Wind turbine power curve modelling under wake conditions using measurements from a spinner-mounted lidar

Most wind turbines are installed inside wind farms, where they often operate under wake-affected inflow conditions. New methods are required to evaluate the power performance of a wind turbine in wake, as the International Electrotechnical Commission (IEC) standard procedure is applicable only to wake-free turbines. In this work, we investigate the accuracy of a multivariate power curve acquired through a polynomial regression, whose input variables are wind speed and turbulence measurements retrieved upstream of the turbine's rotor. For this purpose, we use measurements from the SpinnerLidar, a continuous-wave, scanning Doppler lidar measuring the turbine inflow. The SpinnerLidar was mounted in the spinner of a Neg Micon 80 wind turbine located within an onshore wind farm in western Denmark. The input variables are selected among the available lidar measurements with a feature-selection algorithm, resulting in seven input variables, distributed in different locations along the rotor area: six wind speed and one turbulence measurements. The multivariate power curve is tested and compared with IEC-similar power curves under both wake-affected and wake-free conditions. Results show that the multivariate power curve estimates the turbine's power output more accurately than the IEC-similar power curves, with error reductions up to 66.5% and 34.2% under wake-affected and wake-free conditions, respectively. Furthermore, the multivariate power curve estimates have an accuracy of the same order under both wake-affected and wake-free conditions. Finally, we show that the multivariate model accurately predicts the power even when a simple measuring geometry is used, such as circular scanning pattern with a diameter equal to 90% of the rotor.