Machine learning methods to model Europe’s crowded skies The computational power needed in air traffic modelling is increasing. The EU- and industry-funded SIMBAD project developed powerful artificial intelligence-based simulation models that make evaluating new air traffic scenarios far easier. Transport and Mobility © Irina Anosova/Shutterstock.com With ever more planes travelling across Europe, air traffic management (ATM) is becoming increasingly complex. New technologies and concepts hold the promise of making Europe’s air transport industry more efficient, safe and sustainable. The SIMBAD (Combining Simulation Models and Big Data Analytics for ATM Performance Analysis) project has developed state-of-the-art modelling techniques based on artificial intelligence, using new machine learning approaches to improve on current air traffic microsimulations, the highly detailed models necessary to integrate new technologies into ATM systems. “Given the complexity and computational cost of large-scale, microscopic air traffic simulation tools, simulations are necessarily restricted to a limited number of scenarios, which are often insufficient to obtain conclusive results,” explains David Mocholí, aviation director at Nommon Solutions and Technologies. “The machine learning techniques investigated by SIMBAD help us overcome these shortcomings,” says Mocholí, who coordinated the project. Creating hyperrealistic air traffic scenarios Funded within the framework of the SESAR Joint Undertaking, a public-private partnership set up to modernise Europe’s ATM system, SIMBAD focuses on three fundamental problems with current performance evaluation models. The first is how to estimate hidden variables in flight, such as the aircraft weight on take-off. While not directly observable, these have a tangible effect on flight trajectories in simulations. Secondly, the project seeks to make the simulations more effective and efficient using a variety of machine learning clustering techniques that can establish a representative set of air traffic scenarios. Finally, SIMBAD is applying active learning techniques to build metamodels. These metamodels are simpler and less computationally costly approximations of microsimulations, which leads to more efficient and insightful evaluation of new ATM technologies. SIMBAD metamodels are being developed in collaboration with the EU-funded NOSTROMO project, which has developed an API that facilitates the process of building these metamodels. Modelling air traffic at different spatial and temporal scales While the SIMBAD project is ongoing at the time of writing, there have been several key developments already. The new algorithms for traffic pattern characterisation have shown their ability to identify representative traffic scenarios at different spatial and temporal scales, which will result in more comprehensive simulation experiments. Additionally, SIMBAD’s metamodels have proven to be more efficient and faster than existing simulations when modelling new ATM technologies. The team were also able to find hidden variables through the analysis of historical air traffic data. “We have accurately estimated two hidden variables related to airspace users – cost index and landing weight – for a set of trajectories in different weather conditions,” remarks Mocholí. The project is now moving into the evaluation phase, after which more conclusive results are expected. “The first tests and validations performed so far have been very promising, and we are confident that SIMBAD will make valuable contributions to ATM performance analysis,” he says. The team hopes to further develop the work started in SIMBAD under the SESAR 3 Joint Undertaking. “This would be a great opportunity for getting closer to the ultimate goal of integrating this solution into Europe’s air transport management system,” Mocholí concludes. Keywords SIMBAD, microsimulations, ATM, machine learning, artificial intelligence, air, traffic, data, management, hidden, variables
