Reinforcement Learning for Collaborative Wind Farm Control and Power Optimization
ORAL
Abstract
Maximizing the efficiency and reliability of renewable energy systems is essential for accelerating the decarbonization of energy production. Wind farms frequently underperform due to aerodynamic interactions between wind turbines. This includes wake effects, where downstream turbines experience reduced wind speeds and increased turbulence, lowering the overall farm efficiency.
Traditional wind farm control strategies focus on static, pre-defined turbine settings, which are unresponsive to the dynamic and turbulent nature of the wind field. In this study, we present an approach to active wind farm control that integrates high-fidelity Large Eddy Simulations (LES) of wind farms with Reinforcement Learning (RL) to optimize collaborative wind farm performance dynamically via wake steering. Wake steering is a wind farm control technique where upstream turbines are intentionally misaligned with the wind direction to redirect their wakes away from downstream turbines, reducing wake losses and increasing overall power production.
To simulate the wakes generated by the wind turbines, we use Winc3d, the wind farm simulator of the high-order finite difference framework Xcompact3d. It is coupled to a multi-layer perceptron based controller using the SmartSim and SmartRedis frameworks. This configuration enables efficient controller training across multiple parallel environments on high-performance computing resources. The controller continuously adjusts each turbine's yaw angle based on real-time wind velocity measurements in the turbine's vicinity, with parameters optimized through reinforcement learning to maximize total farm power output.
Testing on a three-turbine wind farm array demonstrated that our RL-based dynamic yaw steering strategies significantly improve total power production. The RL controller achieved a 4.30% increase in wind farm power output compared to baseline operation, nearly double the 2.19% improvement from static optimal yaw control using Bayesian optimization.
Traditional wind farm control strategies focus on static, pre-defined turbine settings, which are unresponsive to the dynamic and turbulent nature of the wind field. In this study, we present an approach to active wind farm control that integrates high-fidelity Large Eddy Simulations (LES) of wind farms with Reinforcement Learning (RL) to optimize collaborative wind farm performance dynamically via wake steering. Wake steering is a wind farm control technique where upstream turbines are intentionally misaligned with the wind direction to redirect their wakes away from downstream turbines, reducing wake losses and increasing overall power production.
To simulate the wakes generated by the wind turbines, we use Winc3d, the wind farm simulator of the high-order finite difference framework Xcompact3d. It is coupled to a multi-layer perceptron based controller using the SmartSim and SmartRedis frameworks. This configuration enables efficient controller training across multiple parallel environments on high-performance computing resources. The controller continuously adjusts each turbine's yaw angle based on real-time wind velocity measurements in the turbine's vicinity, with parameters optimized through reinforcement learning to maximize total farm power output.
Testing on a three-turbine wind farm array demonstrated that our RL-based dynamic yaw steering strategies significantly improve total power production. The RL controller achieved a 4.30% increase in wind farm power output compared to baseline operation, nearly double the 2.19% improvement from static optimal yaw control using Bayesian optimization.
–
Publication: Mole, A., Weissenbacher, M., Rigas, G., & Laizet, S. (2025). Reinforcement Learning Increases Wind Farm Power Production by Enabling Closed-Loop Collaborative Control. arXiv preprint arXiv:2506.20554.
Presenters
-
Sylvain Laizet
Imperial College London
Authors
-
Sylvain Laizet
Imperial College London
-
Andrew Mole
Imperial College London
-
Max Weissenbacher
Imperial College London
-
Georgios Rigas
Imperial College London