Enhancing Model Predictive Control of Wind Turbines with Machine Learning
Approach Helps Reduce Dynamic Loads on Large Turbines
“The advantages of a machine learning–enhanced model predictive controller can be demonstrated both in simulation and in a real-life field-testing environment.”
Key Outcomes
- The machine learning–enhanced model predictive controller was successfully validated through scenario-based testing, including both high-fidelity simulations and field tests conducted on a full-scale wind turbine.
- Simulations showed that the controller significantly reduced thrust force fluctuations, especially around key frequencies.
- MATLAB and Simulink enabled fast preprocessing, model development, simulation, and automated code generation for deployment on the real turbine controller.
Engineers at RWTH Aachen University and W2E Wind to Energy GmbH are addressing the challenges posed by increasingly large and lightweight wind turbines, which are more susceptible to dynamic loads and structural oscillations. To enhance control strategies, they developed an advanced model predictive controller (MPC) augmented with a machine learning component to better predict and mitigate thrust force fluctuations.
Utilizing MATLAB® and Simulink®, the engineers modeled the wind turbine’s dynamics and designed the MPC. They used MATLAB for data preprocessing and trained a local linear neuro-fuzzy model to predict changes in thrust force. The controller was validated through simulations and software-in-the-loop testing, then deployed on a full-scale 3-megawatt wind turbine using Simulink Coder™ for automated code generation. This integrated approach improved load reduction and demonstrated the practical viability of combining machine learning with Model-Based Design for wind turbine control.
Products Used
Related Resources
