A novel multi-class fault diagnosis approach based on support vector machine of particle swarm Optimization and Huffman tree

Based on VC dimension theory and structural risk minimization principle of statistical learning theory, Support vector machine (SVM) has a prominent advantage in solving classification and fault prediction problems, specifically suitable for small sample, nonlinear and high dimensional pattern recognition problems. However, SVM is originally created for solving binary classification problems. The efficient application of SVM on multi-classification has always been a hotspot. This paper represents a novel approach to the multi-class fault diagnosis based on support vector machine of particle swarm optimization method. Besides the one-against-one, one-against-other, directed acyclic graph and binary tree, the Huffman tree is introduced, and the priority of the classification is determined by calculating dissimilarity degree of each two class. Thus, a multi-classification model based on Huffman tree is built. When the sample amount of each class varies greatly, using the same penalty parameter for each class will lower the classification accuracy. Thus, the penalty parameters of different class is optimized by particle swarm optimization method, which guarantee each SVM is the optimal result. Finally, a database of power transformer is used to demonstrate the superiority of this new method.