Evaluation of LES-based time-decoupled model-predictive control in different wind farm layouts

Abstract

Abstract Very recently, Janssens and Meyers (Wind Energy Sci., vol. 9, 65–95, 2024) proposed a time-decoupled model-predictive control (TD-MPC) framework for wind farm power optimization using a control model based on coarse-grid large-eddy simulation (LES). In a receding-horizon fashion, by computing the controls for the next time window based on a prediction of the flow (so ahead of time), they achieved (almost) real-time computational speed. The current paper further explores the potential of this approach in the context of dynamic yaw control, under the assumption of exact state knowledge. To that end, the LES-based, TD-MPC framework is validated on a reference wind farm for three different wind directions, for which we observe gains ranging from 4 (for staggered configurations) to 42% (for aligned configurations w.r.t. the wind direction). The LES-based controller always outperforms a Betz optimal controller and a static (wake model based) yaw controller at near real-time speed. An analysis of the yaw angles reveals that the framework is very effective in steering the wakes away from downstream turbines in upstream and downstream regions, synchronized to the turbulent inflow. By comparing the performance of the LES-based controllers (based on the actuator disc turbine model) against a reference simulation that uses a more accurate actuator sector model, we illustrate the robustness of the approach in the presence of the additional model error.