Based on the computational fluid dynamics, a numerical model of deep-sea landing vehicle was established to meet the requirements of deep-sea landing vehicle for large scale and long period submarine operation. The key design variables were determined by parametric analysis of the control point of the landing vehicle bow modeled line, which was carried out in the Isight software through the combination with SolidWorks and Fluent. The optimal latin Hypercube method is used to select sample points for the direct navigation resistance calculation, an approximate model of design variable-resistance was established based on the radial basis function neural network, and the optimal design of landing vehicle bow modeled line was carried out by using the adaptive simulated annealing algorithm. The optimized resistance is reduced by 14.24%, and the hydrodynamic performance is improved. This work provides a basis for the optimization design of the deep-sea landing vehicle and can also provide a reference for the numerical analysis of other submersible vehicles.