In stamping process, springback is always determined by process parameters, such as blank-holder force, mould parameters, material parameters, and so on. Prediction of springback and parameters is a multi-objective optimization problem. Firstly, based on the same quantity of orthogonal experimental samples, prediction accuracy and efficiency of back propagation neural network (BPNN) prediction model and the response surface prediction model (RSPM) for springback of S-Rail forming were compared. As a result, RSPM was adopted benefit to less influence by sample scale and higher accuracy. Furthermore, a self-adaptive global optimizing of probability search algorithm, neighborhood cultivation genetic algorithm (NCGA) was proposed to optimize the prediction of process parameters. Then optimized parameters can be obtained quickly. Finally, valid of optimized parameters set, as well as the feasible of the prediction model based on both RSPM and NCGA were confirmed by the finite element analysis (FEA) test of S-Rail springback.