Research On Partial Discharge Pattern Recognition Of Switchgear Based On UHF Technology

During the production process of switchgear,there may be hidden dangers such as insufficient smoothness of metal parts or defective insulation.In addition,due to the complex operating environment of the switchgear,the long-term exposure to high temperature,condensation,and salt spray in the process of operation will lead to the reduction of insulation performance and the occurrence of partial discharge,which will accelerate the deterioration of insulation and cause damage to the power system.greater economic loss.In order to realize the collection of partial discharge signal of switchgear and the diagnosis of partial discharge type,this paper designs four typical insulation defects by analyzing the mechanism of partial discharge and combining with the common insulation fault types of switchgear.In the laboratory environment,the defect model is placed inside the cable compartment of the switchgear to simulate the occurrence of partial discharge in the switchgear under the working conditions.The partial discharge signal is sampled and uploaded by the partial discharge ultra-high frequency detection device.In this paper,based on a large number of experimental data,the partial discharge signals of 250 power frequency cycles are superimposed to construct partial discharge PRPD（partial discharge phase statistics）and PRPS（partial discharge pulse sequence distribution）maps,and the PRPD map is gridded to obtain a grayscale image with a resolution of 36×30.In the PRPS map,the interval time between adjacent discharge pulses was used to obtain the partial discharge Δt map.By analyzing the distribution law of the spectrum under different discharge types,the feature values of the spectrum are extracted from the image moment feature and statistical feature respectively.Due to the short distance between phases inside the switchgear,it is impossible to collect accurate partial discharge phase information.The partial discharge maps constructed in this paper all use relative phases,that is,take the sampling starting point as the phase 0 degree,and collect 20 ms data backward as a power frequency cycle,and part of the eigenvalues that characterize the phase are discarded during the eigenvalue selection process.Finally,two methods of BP neural network and convolutional neural network are used to diagnose and analyze the discharge types.In order to verify the robustness of the network,random noise interference is added to the discharge map.In the BP neural network architecture,the moment features and statistical feature values of the graph are used as input.In this paper,7 groups of data are obtained by permutation and combination of PRPD spectral moment feature,PRPD spectral statistical feature and Δt spectral feature,which are respectively used as BP neural network input to obtain the recognition accuracy of different eigenvalues and find the optimal combination.In the convolutional neural network,the grayscale image of the PRPD map is directly used as the input of the neural network,and the feature value of the image is automatically extracted through the convolutional neural network to obtain the classification result.The experimental results show that the recognition accuracy of the BP neural network is significantly reduced after being affected by the noise signal.Compared with the BP neural network,the convolutional neural network cancels the manual calculation of eigenvalues in the partial discharge defect recognition,and has higher accuracy and good adaptability to random noise signal interference.