A Novel Non-Linear Model Based on Bootstrapped Aggregated Support Vector Machine for the Prediction of Hourly Global Solar Radiation

Abstract

The prediction of global solar radiation (GSR) for various regions is of great importance as it provides guidance for the design, modeling, and operation of solar energy conversion systems, the selection of suitable regions, and even informs future investment policies for decision-makers. This paper presents the methods for predicting hourly mean solar radiation using support vector machines (SVM), which is a machine learning algorithm based on statistical learning theory. The primary objective of this paper was to investigate the use of a support vector machine (SVM) based on quantitative structure–activity relationships, specifically a single support vector machine (SSVM) and a bootstrap aggregated support vector machine (BASVM), to predict hourly global solar radiation in Bouzareah city. A dataset consisting of 3603 data points was employed to develop both the SSVM and BASVM models. Bootstrap aggregation of SVM is utilized to enhance the accuracy and robustness of SVM models constructed from limited training datasets. The training dataset is resampled using bootstrap resampling with replacement to create an ensemble of SVM models, each trained on a different sample from the training set. A support vector machine model is developed, and individual Support Vector Machines (SVMs) are then combined to form a Bootstrap Aggregated Support Vector Machine (BASVM). Experimental data for global solar radiation (GSR) were compared to the calculated GSR, and excellent correlation coefficients (R) were found (0.9913) during the testing phase. This novel BASVM model could be utilized by researchers and scientists to design high-efficiency solar devices.