The Department of Defense uses complex high-dimensional simulation models as an important tool in its decision-making process. To improve on the ability to efficiently explore larger subspaces of these models, this dissertation develops a set of experimental designs for searching over as many as 22 variables in as few as 129 runs. These new designs combine orthogonal Latin hypercubes and uniform designs to create designs having near orthogonality and excellent space-filling properties. Multiple measures are used to assess the quality of candidate designs and to identify the best one. For situations in which more than the minimum number of required runs are available, the designs can be permuted and appended to create additional design points that improve upon the design's orthogonality and space-filling. The designs are used to explore two surfaces. For a known 11 dimensional stochastic response function containing nonlinear and interaction terms, it is shown that the near orthogonal Latin hypercube is substantially better than the orthogonal Latin hypercube in estimating model coefficients. The other exploration uses the agent-based simulation MANA to analyze 22 variables in a complex military peace enforcement operation. The need for maintaining the initiative and speed of execution during these peace enforcement operations is identified.
Stored only when both papers are in the corpus — 65 citations exist upstream.