Power transformer fault diagnosis based on support vector machine with cross validation and genetic algorithm

Abstract

Support vector machine (SVM) classifier has been successfully applied to power transformer fault diagnosis. However, there is no theoretical basis or effective method to select appropriate SVM classifier parameters which have a crucial influence on the classification accuracy. Currently, the main method is cut and try based on experience. In this study, genetic algorithm (GA) is employed to optimize the SVM classifier parameters. Cross validation (CV) is used to estimate the performance of SVM classifier with different parameters during the optimizing process and the estimation result is used as the fitness function of GA. It ensures that the SVM classifier has better generalization. The SVM classifier with parameters optimized by GA and CV is applied to fault diagnosis of power transformer (CVGA-SVM). The experimental results indicate that CVGA-SVM has more excellent diagnostic performance compared with Grid-SVM, CVGrid-SVM and GA-SVM.