항적 패턴 클러스터링을 이용한 항적 예측 연구

초록

Trajectory prediction is one of the most important subjects in the field of air traffic management. Recently, data-driven approaches are becoming popular for predicting estimated time of arrivals or future conflicts. However, the movements of aircraft inside busy terminal areas are complex due to frequent radar vectoring by the air traffic controllers. Consequently, future trajectory of each aircraft is affected by all the other aircraft. In this study, instead of trying to predict individual trajectory, an attempt is made to predict the whole snapshot of the given airspace using clustering. One year trajectory data around Incheon International Airport and Gimpo International Airport are converted to images. For a given time interval, 365 days of images are clustered to around ten patterns and the transition probabilities network is constructed among the clusters. The network was able to generate a reasonable prediction with data not used for the clustering, which demonstrates the potential of the approach described in this paper.