Nonlinear long-term vibration response prediction of offshore wind turbines under full operating conditions based on deep learning

Efficient prediction of the vibration response of offshore wind turbines plays a crucial role in proactively identifying potential vibration hazards and enabling real-time adjustments to operational strategies. The simulation methods are limited by lengthy computation times. In addition, pure data-driven prediction models suffer from limited adaptability to unseen conditions and lack constraints based on physical mechanisms. This paper presents a new method combining deep learning and OpenFAST simulation to predict the nonlinear long-term vibration response of offshore wind turbines. The approach is designed to encompass all operating conditions. Firstly, a multi-layer stacked BiLSTM architecture is designed to capture long sequences of time-series data. Recursive calculations are implemented using a sliding time window approach, while independent parallel computations are achieved through the multiprocessing technology. Subsequently, the time series data of 68 different wind-wave load scenarios are obtained through OpenFAST analysis, and the vibration response is predicted using the deep learning framework. Furthermore, the multi-input recursive BiLSTM obtained from the novel method is compared with the existing time series model. The results demonstrate that the proposed model accurately replicates both global and local features of time-history responses across diverse offshore wind turbine datasets. The average computation time is only 1/744.97 of that required by simulation models. Moreover, within a 10 s forecast duration, the model maintains an average online prediction accuracy of 91.94 % across all operational conditions. Under extreme conditions, the prediction accuracy is 29.71 % and 17.52 % higher than those of the end-to-end BiLSTM and RNN models, respectively. This proposed method is particularly suitable for applications where traditional numerical methods are limited, such as rapid simulations under real-time changes in operating conditions within complex marine environments.