Abstract for:Reinforcement Learning in a Strategic Planning Simulation: A Case Study on Challenges and Approaches

Strategic planning simulations can be an advantageous technique to cope with the complex dynamics of realizing business goals in a multi entity environment. Traditionally, the underlying system dynamics models have been optimized using either optimization theory or the modal control theory. We investigate whether reinforcement learning can overcome the limitations faced by these traditional approaches. In this paper we present a case study on the performance of a Q-learning based reinforcement learning agent in a strategic planning simulation within predefined scenarios. Furthermore, we discuss means of extending the agent to react to a competing agent, whose actions adapt dynamically. We show that our approach is a valid means of optimizing behaviour in scenarios where the competitor follows a fixed strategy based on the agent’s actions.  Furthermore, we show an easily adaptable toolkit for testing system dynamics models with different reinforcement agents.