演化排列优化拉丁超立方设计方法
2024,46(3):150-157
李国盛
国防科技大学 空天科学学院, 湖南 长沙 410073，liguosheng126@163.com
张锡
国防科技大学 空天科学学院, 湖南 长沙 410073
文谦
国防科技大学 空天科学学院, 湖南 长沙 410073
杨家伟
国防科技大学 空天科学学院, 湖南 长沙 410073
张为华
国防科技大学 空天科学学院, 湖南 长沙 410073
武泽平
国防科技大学 空天科学学院, 湖南 长沙 410073
国防科技大学 空天科学学院, 湖南 长沙 410073，liguosheng126@163.com
张锡
国防科技大学 空天科学学院, 湖南 长沙 410073
文谦
国防科技大学 空天科学学院, 湖南 长沙 410073
杨家伟
国防科技大学 空天科学学院, 湖南 长沙 410073
张为华
国防科技大学 空天科学学院, 湖南 长沙 410073
武泽平
国防科技大学 空天科学学院, 湖南 长沙 410073
摘要：
拉丁超立方设计是最常用的计算机试验设计方法之一,针对现有拉丁超立方设计方法采样一次性且难以兼顾设计的空间均匀性和计算效率的问题,提出了一种演化排列拉丁超立方试验设计方法。通过对小样本设计的演化、排列信息继承和扩充等操作,以较小的计算量实现了样本的扩充与优化。此外,所提方法可以兼顾现有样本和新增采样点之间的关系,实现样本的序列扩充,这在实际近似建模过程中十分方便。通过多组数值试验,验证了本文方法在空间均匀性和计算效率等方面的优越性。
拉丁超立方设计是最常用的计算机试验设计方法之一,针对现有拉丁超立方设计方法采样一次性且难以兼顾设计的空间均匀性和计算效率的问题,提出了一种演化排列拉丁超立方试验设计方法。通过对小样本设计的演化、排列信息继承和扩充等操作,以较小的计算量实现了样本的扩充与优化。此外,所提方法可以兼顾现有样本和新增采样点之间的关系,实现样本的序列扩充,这在实际近似建模过程中十分方便。通过多组数值试验,验证了本文方法在空间均匀性和计算效率等方面的优越性。
基金项目：
国家自然科学基金资助项目(52005502)；国防科技大学科研计划资助项目(ZK19-11)
国家自然科学基金资助项目(52005502)；国防科技大学科研计划资助项目(ZK19-11)
Evolution permutation optimal Latin hypercube design method
LI Guosheng
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China,liguosheng126@163.com
ZHANG Xi
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
WEN Qian
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
YANG Jiawei
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
ZHANG Weihua
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
WU Zeping
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China,liguosheng126@163.com
ZHANG Xi
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
WEN Qian
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
YANG Jiawei
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
ZHANG Weihua
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
WU Zeping
College of Aerospace Science and Engineering, National University of Defense Technology, Changsha 410073, China
Abstract：
Latin hypercube design is one of the most commonly used computer experimental design methods, in response to the problem of one-time sampling and difficulty in balancing spatial uniformity and computational efficiency in existing Latin hypercube design methods, an experimental design method of permutation evolution Latin hypercube was proposed. By evolving small-sample designs, inheriting and expanding permutation information, the expansion and optimization of samples were achieved with a relatively small computational effort. In addition, this method gives consideration to the relationship between existing samples and new added samples and realizes the sequence expansion of samples, which is very convenient in the actual approximate modeling process. Through several groups of numerical experiments, the advantages of this method in space filling quality and calculation efficiency were verified.
Latin hypercube design is one of the most commonly used computer experimental design methods, in response to the problem of one-time sampling and difficulty in balancing spatial uniformity and computational efficiency in existing Latin hypercube design methods, an experimental design method of permutation evolution Latin hypercube was proposed. By evolving small-sample designs, inheriting and expanding permutation information, the expansion and optimization of samples were achieved with a relatively small computational effort. In addition, this method gives consideration to the relationship between existing samples and new added samples and realizes the sequence expansion of samples, which is very convenient in the actual approximate modeling process. Through several groups of numerical experiments, the advantages of this method in space filling quality and calculation efficiency were verified.
收稿日期：
2022-04-26
2022-04-26
