Files
Dubois_66871900_2024.pdf
Embargoed access until 2025-08-03 - Adobe PDF
- 23.19 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- The optimization of power production at the scale of wind farms is currently one of the key approaches considered to further alleviate the Levelized Cost Of Energy (LCOE) in the context of wind energy. However, current control approaches have mainly been oriented toward developing individualistic controllers, maximizing the efficiency of isolated wind turbines. This approach, though comparatively simple, fails to consider the wake effects thereby leading to a suboptimal operating point of the wind farm taken as a whole. Indeed, a wind turbine can essentially be considered an energy-extracting flow device that converts the wind's kinetic energy into electrical power. Therefore, it sheds a specific signature in its trail: its wake, a zone of low velocity and high turbulence. The latter then propagates over large distances downstream, degrading the performances of downstream impinged wind turbines. This thesis was devoted to developing a robust controller to maximize wind farm power output through wake steering. A model predictive control (MPC) framework was formulated to achieve this goal, leveraging a high-fidelity flow solver in conjunction with an engineering wake model. In this setup, the flow solver was treated as the controlled system, while the engineering wake model served as a surrogate, capable of making predictions at a low computational cost to determine optimal control strategies. This surrogate model was integrated with a gradient-free optimization technique, specifically the Covariance Matrix Adaptation - Evolution Strategy (CMA-ES). The model predictive control approach was validated on a two-turbine configuration, where the yaw angle of the upstream turbine was the control variable, and the wind farm power output was the performance metric.