Predicting Taxi Times Using Airport Surface Movement Data

초록

Taxi time prediction is an essential component of efficient airport surface traffic management, especially for arrival and departure scheduling. In this study, the surface movement data at Incheon International Airport is analyzed to find taxi routes and taxi speed distributions, and then a taxi time prediction model is developed that can reflect the current surface traffic conditions. The airport surface is modeled using a node-link structure so that a route is expressed by a sequence of nodes and links. Previous studies have mainly estimated taxi times based on the average speed of each link, or developed prediction models using regression or machine learning techniques with limited traffic variables. Building on these approaches, this study uses a decision tree-based model, eXtreme Gradient Boosting, to more accurately predict the taxi time by incorporating the ground traffic conditions. Features are divided into two categories, one is the real-time observational features that are captured at the initial time snapshot of the target aircraft such as total number of aircraft on the surface and the number of aircraft along the taxi route. The other is the route-specific static features such as route lengths and number of links. In addition, surface movement statistics obtained from the dataset such as average taxi speeds are included in these static features. The importance of each feature is investigated and with the final feature set of 13 features, an R-2 value of 0.74 was achieved, where the taxi times are predicted within +/- 20% for most of the routes. The proposed method is expected to be an effective tool for improving the surface traffic management such as arrival and departure scheduling or traffic flow management.

키워드