Fatigue Load Prediction of Wind Turbine Drive Train Based on CNN-BiLSTM

The fatigue loads of operational wind turbine drivetrain systems are typically quantified using the rainflow counting method based on stress measurements at critical components, a process that is time-consuming and costly. This paper addresses the significant deviations observed in traditional fatigue load quantification models employed for control strategies and parameter optimization in operational wind turbines. We propose a fatigue load prediction model for the drivetrain system based on a convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) architecture, utilizing state data from wind turbines. First, we construct a fatigue load feature database using simulation data from OpenFAST under rated wind speed conditions and above, which is subsequently used for training and testing the model. We then compare the model's predicted data with actual data, employing relevant evaluation metrics to assess the predictive performance of the model, thereby validating its effectiveness. Finally, by comparing the prediction results with those from long short-term memory and deep neural network models, we demonstrate that the CNN-BiLSTM load prediction model significantly enhances the accuracy of load predictions for wind turbine drivetrain systems.