This Grant Opportunity for Academic Liaison with Industry (GOALI) project is concerned with the construction of highly efficient and statistically valid procedures for comparing systems in the presence of stochastic constraints on the system behavior, assuming that both the objective and the constraints are implicit in a simulation model with few known properties. The goal is to develop statistically valid comparison procedures for solving such simulation optimization problems, and to improve the computational efficiency of the procedures to the point where they can handle up to several thousand alternatives. This involves the use of better variance estimators, common random numbers, and tighter boundaries for comparing systems. The focus of the project will be on optimization problems with discrete decision variables using fully sequential comparison procedures; this is because decision variables in industrial problems are likely to be discrete (e.g., choosing the optimal number of workers and machines) and because sequential procedures have been shown to be extremely efficient relative to other existing comparison approaches.
If successful, the research will enhance the use of simulation not only as an assessment tool but also as an optimization tool. Ideally, this project would lead to the implementation of statistically valid and computationally efficient comparison procedures in simulation software; this would be achieved through close collaboration between researchers at Georgia Tech and software vendors (primarily Imagine That!). The addition of such optimization features to popular simulation software packages would be highly valuable to practitioners with respect to identifying process improvements and improved system designs in the presence of constraints about the system behavior.
