Research On Aerodynamic Shape Optimization Of Trains With Different Dimensional Design Variables

The traditional Kriging method is not efficient in handling high-dimensional optimization problems. In this paper, a neural network method is used as the surrogate model to optimize the aerodynamic performance of a simplified train. In order to study the mapping performance of the surrogate model, two sets of design variables are used. One is high-dimensional, the other is low-dimensional. The results indicate that the structure of the neural network should be changed according to the design variables. When the parameters are appropriate, the accuracy of the neural network exceeds the traditional Kriging method, especially in predicting lift force and in handling high-dimensional variables. In addition, making the train's bottom surface arc-shaped can reduce the upward lift force when the train encounters crosswinds. Finally, the drag, lift, and side forces of the optimized model were reduced by 2.409%, 20.712%, and 5.368%, respectively.