|Title: "Prediction of Taxi-out time using a novel Q-learning approach "|
Abstract— The International Air Transport Association (IATA) classifies flight delays into various categories like Passenger and baggage delays, Aircraft and ramp handling delays, Weather delays, Air Traffic Flow Management Restrictions (ATFM), Reactionary delays, etc. The Federal Aviation Administration (FAA) reports that among these delays, ATFM, due to taxi-out time delays in particular contributes to 20% of total flight delay. Taxi-out time of a flight is the time noted from the gate (chalks off), along the taxi-way up to the takeoff (wheels off) at the runway. Delays in taxi-out time have a noteworthy impact on the airline’s economy. Due to unplanned pushback from the gate, it is difficult to manage fuel consumption, emissions or cost. Moreover since the airport operations are dynamic in nature, prediction of taxi-out time accurately can be challenging. This paper proposes a novel Q-learning approach to predict the accurate taxi-out times, by developing a Markov Decision Process (MDP) for analysis of the operational data received. When seeking more accurate results, this new approach, Q-learning with Temporal Difference (TD) algorithm is useful to formalize the prediction of taxi-out time of flights. The convergence of Q-learning and TD gives a learned policy, the results of which can be compared against the individual algorithms.
Keywords —Flight delays, Q-learning, Taxi-out time, Temporal Difference.