序列Kriging仿真优化方法综述

Review on Simulation Optimization Method of Sequential Kriging

通过深度学习来挖掘设计变量、目标参数与Kriging模型之间内在关系的序列Kriging仿真优化方法已成为基于元模型优化的研究前沿和热点。但仿真优化过程中存在建模效率较低、收敛精度不高、多点采样的并行仿真难以实现等问题。如何在少量昂贵仿真估值条件下提高优化效率和收敛精度是序列Kriging仿真优化方法研究的主要内容。为此,对序列Kriging的近似建模方法、无约束优化、多点并行优化以及约束优化进行综述,介绍经典优化方法、若干改进及相应工具包,并展望所面临的问题和挑战。

Sequential Kriging simulation optimization method for mining the intrinsic relationship between design variables,tar⁃get parameters and Kriging through deep learning has become a research frontier and hotspot.However,there are problems such as low modeling efficiency,convergence precision,and parallel simulation of multi-point sampling.How to improve the optimiza⁃tion efficiency and convergence precision under a few expensive evaluation conditions is the main research content.To this end,the approximate modeling of Kriging,unconstrained,multi-point parallel and constrained optimization are reviewed.The classi⁃cal optimization method,several improvements and corresponding toolkits are introduced,and the problems and challenges are prospected.