A new radial basis function neural network-based method for denoising of partial discharge signals

Abstract

Partial discharge (PD) assessment is one of the most prevalent techniques in the condition monitoring of high voltage (HV) equipment. In this regard, one of the main stages in the procedure of PD assessment is noise reduction of PD signals. Hence in this paper, a novel artificial neural network (ANN) based denoising method is introduced. Unlike the existing denoising algorithms, the proposed technique treats the noisy PD signal, in a data window, as a set of points that can be estimated by a curve. Hence, radial basis function (RBF)-ANN is utilized being a self-structured algorithm compared to multi-layer perceptron (MLP)-ANN that is influenced by the number of neurons. The performance of the proposed method has been compared with MLP-ANN, wavelet transform (WT) and empirical mode decomposition (EMD) based algorithms for different laboratory generated PD signals suppressed with white noise. The obtained results have proved the superiority of the proposed method.