A Short-Term Photovoltaic Power Prediction Considering Satellite Cloud Imagery

In order to address the significant impact of cloud cover on photovoltaic (PV) power due to its obscuring effect, and the challenge of accurately mapping the relationship between cloud information and PV power. Traditional PV power prediction methods often overlook the influence of cloud structure and meteorological elements on PV power. A proposed solution is a short-term PV power prediction method that considers satellite cloud imagery. Firstly, standardization and de-biasing processes are applied to visible light cloud images obtained from satellites, eliminating the intraday variability of these images. Subsequently, a gated recurrent neural network is employed to capture the impact features of cloud cover. Finally, by integrating cloud cover features with other influencing factors, a mapping relationship with PV power is established for prediction. Results indicate that the proposed model effectively addresses the intraday variability of cloud images, achieves precise localization of cloud feature regions, and demonstrates good predictive performance. This research provides valuable insights for cloud-based PV power prediction.