Airspace management is intrinsically difficult to model due to the interaction of multiple different elements in an ever-changing environment. One of the most challenging modelling problems is the assessment of the performance impact of new solutions at a system-wide level, which has been a long-time objective of the airspace research community.

The NOSTROMO project set out to develop a new methodology that could be applied to measure performance of SESAR solutions. It copied the construction of metamodels, which approximate the behaviour of simulation models, and additionally deliver model transparency, computational tractability and ease of use.  NOSTROMO’s approach essentially integrated two well-established techniques: Active learning (to achieve greater performance with fewer data points); and simulation metamodels (functional approximators of the input-output relationships of a simulator).

The active learning metamodeling approach was then applied to two air traffic management simulators, Mercury and FLITAN, each modelling up to two SESAR solutions,  assessed at the network level. The training of the metamodels required several hours (eight for Mercury, five for FLITAN), after which their prediction performance was more than 1 000 times faster, with an error lower than 11%,  compared with the output values generated by the simulators. 

To extend NOSTROMO’s architecture to enable it to be used to assess other airspace management solutions, an online application programming interface (API) was developed to allow the training and creation of a new metamodel in a simulator-agnostic context;  the possibility to explore the entire metamodel space; and to measure confidence in the metamodel predictions.

A NOSTROMO simulator-agnostic dashboard was developed to facilitate performance assessment, allowing real-time dynamic communication with the API, visual exploration of the variables of interest (inputs and KPIs), and user-exportable reports.  

Download Nostromo project's white paper on lessons learned

Benefits

• Simulation-based analyses without full simulation runs
• Visual exploration of assessment variables in real-time
• Potential to save significant computational time