Airport capacity prediction and optimal allocation for strategic air traffic flow management at Sao Paulo/Guarulhos International Airport

Abstract

Demand–capacity imbalances are a major cause of inefficiencies in the air transportation system, accounting for a significant part of flight delays and disruptions. Efficiently addressing these imbalances through air traffic flow management is key for improving system performance. If taken at the strategic level, flow management decisions have the potential to increase operational predictability and reduce costs for airspace users by transferring expected delays to the ground before departure, where it is safer and cheaper to absorb them. However, strategic flow management planning is a challenge for traffic managers due to the stochastic and dynamic nature of airport and airspace capacity. This paper leverages machine learning and optimization methods to develop a decision support framework for strategic air traffic flow management at Sao Paulo/Guarulhos International Airport (GRU) in Brazil. First, an airport capacity prediction model is learned from historical meteorological and throughput data. Random Forests is used in a regression setting for the supervised learning problem, being able to provide not only a point prediction of arrival capacity but also an empirical predictive distribution based on the Quantile Regression Forests approach. An analysis of feature importance reveals that ceiling and convective weather are the most important factors affecting arrival capacity at GRU. A stochastic optimization model for capacity allocation is then used to prescribe the optimal airport acceptance rates for Ground Delay Program planning based on the capacity forecasts and their estimated uncertainty. When applied to an actual test case of demand–capacity imbalance at GRU, the solution provided by the framework is found to generate a reduction in total delay costs of up to 10%, revealing an improvement over the current practice solely based on tactical airborne delays.