In order to solve the complicated nonlinear problem of the warhead preformed fragments dispersing direction control, based on a large number of experimental and simulation data, BP neural network technology was used to predict the design modification value of preformed fragments dispersing direction, and the fragment focusing curve optimization design system was established, which could automatically optimize the fragment focusing curve designed by the traditional analytical model. The test results show that the optimized focusing curve has high fragment focusing efficiency and can effectively reduce the iteration tests of fragment focusing curve.