The fixed BP (Back-Propagation) neural network structure can hardly play to its advantage when the input information become complicated and variable. So the decision- making algorithm is proposed, which combines information fusion with BP neural network. That is, using Dempster-Shafer(D-S) evidence theory to select the structure of BP neural network according to the changing input information. Simultaneously, the initial values are optimized by the Particle Swarm Optimization (PSO) algorithm to improve the problem of BP Neural Network's easily trapping into the local minimum and slow convergence rate. The simulation result shows that through the optimization of combined information fusion with BP neural network, the training time and prediction accuracy are more effective than that only using BP neural network, which has certain advantage of adapting to the complex and varied input information.