A Critical Review and Strategic Roadmap of PV Power Forecasting (2016–2026): Addressing Temporal Leakage and Operational Integration Gaps

Photovoltaic (PV) power forecasting plays a central role in power system operation, electricity markets, and the integration of high shares of renewable energy. Over the past decade, forecasting approaches have evolved from classical statistical time-series models to advanced machine learning and deep learning architectures. This review analyses 119 studies published between 2016 and 2026, providing a structured assessment of PV forecasting methodologies, including model types, data requirements, validation strategies, and performance evaluation practices. Beyond summarizing existing approaches, the paper identifies three major methodological gaps in the literature: (i) fragmentation of evaluation metrics, which limits cross-study comparability; (ii) insufficient reporting of data preprocessing procedures and temporal leakage prevention; and (iii) limited integration of forecasting accuracy with economic and operational performance metrics. A systematic comparison of representative studies is conducted to highlight dominant modelling trends and persistent limitations. Beyond a descriptive summary, this review highlights significant limitations in methodological reporting across the 119 studies analysed, particularly regarding temporal leakage prevention in Deep Learning-based forecasting. To address these issues, we introduce a reproducibility checklist and propose a strategic roadmap aimed at strengthening the link between statistical accuracy (e.g., RMSE/MAE) and operational relevance in electricity markets.