TY  - EJOU
AU  - Du, Yunlong 
AU  - Zhuang, Shuyi 
AU  - Ye, Zhigang 
AU  - Bu, Qiangsheng 
AU  - Chai, Yun 
AU  - Wang, Yuanbing 

TI  - Short-Term Solar Radiation Forecasting System for Jiangsu Province Based on FY-4A Multispectral Data-Regional Applicability Validation for High-Penetration Photovoltaic Grid Integration
T2  - Energy Engineering

PY  - 
VL  - 
IS  - 
SN  - 1546-0118

AB  - Under the dual challenges of global warming and energy transition, improving the short-term forecasting accuracy of surface solar radiation is of great practical importance for photovoltaic (PV) power integration. In this study, a short-term solar radiation forecasting model based on the XGBoost machine learning algorithm was developed for Jiangsu Province by integrating multispectral data from the Fengyun-4A (FY-4A) geostationary satellite with ground-based meteorological observations. The model incorporated 18 input features—including satellite reflectance, solar zenith angle, normalized difference vegetation index (NDVI), elevation, and land-cover data—to dynamically predict ground horizontal irradiance (GHI) with 0–4 h lead times. Systematic validation using data from 17 provincial stations during 2023 demonstrated that the model achieved a root mean square error (RMSE) of 165.62 W/m<sup>2</sup> and a correlation coefficient (R) of 0.82 for 1-h forecasts, representing an approximately 23% reduction in RMSE compared with traditional numerical weather prediction (NWP) models. Forecast errors exhibited distinct seasonal variability, with spring and summer forecasts (RMSE: 182.76 and 184.86 W/m<sup>2</sup>) performing significantly better than autumn (223.58 W/m<sup>2</sup>) and winter (194.07 W/m<sup>2</sup>), a pattern consistent with seasonal changes in cloud–aerosol interactions under the East Asian monsoon regime. Three anomalous stations with elevated RMSEs (&gt;300 W/m<sup>2</sup>) were further identified, primarily attributed to localized radiation attenuation over coastal wetlands, urban agglomerations, and mining surfaces. Nevertheless, correlation coefficients between predicted and observed values remained above 0.70, confirming the capability of FY-4A multispectral data to capture short-term radiation dynamics. Overall, this study establishes an innovative short-term forecasting framework suitable for the complex meteorological conditions of China’s eastern coastal regions. The achieved 1-h forecasting accuracy satisfies the ≤20% relative error requirement specified in the Technical Specifications for Power Forecasting Systems of Photovoltaic Power Stations, providing robust technical support for renewable energy integration across the Yangtze River Delta.
KW  - Fengyun-4A (FY-4A); Jiangsu Province; XGBoost machine learning short-term forecasting model; dynamic prediction

DO  - 10.32604/ee.2025.074702