SARIMA-SVM hybrid model for the prediction of daily global solar radiation time series

The measured solar radiation at the ground level is behaving like a very random variable, because of various disturbances encountered throughout its trajectory through the atmosphere in particular; winds, cloudiness, humidity and temperature variations... etc. As the accumulated daily global solar energy is an indispensable parameter for designing of solar systems (photovoltaic and thermal) and its prior knowledge (prediction) is required to control and manage all these solar systems, so we thought improve the predictive model of daily global solar radiation (DGSR) time series. A hybrid model based on the methodology of Box & Jenkins (SARIMA; Seasonal Auto-Regressive Integrated Moving Average) and SVM (Support Vector Machine) is elaborated to predict the clearness index K
_t
(ratio of daily global solar radiation on the extraterrestrial solar radiation). Then, by a simple calculation, the DGSR is deducted from the K
_t
. The results showed that the developed hybrid model SARIMA_SVM improves slightly the prediction of DGSR compared to the conventional SARIMA model.