EUROCONTROL will use artificial intelligence to anticipate disruptions

Blog Research and Innovation Call for tenders

Artificial Intelligence (AI) and Machine Learning (ML) are a set of techniques which apply particularly well to the identification of patterns in large datasets. Long story short, if you have enough qualified data, you can train a computer to recognise some characteristics in data it has never seen before. Show the algorithm one million flight plans and the associated measured take-off and landing times and it will be capable of predicting delays for flight plans it has never seen before. If you want to read the longer version, check our report on Artificial Intelligence in ATC which also lists many other use cases.

As the Network Manager, EUROCONTROL has a treasure of data: flight plans and their updates but also surveillance data and A-CDM messages, plus all slots and  Flight Update Messages they issue via ETFMS. The flight plans show the initial plans by the airlines and the A-CDM milestones, slots and actual departure and arrival messages measure how well the plan worked. Even without being an expert, this looks like a great starting point for using ML techniques.

Another issue with AI and ML is that even if the algorithms are good at producing results, the "how" question is not trivial and such techniques often have a "black-box" aspect which causes a lot of struggle for regulators and certification authorities. Using AI for operational, tactical ATC tasks requires a level of safety and certification which is challenging for such techniques.

There is however more to air traffic management than just tactical use. With this call for tenders called AI support for network state monitoring, EUROCONTROL will explore how good such data can be used to identify and anticipate disruptions. In theory, AI and ML can be used to identify patterns that are not apparent in data and that humans can't easily grasp. Identifying possible disruption at an early stage is key to prepare the best response and reduce the impact. The research that will be done in this project will certainly be fascinating and could bring a lot of improvement in handling the network.

One interesting question to close: will the decisions taken on the base of early detection by an ML algorithm change the network so significantly that the algorithm will need to "re-adapt" to its own effect?

Interested in AI in ATC? Download our report on Artificial Intelligence in ATC which discuss possible usages at pre-tactical and tactical levels but also in training.