A Neural-Queuing Approach to Modeling Airport Surface Traffic at Charlotte-Douglas International Airport

In the pursuit of more powerful analysis tools for Air Traffic Management, training and evaluation of tool effectiveness is typically performed within the context of a simulated system that can adequately replicate the complexity of the National Airspace System. The ground operations of airports is of particular concern for approaches to improving congestion and traffic flow, but due to complex aircraft taxi procedures and variation of air traffic playbooks, forecasting is difficult. This work outlines a framework to aggregate and predict complex airport surface traffic in the context of optimizing air traffic flow and ground operations. In our approach, an Artificial Neural Network is trained on historical data from Charlotte-Douglas International Airport. The Neural Network then informs a Queueing Network for controlling aircraft traffic flow through various congestion points on the tarmac. The resulting system is capable of forecasting arrival and departure traffic over a long time horizon (e.g. 365 days). The simulation environment then provides a basis for deriving useful performance metrics which could be used as feedback to optimization techniques. The output of the surface model we develop shows high prediction accuracy compared to the historical service rates, queue lengths, and taxi times.