Parameter Optimization for Support Vector Machine and Its Application to Fault Diagnosis of Power Transformers

Support Vector Machine classifier is an effective method for the fault diagnosis of power transformer.However,there is no theoretical basis or effective methods to select appropriate SVM classifier parameters which have a great effect on the performance of SVM classifier.Because genetic algorithm(GA) is one of the most common optimization techniques and cross validation(CV) is widely accepted a standard procedure for choosing proper model parameters and estimating model performance.In this paper,SVM classifier with parameters optimized by GA combined with cross validation is applied to power transformer fault diagnosis(CVGA-SVM).In the method,GA is used to search for the optimal parameters of the SVM classifiers and CV is used to estimate the performance of SVM classifier determined by difference parameters and learning set.The method can make full use of the limited power transformers fault sample data and improve the generalization of SVM classifier.Experimental results show that CVGA-SVM has more excellent diagnostic performance compared with the SVM classifier with parameter optimized by Grid,Grid combined with CV and GA.