Forecasting Air Base Operability in a Hostile Environment: Estimating Metamodels from Large-Scale Simulations.

摘要

An on-going Air Force logistics concern is the ability of an A.F. unit to fly aircraft into combat particularly when their air base comes under attack. A.F. units must not only survive attacks but continue to operate afterwards as well. Limited budgets and long procurement and training pipelines magnify the problem, making it imperative to identify and resolve support system deficiencies. We propose a simulation approach which attempts to capture the logistics infrastructure for a single air base. Multiple simulation runs are used to derive a simpler metamodel useful for forecasting future performance or for evaluating policy alternatives. This metamodel can then be used to explore 'what if' analyses. Research issues include estimation of metamodels from large-scale simulation models with highly correlated responses, and experimental design with similar responses and application of variance reduction techniques to large-scale simulation problems. Good variance reduction results are obtained using a classical two-level experimental design with blocking within the fraction based on common random numbers. Two cases, with and without attacks on the air base, are modeled as sub-experiments. Results indicate homogeneity of variance within each case, but heterogeneity between the two cases. A significant difference in the number of sorties flown when the air base is attacked is evident, but this difference dissipates by Day 30. The estimated metamodels indicate that two-way interactions are extremely important and should not be ignored. Theses. (EDC)