Systematic review of nowcasting approaches for solar energy production based upon ground-based cloud imaging

Abstract

Nowcasting of solar energy considering clouds is important for photovoltaic solar plants and distributed systems. Clouds present a challenge for modeling, due to constant changes in shape and size, and are dependent on local atmospheric conditions. Several methods are being used for the automatic assessment of clouds from the surface to predict solar power generation, assisted by camera, side sensors, etc. During our research we did not find a Systematic Literature Review on this topic. This review is intended to search the related scientific articles to find the state of the art in the area from the period of 2011–2020. We found 65 articles to review after the meta-analysis. We look for the main short-term forecasting methods used. The majority of articles rely on classical statistics approaches based on historical data. Yet recent articles show that this trend might be shifting towards Machine Learning approaches. Our analysis shows that most articles found are based on images captured by fish-eye lenses using a single camera. The most common forecasting techniques are Artificial Neural Networks and Convolutional Neural Networks, with the root mean squared error being the most predominant error metric used for model validation among both classical and Machine Learning approaches.