A New Transformer Fault Diagnosis Method Based on Classified Deep Auto-encoder Network

Abstract

The essence of transformer fault diagnosis is a multi-classification problem which has the characteristics of few fault sample data and a lot of uncertainties. According to the present transformer fault diagnosis methods, the back propagation neural network (BPNN) requires a large amount of fault sample data and is computationally intensive. At the same time, the adjustment of the coefficient is difficult for support vector machine (SVM). In order to remedy these defects, the paper proposes a new method of transformer fault diagnosis based on classified deep auto-encoder network(CDAEN), which optimizes the parameters of CDAEN model by the pre-training with massive unlabeled samples and adjusts them with a few labeled samples. The results show that the diagnosis speed of CDAEN can meet the engineering requirements and diagnosis accuracy is higher compared with BPNN and SVM.