Extracting cloud motion vectors from satellite images for solar power forecasting

The high temporal variability of solar power is a real issue to achieve a balanced production and consumption. Solar power forecasting is then necessary to better exploit this variability and to increase the penetration of photovoltaic power into the energy mix. Solar energy forecasting involves prediction of cloud property above a given point. For several hour ahead forecasts, using images from meteorological geostationary satellite is the most suitable solution. We propose a forecasting method based on a phase correlation algorithm for motion estimation between subsequent cloud maps derived from Meteosat-9 images. The method is assessed against state-of-the-art over a limited area over South of France for a 4-hour period. Cloud index maps are predicted. Our forecasting are 21 % better than persistence in relative RMSE of cloud index. If state-of-the-art shows better results (23 %), our algorithm reduces computing of 25 % and then minimize operational solar forecasting constraints.