The integration of photovoltaic (PV) systems into modern power grids requires accurate intra-hour forecasting of solar irradiance to address intermittency and ensure grid stability. This review systematically examines recent advances in deep learning-based ground-sky imagery methods for intra-hour solar forecasting (DL-GSI-IHSF), with a particular emphasis on the end-to-end forecasting workflow. The literature is organised using a unified Data–Model–Analysis pipeline, which spans ground-based sky image acquisition, deep learning model architectures, and forecast evaluation considerations. Ground-based whole-sky imagers, in combination with neural architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Vision Transformers (ViTs), are reviewed in the context of their ability to capture fine-scale spatio-temporal cloud dynamics, offering enhanced temporal and spatial resolution compared to satellite imagery and numerical weather prediction models. The review further categorises existing studies by data modalities, architectural design, and forecast mechanisms, highlighting emerging trends such as multimodal data fusion and hybrid modelling strategies. Key evaluation aspects, including forecast skill, ramp event characterisation, and probabilistic outputs, are discussed alongside unresolved challenges such as data imbalance, model interpretability, and domain generalisation. By structuring current research through a workflow-oriented framework, this review provides both a theoretical foundation and practical perspective for advancing high-resolution, real-time solar forecasting.
Zhang et al. (Fri,) studied this question.