DTU Risø All-Sky Imager Testbed: One-Week Sky Image and Calibration Dataset

This research focuses on developing and testing a ground-based all-sky imager (ASI) network at the DTU Risø Campus to improve very short-term solar irradiance forecasting. Clouds are the primary source of intra-hour variability in solar power, and their rapid movement across the sky presents challenges for predicting Global Horizontal Irradiance (GHI). Existing methods such as satellite-based forecasts provide regional coverage but lack the temporal and spatial resolution needed for second-to-minute scale predictions at PV plant level.To address this gap, three high-resolution ASIs were installed at different locations on the campus, separated by 400–1200 meters. Each camera was carefully calibrated using a fisheye lens model, producing lookup tables that map every image pixel to its corresponding solar azimuth and zenith angles. These calibrations enable geometric correction of the raw images and equidistant projections of the sky dome.The system has been tested in a spatial casting configuration (no forecasting horizon), where irradiance maps generated by one camera are compared to pyranometer measurements at other locations across the campus, including a solar meteorological station. Results from a full day of 30-second resolution data show that the ASI-based method reproduces the timing and magnitude of irradiance ramps caused by passing clouds, though discrepancies remain in cases of fast cloud movement or complex multi-layer cloud fields.The research provides a testbed for validating ASI-based solar forecasting methods, where controlled experiments can quantify the effect of camera distance to PV plants, cloud fraction, and seasonal variation on forecast accuracy. This contributes to the broader goal of enabling cost-effective, high-frequency solar forecasting solutions that can be scaled for grid integration, PV plant operation, and energy market participation.

本研究聚焦于在丹麦技术大学（DTU）里斯校区搭建并测试地基全天空成像仪（All-Sky Imager, ASI）组网系统，以提升极短期太阳辐照度预测精度。云是造成光伏发电功率小时内波动的主要来源，其在天空中的快速移动给全球水平辐照度（Global Horizontal Irradiance, GHI）的预测带来了严峻挑战。现有基于卫星的预测方法虽可实现区域覆盖，但无法满足光伏电站秒至分钟级预测所需的时空分辨率。为填补这一研究空白，研究团队在校区内不同点位部署了三台高分辨率全天空成像仪，各点位间距为400至1200米。每台相机均通过鱼眼镜头模型完成精密校准，生成了将每个图像像素映射至对应太阳方位角与天顶角的查找表。该校准流程可实现原始图像的几何校正与天球穹顶的等距投影。该系统已通过空间投射配置（无预测提前量）完成测试：将单台相机生成的辐照度地图，与校区内其他点位（包括一座太阳气象站）的总辐射表测量结果进行对比。基于单日30秒分辨率数据的测试结果表明，基于全天空成像仪的方法可复现过境云层引发的辐照度斜坡的时序与幅值，但在云层快速移动或复杂多层云系场景下仍存在偏差。本研究搭建了可用于验证基于全天空成像仪的太阳辐照度预测方法的试验平台，通过控制变量实验可量化相机与光伏电站的间距、云量占比以及季节变化对预测精度的影响。该研究有助于实现具备成本效益的高频太阳辐照度预测解决方案，为该方案应用于电网并网、光伏电站运维以及能源市场参与的规模化推广提供支撑。