Solar Radiation Nowcasting Using a Markov Chain Multi-Model Approach

Abstract

Solar energy has found increasing applications in recent years, and the demand will continue to grow as society redirects to a more renewable development path. However, the required high-frequency solar irradiance data are not yet readily available everywhere. There have been endeavors to improve its forecasting in order to facilitate grid integration, such as with photovoltaic power planning. The objective of this study is to develop a hybrid approach to improve the accuracy of solar nowcasting with a lead time of up to one hour. The proposed method utilizes irradiance data from the Copernicus Atmospheric Monitoring Service for four European cities with various cloud conditions. The approach effectively improves the prediction accuracy in all four cities. In the prediction of global horizontal irradiance for Berlin, the reduction in the mean daily error amounts to 2.5 Wh m−2 over the period of a month, and the relative monthly improvement reaches nearly 5% compared with the traditional persistence method. Accuracy improvements can also be observed in the other three cities. Furthermore, since the required model inputs of the proposed approach are solar radiation data, which can be conveniently obtained from CAMS, this approach possesses the potential for upscaling at a regional level in response to the needs of the pan-EU energy transition.