Research On GIS Partial Discharge Pattern Recognition And Localization Method

Gas insulated switchgear(GIS)has been widely used in power systems because of its small footprint and less interference from the outside environm ent,and its safe and reliable operation is crucial to the stability of the power systems.However,GIS inevitably has defects in the manufacturing,transportation and installation processes,all of which may produce varying degrees of partial discharge(PD).PD is an important parameter reflecting the insulation status of equipment,which is closely related to the insulation deterioration and insulation breakdown of the equipment.Therefore,PD detection of GIS can effectively obtain the insulation status of equipment to avoid the expansion of hidden faults and prevent the occurrence of major accidents.The Ultra-high frequency(UHF)method has been widely used in the field of PD detection for its advantages of strong anti-interference ability and high sensitivity.However,most studies of the propagation characteristics of UHF electromagnetic wave signals have used Gaussian pulses instead,which cannot adequately restore the electromagnetic wave signals excited by different defects;Secondly,the traditional pattern recognition method cannot directly identify the voltage signal output from the sensor causing a large amount of data waste,and the existing positioning methods generally face problems such as high cost and complex topology.Therefore,the UHF electromagnetic wave signal propagation characteristics in GIS,the reasonableness of the pattern identification method and the accuracy of PD source location are the key issues that need to be solved urgently.This thesis focuses on GIS UHF electromagnetic wave signal propagation characteristics,pattern recognition and PD source localization methods.Firstly,a simulation model is established based on the finite difference time domain method,and the measured PD current pulse is used as the excitation source to study the propagation characteristics of the UHF electromagnetic wave signal inside the GIS.The signal components in different directions were analyzed by voltage peak-to-peak and accumulated energy,and it was concluded that the radial component was the largest,indicating that the UHF sensor should be set along the radial direction.Then the effects of different radial positions on the UHF electromagnetic wave signal are compared,and it is pointed out that the sensor should be set along the radial position near the inner wall of the cavity.Next,the effect of cavity size on electromagnetic waves was investigated,and the discharge peak was reduced by increasing either the outer cavity size or the inner conductor size.Secondly,this thesis adopts the convolutional neural network model to study the pattern recognition of UHF time-domain waveform spectrograms directly,and optimizes the performance of the model in terms of optimization algorithm and network model respectively,the results of the study show that the DenseNet model has the best performance with recognition accuracy of 97.625%.Then the effect of sample size on the model identification results was investigated.Next,the pattern recognition method based on deep transfer learning is proposed for the problem of few samples faced by the equipment in the new working environment,which effectively improves the recognition accuracy of the model under few samples,the study results show that the recognition accuracy of the deep transfer learning model can reach 96%,and the training time is significantly reduced which ensures the effectiveness and real-time performance of the model.Finally,the PD localization method is studied based on Complementary Ensemble Empirical Mode Decomposition(CEEMD)and energy ratio method.The first step is to filter and denoise the UHF time domain signal using CEEMD to make the signal features more obvious.The next step is to obtain the first-wave arrival time of the signal using the energy ratio method.In the last step,a neural network is used to linearly fit the first wave arrival time and the axial distance between the sensor and the PD source to achieve the localization of the PD source,and the relative error of localization is basically maintained within 10%,which verifies the feasibility of the localization method.