On 1st October 2026, Dr. Diaa Salman and Dr. Imad Alzeer published a new study in Neural Computing and Applications titled “A systematic evaluation of classical, transformer-based, and foundation models for photovoltaic power time-series forecasting.” The article aimed to systematically benchmark classical deep learning, transformer-based, foundation, and hybrid feature-fusion architectures for short-term photovoltaic power forecasting under identical experimental conditions, and to evaluate a proposed CNN–Foundation–Transformer framework. The results of this study demonstrated clear performance differences between architectural families, with the proposed CNN–Foundation–Transformer architecture achieving the highest overall forecasting performance (MAE = 0.071, RMSE = 0.106, nRMSE = 12.1%, and skill score = 0.64 compared with persistence forecasting). The findings indicate that combining convolutional feature extraction, foundation-level temporal encoding, and multi-head attention can provide a powerful and scalable architecture for high-accuracy PV power forecasting. For more information about the study, please contact the main author at This email address is being protected from spambots. You need JavaScript enabled to view it. or the Scientific Research Office at This email address is being protected from spambots. You need JavaScript enabled to view it. .
