一种基于数独分组的拉丁超立方试验设计方法

为解决拉丁超立方设计在复杂因子的实验设计空间中存在优化时间长和因子水平组合爆炸的问题。提出一种基于数独分组的拉丁超立方设计方法。该方法对因子通过数独分组降维,在降维后的各个子空间中平移种子设计。将基于数独分组的拉丁超立方设计与另外两种常用的拉丁超立方设计方法在中高维度试验空间中进行对比,结果表明,基于数独分组的拉丁超立方设计有较优的空间填充性和最短的运行时间。

A Fast Latin Hyper Cube Experiment Design Method Based on Soduku Grouping

In order to solve the long optimization time and parameter level combination explosion in the complex experiment design space of Latin Hypercube design, which is one of the most popular method in experiment design, a fast Soduku grouping-based method is proposed. The optimal seed design is expanded and transformed in the grouped spaces. Experiments are conducted to compare the Soduku grouping-based Latin Hypercube design method with other two commonly used Latin Hypercube design methods in the middle and high dimension experiment space. The results show that Soduku grouping method is more efficient in computation and has better space-filling performance.