Abstract: Aimed at the disadvantages of the general BP arithmetic, such as convergence speed slowly and easily plunged in the minimum of the local, this article puts forward a method of raising the study speed of BP network and build up a model of magnetic characterisitics Ψ(θ,i) of SRM based on BP neural network . The parameters of neural networks in this article has been optimizedly selected, the training time and step number has greatly been reduced. The procedure runs smoothly, after being trained, recogrized and forecasted three fixes, and it has strong capability of studying generalization, strengthened greatly the real-time and robustness of the system. The magnetic characterisitics model built in this article is critical to the optimized energy conversion and the reduced torque ripple.