File:Breaking barriers to design dimensions in nearly orthogonal Latin hypercubes (IA breakingbarriers1094510306).pdf

A dynamic and extremely complex landscape in security and world events presents problems that challenge all sectors of society to develop efficient means for exploring a wide range of solutions. Similarly, exponential increases in technological capability make it difficult for commercial and governmental leaders to assess those proposed solutions. Computer experimentation is an established method for examining complex models with large numbers of factors. Orthogonal and nearly orthogonal Latin hypercubes are proven techniques for designing simulation experiments. A key property of these efficient, space-filling designs is their ability to explore many factors within a relatively modest number of design points; however, there is a limited inventory of these designs currently available. Those that have been catalogued are usually computationally expensive to produce and have severe restrictions in the number of factors and/or runs that they allow. To remedy this, we present a set of flexible methodologies to create design matrices with little or no correlation-including saturated nearly orthogonal Latin hypercubes. This new family of designs can explore as many factors as there are design points. This research also addresses experiments that include a mixture of continuous and integer variables, some of which have different numbers of value levels.

Subjects: Monte Carlo method; Computer Experiments; Design of Experiments; Mixed Integer Programming; Latin Hypercube; Optimization; Nearly Orthogonal