Automated recognition of partial discharge in oil-immersed insulation

Abstract

This paper presents an application of pattern recognition techniques for identification of partial discharge sources in oil-immersed insulation. Three sources of partial discharge are simulated to generate artificial partial discharge data; bubble wrap to simulate air bubbles, needle to simulate corona discharge, and metal particles. Fingerprints from phase resolved partial discharge patterns are extracted. Dimension reduction techniques are employed to reduce the size of the collected data. Two classifiers (k Nearest Neighbor and Support Vector Machine) are developed for partial discharge source identification. The results show that the proposed classifiers are well able to identify the sources of partial discharge.