Aircraft Taxi Time Prediction Using Machine Learning and its Application for Departure Metering

The growing air traffic demand requires safer and more efficient airport operations. Especially, long take-off queues of departure aircraft need to be reduced, since a significant amount of fuel is wasted because of the queues. The key to take-off queue reduction is Departure Metering (DM), which applies suitable gate-hold times for departure aircraft when departure runways are crowded. DM will be a suitable solution for take-off queue reduction, but accurate taxi-time predictions are required to ensure DM performances. For this reason, we first developed taxi-time prediction models using Machine Learning (ML) methods. We tested 6 ML methods to find the most accurate one. Then, we modeled the traffic flows of our case-study airport, Tokyo International Airport, using a fast-time simulator. The DM operations based on the taxi-time prediction results are simulated on this simulator and its effectiveness on take-off queue reduction was evaluated. The result showed that combining ML taxi-time predictions and DM can reduce the total queuing time by around 45% if there is no arrival traffic. Interference with arrival traffic deteriorated the effect, but DM still reduced the take-off queues by about 29%.