Machine learning based surrogate model to analyze wind tunnel experiment data of Darrieus wind turbines

Abstract

Wind tunnel experiment is one of the most common research methods of wind turbines. However, the processing of experiment data towards Vertical Axis Wind Turbines (VAWT) are inefficient. In this work, a data-driven surrogated model based on Artificial Neural Networks (ANN), Support Vector Regression (SVR), Gaussian Process Regression (GPR) and Decision Tree Regression (DTR) is proposed to regress the experimental data of aerodynamic forces. Meanwhile, a novel U-typed Darrieus Wind Turbine (UDWT) and two H-typed VAWTs are manufactured in this work. A measurement system combining the magnetic remanence brake, six force sensor, and data processing subsystem is assembled to investigate the aerodynamic forces of wind turbines. A Spin-Down measurement procedure is adopted to collect experiment data in various operating conditions. Results showed that SVM- and GPR-based regression models feature the better fitting ability with R2 larger than 0.99, which can regress the experiment data accurately. ANN-based surrogated model can reproduce the fluctuation of experiment data because of the best learning ability. Aerodynamic forces of UDWT outperform H-type wind turbines at azimuth angle of 90° and 270°.