A multi-criteria policy set optimization framework for large scale simulation models

Simulation modeling is a common analysis approach for large and complex systems. The policy set optimization (PSO) problem is the process of selecting a small subset of inputs of the simulation model to modify from their default settings. As a simulation model gets larger, standard analysis techniques such as simulation optimization, evolutionary optimization or design of experiments require a large amount of computation time after a decision maker (DM) selects their outputs of interest.;This research takes two approaches for solving this multi-criteria PSO problem. The first approach is called the model decomposition approach (MDA). This approach requires the decomposition of the large simulation model into smaller black-box simulation models. These smaller models (nodes) are sampled and that sampling data is regressed upon to form constraints in a MDA mathematical model. This nonlinear program is solved via the penalty-successive linear programming (PSLP) method and the solution to this is a solution policy.;The second approach is the model sampling approach (MSA). This approach samples the entire model simultaneously and the data that is generated creates the MSA mathematical model. The benefit of this integer program is that it provides a method to limit the number of inputs in a solution policy. The resulting solution can then be used as an input to a local-search method. Solution time reduction techniques are examined for this mathematical model.;Both approaches are then used on a sample simulation model and analysis of these results show that the ability of the MSA to limit the number of inputs is integral for selecting usable policies for a DM. Visualization of all validated policies is discussed and a description of how this visualization approach aids the DM in selecting their preferred policy is provided.