Simulation optimization using metamodeling: a study in medical cases

Abstract

The combination of discrete event simulation and optimization techniques is growing in recent years. However, such a combination isn’t widely used due mainly to high computational costs. Due to this difficulty, some studies are focused on developing more computationally efficient optimization methods. This article uses a metamodeling technique to create a mathematical model for an output of interest, by running a limited number of scenarios of a simulation model and then proceed to optimization of this output model. We argue that this approach results in significantly reduced computational burden compared to optimizing directly on scenario-specific output values obtained from the simulation model. We illustrate this method using simulation models of distinct functional areas across hospitals in Brazil. The use of metamodeling allows optimization of the simulation models, reaching better solutions than those obtained by a commercial optimizer, with the additional advantage of reducing the computational time by more than 68% on average.