Development of time-decoupled MPC for LES-based real-time wind farm control

Over the years, an optimal control framework based on LES was developed at KU Leuven to maximize wind farm energy extraction by mitigating effects of turbine wake interactions (see e.g. Ref. [1]). However, this work was intended as benchmarking study, since LES is commonly considered too slow for practical purposes. Using coarse grid resolutions, this study is a first investigation on the feasibility of using LES as real-time plant model for receding-horizon wind farm control. To account for computational times, we propose a time-decoupled model-predictive control (MPC) loop, where controls are computed ahead of time based on a prediction of the future flow. The methodology is validated on the TotalControl Reference Wind Power Plant. By varying the MPC horizon and update time and the spatio-temporal resolution of the LES control models, we investigate the trade-off between computational time and performance. The results indicate that power extraction is mostly governed by the MPC parameters, whereas grid resolution has a minor impact. By leveraging these insights, we achieve near real-time computational speed while maintaining competitive power gains up to 40%.

[1] Munters W & Meyers J (2018). Dynamic strategies for yaw and induction control of wind farms based on large-eddy simulation and optimization. Energies (Basel), 11(1), p177.