TY - GEN
T1 - Data-Driven Modeling Techniques for Wind Turbine Aerodynamic Loading
AU - Roetzer, James
AU - Hall, John
AU - Li, Xingjie
AU - Maldonado, Claudia
AU - Boik, Hunter
AU - Janke, Ben
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.
PY - 2025
Y1 - 2025
N2 - Adaptive aerostructures, such as morphing blades, provide a crucial improvement to the control of a wind turbine beyond what is possible with pitch and torque control. However, using these new axes of control proves difficult due to computational expense in real-time applications. In this paper, the DTU 10 MW reference turbine is used to create data-driven models for a variety of load-related turbine performance parameters on a turbine with twist morphing capability. Data is gathered using OpenFAST to simulate the twist morphing under steady-state conditions. This data is then used to train decision trees, Support Vector Machines (SVMs), Neural Networks (NNs), and Gaussian Process Regression (GPR) data-driven models for each target performance parameter. The models are then compared in terms of accuracy and prediction speed to evaluate which model is most suitable for the application. In all examined cases, the models created with GPR were optimal in terms of accuracy, with reductions in Root-Mean-Square Error (RMSE) of between 61 and 86% as compared to the second-best performing model in each case. Prediction speed for these GPR models were at least comparable to competing models in all cases, and in one case was the fastest model examined. Overall, GPR performed the best of the examined modeling techniques for all cases, indicating its suitability for model loading in this application in future work.
AB - Adaptive aerostructures, such as morphing blades, provide a crucial improvement to the control of a wind turbine beyond what is possible with pitch and torque control. However, using these new axes of control proves difficult due to computational expense in real-time applications. In this paper, the DTU 10 MW reference turbine is used to create data-driven models for a variety of load-related turbine performance parameters on a turbine with twist morphing capability. Data is gathered using OpenFAST to simulate the twist morphing under steady-state conditions. This data is then used to train decision trees, Support Vector Machines (SVMs), Neural Networks (NNs), and Gaussian Process Regression (GPR) data-driven models for each target performance parameter. The models are then compared in terms of accuracy and prediction speed to evaluate which model is most suitable for the application. In all examined cases, the models created with GPR were optimal in terms of accuracy, with reductions in Root-Mean-Square Error (RMSE) of between 61 and 86% as compared to the second-best performing model in each case. Prediction speed for these GPR models were at least comparable to competing models in all cases, and in one case was the fastest model examined. Overall, GPR performed the best of the examined modeling techniques for all cases, indicating its suitability for model loading in this application in future work.
KW - Data-driven model
KW - Gaussian process regression
KW - Machine learning
KW - Neural network
KW - Regression modeling
KW - Wind energy
KW - Wind turbine
UR - https://www.scopus.com/pages/publications/105018580333
U2 - 10.1007/978-981-96-5495-6_26
DO - 10.1007/978-981-96-5495-6_26
M3 - Conference contribution
AN - SCOPUS:105018580333
SN - 9789819654949
T3 - Lecture Notes in Mechanical Engineering
SP - 323
EP - 336
BT - Responsible and Resilient Design for Society - Proceedings of ICoRD 2025
A2 - Chakrabarti, Amaresh
A2 - Singh, Vishal
A2 - Onkar, Prasad S.
A2 - Shahid, Mohammad
PB - Springer Science and Business Media Deutschland GmbH
T2 - 10th International Conference on Research into Design, ICoRD 2025
Y2 - 8 January 2025 through 10 January 2025
ER -