Research On Feature Extraction And Defect Recognition Method Of PD In Cable Under DC Voltage

With the steady development of the power industry,the voltage level of DC cable operation has also increased.However,due to the collision of the DC cable during production,transportation,installation and operation,defects are formed in the cable.Any slight defects that develop under the high DC operating voltage may damage the insulation of the DC cable,even causes a power outage,so small defects are a great threat to the safe and stable operation of the DC cable system.Partial discharge(PD)can reflect the insulation state of cable equipment to a large extent.Therefore,it is of great significance to study the method of feature extraction and defect identification of PD signals.PD detection is usually performed on the AC XLPE cable to determine if it is faulty and to confirm its operating status.For DC cables,the current research on PD signal feature extraction and defect identification is in its infancy.In this paper,the feature extraction and defect recognition technology of DC XLPE cable PD images are studied.The DC XLPE cable PD test platform is built,the cable insulation defect model is designed,and the corresponding PD images are collected.In this paper,the enhancement of PD images,feature extraction,feature space dimensionality reduction,defect type recognition and other aspects are studied and practiced,and some achievements are obtained.The PD images generally have poor contrast,low edge definition,and insufficient feature information.The paper proposes a DC cable PD image enhancement method based on multi-scale Retinex method based on cuckoo search optimization(CS-MSR)and brightness preserving dynamic fuzzy histogram equalization(BPDFHE)algorithm.Firstly,the non-subsampled shearlet transform(NSST)is decomposed to form a low-frequency sub-band diagram and a high-frequency sub-band diagram,respectively,using different enhancement methods,using the CS-MSR method to process the former,and using the BPDFHE method to process the latter,and performing weighted reconstruction to form a more prominent PD image.Combining the two methods for PD image enhancement is better than applying only one algorithm,and has better effects in definition,edge intensity,grayscale mean,contrast,and the like.The wavelet domain lacks directionality,has sparse representations only for discontinuous points,and cannot effectively capture the line and surface singularities present in the PD images.In this paper,the feature parameter extraction method based on non-subsampled contourlet transform(NSCT)domain is proposed.Firstly,the PD image is subjected to NSCT processing to obtain subimage.The subimage is extracted with a large number of characteristic parameters,such as maximum discharge times,average discharge time interval,average discharge amount and other parameters.The method can effectively extract features from defects.Compared with statistical features,entropy features and two-dimensional wavelet transform(2DWT)feature extraction methods,the average recognition accuracy is improved,and the variance of recognition accuracy is also significantly reduced.The traditional clustering dimension reduction method can not effectively reduce the sample size,eliminate the redundancy between samples,and the computational efficiency is relatively low.In this paper,a feature space reduction method based on improved immune algorithm optimized affinity propagation(IA-AP)algorithm is proposed.The feature set of DC cable PD image based on NSCT domain is reduced and processed into Multi-kernel support vector machine(MSVM)and compared with other cluster dimensionality reduction algorithms.analysis.The average recognition accuracy of this method is higher than other methods.In addition,under the condition of the same cluster reduction algorithm and the same number of feature parameters,the average recognition accuracy increases with the percentage of the number of training samples in the total number of samples,which also indicates that the feature subset has good performance.In order to solve the problem that the recognition accuracy of the traditional recognition method is low and it is difficult to adaptively optimize the parameters of the defect recognition model.The paper proposes a defect type identification method based on improved error-correcting output codes(ECOC)classifier,and extracts the feature parameter of PD image by using improved cuckoo search optimized sparse coding matrix(CS-SR)-ECOC to identify the DC cable insulation defect.The experimental results show that the coding matrix selection proposed in this paper is more effective,and the recognition result is higher than the traditional ECOC classifier.