Kernel statistical uncorrelated optimum discriminant vectors algorithm for GIS PD recognition

The fault diagnosis of gas insulated switchgear (GIS) partial discharge (PD) is significant for mastering the essence of defects within the GIS accurately and guiding its maintenance. This paper designed four typical single defects and three mixed defects in GIS, then PD 3-dimensional patterns were constructed based on mass sample gathered by the very-high frequency and high speeds sampling systems. As to the problems in linear Fisher analysis used for diagnosing the failures of PD, this paper puts forward kernel statistical uncorrelated optimum discriminant vectors (KSUODV) algorithm to solve the problem of non-linear feature extraction in High-dimensional feature space based on kernel method and to eliminate statistical correlation between transformed sample features. The recognition result of the PD patterns of seven defects obtained in the lab proves that KSUODV algorithm for recognition is prior to SUODV algorithm.