The Operational Disruption Forecasting System (ODFS) predicts the uncertainties in the arrival process by integrating a suite of deep learning models with live data streams and streaming analytics to forecast the possible evolution of arrivals into the terminal airspace over the next 15 hours. The innovation accomplishes that objective by leveraging knowledge of the airspace structure and procedures, assessment of weather forecast accuracy, predictions of FAA-imposed constraints, and airport capacity forecasts to present a complete assessment of the future evolutionary paths of the terminal airspace.
Beginning 15 hours ahead, ODFS generates probabilistic airport capacity scenarios based on weather forecasts, arrival and departure demand, and predictive models of weather forecast uncertainty. As the forecast window shortens, more information is incorporated into the prediction system. Predictions of possible convective weather impacts on arrival flows and runway capacity, and the potential for FAA-imposed traffic management initiatives (TMI), add more potential for disruption to planned operations. Enroute and terminal convective weather and changes to airline schedules may create congestion at arrival fixes. A trajectory prediction model integrates those data and forecasts, generating higher fidelity predictions of aircraft ETAs to the runways.
The Robust Analytics solution develops a multi-stage analytical process that generates probabilistic scenarios at each stage. Once integrated, the scenarios form a comprehensive picture of the uncertainties along the entire arrival process. Over time, higher fidelity datasets continuously update the predictions allowing dispatchers and controllers to adjust. The core modules of ODFS are airport capacity forecasts and a trajectory prediction model. Potential impacts of other constraints ‒ convective weather forecasts, TMI predictions, and arrival fix demand ‒ serve as inputs to the core models.
This project supports NASA research objectives by developing algorithms and applications to predict how airport arrival flows may change with evolving and uncertain weather and traffic conditions. Our approach differs significantly from previous research efforts in its explicit treatment of uncertainty and the aim to develop tools to support decision making in an uncertain environment. Our approach illustrates the benefit of applying deep learning methods to long-standing technical challenges in traditional airspace operations.
In separate meetings held with Robust Analytics from July 2022 to February 2023, three airlines stated their top priority to be improved operational efficiency at their busiest airports. Achieving that objective was hampered by uncertainties in forecasting airport conditions over the next several hours and lack of insight into FAA decisions. And that they would be interested in such a service.