Determining policy for a system dynamics model using reinforcement learning

Abstract

System dynamics allows managers and policy makers to analyze problems with non-linear feedback structures and thus counter-intuitive behavior. A main tool of system dynamics is to build a computational model of a system and analyze it to determine suitable policies to move the system to a desired goal. This work aims at using methods and algorithms from reinforcement learning to determine suitable policies for a system dynamics model. We introduce the techniques, methods and algorithms of reinforcement learning and apply them to a classical model from the system dynamics literature.