Fatigue Load Prediction of Large Wind Turbine by Big Data and Deep Learning

This paper introduces a fatigue load prediction framework for the nacelle of wind turbines utilizing operational data sourced from the supervisory control and data acquisition (SCADA) system of a Wind Farm. The primary objective revolves around achieving precise fatigue load predictions for the nacelle, forecasting 15 minutes in advance. Firstly, a series of data preprocessing steps were implemented on the wind turbine data procured from the SCADA system to enhance the model's generalization capability and robustness, including 3-Sigma outlier detection, sample point selection, as well as filtering and de-noising techniques. Then, the gate recurrent unit (GRU) was tuned and trained as the optimal model for long time-series forecasting of SCADA parameter channels. Subsequently, the support vector regression (SVR) model, refined with grid search methodology, was employed to establish the intricate relationship between SCADA parameters and fatigue loads. Finally, the predicted SCADA parameter values were fed into the optimized SVR model to derive fatigue load prediction results for wind turbines. Experimental results indicated that the proposed model framework can effectively predict fatigue loads on the nacelle of wind turbines utilizing the SCADA data in real scenarios.