Estimation of solar radiation using neighboring stations through hybrid support vector regression boosted by Krill Herd algorithm

Abstract

Solar radiation is a basic input in many fields of studies and models. However, the low density of solar network stations; the improper distribution of these stations; high cost of purchasing, maintaining, and calibrating solar radiation measurement instruments; and frequent errors in the available data are the most important deficiencies in this regard. Thus, researchers are seeking for new and practical methods to estimate solar radiation accurately. The present study aimed to estimate the solar radiation values based on a new hybrid support vector regression model. To this aim, the solar radiation values of all eight target synoptic stations during 1974–2014 were estimated by using Krill-Herd hybrid algorithm (SVR-KHA) method based on support vector regression and implementing neighboring station data. Results indicated that the testing performance of SVR-KHA has a more precision and lower error for all target stations, compared with classical SVR. In addition, the best results were obtained for SVR-KHA3 hybrid model (Isfahan station). Further, the RMSE, MAPE, and R2 values for this model were 1.98 MJ/m2/day, 7.4%, and 0.93, respectively. In accordance with the results, Krill-Herd algorithm method coupled with support vector regression had a high performance and capability for solar radiation estimation in Iran. In other words, the hybrid SVR-KHA model is more flexible and has less error in modeling the nonlinear and complex systems. Finally, the new method of using neighboring stations can be regarded as an appropriate method for estimating nonlinear phenomenon such as solar radiation.