Master's Thesis
Development of a Surrogate Model for Predicting Turbine Loads and Power Production
— 2026
Key information
Authors:
Supervisors:
Published in
July 3, 2026
Abstract
The computational cost associated with high-fidelity aeroelastic simulations and experimental measurements for estimating wind turbine power production and fatigue damage limit their applicability in control strategies and real-time applications. This problem can be mitigated by using "black-box" models that, at a fraction of the computational cost of traditional methods, provide approximate predictions of the desired quantities. This thesis proposes developing surrogate models using three machine learning algorithms: Polynomial Regression (PR), Gaussian Process Regression (GPR), and Artificial Neural Networks (ANN). The datasets used for training were generated using the mid-fidelity wind farm simulation tool FAST.Farm. The primary conclusion is that the surrogate models are more accurate in estimating power output than fatigue damage. Polynomial Regression was the worst performing model, particularly for fatigue damage at the tower base. Both GPR and ANN models presented similar performances, with a slight advantage generally observed for the former. For power output, the lowest MAPE obtained was 0.15%, achieved by the GPR model. For fatigue damage, the GPR model also yielded the best results, with minimum MAPE values of 9.63% for the tower base and 7.98% for the blade root. Sensitivity analysis revealed that the most important features are wind speed and turbulence intensity, followed by the remaining features. The analysis also indicated that power output is predominantly determined by wind speed, whereas fatigue damage quantities are influenced by all features.
Publication details
Authors in the community:
Raquel Sofia Coelho Vieira Monteiro
ist199548
Supervisors of this institution:
Tiago José Rosário Lucas
ist425651
Ricardo Balbino Santos Pereira
ist151719
Fields of Science and Technology (FOS)
mechanical-engineering - Mechanical engineering
Publication language (ISO code)
eng - English
Rights type:
Embargoed access
Date available:
April 30, 2027
Institution name
Instituto Superior Técnico