20 series of empirical parameters were obtained by applying optimal Latin Hypercube method in Isight, and then the external flow field around Ahmed model was simulated with these parameters in Fluent. On this basis, Kriging model was used to create approximate model. Taking the results from wind tunnel experiments as the optimization goal, the empirical parameters in the Realizable kε turbulence model were optimized with the Multiisland Genetic Algorithm. Finally, these parameters were applied to the development of some real road vehicles, and were verified with the wind tunnel experiment. The results have shown that, by using the optimized empirical parameters, the contour of velocity from simulation is much closer to that of the experiment, and the drag force coefficient error also decreases by 2.8% with fast convergence velocity.