Transformer Fault Diagnosis Based on Hybrid Sampling and Support Vector Machines

Abstract

Aiming at the impact of transformer imbalanced data set on transformer fault diagnosis model. A transformer fault diagnosis method based on hybrid sampling and support vector machines (SVM) is proposed. It uses synthetic minority oversampling technique (SMOTE) and under sampling method based on nearest neighbor rules to underestimate transformer fault data and normal data, respectively. Sampling and oversampling, and then using the balanced data obtained by hybrid sampling training based on support vector machines transformer fault diagnosis model. The performance of the SVM-based transformer fault diagnosis model is compared through the test set under imbalanced data and balanced data. Finally, the influence of sampling rate on the diagnostic accuracy of transformer fault diagnosis model is analyzed. Experimental results show that this method can effectively reduce the impact of imbalanced data on the diagnostic model and improve the diagnostic accuracy of the transformer fault diagnostic model.