Proposal of a novel fitness function for evaluation of wavelet shrinkage parameters on partial discharge denoising
Partial discharge (PD) measurement is an efficient method for monitoring the integrity of the high-voltage equipment. However, extensive interferences can compromise the measured PD signals. After the acquisition, a post-processing algorithm is required for the effective extraction of PD signals from noise. The wavelet shrinkage is one of the most popular tools used in such application. The wavelet processing involves the choice of parameters that will determine the performance of denoising. With the purpose of finding suitable parameters, optimisation techniques, such as genetic algorithms (GAs), can be used in conjunction with a fitness function to evaluate how good a solution is. This work proposes a new fitness function, based on a composition of local variables and statistics usually employed for the evaluation of PD denoising algorithms. The proposed function was applied to a set of PD signals and proved effective in the identification of optimal parameter values by a GA, thereby producing better PD denoising.