Real-time departure slotting in mixed-mode operations using deep reinforcement learning : a case study of Zurich airport

A mixed-mode runway operation increases the runway capacity by allowing simultaneous arrival and departure operations on the same runway. However, this requires careful evaluation of safe separation by experienced Air Traffic Controllers (ATCOs). In daily operation, ATCOs need to make real-time decisions for departure slotting. However, an increase in runway capacity is not always guaranteed due to the stochastic nature of arrivals and departures and associated environmental parameters. To support ATCOs in making real-time departure slotting decisions, this paper proposes a Deep Reinforcement Learning approach to suggest departure slots within an incoming stream of arrivals while considering operational constraints and uncertainties. In this work, novel state representation and reward mechanism are designed to facilitate the learning process. Experimentation on A-SMGCS data from Zurich airport shows that the proposed approach achieves an efficiency ratio of more than 83.8% of the expected runway capacity while maintaining safe separation distances in mixed-mode operations. The results of this work have demonstrated the potentials of Deep Reinforcement Learning in solving decision-making problems in Air Traffic Management.Civil Aviation Authority of Singapore (CAAS)National Research Foundation (NRF)Published versionThis research is supported by the National Research Foundation, Singapore, and the Civil Aviation Authority of Singapore, under the Aviation Transformation Programme