Study On Partial Discharge Characteristic Analysis And Recognition Of High Voltage DC Cables

HVDC cable transmission is one of the major forms of power transmission.Cross-linked polyethylene(XLPE)cables have become the mainstream of HVDC cables due to their good electrical and mechanical properties.In recent years,the manufacturing technology of HVDC cables has been developing rapidly.However,research on detection and fault diagnosis of the insulation status of DC cables lags behind relatively.Partial Discharge(PD)is one of the main forms leading to degradation and aging of power cable insulation.And PD detection is an important method for insulation condition detection and early fault diagnosis of cables.Studies on PD characteristics,detection standard,and defect recognition of PDs in AC XLPE cables have been very rich.However,there is few research on similar aspects of HVDC cables.On the one hand,research on DC cables started later.On the other hand,studies on the characteristics of DC PDs are more complicated than AC PDs because of their lack of phase information.In response to the above problems,this paper has achieved the following results:The process of DC PD is studied based on the equivalent circuit model.Correlations among key parameters such as discharge magnitude,discharge repetition rate and applied voltages are analyzed theoretically.In addition,the correlation between two successive discharges is studied based on the time delay/ recovery model.By summarizing the common types of insulation defects in HV power cables and their causes,four typical insulation defect models are designed: a scratch on the surface of the main insulation,a metal burr inside insulation,a piece of semi-conductive layer material remains on XLPE,and the defect at the stress cone.The electric field distortion characteristic of each defect is studied using ComsolMulti-physics software and MATLAB.The above research mainly provides theoretical support for the following analysis of the test results.The testing and detecting platform of DC PD is set up.Firstly,the sensitivity and applicability of several detection methods are analyzed and compared,including the High Frequency Current Transformer(HFCT)detection,the Ultra High Frequency detection and so on.After that,each cable defect model is tested under step-wise DC voltage.The features of PD development are studied based on the correlations among the key parameters: discharge magnitude,discharge repetition rate and applied voltages.Each discharge process is divided into three discharge stages: the initial stage,the development stage and the near breakdown stage.By assessing the severity of each stage,the discharge process is divided into two severity degrees as well.A variety of statistical characteristic plots are obtained based on the continuous pulse data which are detected by HFCT,including the density histogram of discharge magnitude and time interval H(q,?t),and four kinds of scatter plots reflecting the continuous discharge relationship.The plots of each severity degree are comparatively analyzed.The DC PDs development and evolution characteristics of the typical defects,as well as the statistical rules of key parameters are revealed.Different types of defect,as well as different degrees of each kind of discharge are regarded as different discharge patterns.The above mentioned statistical plots contain rich pattern information.A variety of feature parameters are extracted for quantitative analysis of PDs.Firstly,the scatter plots are used.On the one hand,parameters such as Spearman coefficient are extracted to represent the linear correlation.On the other hand,the nonlinear correlation are characterized using Mutual Information,Maximum Information Coefficient and its extended parameters.Secondly,six types of 2-D statistical histograms are used.Operators such as skewness are introduced to reflect their shape features.Based on a large number of experimental data,the distribution range and features of each parameter for different discharge patterns are statistically analyzed.Furthermore,the performances of eachparameter are quantitatively evaluated by using the improved Maximum Relevance and Minimum Redundancy algorithm,and the “Optimal Fingerprint” is constructed after screening.Considering the features of H(q,?t)and the characteristics of time-domain pulse,a high-dimensional feature space is composed of the discharge repetition rate vector and the norm eigenvector of time domain signals.All the high-dimensional eigenvectors of training samples compose an overcomplete dictionaries.Based on the sparse representation of the eigenvector of the test sample,Compressed Sensing(CS)technique is used to realize the DC PD pattern recognition.Further,the recognition based on CS and the recognition based on the fingerprint of scatter plots are respectively taken as two evidence bodies.The decision information of these evidence bodies is fused.On the one hand,the restrictions of CS on the underdetermined conditions and the training samples' number are reduced.On the other hand,multi-source PD pattern information can be used efficiently,and a better recognition results can be obtained.