Evaluation of Different Models for Global Solar Radiation Components Assessment

This work presents a comparative study between different machine learning techniques for solar radiation components assessment. The developed models are evaluated using a valid measured data of daily global solar radiation recorded during the period from May 2013 to December 2015 in a semi-arid region, south of Algeria. To this end, the collected data were divided into learning and testing sets to validate the models’ performance. Therefore, the main objective of this study is to assess the performance of different techniques and their hybridization with a linear model for direct and diffuse solar radiation estimation using only global solar radiation components as measured input data. The results of different metrics of the evaluation show that the gaussian process regression (GPR) and least-squares support vector machine (LS–SVM) models provide a high performance compared to other studied models. We have obtained a root mean square error of 0.074–0.075 and 0.093–0.093 for GPR and LS–SVM models respectively, for both direct and diffuse solar radiation estimation.