A systematic method to compose over complete dictionaries for the processing of partial discharge signals

On-site measurement of partial discharges (PD) can provide valuable information about electrical equipment insulation degradation. Classical signal processing techniques have been used for PD signal denoising, but few approaches evaluate or present effective methods to attenuate impulsive and amplitude modulated (AM) noises, which are commonly found in on-site measurements. In this paper, a PD denoising method based on an over complete wavelet dictionary and Basis Pursuit Denoising (BPD) is presented. We describe the overall denoising algorithm and propose a methodology to compose the over complete dictionaries, starting from candidate wavelet families. The procedure starts by identifying the best families, followed by a method to define the number of families and their best combination. We show that the procedure gives better results when compared to a random choice of families, and present several processing cases based on simulated and measured signals.