In this study, a systematic optimization method for the thermal management problem of a passenger vehicle was proposed. This article addressed the problem of the drive shaft sheath surface temperature exceeded allowable value. Initially, the causes and initial measures of the thermal problem were studied through computational fluid dynamics (CFD) simulation. The key measures and their parameters were determined through the Taguchi method and significance analysis. A prediction model between the parameters and optimization objective was built by the radial basis function neural network (RBFNN). The prediction model and particle swarm optimization (PSO) algorithm were combined to calculate the optimal solution, and the optimal solution was selected for simulation and experimental verification. Experiment results indicated that this method reduced the drive shaft sheath surface temperature promptly, and the decreasing amplitude was 22%, which was met the experimental requirements.