Physics-Enhanced Data-Driven Approach for Wind Turbine Aeroelastic Damping Prediction Based on LSTM RNN

Real-time monitoring of wind turbine aeroelastic damping is crucial for dynamically adjusting operational strategies and enhancing turbine stability and economic efficiency. However, since aeroelastic damping cannot be directly measured and effective industrial methodologies remain limited, this study proposes an innovative hybrid prediction framework for aeroelastic damping of wind turbine blades based on field-measured turbine data. First, a general model for calculating blade root reaction forces was developed using blade element momentum theory, considering multiple influencing factors such as motor torque, gravitational force, and centrifugal force. Linear regression and decision tree algorithms were employed to identify key coefficients in the theoretical model, thereby providing accurate hub load inputs for finite element (FE) calculations of tower aeroelastic damping through blade physical modeling. Second, a full-scale FE model of the wind turbine was constructed in Abaqus, where dynamic responses were computed using hub axial forces as inputs and compared with field data to obtain aeroelastic damping values, yielding high-quality labeled data for training machine learning models. Finally, an aeroelastic damping dataset was generated through data analysis and downsampling, and a long short-term memory recurrent neural network was trained as the prediction model. Simulation results demonstrated high accuracy, with mean prediction errors below 2.2% and maximum errors below 2.5% on real turbine datasets. In addition, experimental validation further confirmed the effectiveness of the proposed method. The model features a computationally efficient architecture, strong real-time applicability for high-dimensional inputs, and considerable potential for practical implementation.