Abstract: Multi-objective optimization design method of the streamlined head of high-speed trains is proposed to improve the aerodynamic performance of high-speed trains under crosswinds. The side force and lift force of high-speed trains under crosswinds are set as optimization objectives and the automatic multi-objective optimization design of the streamlined head of high-speed trains is carried out. The parametric model of the streamlined head of high-speed trains is established and 5 optimization design variables are extracted. The flow field around high-speed trains under crosswinds is computed based on computational fluid dynamics(CFD) method. The multi-objective genetic algorithm is used to update optimization design variables to achieve the automatic optimization design of the streamlined head of high-speed trains. The correlation between the optimization objectives and optimization design variables is analyzed to obtain the most important optimization design variables, and further analysis of the nonlinear relationship between the key optimization design variables and the optimization objectives is obtained. After the multi-objective optimization design, a series of Pareto-optimal head types can be obtained, and the aerodynamic performance of Pareto-optimal head types under crosswinds has been significantly improved. Meanwhile, in order to ensure that the basic aerodynamic performance of high-speed trains with zero wind condition does not deteriorate, 8 Pareto-optimal head types are selected. For these 8 Pareto-optimal head types, compared with the aerodynamic performance under crosswinds and the basic aerodynamic performance with zero wind condition of the original train, the side force under crosswinds is reduced up to 3.06%, the lift force under crosswinds is reduced up to 19.60%, the drag force with zero wind condition is reduced up to 4.51% and the lift force with zero condition is reduced up to 9.68%.

Key words: crosswinds, genetic algorithm, high-speed train, multi-objective optimization, parametric model