Multiagent Reinforcement Learning Methods for Resolving Demand - Capacity Imbalances

In this article, we explore the computation of joint policies for autonomous agents, representing flights, to resolve congestions problems in the Air Traffic Management (ATM) domain in the context of Demand-Capacity Balance (DCB) process. We formalize the problem as a multi-agent Markov Decision Process (MDP) towards deciding flight ground delays to resolve imbalances, during the pre-tactical phase. To this end, we present and evaluate multi-agent reinforcement learning methods. An experimental study on real-world cases confirms the effectiveness of our approach.