Genetically Aerodynamic Optimization of High-Speed Train Based on the Artificial Neural Network Method

Abstract

The optimization of the car body cross-section shape of high-speed train under cross winds has been performed using genetic algorithms (GA), wherein the computational fluid dynamics (CFD) simulation model was established based on Design of Experiments (DOE), and then the aerodynamic performance of a train subjected to crosswind was calculated by CFD. However, directly using the CFD to compute the high-speed trains performances in an optimization scheme is not suitable for the optimization process. This is because the GA requires a large number of CFD solver calls, which increase the computational cost and time-consuming of the optimization process. To decrease the CFD simulation computational cost, an approximate substituted Artificial Neural Network (ANN) model was proposed. The ANN models were trained and tested using the data obtained from the CFD simulation model, and its correlation coefficient values are considerably close to 1, which indicates fine accuracy and prediction capability. Meanwhile, the impact of the car body cross-section design parameters on the aerodynamic performance has been also investigated, and the result show that the design parameters interrelate each other and jointly impact the aerodynamic performance of high-speed trains. After the optimization, the overturn moment coefficient has been reduced by 17.5%, and the lift force and slid force coefficient have been reduced by 11.5% and 8.05%, respectively. Compared to the original shape, the pressure coefficient of optimal shape has great improvement. The analysis method can provide a frame of reference for the high-speed train safe operation running on the track under crosswind conditions.