Application of neural nets for modelling partial discharge phenomenon

The classical approach to modelling is a quasiempirical relationship based on experiments on single artificial voids of well defined geometry. Such methods restrict the validity to the range of inputs considered. Keeping all this in view, this work attempts to apply an artificial neural network (ANN) for the modelling in order to exploit flexibility of ANN modelling with a short time for development and reasonably high accuracy. The results indicate good agreement of the estimates with the published values with a MAE (mean absolute error) of as low as 1%.