Simulation Optimization with Multiple Attempts

Simulation optimization is a widely utilized approach that allows decision-makers to test various decision variable settings in simulators before implementing a final recommended action on the real systems. In some real-world scenarios, the recommended action can be executed multiple times and the performance is evaluated as the best one among these multiple attempts. In this paper, we introduce such simulation optimization problems with multiple attempts and provide insights into the problem through comparison to risk-averse decision-making problems. We propose a surrogate-assisted algorithm based on the Gaussian process model and the upper confidence bound criterion for efficiently solving such problems. We demonstrate the efficiency and effectiveness of the proposed approach with several numerical examples.