Distributionally robust optimisation approach for aircraft sequencing and scheduling with learning-driven arrival and departure time predictions

Abstract

As air traffic demand grows, some airspace systems are nearing capacity. Optimising runway utilisation is a key strategy for increasing capacity. To enhance efficiency and robustness in aircraft sequencing and scheduling under uncertainty, we introduce two prescriptive analytics approaches. First, the estimate-then-optimise (ETO) approach uses a machine learning method to estimate probability distributions, which inform a stochastic programming (SP) model for the aircraft sequencing and scheduling problem (ASSP). However, prediction and sampling errors may affect decision quality. To mitigate this, we replace the SP model with a distributionally robust optimisation (DRO) model, proposing the estimate-then-distributionally-robust-optimise (ETDRO) approach. Given the complexity of solving DRO models, we develop decomposition methods to improve computational efficiency. Numerical experiments show that ETDRO consistently delivers high-quality decisions, outperforming benchmark optimisation approaches. Meanwhile, the proposed inexact decomposition methods significantly improve computational performance, enabling the real-world implementation of ETDRO.