IMPROVING BETTAIR MAPS
The experiment aims to train AI Generative Adversarial Networks to mimic the output of HPC-CFD simulations at an affordable cost, adding them to Bettair’s map generation pipeline, to better understand the distribution of pollutants at street and urban levels.
Start date: 01/06/2021
Duration in months: 18
Problem Description
The 2021 WHO Air Quality Guidelines highlight the significant health risks of air pollution, with research revealing its adverse effects begin at lower concentrations. This experiment aims to reduce computational costs and provide affordable real-time air quality modelling tools for Bettair Cities.
Goals
New services
Challenges
Urban air quality is influenced by atmospheric dynamics, geometry, land use, and traffic patterns, resulting in varying pollutant distributions at microscales. Most problems are hyper-local, requiring high-accuracy maps using dense sensors and computational resources for high resolution.
Innovation results
Using 30 3D models of European capitals a dataset of 30.000, 256m by 256m areas was built. CFD simulations were performed on them, and the results used to train an AI model now integrated into Bettair cities platform for real-time air quality and emissions information.
Business impact
Simulation cost per km2 reduced from €1.850 to less than €1, spatial resolution of the real time modeling capabilities enhanced from 100 m2 to 1 m2, experimental setup time for new cities reduced by 80% from 3 weeks to 4 days, Bettair cities expected turnover for 2023 increase by a factor of two.