Characterizing forecastability of wind sites in the United States. (April 2019)

• Record Type: Journal Article
• Authors: Feng, Cong; Sun, Mucun; Cui, Mingjian; Chartan, Erol Kevin; Hodge, Bri-Mathias; Zhang, Jie
• Is Part Of: Renewable energy. Volume 133(2019)
• Page Start: 1352
• Page End: 1365
• Publication Date: 2019-04
• Subjects: Forecastability -- Time series analysis -- Wind power forecasting; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2018.08.085
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: With the rapid growth of wind power, managing its uncertainty and variability becomes critical in power system operations. Wind forecasting is one of the enablers to partially tackle challenges associated with wind power uncertainty. To improve the 'forecasting ability', defined as forecastability, different forecasting methods have been developed to assist grid integration of wind energy. However, forecasting performance not only relies on the power of forecasting models, but is also related to local weather conditions and (known as wind characteristics) wind farm properties. In this study, geospatial and instance spatial distributions of six wind characteristics and two forecasting error metrics are first analyzed based on 126, 000 + wind sites in the United States. Forecasts in different look-ahead times are generated by using a machine learning based multi-model forecasting framework and the Weather Research and Forecasting model. A forecastability quantification method is developed by characterizing the relationship between forecastability and wind series entropy using three regression methods, i.e., linear approximation, locally weighted scatterplot smoother nonlinear nonparametric regression, and quantile regression. It is found that the forecastability of a wind site can be successfully characterized by wind series characteristics, thereby providing valuable information at different stages of wind energy projects. Highlights: Characterize weather conditions and wind site properties of 126, 000 + wind sites in the United States. The forecastability of wind sites is assessed by time series characteristics in different look-ahead times. Both numerical weather prediction methods and machine learning methods are adopted to generate forecasts.