Developing Neural Network Models for Partial Discharge Analysis

Neural networks in recent years have seen a rise in partial discharge related applications, with efforts mainly focused on measurements from ultra-high frequency sensors or high frequency current transformers. Existing works do not include neural network analysis of time-resolved partial discharge measurements on in-service cables generated with an external energising source. The inherent convoluted nature of these waveforms is a complicated recognition task which traditionally requires costly domain expert interpretation. This paper compares several neural network models and proposes a method that performs highly accurate recognition whilst reducing cost. The effectiveness of the proposed procedure is demonstrated by evaluating the performance across statistical measures.