A surrogate model for capturing wave power farm dynamics using spatial-temporal attention

Wave energy converters deployed in farms can experience intense hydrodynamic interactions due to the scattered and radiated waves on the free surface, making farm modeling challenging in realistic sea states. This study introduces a spatial-temporal surrogate model based on a transformer encoder architecture to predict the motion of multiple interacting wave energy converters in various sea states. The framework leverages experimental data from the SWELL dataset, predicting array responses in a previously unseen layout, i.e., a farm configuration, not available during the model's training phase. The model embeds incident wave time series together with device coordinates into a unified spatial-temporal representation. Self-attention then jointly captures the temporal evolution of motion dynamics and inter-device spatial dependencies. Across three irregular sea states, the model predicts device responses with high accuracy, showing close agreement with experimental measurements. These findings provide an initial proof-of-concept, highlighting the potential of an attention-based spatial-temporal surrogate model as a building block for predicting the dynamics of several interacting wave energy converters in previously unseen array configurations.