Fingerprint Extraction And Optimization Identification Of DC XLPE Cable Partial Discharge

Cross-linked polyethylene(XLPE)insulating power cables are widely used in the field of flexible HVDC transmission.Because the manufacturing and installation process of XLPE cable is relatively complex,and the operating environment is more stringent,different types of insulation defects are often produced at different locations of the cable.Partial discharge(PD)can cause insulation damage of cable equipment,but it also contains the information of defect types at the fault site.Therefore,PD measurement is considered as an effective means to detect and identify insulation defects of XLPE cable.Aiming at the four common types of insulation defects of DC XLPE cable,i.e.inner semi-conductive layer damage,inner air gap defect,surface scratch defect and creeping defect of outer semi-conductive layer,this paper establishes the corresponding physical test model.In the test process,three kinds of impact factors,voltage amplitude,voltage polarity and insulation aging time,are introduced,and PD tests are carried out under different conditions.Partial discharge test is used to study the influence of various factors on DC PD characteristics,and on this basis,the identification method of cable defect types under DC voltage conditions is explored.In the study of the influence of voltage amplitude and voltage polarity on PD,three kinds of insulation defects,i.e.inner semiconducting layer breakage,internal air gap and defective insulating surface scratch,were used to test the DC PD characteristics under different voltage amplitudes and polarities.It was found that the discharge repetition rate of inner semiconducting layer breakage defect under positive polarity was higher than that under negative polarity.The average discharge amount is higher than the corresponding value under negative polarity.The discharge repetition rate of air gap defect in insulation is higher under negative polarity.The average discharge amount is basically the same as that of positive polarity.The discharge repetition rate of scratch defect on insulation surface varies greatly with the average discharge amount,and the discharge repetition rate is obviously higher than that of the other two.When the voltage condition changes,the parameters of partial discharge fingerprint corresponding to the three defects fluctuate obviously.Regardless of the fluctuation of fingerprint parameters,the recognition accuracy of cable defect type is 72.93% by using backward transmission neural network.In order to improve the efficiency of defect identification under the influence of voltage,this paper proposes an optimization method of fingerprint parameters extracted by GA-BP algorithm.The model considers the fluctuation of fingerprint parameters caused by voltage amplitude and voltage polarity.Through screening the correlation proportion of each feature parameter under different test conditions,the feature parameters that do not contain or contain less information about defect types are eliminated.The traditional fingerprint and the optimized fingerprint are used to identify the fault types respectively.It is found that the recognition rate of the traditional fingerprint will decrease significantly after the voltage changes are introduced,while the identification rate of the optimized fingerprint is relatively stable,and remains above 97.30%.In order to study the influence of insulation aging time on partial discharge,four defect models,i.e.internal semi-conductive layer damage,internal air gap,surface scratch and creep of external semi-conductive layer,were used to carry out the experiments.The results show that the partial discharge characteristics of different XLPE cable defect models are various under long-term insulation aging test,and the partial discharge characteristics of single defect type also fluctuate obviously under different insulation aging stages,which affects the extraction of fingerprint parameters.For the influence of insulation aging on fingerprint parameter fluctuation,an optimal identification model for defects of DC XLPE cable based on bidirectional cyclic neural network is proposed.Based on the phased processing of partial discharge process,the model takes both the fingerprint parameters and phase information of each stage as the input of the model,and the output of each stage is not only related to the current stage information,but also affected by the input information of the front and back stages.Therefore,the model can effectively reflect the time-sequence characteristics of partial discharge data and improve the identification accuracy rate of defect types.By calculating the data set obtained from four kinds of insulation defects,it is found that the recognition rate of bi-directional cyclic neural network model is 93.71%,while the recognition rate of defect type based on traditional fingerprint parameters is only 72.93%.The results show that the defect recognition model based on bi-directional cyclic neural network algorithm can effectively eliminate the influence of insulation aging on defect identification using partial discharge fingerprints,thus improving the recognition efficiency.