通过多角度云层三维重建优化区域太阳辐照度的时空预测精度

Abstract Accurate real-time regional solar irradiance estimation is crucial for optimizing photovoltaic systems and managing power grids. However, traditional methods suffer from significant limitations in dynamic responsiveness, spatial resolution, and economic feasibility, making them inadequate for high-precision applications under complex weather conditions. To address these challenges, this study proposes a high-resolution real-time irradiance estimation method based on an all-sky imaging network. By deploying ten fisheye all-sky cameras, a regional panoramic cloud map is constructed using a multi-view 3D cloud reconstruction technique. Furthermore, an innovative irradiance separation modeling strategy is introduced, where direct irradiance is computed using a cloud-shadow model, and scattered irradiance is predicted via a spatiotemporal convolutional Transformer. This approach comprehensively accounts for both cloud occlusion and scattering effects, thereby enhancing the accuracy and robustness of irradiance estimation. Experimental results demonstrate that, compared to traditional Kriging interpolation and seven baseline methods, the proposed method consistently achieves the lowest root mean square error (RMSE) and the highest change-point detection rate across four representative cloud transition scenarios: clear with sparse clouds, overcast with showers, morning cloudy, and afternoon cloudy. This highlights its superior dynamic responsiveness and precise tracking of rapid irradiance fluctuations. Additionally, the method significantly enhances spatial resolution, achieving 4.39 m-19.45 m under cloudy conditions, outperforming conventional static approaches. The computational framework supports efficient offline training and real-time prediction, ensuring strong adaptability. With cost-effective hardware, minimal maintenance requirements, and high spatial scalability, this method offers a practical and economically viable solution for high-resolution regional solar irradiance estimation.