Claudio Maffi and Jens Engel, "Evaluation and comparison of PV prediction quality of a sky imager", HRI-EU, 2023.

On the premises of Honda R&D Germany in Offenbach, Main, a large photovoltaic (PV) system with a peak power of 750kW is installed. The PV system is connected to an energy management system (EMS), which manages how electric energy is distributed and stored in the building. In order to use efficiently the green energy provided by the PV system into the EMS, accurate predictions about the accessibile PV power are requested. Currently a meteorological model-based provides a prediction for the PV power twice a day with a prediction horizon of 14 days. While these predictions have high accuracy in terms of overall available energy, due to the nature of how these predictions are derived, the accuracy of the short-term forecasts is comparatively low. To solve this problem, we evaluated different prediction sources, comparing the performances of two additional forecasting model. The first alternative prediction model that we analyzed uses a sky imager to derive the short-term PV power predictions. The sky imager takes a picture of the sky over the building every minute, and provides us a prediciton horizon of 30 minutes every minute. The second alternative is a satellite and meteorological model-based, that provides us a prediciton horizon of 3 days every 15 minutes. Using suitable metrics, the sky imager model showed better results in terms of PV power and global horizontal irradiance short-term accuracy, resulting an important benefit for EMS.