Transformer Fault Diagnosis Using Deep Neural Network

Abstract

Analysis of dissolved gases in transformer oil is one of the practical methods for identifying the different types of faults in oil-insulated power transformers. Dissolved gas analysis (DGA) is often exercised as part of the maintenance process, and the Duval Triangle is a commonly applied method for classifying transformer faults. This paper proposes using the deep neural network to identify transformer fault type. Due to limited availability of field data, we simulate DGA data samples along with the fault type determined by Duval Triangle. Numerical results show that the deep neutral network provides very high accuracy in fault type identification and outperforms other learning methods such as k-nearest neighbor (k-NN) algorithm and random forest classifier method.