Data-Driven Robust Aircraft Assignment to Minimize Delay Propagation

Abstract

Flight delays have a significant impact on an airline's operating cost including increased expenses for crew, fuel, and maintenance. Propagated delays due to late arriving aircraft contribute to 40% of all flight delays as reported by the Bureau of Transportation Statistics. The robust aircraft assignment problem is to assign tail numbers on scheduled arriving flights at an airport to scheduled departing flights at the same airport with the objective of minimizing propagated delays. In this paper, we propose a new data-driven approach for the robust aircraft assignment problem by formulating it as a balanced assignment problem between incoming and outgoing flights flown by the same aircraft type at a single airport. We consider both deterministic and stochastic versions of the aircraft assignment problem. In the deterministic case, we prove the optimality of the First-in-First-out (FIFO) assignment policy under two different performance measures, justifying the use of the FIFO policy as a benchmark. In the stochastic case, we show that the FIFO assignment policy is no longer optimal and propose the rFIFO and stochastic assignment formulations based on the mean and empirical distribution of arrival delay, respectively. We propose a data-driven approach to estimate the assignment costs by using empirical observations of arrival delays from prior years' flight records to compute the empirical propagated delay distribution. We propose a data-driven clustering method to account for factors such as originating airport, time of day, and aircraft type that affect the arrival delay distribution. These empirical cluster based aircraft assignment costs serve as an input to our stochastic assignment model. These assignment costs are then used to derive the optimal aircraft assignment for the rFIFO and stochastic assignment policies for an out of sample data set from 2018 for Delta at Atlanta airport. We show that both the rFIFO and stochastic assignment policies derived from the data-driven approach perform better than the benchmark FIFO assignment. We also show that for all Delta airline flights at Atlanta airport from July to September in 2018, the stochastic assignment policy yields a roughly 40% improvement in total actual propagated delay over the actual airline assignment while reducing the fraction of delayed flights due to propagated delays from 6.58% to 2.62%, thus potentially saving approximately 14 million dollars in flight delay related annual operating costs. We conclude that incorporating the stochastic nature of arrival delays in solving the robust aircraft assignment problem, and combining it with a data-driven approach can potentially help an airline significantly reduce its operating costs due to flight delays, improve passenger convenience and experience, and help the environment by reducing emissions.