With the development of artificial intelligence technology, neural networks, fuzzy technology, expert systems, grey system theory, fuzzy clustering and other methods have gradually been applied to transformer fault diagnosis, and have achieved better diagnostic results. However, the above methods all have certain shortcomings. For example, knowledge-based methods such as artificial neural networks need to obtain an infinite number of fault samples, and the training time is long, and there are problems such as local optimal solutions; This paper proposes a transformer fault diagnosis method based on the improved DDAG-SVM, which enriches the transformer fault diagnosis information, and combines the dissolved gas composition, content and change in the oil, operating voltage and current and other information to establish a comprehensive fault diagnosis system based on the transformer. It is helpful to guide the efficient development of maintenance.

Transformer fault diagnosis , Support vector machines, Directed acyclic graph, Overheating failure , Decision tree .