A Multivariate Machine Learning Approach for the Prediction of Wind Turbine Blade Structural Dynamics

Wind turbine blade structural dynamics are crucial in the turbine structural design phase. Blade deflections and loads can affect the weight of the rotor as well as the power performance of a wind turbine if the deflections are extremely high. Predictions of the turbine’s blade deflections and loads can lead to informative decisions on optimizing the design of the blade. In this work, a multivariate machine learning (ML) approach is used to predict the blade’s dynamics based on the wind flow conditions and control actions of the turbine. Three different datasets were generated using the OpenFAST software tool for three different wind turbulence classes. Various ML algorithms were trained to predict the blade deflections at the tip and blade loads at the root in the edgewise and flapwise directions. The ML models were tested for generalization of the model to different flow conditions. A model is trained for one dataset with one of the turbulence classes and then used to predict the outputs of the other two datasets. The random forest ML algorithm gave the best accuracy for predicting the outputs for the dataset it was trained for, as well as the other two datasets. The accuracy of predictions was found to be higher in the edgewise direction for both load and deflection outputs. In the flapwise direction, the model could predict the outputs of the data it was trained for with an accuracy of around 99% and for the other two datasets with an accuracy of over 75%. While in the edgewise direction, the model trained on only one dataset gave a prediction accuracy above 95% for all three datasets.