The power transformer is an important component of the power system, its safe operation has great influence on the reliable power supply of power system.Therefore, the paper focuses on the method for fault diagnosis of power transformer.The dissolved gas analysis of power transformer is an effective means of diagnosing internal faults of power transformer.The gradient descent principle is a general local optimization algorithm of neural network, which has slow convergence speed in fault diagnosis of power transformer, and is easy to fall into local minimum.While the genetic membrane algorithm has the characteristics of parallel computing, and can be used to effectively prevent neural network from converging to the local optimal solution, and has better fault diagnosis speed.Finally numerical simulations show the effectiveness of the given method.