Investigation and classification of field leakage current waveforms

Abstract

Leakage current (LC) monitoring is a widely employed tool for the investigation of surface electrical activity and the performance of high voltage insulators. Surface activity is correlated to the shape of LC waveforms. Although field monitoring is necessary in order to acquire an exact view of activity and insulators' performance, field waveforms are not often recorded due to the required long term monitoring and the accumulation of data. Instead, extracted values, such as the peak value, charge and number of pulses exceeding predefined thresholds, are recorded, with actual waveforms either being recorded occasionally or not at all. However, a fully representative extracted value is yet to be determined. In this paper, 1540 field waveforms are investigated to acquire a detailed image of the waveforms' shape in the field. Simple classification rules are employed to distinguish between basic groups. Discharge waveforms are further classified based on the duration of discharges. Twenty different features, from time and frequency domain, two feature extraction algorithms (student t-test and mRMR) and three classification algorithms (knn, Naive Bayes, Support Vector Machines) are employed for the classification. Results described in this paper can be used to maximize the efficiency of field LC monitoring.