Neural Network-Guided Active Yaw Control in a Two-Turbine Wind Farm

Abstract

To optimize the power production of wind farms, the wakes can be manipulated and their adverse effects mitigated. A promising wake deflection method is yaw misalignment. By deflecting the wake away from downwind turbines, the yaw-based approach seeks to increase the collective power production of the grouped wind turbines by sacrificing some power output of the upwind turbine in yaw. Using an artificial neural network (ANN), specifically a feedforward neural network, this work aims to develop an active yaw control (AYC) scheme for a two-turbine wind farm. To aid in the development of the AYC, FAST.Farm is utilized, which is a newly developed midfidelty tool by the National Renewable Energy Laboratory (NREL). FAST.Farm is calibrated by comparing several of its predicted wake properties to results of a Large-Eddy Simulation (LES) for which a 1D actuator disk method is used. Subsequently, with the purpose of gathering data to train an ANN, the wind farm is simulated in a multitude of operating conditions. Specifically, the operating conditions are combinations of various turbulence intensities, wind shear exponents, and yaw angles for the upwind turbine. The ANN is then used to predict total power production in the wind farm, which informs the decisions of the AYC. It is shown that, when the turbulence intensity (TI) is low (5 %), the AYC increases wind farm power production by 5-6 %, depending on the wind shear exponent (a smaller wind shear exponent yields a larger gain in power production). As the TI increases, the gain in power production goes to zero. Moreover, it is shown that the AYC increases the structural loads (up to 20 %) on both the upwind turbine and downwind turbine with respect to the blade root out-of-plane bending moment and tower base fore-aft moment. With respect to the yaw bearing moment, the AYC greatly increases it for the upwind turbine (up to 7 times) and slightly lowers it for the downwind turbine (roughly 5 %).