Methodology for Optimal Latin Hypercube Design Generation

This paper is a technical report that presents a new methodology for creating approximations to Optimal Latin Hypercube designs without the need for formal optimization. The proposed approach allows for the generation of these designs with minimal computational effort and in real time. It utilizes patterns of point locations and small building blocks to recreate the necessary configurations based on the dimensionality of the problem and the required number of design points. The report outlines the significance of the location of experimental data points in generating accurate metamodels while maintaining a manageable number of points. It discusses the limitations of traditional methods, such as Genetic Algorithms, which often result in high computational costs. The new methodology is compared to existing methods, particularly the Enhanced Stochastic Evolutionary Algorithm, highlighting its efficiency in producing Optimal Latin Hypercube designs. The report concludes that the proposed methodology serves as a powerful tool for obtaining good approximations in real-time scenarios.