Study of Fault Diagnosis Model of Oil-Immersed Transformer Based on SVM Binary Tree with Combinatorial FKCM

This paper presents an improved binary tree algorithm for the compactness characteristics of the data sets of oil-immersed transformer in the pattern feature space. In order to improve the classification accuracy, the conception of combination is introduced in the training process of classifiers. The FKCM1 (fuzzy means kernel clustering) is used to choose the training sample of the normal mode and the fault mode; The FKCM2 is used to choose the training sample of the discharge fault mode and the thermal fault mode; The FKCM3 is used to choose the training sample of the thermal fault of low temperature and the thermal fault of high temperature; The FKCM4 is used to choose the training sample of the discharge of low energy and the discharge of high energy. This method overcomes the disadvantage that the traditional binary tree, which doesn´t consider the distributing situation of the data set, constructs directly the SVM classifier. The simulation experiments show that the SVM classifier constructed by this paper have higher testing accuracy than the traditional SVM multi-class algorithms and the binary tree.