Abstract:The dissolved gas analysis method (DGA) in oil is an important method for internal fault diagnosis of transformers. It is widely used in transformer on-line monitoring and periodical test detection. Traditional diagnostic methods such as characteristic gas method and three-ratio method are commonly used in practical applications. These methods have some limitations which caused the fault diagnosis accuracy to be low. In order to solve this problem, this paper proposes a transformer fault diagnosis method based on multi-layer perceptron in deep learning technology. It uses the open source Scikit-learn machine learning framework and TensorFlow deep learning framework to construct a transformer fault diagnosis model. The fault diagnosis model was trained and tested by using the fault sample data in practical engineering. The test results show that the transformer fault diagnosis model based on multi-layer perceptron technology can correctly diagnose the transformer fault. Compared with the traditional three-ratio method and the support vector machine techniques, the multi-layer perceptron has a higher diagnostic accuracy and better fault diagnosis performance, which can provide more accurate reference information for transformer overhaul.