Publication
Study summary
The paper investigates model-free reinforcement learning for runway capacity management under uncertainty, combining runway-configuration selection with airport service-rate allocation. The problem is formulated as a Markov Decision Process and evaluated through both data-based and forecast-based approaches. Maskable Proximal Policy Optimization is compared with Deep Q-Network learning, with historical and forecast information used to represent changing operational and weather conditions. In the São Paulo/Guarulhos case study, the stochastic forecast-based and incremental data-driven approaches outperform traditional baselines and achieve performance comparable with strong optimization benchmarks. The work emphasizes practical decision support for air traffic flow management without requiring an explicit model of the environmental dynamics, and shows how model-free RL can adapt runway-capacity decisions to dynamic and stochastic conditions.
Keywords: air traffic flow management; runway capacity management; reinforcement learning; stochastic optimization; weather uncertainty.
BibTeX
Please cite the published journal article through its DOI.