Automated Data-Driven Prediction on Aircraft Estimated Time of Arrival

Zhengyi Wang (1) , Man Liang (2) , Daniel Delahaye (1)

1 ENAC - Ecole Nationale de l'Aviation Civile
2 University of South Australia [Adelaide]

—4Dtrajectorypredictionisthecoreelementoffuture air transportation system, which is aimed at improving the operational ability and the predictability of air traffic. In this paper, we introduce a novel automated data-driven model to deal with the short-term trajectory prediction problem in Terminal Manoeuvring Area (TMA). The proposed model consists of data mining and Deep Neural Networks (DNNs). Firstly, the dataset is analyzed and cleaned by several criterions. Then, the flights in the dataset are split into partitions according to the runway in use (QFU). Prediction models of each QFU will be trained by the corresponding dataset. The experiments were firstly performed on real traffic data in Beijing TMA for 5 Neural Networks (NNs) models with nested cross validation. The results demonstrate that the DNNs perform better than shallow NNs. In addition, comparative study on data mining is conducted and proves that thedataminingoperationisrobustinprocessingoutliers,missing point and noise, which greatly improves the prediction accuracy intermsofMeanAbsoluteError(MAE)andRootMeanSquared Error (RMSE). We finally introduced ensemble learning model by combining well-performed individual models with certain strategies. Compared to other models, the minimum rule and mean rule applied to Deep Forward Neural Networks (DFNNs) with 3 hidden layers and DFNNs with 4 hidden layers after data mining performs the best in terms of MAE and RMSE.

Mots clés: Air Traffic Management; 4D Trajectory Prediction; Data mining; Machine Learning; Deep Neural Networks

Domaines: Optimisation et contrôle [math.OC]

Zhengyi Wang, Man Liang, Daniel Delahaye. Automated Data-Driven Prediction on Aircraft Estimated Time of Arrival. SID 2018, 8th Sesar Innovations Days, Dec 2018, Salzburg, Austria. ⟨hal-01944608⟩