Solar irradiance forecasting using multi-layer cloud tracking and numerical weather prediction

The advances in photovoltaic technology make solar energy one of the top three renewable energy sources. However, predicting the variability of solar penetration caused by cloud cover is the biggest hurdle for the effective use of solar energy. Grid operators enforce regulations that require ramp events to be within a certain range, which makes short term forecasting essential. The Total Sky Imager (TSI) is one of the best instruments for accurate shortterm irradiance forecasting but is limited to a forecast of approximately five minutes for low altitude clouds, which usually cause large ground irradiance fluctuations. To extend the forecasting horizon to 15 minutes, we propose to incorporate NWP (Numerical Weather Prediction) based weather categories (every 15 minutes) into a short-term irradiance forecasting model. This advanced Support Vector Regression (SVR) is the product of our novel multilayer cloud image processing pipeline, which can handle complex cloud scenarios. We observe an average of 21% improvement over the baseline model in our systematic validations for 1-15 minute forecasts.