A two-stage stochastic optimization approach for mega-airport departure metering under data-driven taxi-time uncertainty predictions

Over the past decade, mega-airports have experienced a surge in air traffic demand, physical expansion, and increased complexity in apron layouts, leading to a high level of aircraft taxi-time uncertainty and shifting the airport surface management from integrated tower control to dedicated apron control. In this study, a two-stage stochastic optimization framework is developed for mega-airport departure metering (DM), which specializes apron-centric and tower-centric optimization in different stages. Moreover, a data-driven Mixture Density Network (MDN) is built to predict the aircraft taxi-time distribution and characterize the uncertainty levels. A large-scale trajectory dataset is collected from a representative mega-airport in China to illustrate the procedure. The results indicate that the developed two-stage stochastic optimization framework distinguishes tower control and apron control in the DM process, improving the overall flexibility of airport airside operations. The data-driven neural network could better predict the taxi-time uncertainty levels through multimodal probability distributions especially at mega-airport with volatile traffic situations. Furthermore, compared with state-of-the-art DM methods, the two-stage stochastic optimization framework could achieve more robust performance of airport departure management and better trade-off between gate-holding and runway throughput.