Research on the Application of Reinforcement Learning in Arrival Flight Sequencing Problems

This paper presents a reinforcement learning (RL) model for sequencing arrival flights. In the proposed model, the state represents the current assignment of arrival flights’ positions, the action involves adjusting the arrival sequence, the agent adjusts the arrival sequence of flights, and the environment responds to actions. The model prioritizes delay time as the primary component of the reward function and incorporates constraints such as constraint position shifting. The model was validated using simulated flight arrival data, and the total delay time of the reinforcement learning (RL) model was compared with that of the First-Come-First-Served (FCFS) scheduling strategy.The total delay under the FCFS strategy was 1244, whereas the RL model achieved total delays of 1061 (cps=2) and 1140 (cps=5). The experimental results show that the RL model is superior, and the model can provide decision support for air traffic controllers in sequencing arrival flights.