A Flight Arrival Time Prediction Method Based on Cluster Clustering-Based Modular With Deep Neural Network

With the rapid development of the air transportation industry, air traffic is facing a severe test. The accurate prediction of the estimated arrival time (EAT) plays an important role in rational and efficient scheduling of flights. Aircraft EAT predictions often rely on aircraft performance parameters or physical trajectories. This gives poor consideration to the flight environment and take-off and landing conditions. This paper applies a data-driven methodology for ETA prediction. A Cluster clustering-based modular integrated DNN (CC-MIDNN) is proposed. According to the information of flight mission and flight environment, it decomposes complex modeling tasks into multiple distinct subtasks using cluster-based clustering. Then Bayesian optimization is used to design corresponding parallel sub-networks to learn different sub-tasks in order to reduce the complexity of the subsequent sub-networks. Finally, a new integration scheme is proposed to solve the overfitting problem due to the reliance on the training accuracy of sub-modules. To verify the effectiveness of the method, CC-MIDNN performs EAT prediction on flight data of European Airlines LIS airport in 2017. The experimental results show that the CC-MIDNN improves the time prediction accuracy by 5.92 minutes and keeps the error within 13 minutes. Compared with other deep learning algorithms and integration algorithms, the CC-MIDNN has high accuracy and stability.