Posters

Abstract

Standard practice for windfarm interaction model validation typically aggregates production data across the entire wind speed distribution, employing percentage-based error metrics for performance assessment [1]. However, windfarm interaction effects expressed as a loss are physically constrained to specific operating ranges. Including low-wind and high-wind (rated power) ranges introduces "non-informative" data that skews error statistics without offering insight into wind farm interaction physics. Specifically, near-zero and rated power production can dominate aggregate percentage errors, obscuring model fidelity within the active wind farm interaction-loss domain. This study proposes an analytical method for isolating operational data within the informative part of the wind speed spectrum to capture the peak sensitivity of wind farm interaction-induced power deficits. The proposed method analyses freestream wind speed data and predicted interaction losses to determine a set of filtering criteria. We refer to the two approaches as ‘baseline’ and ‘filtered’. Building on a related yield validation exercise using two model chains, this study presents a comparison of these model’s vs public production data for more than 40 wind farms across GB. The focus of this study is on highlighting the error distributions and summary metrics for both baseline and filtered validation methods.  In the filtered example, the error between prediction and observed data is magnified. In some cases it was observed that the error varies by more than 30% between baseline and filtered approaches. We demonstrate that conclusions derived from full-distribution datasets are frequently skewed by operating range where wind farm interaction losses are negligible. Filtering the data for error estimation should not change the relative performance of models, however it helps in a better understanding of the error in prediction and quantification of associated uncertainty.  Our findings suggest that targeted wind speed filtering provides a more physically consistent basis for validation and error estimation. This approach enhances the interpretability of error statistics and ensures that model comparisons are true reflections of wind farm interaction-loss representation rather than operational noise at the extremes of the power curve. Reference 1.    N. G. Nygaard, L. Poulsen, E. Svensson, and J. G. Pedersen, “Large-scale benchmarking of wake models for offshore wind farms,” J. Phys.: Conf. Ser. 2265, 022008 (2022). https://doi.org/10.1088/1742-6596/2265/2/022008