Locating partial discharges in a power generating system using neural networks and wavelets

This paper describes how a neural network can be used to identify the location of a partial discharge event in a Power Generation System consisting of a generator, isophase buss duct, main step up transformers and radio frequency sensors. Wavelets are used to reduce the dimensionality of sensor data before using it as an input to the neural network. The results obtained indicate that a neural network with 165 inputs and 35 hidden units can predict whether a partial discharge has occurred and identify one of twelve possible locations for the discharge with an accuracy of 70%. This neural network tool could reduce the time required for maintenance during power plant outages.