Abstract
•PD activity was measured and analyzed for various defects in gas insulated switchgear.•PD diagnosis based on PD pulse sequence was proposed.•PD diagnosis was implemented using different machine learning classification techniques.•Naïve Bayes classification exhibited the highest diagnostic accuracy.
Partial discharge (PD) diagnostics in Gas Insulated Switchgear (GIS) is important for reliable and secure operation of electrical utilities. Different techniques were used for PD diagnosis in GIS. In this work, PD diagnosis in GIS is proposed based on PD pulse sequence. PD pulse sequence only requires the measurement of PD phase appearance and its corresponding instantaneous voltage. The PD diagnosis of various defect types is implemented using five optimized machine learning classification techniques: decision tree classification, ensemble methods, k-nearest neighbouring, Discriminant analysis, and Naïve Bayes classification. The features used for PD pulse sequence are the voltage change and phase angle change between successive PD pulses. Three scenarios are proposed for predicting the defect types in GIS. The first scenario is built based on the extracted features for two successive PD pulses, the second scenario is built based on the extracted features for three successive PD pulses, while the last scenario is built based on the extracted features for four successive PD pulses. The results illustrate the superior detecting accuracy of the second scenario with the proposed five ML classification techniques. The optimized ML classification techniques are implemented and carried out based on MATLAB software package. The ensemble classification method exhibited the highest accuracy for PD-based diagnosis in GIS with an overall accuracy of 97.1%.