Support vector regression based prediction of global solar radiation on a horizontal surface

Abstract

In this paper, the support vector regression (SVR) methodology was adopted to estimate the horizontal global solar radiation (HGSR) based upon sunshine hours (n) and maximum possible sunshine hours (N) as input parameters. The capability of two SVRs of radial basis function (rbf) and polynomial basis function (poly) was investigated and compared with the conventional sunshine duration-based empirical models. For this purpose, long-term measured data for a city situated in sunny part of Iran was utilized. Exploration was performed on both daily and monthly mean scales to accomplish a more complete analysis. Through a statistical comparative study, using 6 well-known statistical parameters, the results proved the superiority of developed SVR models over the empirical models. Also, SVR-rbf outperformed the SVR-poly in terms of accuracy. For SVR-rbf model on daily estimation, the mean absolute percentage error, mean absolute bias error, root mean square error, relative root mean square error and coefficient of determination were 10.4466%, 1.2524MJ/m2, 2.0046MJ/m2, 9.0343% and 0.9133, respectively. Also, on monthly mean estimation the values were 1.4078%, 0.2845MJ/m2, 0.45044MJ/m2, 2.2576% and 0.9949, respectively. The achieved results conclusively demonstrated that the SVR-rbf is highly qualified for HGSR estimation using n and N.