Taxi-out Prediction using Approximate Dynamic Programming

High taxi-out times (time between gate push-back and wheels off) at major airports is a primary cause for flight delays in the National Airspace System (NAS). These delays have a cascading effect and affect the performance of Air Traffic Control (ATC) System. Accurate prediction of taxi-out time is needed to make downstream schedule adjustments and better departure planning, which mitigates delays, emissions, and congestions on the ground. However, the accurate prediction of taxi-out time is difficult due to the uncertainties associated with them. The primary objective of this paper is to accurately predict taxi-out time at major airports, in the presence of weather and other departure-related uncertainties. This paper presents a novel reinforcement learning (RL) based stochastic approximation scheme for predicting taxi-out times. The prediction problem is cast in a probabilistic framework of stochastic dynamic programming and solved using approximate dynamic programming (ADP) approaches. The strengths of the method is that is it non-parametric unlike the regression models with fixed parameters, highly adaptable to the dynamic airport environment since its learning based, is scalable, is inexpensive since it does not need highly sophisticated surface management system, and effectively handles uncertainties due to the probabilistic framework. The taxi-out prediction performance was tested on data obtained from the FAA’s Aviation System Performance Metrics (ASPM) database on Detroit Metropolitan Wayne County International Airport (DTW), and Washington Reagan National (DCA) airports. Results show that the average prediction error 15 minutes before gate departure for about 80% of the flights was less than 2.9 min.