Daily Direct Normal Irradiance Forecasting by Support Vector Regression Case Study: in Ghardaia-Algeria

Abstract

Concentrated solar thermal plants (CST) generate electricity from the direct normal irradiance (DNI) component of solar irradiance. Accuracy forecasting of DNI can reduce the uncertainty of solar power plant output caused by solar irradiance intermittency, in the objective to increase CST plant profitability. In this study, the support vector regression (SVR) methodology was adopted to forecast the DNI based upon some meteorological and radiometric data such as, measured mean daily values of Temperature (T), Humidity (H), Global Horizontal Irradiation (GHI), sunshine duration (SS) and the calculated Fractal Dimension (FD) which is tested for the first time here. The capability of the SVRs-Radial Basis Function (RBF) constructed with different combinations of the parameters mentioned above are investigated. For this purpose, long-term measured data (one year) for the city of Ghardaia situated in sunny part of Algeria was utilized. The sunshine hours (SS) have been widely endorsed as the most effective parameters in forecasting of the DNI in the horizon of 122 days ahead by an error NRMSE =14.7% and R2=0.87. A slight improvement in the accuracy is performed using other parameters as inputs to get NRMSE =12.41% and R2=0.90.