Application of affinity propagation algorithm based on manifold distance for transformer PD pattern recognition

2017 2nd International Conference on Mechatronics and Electrical Systems

机械制造;无线电电子学

Wei, B.G.^1 ; Huo, K.X.^2 ; Yao, Z.F.^1 ; Lou, J.^2 ; Li, X.Y.^2

Electric Power Research Institute, Shanghai Electric Power Company, No.171 Handan Road, Shanghai, China^1

School of Electrical Engineering, Shandong University, No.17923 Road, Shandong Province, Jinan, China^2

关键词: Affinity propagation;    Condition maintenance;    Fuzzy C-means algorithms;    Information dimensions;    Least square methods;    Oil paper insulation;    Phase resolved partial discharges;    Physical characteristics;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/339/1/012014/pdf
DOI  :  10.1088/1757-899X/339/1/012014

来源: IOP

【 摘 要 】

It is one of the difficult problems encountered in the research of condition maintenance technology of transformers to recognize partial discharge (PD) pattern. According to the main physical characteristics of PD, three models of oil-paper insulation defects were set up in laboratory to study the PD of transformers, and phase resolved partial discharge (PRPD) was constructed. By using least square method, the grey-scale images of PRPD were constructed and features of each grey-scale image were 28 box dimensions and 28 information dimensions. Affinity propagation algorithm based on manifold distance (AP-MD) for transformers PD pattern recognition was established, and the data of box dimension and information dimension were clustered based on AP-MD. Study shows that clustering result of AP-MD is better than the results of affinity propagation (AP), k-means and fuzzy c-means algorithm (FCM). By choosing different k values of k-nearest neighbor, we find clustering accuracy of AP-MD falls when k value is larger or smaller, and the optimal k value depends on sample size.