Optimizing the output of wind farms with micro-siting and wake steering: An LES-informed data-driven approach

We present results from an extensive computational campaign focused on maximizing the power output of wind farms. To achieve this objective, we use a framework based on large-eddy simulations of the flow and a Bayesian approach, where the design space is explored with surrogate models. The proposed framework is applied to two problems: layout optimization (micro-siting) and yaw angle optimization (wake steering). We assess the computational feasibility and potential benefits of the proposed method, and compare its performance with conventional optimization strategies. Moreover, we discuss the role of the fluid mechanical phenomena that are neglected by low-fidelity flow solvers but are effectively utilized by the proposed framework to increase the power output of the farms. We find that the proposed framework shows comparable performance to established optimization strategies. However, considerable performance gains can be achieved when the flow is more complex, such as in the case of the wake steering problem, where several of the assumptions in the simplified flow models become less accurate. This work opens up opportunities for data-driven and high-fidelity optimization of wind farms' power output.