Accurate solar energy forecasting has become increasingly critical with the growing integration of solar power into modern energy systems. In response, artificial intelligence (AI)–based techniques have been widely employed to address the nonlinear, nonstationary, and weather-driven characteristics of solar generation. This scoping review systematically maps the literature on AI-based solar energy forecasting with emphasis on modeling approaches, data practices, and operational implications. A comprehensive search was conducted across Scopus, Web of Science, IEEE Xplore, ScienceDirect, and Google Scholar to identify eligible studies and extract information on model categories, input data characteristics, forecasting horizons, and application contexts. The synthesized evidence reveals a clear temporal evolution from classical machine learning toward deep learning, hybrid ensembles, and transformer-based architectures, with solar irradiance forecasting and short-term to day-ahead horizons dominating the research landscape. Ensemble and hybrid models consistently demonstrated improved robustness under variable atmospheric conditions, whereas transformer-based approaches showed potential for multi-step prediction tasks. However, most studies relied on site-specific datasets with limited cross-regional validation, highlighting persistent challenges in model transferability. Importantly, forecasting accuracy was closely linked to operational objectives, including grid stability, reserve management, and market participation, yet deployment-related considerations such as uncertainty quantification, retraining strategies, and explainability were rarely evaluated. This review identifies a key evolutionary transition in solar forecasting research, characterized by the progression from classical machine learning approaches toward deep learning, hybrid, and transformer-based architectures.
Jadapalli et al. (2026) studied this question.