The motion forecasting study of floating offshore wind turbine using self-attention long short-term memory method

Abstract

The motion response of a floating offshore wind turbine (FOWT) serves as a critical indicator for the safe operation of offshore wind energy systems. It is significant to predict these responses accurately to enhance both operation and maintenance requirements. This study is thus motivated to investigate a novel hybrid model integrating the self-attention (SA) method with the long short-term memory (LSTM) method for improved accuracy prediction of motion response of a FOWT in wind-wave coupled environments. The novel SA-LSTM model synergizes the LSTM capability to remember the historical data in the sequence coupled with the self-attention ability to consider the important data points in a global context, addressing the long-term temporal dependencies limitation and enhancing the model accuracy. Numerical simulations of a 5 MW semi-submersible platform are conducted under random waves and winds. The forecast model collects numerical data on platform response and environmental parameters as input variables. Subsequently, the data is post-processed to establish training and test groups. The LSTM and SA-LSTM models are then applied to learn from the training data. Comparing results generated from the two models, it is found that the SA-LSTM model outperforms the traditional LSTM method in terms of discreteness and overall accuracy. The R2 value for yaw motion calculated using the SA-LSTM model is 97.8%, significantly higher than the LSTM model R2 value of 71.6%. This study demonstrates that a self-attention module can significantly enhance FOWT motion forecasting. As the predicted advance time (PAT) increases, the veracity decreases. Thus, balancing model veracity and PAT is essential. The developed SA-LSTM model is also applied to forecast the platform motion in an environment where the water depth changes. The model shows strong adaptability in forecasting the platform response to variations in water depth, which demonstrates the practicality of the SA-LSTM model. Combining the SA-LSTM model with a global positioning system enables short-term forecasts of FOWTs in various wind-wave coupled environments.