Hybrid CNN-LSTM Model for Solar Radiation Nowcasting Using All-Sky Camera Images and Meteorological Measurements

Accurate and reliable very short-term solar radiation forecasting is critical for the efficient operation and management of solar energy systems. In this study, we propose an innovative hybrid model combining Convolutional Neural Networks and Long Short-Term Memory networks designed to leverage spatiotemporal data for precise solar radiation predictions. The model integrates a CNN to extract spatial features from sequential infrared all-sky images and an LSTM to capture temporal dependencies in both image sequences and meteorological data. By combining these complementary approaches, the model effectively learns complex patterns in solar radiation dynamics. The architecture dynamically adapts to varying input dimensions, making it robust for diverse datasets. Experimental findings indicate that the proposed model attains a forecast skill of 10.7 % compared to the persistence model, with lower RMSE (86.38 W/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> vs. 96.74 W/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) and MAE (41.51 W/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> vs. 44.30 W/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), demonstrating improved reliability in capturing solar radiation variability over a 5-minute horizon. This research emphasizes the capabilities of hybrid deep learning approaches in advancing renewable energy forecasting and provides a solid basis for advancing future studies in spatiotemporal prediction tasks.