Artificial intelligence applications for optimization, control, and decision-making in photovoltaic systems

Artificial intelligence (AI) is increasingly applied to photovoltaic (PV) systems for power forecasting, maximum power point tracking, converter and storage coordination, fault diagnosis, and operational decision-making. However, reported gains are difficult to compare because studies differ in datasets, baselines, climates, system scales, metrics, and validation environments. This systematic review evaluates not only whether AI methods perform well, but also where, against which comparators, under what conditions, and with what deployment burden. Searches conducted from January to July 2026 across eight scholarly databases identified 1284 records; 1012 unique records were screened, 286 full texts assessed, and 71 studies included. Evidence was coded by task, AI family, application level, data source, metric, validation fidelity, computational reporting, and limitation. No AI family was universally superior. Machine-learning and fuzzy methods suit resource-constrained settings; deep learning performs strongly with high-dimensional data but compute-intensive; evolutionary methods support offline optimization; and reinforcement learning is promising for sequential control but needs stronger safety and field validation. Key gaps include external validation, explainability, computational reporting, cybersecurity, and long-term field evidence.