Diagnosis of Transformer Dissolved Gas Analysis Based on Multi-class SVM

Support Vector Machine (SVM) is based on statistical learning theory which developed from the common machine learning. It is an effective tool to deal with limited samples. This paper proposes a model of the dissolved gas analysis (DGA) of transformer based on Multi-class SVM. Firstly, with the combination of SVM multi-class classification methods one-versus-rest (1-v-r) and one-versus-one (1-v-1), the normal state and fault state of transformers are diagnosed. Then the fault data is diagnosed in detail with 1-v-1 method. The grid search method based on cross-validation, which is effective in solving the problem of designing SVM parameters, is used to determine model parameters. The results of case study show that the model ensures a very high accuracy rate of classification and has upstanding generalization ability.