Power transformer fault diagnosis based on a support vector machine and a genetic algorithm

Abstract

A fault diagnosis method was developed based on a support vector machine (SVM) and a genetic algorithm (GA) to improve the accuracy of power transformer fault diagnoses. The system receives 20 different inputs from 5 common dissolved gas analysis (DGA) approaches to create the original feature set. Then, the penalty parameters, the SVM kernel function parameters and feature subsets are randomly encoded into the GA chromosome using a binary code technique with the 5-fold cross validation accuracy of the training set used as the fitness function. The SVM parameters and the feature subsets are then simultaneously optimized by the genetic algorithm. Finally, DGA samples from the testing set are examined by the model trained with the optimal parameters and the selected feature subsets. The results demonstrate that this method is able to accurately distinguish power transformer faults. This method has fault diagnosis accuracy than GA-SVM models with a non-optimal feature subset, the IEC method, the back propagation neuro network (BPNN) and the Naïve Bayes method.