Abstract

The solar photovoltaic (PV) sector is expanding rapidly within smart grid ecosystems, yet the inherent variability of solar irradiance and weather conditions makes accurate forecasting of PV power generation a fundamental challenge for efficient energy management systems. This paper presents a systematic review of 57 studies that employ deep learning and related artificial intelligence models for solar power forecasting and its integration into smart energy management, spanning convolutional neural networks (CNNs), recurrent architectures (RNN, LSTM, GRU), attention-based and Transformer models, hybrid architectures, and reinforcement-learning-based control. A structured methodology was followed to identify, screen, and select the literature according to clear inclusion and exclusion criteria; the selected studies were then analyzed in terms of forecasting accuracy, data requirements, computational complexity, and interpretability. The review further discusses how forecasting outputs are integrated within smart energy management systems to support decisions on storage dispatch, demand response, and grid interaction. The paper concludes by outlining key open challenges — including data scarcity, limited geographic generalization, and the need for interpretable models — alongside promising future research directions such as transfer learning, federated learning, physics-informed hybrid models, and reinforcement-learning-based energy management.