Research On Feature Optimization And Type Recognition Of Partial Discharge In GIS Based On UHF

Gas Insulated Switchgear (GIS) has been widely used in the electric power system, due to its significant advantages of compact structure, small floor area, free maintenance and reliable operation. But in the manufacturing, transportation, installation, maintenance and long-term operation of the GIS, it may cause partial discharge. The dangers and corresponding treatment measures of the different discharge types there are some differences. So the research of partial discharge characteristics and diagnostic method for GIS has great significance to the safe and stable operation of the GIS and guide the on-site processing strategy. At present, there are a lot of researches on the UHF (Ultra High Frequency) partial discharge pattern recognition. However, the statistics show that the low recognition accuracy rate problem in the field application remains a serious bottle neck for UHF partial discharge detection and diagnosis.In order to realize the effective and accurate diagnosis of the discharge type for UHF partial discharge in GIS, and solve the problem of difficult to obtain accurate information on the discharge phase in the field, five kinds of typical insulation defect models in GIS are designed in this paper, specifically, on the basis of keeping the basic characteristics for each type model, by changing the material, size and shape, some styles were designed. The UHF partial discharge model test was carried out on the GIS platform and a large amount of data can be obtained. Based on these data, the PRPD (phase-resolved partial discharge) spectrogram, PRPS(Phase Resolved Pulse Sequence) spectrogram, △u spectrogram, △t spectrogram, N-V spectrogram and N-△t spectrogram are showed. And then, by analyzing the spectras of different discharge types,30 features that phase independent are extracted to describe the information of PD, these features include the statistical features, fractal features and image features. In the thesis, the method that calculating the features after accumulated discharge pulse reaches a certain number is proposed. Finally, with the hierarchical PD identification method and BP artificial neural network, the diagnostic results of different features are analyzed and compared.The results show that, based on the statistical features extracted in this paper, by using the hierarchical PD identification method and the method that calculating the features after accumulated discharge pulse reaches a certain number, the recognition accuracy rate can be up to 95%.