Improved particle swarm optimization algorithms for aerodynamic shape optimization of high-speed train

Abstract

In this paper, improved particle swarm optimization algorithms are presented for improving the aerodynamic optimization efficiency of a high-speed train head shape. A hybrid particle swarm optimization algorithm, which employs an artificial fish swarm algorithm (AFSA) and a backward learning strategy in particle swarm optimization, is first proposed for constructing an optimal least squares support vector regression (OPT-LSSVR) model. The prediction accuracy of various surrogate models was evaluated, and the results indicate that the OPT-LSSVR model has the smallest prediction errors, where the prediction error for the total aerodynamic drag coefficient is reduced to about 0.45% and reduced to within 5% for the lift coefficient. Besides, an elite-evolved multi-objective particle swarm optimizer (EMPSO), which employs a grouping-based stochastic elite competition mechanism and elite gathering behavior, is proposed to improve the multi-objective optimization efficiency. The verified results indicate that the EMPSO algorithm is more efficient and capable of tackling complex multi-objective problems. Based on the OPT-LSSVR model and the EMPSO algorithm, the aerodynamic shape optimization of the high-speed train is performed. After optimization, the aerodynamic drag coefficient and the aerodynamic lift coefficient are reduced by 3.63% and 10.59%, respectively. Improved algorithms are simple yet efficient, and they have significant implications for the research and design of high-speed trains at higher speeds.