Using response surfaces to improve the search for satisfactory behavior in system dynamics models

Traditional system dynamics studies rely heavily upon heuristics and experience. Nevertheless, mathematical exploration techniques have been introduced as important elements for a successful study. We argue that the role of optimization in system dynamics studies is not to replace experience‐based knowledge, but instead to augment, facilitate, and expand the heuristic exploration of a model. Accordingly, our approach involves narrowing the design space (using response surfaces) and the subsequent direct investigation of the simulation model (using heuristics). Response surfaces have received considerable attention in optimization because of their capability to replace complex models with analytic equations, thereby increasing computational efficiency. However, doubts exist as to the usefulness of a response‐surface approximation of an approximation of reality (i.e., a system dynamics model). We demonstrate the usefulness of response surfaces in system dynamics studies with a case study involving a high‐level model of an industrial ecosystem; our intent in using response surfaces is not to replace the simulation models with analytic equations, but instead to direct attention to regions within the design space of the original simulation with the most desirable performance. Recommended changes to a system are based directly on the simulation model, not on response surfaces, avoiding the added level of approximation inherent in response surfaces. The primary focus of the article is on the concept exploration approach , which is presented first. The case study towards the end is offered as supporting evidence. Copyright © 2000 John Wiley & Sons, Ltd.