Budget Constrained Optimization of Simulation Models via Estimation of Their Response Surfaces

This paper is concerned with optimizing simulation models under the condition that a budget on simulation runs exists. A multistage search procedure is developed that allocates the total experimental budget across stages so as to maximize the expected gain in information. At each stage the experiments are allocated to the search region so as to maximize the amount of information provided. A response surface is estimated and used to reduce the search region and predict the optimum, which is then used in the next stage. The procedure continues until the experimental budget is exhausted or a stopping criterion is met.