Robust energy production optimization of a wind farm using wake steering by calibrating the FLORIS model on SCADA data

Abstract

Wind farms experience significant efficiency losses due to the aerodynamic interaction between turbines. A possible control technique to reduce these losses to a minimum is yaw-based wake steering. This thesis investigates the feasibility of this technique by calibrating a surrogate model called the FLOw Redirection and Induction in Steady-state (FLORIS) model on a data set from the Lillgrund wind farm and using it to estimate the potential energy gain. The data set available is processed methodically to remove outliers and erroneous data points, resulting in a reliable and useful data set. It is used to obtain free stream wind conditions per time step and relate those to power measurements. The data set is consequently used to calibrate the tuning parameters of the FLORIS model. The calibration is done using a newly proposed method that determines the tuning parameters per combination of wind speed and turbine spacing. A difference with commonly applied calibration methods is that power measurements are used instead of predicted powers or flow field data from high-fidelity models. The performance of the calibrated model is tested through multiple uncertainty analyses. It is found that the model has a significant bias but low uncertainty by comparing the predicted wake losses with measured wake losses. This bias can potentially be reduced if atmospheric stability is taken into account. With the bias and uncertainty quantified, the FLORIS model is used to optimize the annual energy production of the Lillgrund wind farm by finding the ideal yaw angles for specific inflow conditions. A significant energy gain can be achieved when the optimal yaw angles are determined deterministically. However, the energy gain decreases drastically when uncertainty in input conditions is considered, showing that these yaw angles are not robust in terms of performance under uncertainty. More robust yaw angles can be obtained when the input uncertainty is taken into account during the yaw optimization. The energy gain achievable with these more robust yaw angles is approximately 3.4%. Therefore, it can be concluded that achieving an energy gain using yaw-based wake steering is feasible for the Lillgrund wind farm.