Machine learning-based improvement of empiric models for an accurate estimating process of global solar radiation

Abstract

The change of the solar radiation reaching the earth depending on specific conditions brings the execution of system planning meticulously and optimally by solar power researchers to the fore. For the estimation of the solar radiation, the most frequently used model is the Angtrom-Prescott model. In this model, sunshine ratio plays an important role. In the study, it is attempted to enhance the annual and semi-annual models developed for the city of Mugla, Turkey and to congregate the semi-annual models in a single model by using the Artificial Bee Colony (ABC) algorithm. The results obtained have revealed that in the multiple model relying on only the sunshine duration, the statistical error values were not reduced to very low levels. In order to cope with this problem, the multiple model relying on both the sunshine duration and the sunset-sunrise hour angle has been proposed. In this way, the statistical errors are found to be reduced by about 40% using the ABC algorithm and the multiple model. It was seen that the models recommended are superior to all the models especially in summer and spring months when there is plenty of sunshine.