Solar forecasting model combines weather data, satellite images, and time series — first to fuse all three at once

Researchers built a system that pulls cloud cover from satellite images, weather descriptions from text, and power readings from actual solar sites to predict PV output hours ahead. This matters because solar grids swing wildly with cloud motion, and forecasting accuracy directly affects whether a grid operator can balance supply and demand without buying expensive backup power.

Solar plants are unpredictable — clouds change output faster than grid operators can respond, which costs money and wastes generation. Every percentage point of forecasting accuracy saves dispatch costs and makes solar more reliable for grid operators. Right now most forecasting systems treat satellite images, weather text, and time series as separate problems and stitch them together. This model processes all three at once, which lets it learn patterns that separate systems miss — like how particular cloud formations affect specific sites differently.