A Novel Random Forest Model Approach for Accurate Classification of Single Partial Discharge Sources of HV Transformer Insulation Faults

摘要

Transformers play a critical role in maintaining the reliability of the power transmission and distribution system. Failure of high voltage transformers due to insulation problems is a major issue faced by electrical utilities. In recent times, partial discharge (PD) detection and analysisis a well recognized insulation condition monitoring technique for power equipments. However, accurate classification of PD signals originating from different PD sources is always a vital and hot research issue. This paper attempts to use Random Forest (RF) classifier, which produces accurateresults on large data bases, to deal with partial discharge classification. In this work, PD sources normally experienced in transformers such as corona discharges, internal discharges and surface discharges were simulated in the laboratory using different electrode configurations and thecorresponding PD signals were acquired using wide band detection system. Phase resolved PD pattern and time-frequency characteristics of PD pulses were analyzed and important statistical features were extracted. RF model was trained and tested using the extracted statistical features of PDsignals and reported results show that the proposed RF based PD source classifier is very efficient and will be useful for understanding PD sources of transformers.