A novel frequency-domain physics-informed neural network for accurate prediction of 3D spatio-temporal wind fields in wind turbine applications

Wind power is a pivotal clean energy source worldwide. The structural safety and dynamic response analysis of wind turbines is significantly impacted by the availability and precision of wind speed data at their location. However, the sparse distribution of meteorological stations often makes it difficult to obtain high-resolution spatial wind speed data. This necessitates the application of conditional simulation to supplement low-resolution observational data. This study addresses this challenge by developing a frequency-domain physics-informed neural network (FD-PINN) designed to predict three-dimensional (3D) spatio-temporal wind fields for wind turbines by leveraging frequency-domain information. This approach involves constructing a deep neural network and integrating it with key physical models, including wind spectra, wind field coherence functions, and wind profiles. This integration allows the network to accurately predict wind conditions in environments with sparse wind field samples. The efficacy of our proposed methodology is assessed by comparing its predictive performance against traditional neural network approaches and actual observation data. Our findings demonstrate that integrating frequency-domain information significantly enhances the accuracy of spatial wind speed distribution predictions for wind turbines, compared to conventional methods. Additionally, this approach reduces spatial dependency issues with wind speed. Validation against real-world wind fields further confirms the feasibility and precision of this FD-PINN model in predicting 3D spatio-temporal wind fields.