Classification of Partial Discharge Sources in HVDC Gas Insulated Switchgear using Neural Networks

Partial Discharge (PD) activity in HVDC Gas Insulated Switchgear (GIS) is caused by various defects in the insulation system. Therefore, it becomes necessary to correctly identify the type of defect causing PD activity and then rectify the problem. In this contribution, convolutional neural networks (CNN) are investigated to classify the type of PD source in the ultra-high frequency (UHF) PD measurement data obtained from an HVDC GIS. This analysis considers three types of defects: fixed protrusions, free-moving particles and electrodes at a floating potential. Both one-dimensional (1D) and two-dimensional (2D) CNNs are implemented, and their accuracy is compared. The total dataset consists of 6923 PD signals from the aforementioned defects. The training dataset consists of 80% of the whole dataset, and the test dataset consists of the remaining 20%. It is found that the CNNs can successfully distinguish between three kinds of PD types. Additionally, the floating potential PD can also be distinguished based on the electrode’s polarity where it occurs (i.e., positive or negative electrode), thus classifying four types of PD.