A reference approach for policy optimization in system dynamics models

In this article we develop a formal method, the reference approach, that can be used as a tool for policy optimization in a system dynamics model. The obtained policy is expressed in terms of feedback loops, which master the behavior of the system so that it reaches a dynamic according to the objectives of the designer. The reference approach utilizes three models: a system dynamics model, a reference model, and a control model. The first one is built using Forrester's principles. The second is a large optimization nonlinear model whose solution yields the dynamic desired by the planner. The third is a linear quadratic optimal control model with closed‐loop solution, which minimizes the gap between the observed dynamic (in the system dynamics model) and the desired dynamic expressed by the reference model. We illustrate the operation of the reference approach by using it to find the optimal policy of a well‐known system dynamics model. For this same model the best policy obtained with a heuristic approach is known. Finally, we compare the performance of these two policies.