期刊
ENERGY CONVERSION AND MANAGEMENT
卷 210, 期 -, 页码 -出版社
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.enconman.2020.112733
关键词
Wind power; Wind energy; Wind resource; Machine learning; L-moments; GloWiSMo
A new model for mapping the near-surface wind speed L-moments on a high spatial resolution scale (250 m x 250 m) is introduced (GloWiSMo). The target variables are the first five L-moments of 6146 globally distributed wind speed time series. ERA5 reanalysis wind speed available on a 0.25 degrees x 0.25 degrees grid was used as predictor representing the large-scale wind field. Eleven predictors derived from a land cover model and a digital elevation model were applied to integrate the influence of small-scale surface properties on the wind field. The model is based on a least-squares boosting approach which is a machine learning algorithm. The parameters of the Kappa and Wakeby distribution were estimated based on the modeled L-moments. By applying the power law, the near-surface wind speed distribution can be extrapolated to any hub height. Here, we selected a typical wind turbine hub height of 120 m to demonstrate the potential of GloWiSMo. It was found that the relevance of a predictor on the spatial variability of the wind resource changes with the size of the investigation area. While the roughness length is a decisive factor for the large-scale spatial variability of the wind resource, the relative elevation is an important factor for the small-scale spatial variability. Rigorous model evaluation was performed using a validation dataset containing 598 globally distributed wind speed time series. The coefficient of determination calculated for the first L-moment was found to be 0.83. Based on the evaluation results, we argue that the developed model enables accurate and spatially explicit wind resource estimates at a very high spatial resolution.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据