Partial Discharge Defect Identification Of Power Cables Based On Sample Augmentation

Due to the long-term operation of high voltage,high voltage cables are prone to operation failures due to external force damage,equipment defects,water tree invasion and other factors.Through partial discharge monitoring,cable faults can be found in time and given early warning,which is beneficial to ensure the safety of power supply.Partial discharge pattern recognition is one of the important links in partial discharge status monitoring,which is difficult to recognize.On the one hand,partial discharge signals are mixed with complex noise and interference signals in industrial field,and on the other hand,the similarity of partial discharge types is very high,which increases the difficulty of partial discharge recognition.Studying the pattern recognition of partial discharge is conducive to timely and accurate judgment of fault types,and is of great significance.In this paper,firstly,the causes of cable partial discharge are analyzed.Aiming at four typical cable fault types,the cable defect models are made respectively,and the partial discharge experimental platform is built.The PD waveform and PRPD three-dimensional map under each experimental voltage are generated by gradually pressurizing the defects,and the waveform characteristics are analyzed in detail.In the follow-up,the time-domain partial discharge pulse or PRPD map of various faults can be taken as the starting point of feature extraction,which lays a foundation for later pattern recognition.The working environment of cables is complex,and the partial discharge signals detected by electromagnetic coupling elements are seriously disturbed by noise.Extracting the partial discharge signals from noisy signals effectively is the key link to realize the preprocessing of partial discharge signals and the pattern recognition of power cables.A PD signal denoising method based on adaptive variational modal decomposition(AVMD)-adaptive wavelet packet is proposed.Firstly,AVMD is used to decompose periodic narrowband interference,Gaussian white noise and local broadcast signal into different basic modal components,and the periodic narrowband interference is filtered out to obtain PD signal containing only white noise;Then,the adaptive wavelet packet method is used to filter out Gaussian white noise,and a relatively pure PD signal is obtained.Simulation results show that this method has more obvious noise suppression effect and the highest similarity with the simulation signal.Aiming at the problem that the ratio of positive and negative samples in cable partial discharge data sets is very different,and it is difficult to extract the features of partial discharge signals manually and is easily affected by subjective uncertainty,a method of cable partial discharge pattern recognition based on GAN-CNN is proposed.The original one-dimensional time domain signal is converted into two-dimensional image by signal-image conversion method,and then the data is enhanced by GAN.Finally,the cable defect type is pattern recognized by CNN.The results show that GAN-CNN pattern recognition method solves the defect that traditional shallow pattern recognition methods need to extract features artificially,at the same time,it also greatly improves the recognition accuracy,and when the sample capacity is insufficient,the data is enhanced by GAN to solve the defect of insufficient samples.