Historical reconstruction dataset of hourly expected wind generation based on dynamically downscaled atmospheric reanalysis for assessing spatio-temporal impact of on-shore wind in Japan

Abstract

Wind power is crucial for achieving carbon neutrality, but its output can vary due to local wind conditions. The spatio-temporal behavior of wind power generation connected to the power grid can have a significant impact on system operations. To assess this impact, the use of long-term reanalysis results of wind data based on a numerical weather prediction (NWP) model is considered valid. However, in Japan, the behavior of on-shore wind power generation is influenced by diverse topographical and meteorological features (TMFs) of the installation site, making it challenging to assess possible operational impacts based solely on power curve-based estimates using a popular conversion equation. In this study, a nonparametric machine learning-based post-processing model that learns the statistical relationship between the TMFs at the target location and the actual wind farm (WF) output was developed to represent the expected per-unit output at each location. Focusing on historical reconstruction results and using this post-processing model to reproduce the real-world WF output behavior created a set of expected wind power generation profiles. The dataset includes hourly long term (1958–2012) wind power generation profiles expected under the WF installation assumptions at various on-shore locations in Japan with a 5 km spatial resolution and is expected to contribute to an accurate understanding of the impact of spatio-temporal wind power behavior. The dataset is publicly accessible at https://doi.org/10.5281/zenodo.11496867 (Fujimoto et al., 2024).

Keywords

• On-shore wind power
• numerical weather prediction
• machine learning
• post-processing
• dataset