Prediction Method of Aircraft Dynamic Taxi Time Based on XGBoost

Abstract

Accurate and dynamic prediction of aircraft arrival and departure taxi time can effectively improve the operation efficiency of the airport. An aircraft dynamic taxi time prediction method based on XGBoost is proposed for the first time. Firstly,the key characteristic index of variable taxi time prediction is constructed by analyzing various factors affecting the taxi time. Then,XGBoost algorithm is selected to establish the variable taxi time prediction model,and the key input parameters of the model are adjusted and tested. Finally,the prediction effect of XGBoost algorithm is compared with random forest and support vector regression. At the same time,the correlation between sample data size and prediction accuracy of taxi time is analyzed for the first time. Guangzhou Baiyun International Airport is taken as the analysis object. The results show that the prediction accuracy of the taxi-in and the taxi-out time reaches 94.1% and 96.6% by using the XGBoost algorithm,which are better than the mainstream algorithms of random forest and support vector regression. In addition,more than 32 000 samples are needed for accurate and stable prediction of dynamic taxi time at Baiyun Airport.

Keywords

• air transportation
• dynamic taxi time
• machine learning
• xgboost
• sample size