Analysis of Model-Free Control of Wind Farms Using Large-Eddy Simulations

Wind farms are clusters of wind turbines deployed over a relatively small area. During operations, the wake from upstream turbines may impinge on trailing turbines causing a decrease in power production. Wind farm control strategies aim at mitigating the effect of wake interactions. In this dissertation, model-free control strategies for wind farm power maximization have been evaluated using numerical simulations of the flow through wind farms. A model-free approach does not require a priori assumptions on the physical system, but learns on-line the system dynamics, avoiding modeling uncertainties. The control strategies are based on extremum-seeking control (ESC), a real-time gradient-based optimization algorithm. Either the turbine generator torque or the rotor yaw angle is used as the control parameter tuned by ESC to optimize the wind farm power production. The generator torque adjusts the turbine angular speed and the momentum deficit in the trailing wake, while the yaw angle serves to vary the direction of the wake and avoid trailing turbines. We first consider several implementations of ESC and assess their performances and practical feasibility. Both torque- and yaw-based ESC enhance power production, but the latter has a larger margin for improvement. For idealised turbine arrays, ESC achieves a potential power improvement of at least 7–8% compared to operations with design settings for an isolated turbine. After this calibration, we perform an optimization study for a real wind farm and obtain a quantitative evaluation of the impact of the control strategy in annual energy production. Large-eddy simulations with rotating actuator disk are used, in the first place, to provide a virtual wind farm to test the control algorithms. Additionally, the numerical data are investigated to gain a physical insight on the mechanisms underlying the performance improvement and broaden the impact of the optimization.