Solar Radiation Forecast Using Artificial ...

The fast increase in importance of the solar energy resource as viable and promising source of renewable energy has boosted research in methods to evaluate the short-term forecasts of the solar energy resource. There is an increase on demand from the energy sector for accurate short-term forecasts of solar energy resources in order to support the planning and management of the electricity generation and distribution systems. The Eta model is the mesoscale model running at CPTEC/INPE for weather forecasts and climate studies. It provides outputs for solar radiation flux at the surface, but these solar radiation forecasts are greatly overestimated. In order to achieve more reliable information, Artificial Neural Networks (ANN) were used to refine short-term forecast for the downward solar radiation flux at the surfaceprovided by Eta/CPTEC model. Ground measurements of downward solar radiation flux acquired in two SONDA sites located in Southern region of Brazil (Florianópolis and S?o Martinho da Serra) were used for ANN training and validation. The short-term forecasts produced by ANN have presented higher correlation coefficients and lower deviations. The ANN removed the bias observed in solar radiation forecasts provided by Eta/CPTEC model. The skill improvement in RMSE was higher than 30%when ANN was used to provide short-term forecasts of solar radiation at the surface in both measurement sites.