约束空间近似正交的空间填充试验设计方法

试验设计空间存在约束是一种广泛存在的实际工程问题, 针对现有的约束空间试验设计方法存在生成设计点数不灵活, 算法优化时间长、效率低, 适用约束类型有限, 设计准则单一等问题, 提出一种约束空间近似正交的空间填充试验设计方法。基于设计点之间的距离和相关系数值构造试验设计准则, 通过改进的随机坐标交换算法进行方案求解。所提算法适合凸约束、非凸约束、解析约束、非解析约束等多种类型的约束, 而且适用于多维度的不规则试验设计空间。示例分析表明, 与现有方法相比, 所提算法具有优良的空间填充特性和较好的正交性。

Nearly-orthogonal space-filling experimental design method for constrained spaces

The existence of constraints in experimental design space is a widespread practical engineering problem. In view of the problems in the existing experimental design methods for constrained spaces, such as inflexibility in generating design points, long optimization time and low efficiency, limited applicability for different types of constraints, single design criteria, and so on, an experimental design method of nearly-orthogonal space-filling for constrained spaces is proposed. The experimental design criterion is constructed based on the distance between design points and the correlation coefficient, and the scheme is solved by the improved stochastic coordinate-exchange algorithm. The proposed algorithm is suitable for convex constraints, non-convex constraints, analytical constraints, non-analytical constraints and other types of constraints, and it is suitable for multi-dimensional irregular experimental design space. The analysis of the example shows that the algorithm proposed in this paper has better space-filling characteristics and orthogonality compared with the existing methods.