Partial discharge defect recognition in power transformer using random forest

Partial Discharge (PD) diagnostic become more important for high voltage (HV) equipment condition monitoring. PD phenomenon in power transformer could indicate insulation aging or degradation, which in long term could reduce the integrity of the insulation and leading to transformer failure. High accuracy of recognition rate for different PD defect is necessary for a successful PD diagnostic. This paper presents Random Forest (RF) method for PD defect recognition in power transformer. RF is one of supervised learning algorithm in machine learning. RF known as an ensemble classifier build using many decision trees. The majority vote of each three will determine the PD type. There are three defects used in this paper, protrusion, floating metal, and void. A commercial PD measurement system and detecting impedance was used to record the phase resolved partial discharge (PRPD) patterns of different defects. 8 PD statistical features extracted from PRPD patterns to identify each defect. To calculate the accuracy of RF method, different amount of PD features was use for recognition and then compare with other methods.