Classification of partial discharge events in gas-insulated substations using wavelet packet transform and neural network approaches

Abstract

To ensure the safe and reliable operation of a gas-insulated substation (GIS), it is crucial to quickly identify partial discharge (PD) sources to prevent the occurrance of breakdowns. A method based on wavelet packet transform techniques is developed to meet this requirement. The proposed method extracts is able to extract features from ultra-high frequency resonance signals measured from a test GIS section. These features are subsequently used to train a neural network that is then able to quickly and reliably diagnose PD events. A quality-assurance scheme is developed that ensures the robustness of the PD classification to changes in the background noise level and the location of the PD event within the test GIS section.