Study On Feature Extraction And Pattern Recognition Of Partial Discharge In Xlpe Cable Joint

Partial discharge is one of the main performance of cable accessories'insulation early fault and sudden breakdown, it is not only the reason which cause cable accessories insulation's further degradation, but also the main characteristics of cable accessories insulation situation. The inner insulation condition of cable accessories and defects can be discovered timely and accurate by developing the research of on-line monitoring and pattern recognition for PD in cable accessories, which has important significance of preventing accident faults of cable lines and insuring safe and stable operation of the power system.In this paper, waveform characteristics and statistical characteristics of different insulation defects PD signals are studied based on analyzing researches about PD detection and pattern recognition home and abroad. Two new cable PD signal pattern recognition methods are developed from different angles of partial discharge pattern. The main achievements are as follows:According to typical fault type and characteristics of PD in 110 kV, in order to simulation of cable joint internal defects, cable physical model of four defects are developed. In the laboratory, the tester use capacitive coupling sensor, a large number of four kinds of insulation defects'PD sample data are acquired by high-speed data acquisition system, construct the ? ? q ? ndistributions and gray image of PD in cable.Cable accessories'PD single waveform feature extraction method are developed, base on discrete wavelet transform and singular value decomposition: First, partial discharge data are processed by discrete wavelet transform, based on Birge-Massart strategy extracted the great value wavelet decomposition coefficients at all levels, form the great value coefficient matrix, condense sample data from the high-dimensional space into the low-dimensional space, then carried singular value decomposition on the coefficient matrix, use the singular values as PD feature values input to the classifier, the recognition results is satisfied.Introduced two-dimensional linear discriminant analysis, The method of extract feature vectors of cable PD gray image basing on two direction two-dimension maximum margin criterion is proposed. This algorithm can keep the original data's topological structure, project the image from two direction, the high-dimensional data is mapped to the corresponding low-dimensional space, eliminate the redundant information, the percentage of PD pattern recognition is improved effectively, and the pattern recognition consuming time is great shortened.