适用于复杂系统仿真试验的试验设计方法

Studies on the Optimization of Latin hypercube designs for Complex System Simulation

拉丁超立方试验设计是"充满空间"试验设计方法的重要领域。传统拉丁超立方试验设计优化以列相关系数最小化或试验点之间最大距离最小为单一优化准则,这种单目标试验设计优化方法存在缺陷。提出了将两类优化准则作为优化目标的多目标优化准则,以及实现多目标优化的改进ESE算法。算例分析证明,提出的试验设计优化算法优于已有典型试验设计方法。

The optimization of Latin hypercube design is an important area of the space filling experimentations. The existing procedures to find optimal LHD by minimizing the pairwise correlations or maximizing the minimum inter-site distance in traditional LHD optimization are intreduced, and their drawbacks are discussed in these methods. This paper proposes a multi-objective optimization approach to find optimal LHD by combining pairwise correlation and the maximin inter-site distance performance measures by modified ESE algorithm. Several examples are presented to show that the optimal designs are good in terms of both the crrelation and distance criteria.