Waveform Analysis and Feature Extraction of Partial Discharge UHF Signals in Gas Insulated Switchgear

Ultra high frequency (UHF) method gradually becomes an effective approach for partial discharge (PD) detection in gas insulated switchgear (GIS). The waveform of UHF signals can be acquired by digital detection instruments with a high sampling rate. By analyzing the waveform characteristics of UHF signals, the extracted features can be applied to separation of mixed signals originated from multiple PD sources. Consequently, we proposed a signal feature extraction algorithm based on waveform analysis technique. The envelope curves of UHF signals were obtained by using the Gaussian smoothing method, and some features such as rising time and falling time were extracted. The zero-cross rate was extracted to represent the oscillating characteristics of UHF signals. Moreover, the mathematical morphology gradient (MMG) of cumulative energy function was calculated to characterize its rise steepness. Thereafter, the extracted features were applied to separation of mixed UHF signals. Experiments on four typical defects, including protrusion, free-moving particle, floating electrode, and metal wire, on spacer were performed, and the classification performances of the extracted features were tested with the acquired signals. By clustering the extracted features with the fuzzy maximum likelihood algorithm, mixed UHF signals of two multi-defect models were successfully separated. The results suggest that the proposed method is effective for representing UHF signal characteristics. The extracted features can accurately classify UHF signals generated by different typical defects in GIS, and can effectively separate the PD signals of multiple defect sources.