Realization of partial discharge signals in transformer oils utilizing advanced computational techniques

Abstract

The measurement of acoustic and electrical signals for the partial discharge (PD) activity due to the presence of metallic particles within transformer oil are utilized for the characterization of the incipient hazards. The utilization of phase resolved, in addition, to time resolved partial discharge signals is undertaken to extract numerous features using statistical and frequency analyzers. The extracted features are down scaled to pinpoint the effective attributes that render an intelligent classification means useful for determining contaminating particle type and dimensions. This is accomplished by utilizing feature selection wrapper models undertaking the sequential floating forward selection (SFFS) and particle swarm optimization as alternative search strategies. Support vector machines (SVM) is finally used for the classification of contaminating particles identity. A comprehensive comparison between various selection techniques of the best feature vector for the most efficient classification is tackled based on size of selected feature vector, processing time and success of classification. Results of this study can be integrated into a smart automatable tool based on recent data mining techniques that would provide an efficient and prompt identification for the nature of incipient hazards due to lack in insulation integrity.