Real-Time Estimated Time of Arrival Prediction System using Historical Surveillance Data.

Prediction of Estimated Times of Arrival (ETA) is a challenging problem for the aviation industry. Flights recurrently deviate from their scheduled time of arrival, which has negative downstream consequences that affect the efficiency of operations. Therefore, accurate and up-to-date ETA estimations prior to its landing can help in optimizing the actions to be taken by the different air transportation agents whenever schedule deviations are incurred, and thus reduce the economic, logistic and environmental impact that they cause. This presentation exposes an infrastructure for high-fidelity computation of accurate ETA in real-time, based on a data-driven approach that leverages the use of recorded aircraft trajectories. This infrastructure is composed of different elements: (1) a live ADS-B tracks gathering system embedded in a lambda-architecture cluster, with capabilities for real-time distribution and data lake storage (2) an ETA prediction machine learning model, employing the actual 4D aircraft position as input; and (3) a hybrid cloud architecture to support real-time visualization and distributions of ETA predictions. The proposed infrastructure has been successfully validated in a real environment (Transforming Transport, an European Commission funded project). This infrastructure enables real-time computation and distribution of accurate ETA for any arrival operation of interest. Results supported the envisioned benefits of getting such accurate ETA, which basically turn into a reduction of associated costs for airport authorities and airlines.