PD Pattern Classification for dc System Based on Fractal Dimensions Combined with Statistical Features

Abstract

For the purpose of identifying the defects within the insulation at dc system, a 3-dimensional image of partial discharge (PD), which is based on the theory of wavelet analysis and different from the traditional y-q-n image, is proposed in this paper. Its three parameters are the frequency, the time and the amplitude. Then the fractal dimensions combined with the lacunarity that is a measure of denseness of the fractal surface in the point of probability are computed. In succession the back-propagation neural network (BPNN) is used for the classification. With acoustic PD signals gathered in artificial defect experiments, the final results of the BPNN show that the method performs effectively in recognizing the PD patterns.