Partial Discharge Classification Based On Wavelet And Fractal Theory

In recent years, great improvements have been made in the method of on-line monitoring high voltage power apparatus by means of measuring electrical partial discharge (PD). In this method, it is essential to investigate the characteristics of the PD pulses to evaluate electric insulation conditions. Recognition of PD types is fundamental for insulation system diagnosis. In the past decade, theφ-q-n PD pattern has been investigated for PD recognition and shown encouraging results. However, this pattern is not suitable for DC voltage application. Another prominent disadvantage of using this patterns or statistical operators is that it can be significantly influenced by the applied electric stress.In this research, a novel 3-dimensional pattern of partial discharge was set-up based on the theory of wavelet analysis. The PD pattern could reflect the discharge characteristic in both time-domain and frequency-domain. The parameters which characterize the PD were the frequency, the time and the energy.It is important to identify the type of these defects, e.g. internal discharge, surface discharge, corona, etc, so in this paper, five kinds of typical discharge models are designed and the relevant experiment methods are projected. In order to extract the PD pattern features, the authors investigated the application of fractal features to characterize the new 3-D PD patterns produced from different PD experimental models. PD pattern recognition scheme was put forward. In this scheme,PD fractal dimensions were used as feature vector, Back-propagation (BP) neural network was used as classifier. The discharge experiments have been carried out to validate the method with five types artificial discharge models. The results indicate that the proposed methodology has the recognizing and classifying ability for the PD types.