Machine Learning-Based Photovoltaic Power Prediction from Ultra-Short-Term to Long-Term Forecasting: A Review

Photovoltaic (PV) power generation plays a crucial role in addressing energy shortages and advancing the green economy. Accurate PV power forecasting contributes significantly to grid stability and facilitates the achievement of sustainable development goals. This paper provides a comprehensive review on PV power prediction methods from multiple perspectives, including types of prediction results, methodological processes, data sources, and temporal/spatial scales. It examines the growing adoption of machine learning techniques in this field, driven by advances in computational hardware and software frameworks. The review also highlights the geographical concentration of PV prediction research and emphasizes the importance of time-scale classification for practical applications in power markets. Accordingly, the study summarizes research findings across ultra-short-term, short-term, and medium-to-long-term prediction horizons, classifying methods into general and specialized categories (e.g., image data-driven, regional, online, interval, and indirect prediction). Furthermore, commonly used evaluation metrics for PV power forecasting are summarized. Finally, the paper concludes with a summary of key insights and suggests potential research gaps for improving the accuracy of PV power prediction through machine learning.