Photovoltaic Power Forecasting: A Review on Models and Future Research Directions

ABSTRACT The stability of power grid systems can be significantly affected by the unpredictability and volatility of power generation; however, accurate forecasting of solar energy power can help reduce this impact. This benefits the system through lower operating costs, balanced operation, and optimal dispatch. Over the past decade, extensive research has been published on this topic, exploring physical models, artificial intelligence (AI) techniques, and numerical and probabilistic approaches. Additionally, previous review studies centred their review discussions on a specific event horizon, others focused exclusively on the geographical horizon, and assessed only particular classes of photovoltaic (PV) output power forecasts. They paid little or no attention to other classes. Therefore, a thorough analysis of solar PV output power forecasting methods is required. In this paper, special focus is given to deep learning (DL), machine learning (ML), and hybrid methods, as these AI areas are gaining popularity. This study aims to provide a comprehensive and critical review of the latest AI applications. It also features a statistical analysis of forecasting errors based on over a hundred solar generation forecast studies. Additionally, the paper offers a brief introduction to the metrics used in ML, DL, and hybrid methods and their interpretation. A discussion of factors influencing forecasting errors is included. Future models will be more accurate because of the clarification that has been provided.