Research On Feature Extraction And Severity Assessment Of Partial Discharge In GIS

Gas insulated switchgear(GIS) is used widely in power system for its excellent performance. However, some latent defects are inevitably produced in manufacturing, assembling or operation, which will induce partial discharge(PD) and then result in insulation deterioration or failure. It is significantly important to detect the PD signals effectively. We can explore the relationship between the PD signals and the defects type, so as the relationship between the PD signals and the severity of PD, then identify the defects type and assess the severity of PD, and send out warning signals before faults occur finally.Four typical GIS internal defect models was designed in the laboratory, and a large number of ultra-high frequency(UHF) PD signal samples under different physical conditions were acquired by using the UHF method, and a step voltage method was adopted to study the development process of PD under the four typical protrusion defects. The main work and achievements are as follows.1) Considering the differences of the distribution and complexity of singular information between the wavelet coefficients of each layer, a novel signal processing method was proposed, which combined dual-tree complex wavelet transform(DT-CWT) with singular value decomposition(SVD). An optimal complex wavelet decomposition algorithm was proposed based on the relationship between the singular entropy and the wavelet decomposition level. Besides, two kinds of features, namely the largest singular values and singular entropies, were extracted based on Hankel matrix to diagnosis defects types. Because these features not only contain the differences of the distribution of PD signals in different decomposition level, but also the complexity characteristics of each level, the recognition rates are improved evidently.2) The φ–u and φ–n spectrograms of UHF PD signals in various PD stages were obtained. Furthermore, we extracted night features, such as discharge amplitudes, discharge times, and so on, to characterize the degree of deterioration of insulation. We analyzed their values, variation trends, and change rates, and the relevant physical mechanism by considering the microphysical process of PD formation and development and by incorporating the distortion effect generated by the space charges on the initial field. It proves that it is effective and reliable of using the features to assess PD severity.3) According to the variation law of PD development process and its harmfulness to insulating medium, the development process was divided into four stages: initial discharge stage, discharge developing stage 1, discharge developing stage 2, pre-breakdown stage, and the four stages were defined four different severity states of PD, namely normality, attention, abnormality, and warning state. Fuzzy C means clustering algorithm was used to calculate the value of the nine features under the state centers. Fuzzy membership functions were adopted to solve the problem of dividing the fuzzy boundary of PD developing process. Moreover, the adaptive objective weight method, which is based on maximum deviation theory, was applied to calculate the weights of all features and to solve the problem of weight distribution. To avoid the phenomenon wherein some weights were distributed to small weights and the presentation of incorrect or meaningless results, which result in excessive feature parameters, a two-level fuzzy comprehensive evaluation model was constructed to eliminate this problem. Finally, the effectiveness and reliability of using the features to assess PD severity were proved by testing a large number of PD samples and a random example.