Wind farms are essential to the energy transition, but their optimization remains limited by prediction models that use simplified representations of wind, ignoring the complexity of atmospheric conditions.
The result: inaccurate predictions of energy production and turbine lifespan, hindering optimization and driving up costs.
The objective: to create a fast and accurate digital twin of wind fields at wind farm scale. You will develop generative models (probabilistic diffusion, autoencoders, transformers) capable of reproducing in a matter of seconds what high-fidelity simulations (Méso-NH) compute in several hours.
The challenge: encoding the complex physics of turbulent flows into deep learning architectures while preserving essential spatio-temporal properties. You will work with data from Méso-NH (CNRS/Météo-France) to train and validate your approaches.
The impact: faster feasibility studies, optimized turbine placement, and improved predictive maintenance.
What you will gain: a rare profile at the physics/AI interface, and transferable skills well beyond the energy sector.
Your supervisors: Prof. Taraneh Sayadi (Cnam, M2N), expert in Scientific Machine Learning and model reduction for turbulent flows. Dr. Emeline Noël (IFPEN), specialist in boundary layer/wake interactions and Méso-NH contributor. Dr. Guillaume Enchéry (co-supervisor), expert in model reduction for PDEs.
