GNC-network: a new tool for partial discharge pattern classification

Abstract

A new neural network classifier is presented that was designed to optimize the recognition of partial discharge patterns. PD patterns resulting from various model defects are used to investigate the performance of the classifier. The classification results are compared with results obtained by a neural backpropagation network. It is shown that the classification performance can be improved when applying a suitable PD parameter, different from those commonly used. The results indicate that the new tool presented here is able to overcome typical problems inherent in most neural network based PD pattern classification approaches.