Identification Of Internal Defects In GIS Equipment

At present,gas insulated switch(GIS)equipment has gradually replaced traditional insulated switches due to its high reliability,low maintenance cost and small size,and is an important power equipment widely used in unmanned substations.However,once the GIS insulation is deteriorated,a tip corona discharge will be caused,or an electric tree channel discharge or the like will be formed due to internal discharge,resulting in a malfunction during the operation of the power system.Because different defect types of GIS will produce different partial discharge characteristics,partial discharge detection and analysis can be performed on the running power equipment to achieve the purpose of preventive equipment fault diagnosis.In this thesis,the inverse transfer neural network is used as the identification method of defect type partial discharge data.For three GIS equipments with different internal defects,partial discharge measurement is performed,and the phase analysis map is obtained by extracting the discharge signal,and image processing techniques such as image interpolation and two-dimensional wavelet transform are applied to perform data on the phase analysis map.Compression and feature extraction,then construct a three-layer feedforward inverse transfer neural network,use the local phase analysis map after data reduction and processing as the input neuron vector of the neural network,and use the neural network to perform GIS internal defects.Identification training and testing of the discharge map.This thesis mainly found the following conclusions:(1)After using the two-dimensional wavelet transform,the four sub-image functions are used as input neurons as the defect identification test.The success rate of the defect can be found.Compared with the original phase-analyzed map,when the low-pass is used twice When the filtered c A sub-image function with low-frequency components is used as the input vector of the neural network,the defect recognition success rate can be significantly improved.In contrast,the recognition success rate of c H sub-image function,c V sub-image function and c D sub-image function with high-frequency components is not good.The reason is that the high-frequency components have high similarity and are difficult to distinguish.(2)When the fault identification of the phase analysis map is directly used atthe voltage level of 34 k V and the voltage level of 42 k V,the recognition success rate is not high,which may be caused by the discharge phenomenon caused by the non-defective position when the voltage level is high.The effect of high frequency noise is more serious.(3)The use of multi-order two-dimensional wavelet transform has a limitation on improving the recognition success rate.The higher the order is,the higher the recognition success rate is.At the 26 k V voltage level,the total recognition success rate after the second-order two-dimensional wavelet transform is the highest;at the34 k V voltage level and the 42 k V voltage level,the total recognition success rate after the third-order two-dimensional wavelet transform is the highest.(4)When the number of neurons in the hidden layer is less than 50,the number of neurons has no significant effect on the success rate of defect recognition.(5)In the 26 k V voltage level,after the image interpolation method,the second-order two-dimensional wavelet transform is used to obtain the c A sub-image function with the matrix size of 10×10 as the input neuron.The number of hidden layer neurons is At 10 o'clock,the average recognition success rate of the three different defect types obtained was the highest,which was 96.83%.In addition,under other conditions,the average recognition success rate of defects is above 87%.Therefore,the neural network established in this paper can effectively identify the defects of GIS and provide reference for its maintenance.