Modeling of runway assignment strategy by human controllers using machine learning

With increasing air traffic demands, efficient runway use has become very important. Future implementation of arrival management with optimal runway assignments is being planned. At the Tokyo International Airport, which has four runways, departures and arrivals can use either of two runways. While a nominal runway is basically assigned, an optimal runway assignment can potentially increase runway capacity. On the other hand, optimal assignment could increase the workload of air traffic controllers (ATCos). Therefore, an optimal runway assignment strategy must consider both capacity and workload for operational feasibility. Currently, ATCos sometimes instruct arrival aircraft to switch runways, which actually reduces both departure and arrival queues. As the current assignment strategy favors both runway capacity and workload, we strive to develop a model that can predict landing runways based on the current runway assignment strategy by ATCo. The proposed approach uses a neural network to predict runway assignment. Basic information for the runway assignment is selected and used as input. Considering the characteristics of the runway operation at Tokyo International Airport, four independent neural network models were developed. The accuracy of the models and criticality of each input were examined. It was demonstrated that the accuracy of the model differed widely with respect to the traffic scenarios. It was also indicated that the terminal preference is one of key features to predict runway assignment.