PDE parametric modeling with a two-stage MLP for aerodynamic shape optimization of high-speed train heads

The aerodynamic drag of high-speed trains has a negative effect on their running stability and energy efficiency. Since the shape of the high-speed train head closely influences its surrounding airflow, optimizing the head shape is the primary way to reduce the aerodynamic drag. However, existing optimization methods have limitations in parametrically describing the train head with enough details and fewer parameters. In this paper, we propose a novel parametric modeling method based on the approximate analytical partial differential equation (PDE) for the aerodynamic shape optimization of high-speed train heads. With this method, the detailed shape of the train head is controlled by four design parameters. To enhance the optimization efficiency, a two-stage multilayer perceptron (MLP) surrogate model is proposed to predict the aerodynamic drag coefficients of the high-speed train, and a classic genetic algorithm (GA) is adopted to optimize the total drag coefficient and generate the train head shape with good aerodynamic performance. The effectiveness of the proposed method is demonstrated through several comparison experiments.