Research On PD Pattern Recognition Technology Of Power Cable

Power cable transmission is one of the main forms of power transmission,in which cross-linked polyethylene(XLPE)cable,with its good electrical and mechanical properties,has become the mainstream product of power cable,but the research on its insulation state detection and fault diagnosis and other operation and maintenance technologies is relatively lagging behind.Partial discharge(PDS)is one of the main forms of deterioration and aging of power cable insulation.n the early stage of fault,the discharge signal caused by partial defect is very weak,so it is difficult to detect the abnormal signal in the traditional preventive test project,so the traditional test method has been unable to meet the requirements of cable safe operation.In this paper,the real-time online monitoring technology of cable is proposed,which can stop the discharge at the early stage and eliminate the fault in the bud.In this paper,an on-line partial discharge monitoring system is built.According to the requirements of on-line partial discharge monitoring of power cables,four kinds of typical discharge defect models and test circuits are designed.The four kinds of cable defects are respectively:creepage of the external conductive layer,air gap inside the insulation,scratch of the insulation surface and metal pollution on the insulation surface.When the defective cable is pressurized,the PD signal is detected by coupling partial discharge(PD)pulse with high frequency current sensor(HFCT),and a large number of discharge data are generated and the three-dimensional prpd spectrum is obtained.It not only realizes the real-time on-line monitoring of the cable,but also understands the characteristics of the time-domain partial discharge waveform and prpd spectrum of each defect,which lays the foundation for the later pattern recognition.Because of the complex working environment of the cable,the PD signal detected by the electromagnetic coupling element is seriously interfered by the noise.Extracting the PD signal from the noisy signal is the key link to realize the PD signal preprocessing and power cable pattern recognition.This paper presents a PD signal denoising method based on adaptive variational mode decomposition(AVMD)-adaptive wavelet packet.Firstly,AVMD is used to decompose the periodic narrow-band interference,Gaussian white noise and PD signal into different basic modal components,filter out the periodic narrow-band interference,and obtain the PD signal containing only white noise;then,adaptive wavelet packet method is used to filter out the Gaussian white noise,and obtain the purer PD signal.The experimental results show that this method is compared with one of them.The simulation results show that the noise suppression effect of this method is more obvious,and the similarity with simulation signal is the highest.The data preprocessing of PD signal is completed.In this paper,a feature fusion extraction method based on the combination of statistical and moment features is proposed to obtain the discharge information of cable defects from different perspectives,and the three-dimensional prpd map is constructed by using the PD phase resolved partial discharge mode,according to the three-dimensional map,we can get the?φ-q_p map,?φ-q_a map,?φ-n map and two-dimensional gray-scale map under prpd mode,Five feature parameters,such as skewness,steepness and local peak number,are extracted from the?φ-q_p map and?φ-q_a map in prpd mode as the feature quantity of time-domain signal,which mainly describes the overall geometric characteristics of the map.Two dimensional gray image is used to extract the corresponding moment feature,and two kinds of feature vectors are synthesized as the feature extraction quantity.The experimental results show that the feature fusion extraction method based on the combination of statistical and moment features is better than the feature quantity extracted by statistical analysis.As the feature input of pattern recognition,the accuracy of classification results is higher.In the model of pattern recognition based on support vector machine,different combinations of penalty factor C and kernel parameter g directly determine the final classification result.In this paper,support vector machine(PSO-SVM)based on particle swarm optimization algorithm is proposed as the classifier of pattern recognition.Through the iteration of particle swarm optimization,support vector machine(SVM)obtains the individual extremum pbest and the global extremum gbest.Through the two extremum updating speed and position,the best parameters c and g of SVM are obtained finally,and a more accurate SVM classification model is realized.However,the traditional support vector machine algorithm has no clear classification basis,it can not get the global optimal solution in parameter optimization,and the classification model is not accurate enough.The recognition results show that the final classification result of the pattern recognition classifier based on PSO-SVM algorithm is obviously better than that of traditional SVM.