Abstract: Traction transformer, as an important equipment of railway power supply, plays an important role in power transmission and voltage conversion of electrified railway, and its operation determines the safety and stability operation of the whole traction power supply system. Due to the harsh operating environment of the traction transformer, it is also subjected to various internal and external stresses such as “electric-magnetic-force-thermal”, and complex load characteristics, which makes the difficulty of fault diagnosis of the traction transformer increased. The research on the fault diagnosis of traction transformers can not only give timely warning before the failure of traction transformer, but also can provide sufficient theoretical support for the overhaul of traction transformer, so as to improve the safety and efficiency of railway power supply. A new method for traction transformer fault diagnosis based on dissolved gas in oil is proposed by combining Adam optimization algorithm based on classical momentum concept with PSO algorithm. Firstly, a PSO-RBF traction transformer fault diagnosis model is constructed, through the simulation experiment of nonlinear acceleration factor collocation, the nonlinear exponential decreasing collocation is used to improve the optimization ability of particle swarm optimization. The IPSO-RBF-ADAM traction transformer fault diagnosis model is better than PSO-RBF model in diagnosis accuracy and stability by simulation analysis of traction transformer fault data under the jurisdiction of a railway bureau.