Real-time optimal control of wind-farm atmosphere interaction

Current Wind-Farm flow Control (WFC) focuses on the interaction of wind turbines through their wakes, and relies on fast heuristic wake models that are mostly considered in an open-loop control framework. However, modern wind farms interact with the atmosphere at much larger scales, such, e.g., as their excitation of atmospheric gravity waves. WFC response is thereby not only governed by intra-farm turbine wakes, but possibly even more by the interaction between the larger atmospheric mesoscales and the farm operation. The only models that realistically capture these aspects down to the wake scale are large-eddy simulations (LES), which are generally run on high-performance computers, yet considered orders of magnitude too slow for use in real-time model predictive control.Recently however, we have shown that coarse-grid LES integrated in a time-decoupled model predictive control (TDMPC) framework, is about a factor three too slow only for real-time use, while potentially still being effective at realizing the WFC objective. With wind turbines being the largest manmade “flow actuators” existing today, and smaller-sized systems exhibiting faster time scales, the wind farm will be the first turbulent flow system in which LES can be used as a real-time control model.We aim at inducing a paradigm shift in the use of LES, by developing a first fully integrated LES-based TDMPC and demonstrate it in a high-fidelity emulator environment, as well as, in part, using field data. This raises following fundamental research challenges: diverging sensitivities of perturbations in turbulent flows (chaotic systems) over long time horizons, the sparse nature of measurements in the atmosphere required for state estimation in the control loop, the limited understanding of wind-farm atmosphere interaction in non-neutral stratification, and the efficient emulation of WFC using high-performance computing. My pioneering work in these fields will enable us to tackle these challenges.