Attention based surrogate model to predict load envelope of monopile supporting offshore wind turbines

Monopiles are widely adopted as the foundations for supporting offshore wind turbines (OWTs) in shallow water. In the serviceability limit state design, the maximum allowable rotation of monopiles is 0.25°. The concept of load envelope can help understand the load combinations on monopile satisfying the above criterion. This paper aims to create a surrogate model using the machine learning technique, i.e., an attention-based model, to predict the load envelope and response of monopiles in sand. First, finite element models are established to calculate the monopile responses under different load paths after the model validation, which provide a comprehensive dataset. Then, the surrogate model, called Offshore Wind Turbine Transformer (OWTTransformer), is trained on a set of numerical model outcomes and subsequently validated by unexplored data of monopile responses. By leveraging the OWTTransformer, monopile responses and load envelopes can be rapidly predicted within seconds, thereby significantly reducing the computational expenses associated with determining the load envelopes of monopile-supporting offshore wind turbines. The results indicate that the OWTTransformer model is sufficient to capture the non-linear response of monopiles with excellent accuracy and robustness.