BMDS Uncertainty Quantification

Uncertainty quantification (UQ) can be performed with existing MDA M&S tools if efficient algorithms are developed and deployed. OptTek proposes to explore innovative methods and algorithms to 1) quantify output variation from all input uncertainties providing distributional information and extreme value identification, 2) isolate output variation from individual input uncertainties and from input uncertainty interactions, and 3) implement Inverse UQ algorithms to anchor simulations based on test data or outputs from higher-fidelity simulations Providing these results for even quick running simulation models will be computationally intensive, so a primary focus will be developing efficient algorithms that derive the maximum amount of information from the fewest number of simulation runs. We will leverage and expand the SBIR-derived OptDef simulation optimization and analysis tool for testing, and UQ methods and algorithms that prove effective will be added to OptDef for use by MDA and DOD analysts. The proposed approach to UQ treats the simulation model as a black box, requiring only that simulation inputs can be updated, the model can be executed, and simulation outputs can be collected programmatically. Thus any existing or future MDA simulation model or framework for which UQ analysis is valuable can benefit from this research.