A comparative evaluation for identifying the suitability of extreme learning machine to predict horizontal global solar radiation

Abstract

In this paper, the extreme learning machine (ELM) is employed to predict horizontal global solar radiation (HGSR). For this purpose, the capability of developed ELM method is appraised statistically for prediction of monthly mean daily HGSR using three different types of input parameters: (1) sunshine duration-based (SDB), (2) difference temperature-based (TB) and (3) multiple parameters-based (MPB). The long-term measured data sets collected for city of Shiraz situated in the Fars province of Iran have been utilized as a case study. The predicted HGSR via ELM is compared with those of support vector machine (SVM), genetic programming (GP) and artificial neural network (ANN) to ensure the precision of ELM. It is found that higher accuracy can be obtained by multiple parameters-based estimation of HGSR using all techniques. The computational results prove that ELM is highly accurate and reliable and shows higher performance than SVM, GP and ANN. For multiple parameters-based ELM model, the mean absolute percentage error, mean absolute bias error, root mean square error, relative root mean square error and coefficient of determination are obtained as 2.2518%, 0.4343MJ/m2, 0.5882MJ/m2, 2.9757% and 0.9865, respectively. By conducting a further verification, it is found that the ELM method also offers high superiority over four empirical models established for this study and an intelligent model from the literature. In the final analysis, a proper sensitivity analysis is performed to identify the influence of considered input elements on HGSR prediction in which the results reveal the significance of appropriate selection of input parameters to boost the accuracy of HGSR prediction by the ELM algorithm. In a nutshell, the comparative results clearly specify that ELM technique can provide reliable predictions with further precision compared to the existing techniques.