Reinforcement Learning for Traffic Flow Management Decision Support

Recent advances in Artificial Intelligence and Machine Learning are being harnessed to solve increasingly complex problems across a variety of domains, including Air Traffic Management. Application of these methods to the domain of Traffic Flow Management, however, remains a challenge as it is first necessary to effectively represent the dynamics of weather forecasts – and the uncertainty in the resulting constraints – within the construct of the decision-making process. In this paper, we propose a novel approach for capturing weather forecast uncertainty in a reinforcement learning process that generates Traffic Flow Management strategies in a real-time environment. Specifically, we leverage Monte Carlo Tree Search to explore and evaluate potential traffic management actions against an ensemble of weather futures. The results demonstrate that under the assumptions of the operational environment developed and the objective defined, the algorithm can generate effective solutions for managing uncertain constraints, adapt to changing information, and do so in a real-time context.