Multi-agent Estimated Time of Arrival prediction and dynamic arrival sequencing by Emulating Air Traffic Controllers

Accurate Estimated Time of Arrival (ETA) prediction is critical to the air traffic management system including aircraft sequencing for which Air Traffic Controllers (ATCs) are responsible. Although significant advancements have been achieved in both ETA prediction and arrival sequencing, the development of decision support tools can be further improved by learning the expertise of ATCs and reflecting on their practical considerations. To fill the research gap, in this paper, we propose a multi-agent model for both ETA prediction and arrival sequencing based on the attention mechanism that can account for the current air traffic situation and capture the decisions made by ATCs. The proposed model is demonstrated with real air traffic surveillance data recorded at Incheon International Airport in South Korea and compared with existing models in terms of ETA prediction, sequence similarity, and arrival sequencing performance. The experimental results show that, in a real-time manner, the proposed model can provide landing sequences more acceptable to ATCs as well as more accurate ETAs than those of comparison models. Specifically, sequence similarity is measured by two rank correlation coefficients, which shows the superiority of the proposed model in emulating ATC decisions. Furthermore, important considerations in arrival sequencing are discussed based on actual ATC feedback.