Wind Turbine Airfoil Noise Prediction Method Based on Generalized Airfoil Database and Residual Neural Network

To address the limitations of existing wind turbine airfoil databases, the high computational cost, and low efficiency of noise prediction, this paper proposes a wind turbine airfoil noise prediction method based on generalized airfoil sets and residual neural networks. Firstly, taking a database of 31 commonly used wind turbine airfoils as a reference, a generalized airfoil set with diverse geometric contours was generated. This was achieved by employing airfoil functional integration theory, B-spline curves, and the Class function/Shape function Transformation (CST) method while varying coefficients and control vector parameters. Secondly, the BPM semi-empirical model was used to compute the noise for the generalized airfoil set, which served as the data labels for deep learning. Finally, classical machine learning models were utilized to construct the airfoil noise prediction model. The results demonstrate that the airfoil noise prediction model constructed with the residual neural network (ResNet-18) achieved the highest prediction accuracy, with a mean squared error (MSE) of 0.0282 and a coefficient of determination (R2) of 0.99972. Additionally, the trained model exhibited computational efficiency that was 17.5 times higher than the BPM model.