约束过滤器与代理模型辅助的粒子群优化方法

Considering the unaffordable computational budget and poor convergence to feasible regions,a surrogate-assisted particle swarm optimization method using constraints filter (SAPSO-CF) is proposed for flight vehicle design optimization. A radial basis function is integrated with a particle swarm optimization framework to reduce the computational cost significantly. Then,a dual-filter sampling strategy is developed. The Kreisselmeier-Steinhauser function constraints filter is employed in the global exploration phase,which is subsequently cooperated with a radial basis function subspace based local search. In this way,the optimality and feasibility of the new infilling sample points can be improved simultaneously,which leads to a rapid convergence for the particle swarm. The experimental results on several numerical benchmarks indicate that the proposed SAPSO-CF outperforms the GLoSADE and C²oDE in terms of global convergence, robustness and optimization efficiency. Finally,SAPSO-CF is employed to deal with the solid rocket motor multidisciplinary design optimization problem. The total impulse performance optimized by SAPSO-CF is improved by 15. 3% compared with the initial solution satisfying the constraints of the combustor,nozzle,and other disciplines. Simultaneously,the optimality of SAPSO-CF is better than that of GLoSADE. The optimization results verify the effectiveness and engineering practicability of SAPSO-CF.