Spatiotemporal wind field prediction based on physics-informed deep learning and LIDAR measurements

Spatiotemporal wind field information is of great interest in industry e.g. for resource assessment and turbine/farm monitoring & control. However, its measurement not feasible because only sparse point measurements are available with the current sensor technology such as LIDAR. This work fills gap by developing a method that can achieve spatiotemporal predictions combining LIDAR flow physics. Specifically, deep neural network constructed Navier–Stokes equations, which provide good description atmospheric flows, incorporated employing physics-informed learning technique. The training this physics-incorporated model requires data while whole domain (which cannot be measured) predicted after training. study, discover complex patterns do present dataset, totally distinct from previous machine based prediction studies treat models “black-box” require corresponding input target values to learn relations. numerical results on front turbine show proposed predicts velocity (including both downwind crosswind components) very well wide range scenarios various noises, resolutions, look directions, turbulence levels), promising given line-of-sight speed at locations used.