Mandar Tabib of SINTEF Digital Norway, and researcher on the EU-funded AI4HyDrop project, explains the use of AI to predict urban-scale wind and turbulence for drone operations.
The goal of this project task is to use AI to predict microscale wind and turbulence patterns, which are affected by turbulence induced by buildings and terrain. Current physics-based models take a long time to compute these predictions, whereas AI can provide fast predictions that are essential for drone operations such as route planning and dynamic airspace management.
The AI model was trained using data generated using computational fluid dynamics (CFD) simulations of a segment of the city of Prague. The CFD simulations were performed for different mesoscale wind directions and intensities, providing a dataset of microscale wind and turbulence patterns. Subsequently, an unsupervised machine learning technique was used to train the AI model using this dataset.The AI model has been used in a reinforcement learning model in which drones learn to avoid regions of high turbulence. The fast prediction speed of the AI model allows real-time decisions to be made for drone route planning.