Background
Air traffic control is a crucial part of aviation safety, ensuring that aircraft are safely guided through the airspace and landed or taken off from airports. Effective workload management of air traffic controllers (ATCos) is an important component of maintaining safety within the aviation domain. One component of effective workload management involves relying on accurate ATCo workload predictions. However, workload overhead often occurs when the demands exceed the human operator’s capacity, and can lead to efficiency drop and operational safety concerns.
Invention Description
Researchers at Arizona State University have developed a new method for predicting air traffic controller (ATCos) workload using graph neural networks and conformal prediction techniques. This technology analyzes air traffic data within dynamically evolving graphs and captures the spatiotemporal variations in airspace traffic and controller workload. This data is collected from human-in-the-loop simulations with retired ATCos under various air traffic scenarios to train and validate the prediction model. This approach improves prediction accuracy and enhances the understanding of factors contributing to ATCo workload.
Potential Applications:
- Real-time air traffic management systems for optimizing ATCo staffing and scheduling
- Safety analysis tools for mitigating potential air traffic control operational risks
- Training & simulation platforms for ATCos
Benefits and Advantages:
- Improved prediction accuracy – can accurately predict ATCo workload levels better than traditional methods
- Enhanced learning capabilities – utilizes graph neural networks to directly learn from the spatiotemporal layout of airspace
- Comprehensive analysis – includes traffic conflict features, enabling full view of factors affecting workload
- Enhanced uncertainty indication – uses conformal prediction technique
- Dynamically adaptable – can adapt to varying numbers of aircraft and spatiotemporal changes in air traffic
Related Publication: Air Traffic Controller Workload Level Prediction using Conformalized Dynamical Graph Learning