Abstract: For solving expensive constrained multi-objective optimization problems in parallel computing environments,a multi-objective parallel surrogate-based optimization method based on constraint prediction improved aggregation strategy was proposed,which constructed constrained prediction improvement aggregation based on the decomposition of prediction objectives into single objectives,and the multi-point parallel design through influence function was implemented.This method could fully utilize the rich computing resources in practical engineering problems,which helped to improve the efficiency of optimization design.The test results showed that the proposed method could effectively improve the optimization efficiency of expensive multi-objective constrained optimization problems and shorten the calculation time required for optimization design.Compared with similar techniques,Pareto optimization solutions had good quality characteristics and certain advantages in terms of convergence,spatial distribution and diversity of solutions.

Key words: expensive multi-objective optimization problem, Kriging model, constrained prediction improvement aggregation, parallel surrogate-based optimization, spontaneous electrical buffer back frame design

摘要： 针对并行计算环境下的昂贵约束多目标优化求解高耗时问题,提出了基于约束预测改进聚合策略的多目标并行代理优化方法.该方法在预测改进函数分解的基础上构建约束预测改进聚合策略,采用最大化距离分解函数实现多点并行设计,并在并行计算环境下实现多点仿真的同步估计。该方法一方面充分利用实际工程中丰富的计算资源,实现优化设计效率的进一步提升;另一方面,所构造的约束预测改进聚合策略仅进行一维积分运算,具有计算复杂度低的优势。测试算例及自发电缓冲背架优化结果表明:所提方法可有效提升昂贵多目标约束优化问题的优化效率,进一步缩短优化设计所需计算时间;与同类方法相比,Pareto优化解具有良好的质量特性,在解的收敛性、空间分布性及多样性方面均具有一定优势。

关键词: 昂贵多目标优化问题, Kriging模型, 约束预测改进聚合准则, 并行代理优化, 自发电缓冲背架设计