Based Denoising of Partial Discharge Signals

Abstract

Partial Discharge measurements contain valuable information about the state of the insulation system that is useful for the diagnosis of electrical equipment.Signals that result from on-site PD measurement are often completely buried in noise.One critical task for PD analysis and diagnosis is the recovery of clean PD signals from heavily contaminated measurements.Removing noise from PD measurement remains a serious challenge because the majority of the noise encountered during on-site measurement has similar time and frequency characteristics to the desired PD signals.This paper presents a sophisticated technique for denoising PD measurement.More specifically, we propose to combine Wavelet Transform (WT) and Principal Component Analysis (PCA) to recover PD signals buried in excessive noise.The first phase is based on multiresolution signal decomposition via wavelet.In the second phase, the coefficients that result from signal decomposition are filtered using PCA to retain the most informative signal characteristics of the measured PD signal.PCA automatically discards the uncorrelated noise that remains in the wavelet coefficients.An assessment of the proposed denoising technique is carried out, by processing experimental PD data.The results indicate that the proposed technique not only suppresses all kinds of noise present but also effectively preserves the features of the PD signals.