Power Transformer Fault Diagnosis Method Based on SMOTE and Convolution Soft Threshold Network

Abstract

Power transformers play an important role in the entire power grid. However, the fault diagnosis method based on machine learning suffers from decreased diagnostic performance when faced with redundant information interference and unbalanced data interference. In order to solve the above problems, this paper proposes a power transformer fault diagnosis method based on SMOTE and convolutional threshold neural network. First, a convolutional soft threshold network is proposed, which introduces the soft threshold function into the convolutional network to strengthen the perception of important information and suppress redundant information interference. Then, the SMOTE method is introduced into the proposed method, which can generate minority class samples, making the data set more balanced and alleviating the generalisation performance degradation caused by data imbalance. The proposed method is tested on a real power transformer fault data set, and the experimental findings demonstrate its superiority and efficacy.