Prediction of major air traffic flows using ADS-B data

Abstract

Prediction of air traffic flow, i.e., staggering the air traffic demand over time and space, is a very important activity in Air Traffic Flow Management (ATFM). Accurate air traffic flow prediction can advise ATFM about the forthcoming air traffic in the airspace and help ATFM develop control strategies in advance to address anticipated saturations in the airspace. There are mainly two ways for air traffic flow prediction. One way is propagating the trajectories of flights forward in time and counting the number of aircrafts at a particular sector in the airspace. While making accurate predictions, it does so for a duration of up to 20 minutes which is far from ideal for ATFM. The other way is the aggregated flow prediction which provides the distribution of traffic flows in the airspace. One of the state-of-the-art aggregate prediction approaches is the Linear Dynamic System Model (LDSM) which: i) predicts air traffic flow for a whole day in advance based on historical data, ii) Accounts for uncertainty in departure. iii) makes predictions based on number aircrafts in sector in previous time interval is ideal. The LSTM model has been proposed and applied on the American Airspace. However, it remains to be seen if it can do so for other airspaces. This report aims to examine the accuracy of air traffic flow prediction using the LDSM model and analyze the potential influencing factors of the prediction accuracy. Based on the flight trajectory data, this report has carried out a case study in the Singapore airspace. Results show that the prediction of air traffic flow for the Singapore airspace is not as accurate as the prediction for the American airspace