Numerical evaluation of multivariate power curves for wind turbines in wakes using nacelle lidars

The IEC standards describe how to measure the power performance of an isolated wake-free wind turbine. However, most wind turbines operate under waked conditions for a substantial amount of time, calling for the need of a new methodology for power performance evaluation. We define multivariate power curves in the form of multivariate polynomial regressions, whose input variables are several wind speed and turbulence measurements retrieved with nacelle lidars. We use a dataset of synthetic power performance tests including both waked and wake-free conditions. The dataset is generated through aeroelastic simulations combined with both virtual nacelle lidars and the dynamic wake meandering model. A feature-selection algorithm is used to select the input variables among the available measurements, showing that the optimal model includes four input variables: three correspondent to wind speed and one to turbulence measures. Additionally, we give insights on the optimal nacelle-lidar scanning geometry needed to implement the multivariate power curve. Results show that the multivariate power curves predict the power output with accuracy of the same order under both waked and wake-free operation. For the in-wake cases, the accuracy is much higher than that of the IEC standard power curve, with an error reduction of up to 88%.