Transformer dissolved gas analysis using least square support vector machine and bootstrap

This paper presents a least square support vector machine (LS-SVM) approach to dissolved gas analysis (DGA) problems for power transformers. Two methods are employed to improve the diagnosis accuracy for DGA analysis. Firstly, bootstrap preprocessing is utilised to equalise the sample numbers for different fault types. Then, the preprocessed samples are inputted to a classier for fault classification. For comparison purposes, four classifiers are utilised, i.e. artificial neural network (ANN), K-nearest neighbour (KNN), simple SVM and LS-SVM. The classification accuracy of LS-SVM is then compared with the ones of ANN, KNN and a simple SVM. The results indicate that the LS-SVM approach can significantly improve the diagnosis accuracies for transformer fault classification.