An artificial intelligence algorithm will assist flow management positions in predicting and managing congestion well in advance. It will be developed by the SESAR JU ASTRA project, in which Deep Blue is responsible for defining the human-machine interface (HMI) requirements and leading the validation activities for the solution.
There are several factors that can cause headaches for an air traffic controller, leading to a flight plan not proceeding as scheduled: suddenly adverse weather conditions, technical problems, and unexpected delays which lead to additional ones. Each of these unforeseen events affects air traffic, sometimes causing congestion in a specific sector of the air navigation space. These hotspots are anticipated and "addressed" by the flow management positions (FMP), whose role is precisely to monitor the flow of aircraft by communicating to the air traffic controller supervisor (air traffic control officer supervisor or ATCO Supervisor) the need to delay departures or open new sectors, in order to avoid managing too many aircraft simultaneously, increasing the workload and the risk of accidents.
With the tools currently available, FMPs can predict and resolve hotspots with only 20 minutes notice. Is it possible to improve upon this procedure, without adding to the workload of operators, while enhancing the efficiency, safety, and sustainability of air traffic flow management?
The ASTRA project (an acronym for AI-enabled tactical FMP hotspot prediction and resolution) is set to prove it. Funded by the SESAR JU within the framework of Horizon Europe, the project is coordinated by the University of Malta while Deep Blue takes charge of the research and experimental design, as well as the coordination of dissemination activities. The other consortium partners are the Spanish consulting company Ingenav and the Swiss companies Skysoft-ATM (a company developing innovative solutions for Air Traffic Management) and Skyguide (Swiss Air Navigation Service Provider).
The algorithm that supports Air Traffic Flow Management and improves communications
"ASTRA is an exploratory research project,” explains François Brambati, psychologist, company project manager, and Human Factors consultant at Deep Blue, “which aims to validate the feasibility and development potential of a concept". The concept in focus is an algorithm named ASTRA, designed to identify air traffic congestion areas one hour in advance. "The machine-learning algorithm that we will develop with our technical partners, the University of Malta and Skysoft-ATM, will not only predict hotspots but will also be able to suggest to FMPs how to avoid them”, he adds. This is the truly innovative part of the project: ASTRA will present flow management positions with optimal solutions, considering operational efficiency and safety, while also evaluating environmental impacts such as flight paths and aircraft fuel consumption.
FMPs operate within the control centres, known as ACC (area control centres), for designated airspace sectors. "For example, the airspace over Switzerland is divided into two sectors, Zurich and Geneva, each hosting its FMPs,” explains Brambati. Typically, FMPs of adjacent sectors communicate with each other, but the project wants to “optimise" the network of communications. By analysing air traffic control data from one sector ahead, the algorithm will be able to proactively warn of impending aircraft congestion. "For instance, FMPs at the Geneva ACC will be alerted if incoming flights from Germany could congest their sector’s air traffic,” the psychologist explains. “This allows them, rather than coordinating with the next sector, such as Zurich, to directly collaborate with their German colleagues. Thus, it will no longer be necessary to go through Zurich for a problem that from Germany will affect Geneva, while keeping Zurich informed of the entire process".
The path towards a new algorithm for air traffic flow management
"Talking with FMPs and ATCO Supervisors from Geneva and Zurich, we have already collected needs and requirements for the ASTRA solution,” continues Brambati. “The first phase of the project will be the validation of the operational concept and identified functional and non-functional requirements, which will be conducted with external experts member of the ASTRA advisory board”.
The following phase will consist of the development of the algorithm, which will be "trained" with historical data (from 2018 to today) provided by EUROCONTROL concerning traffic flow in Swiss airspace, as well as the development of the interface, which will be designed taking into account operators' feedback regarding the type and quantity of information to display on the screen, the notifications to present, and so on.
Another important consideration: how explainable must the solutions offered by the algorithm be, without compromising the usability and effectiveness of the tool?
"One of the main aspects raised during interviews with FMPs and ATCO Supervisors is the trust in artificial intelligence, which is a black box due to the complexity of discerning the machine's decision-making process - Brambati highlights - indeed, one of the first requests made by operators is precisely that the tool should be able to explain why a particular solution is provided and what its impact is in terms of effectiveness in optimising traffic".
Once developed, the interface will be tested with its end-users: the FMPs. The project will conclude in 2025, when the final phase of the ASTRA validation exercises will take place: a real-time simulation in Geneva. "We will try to conduct this validation with the same Flow Management Positions who participated in the earlier phases of the project so that they can have the opportunity to test the concept they helped to develop", Brambati concludes. The concept that, if implemented and developed, will contribute to increasing the capacity of the airspace sectors (i.e., the number of flights) without compromising safety and without affecting the workload of those who monitor and manage traffic in the skies.
This article, in its original version, was published in Italian on Deep Blue's blog on 13/02/2024. The translation was provided to by the authors.