An evolutionary-assisted machine learning model for global solar radiation prediction in Minas Gerais region, southeastern Brazil

Solar radiation prediction is necessary for designing photovoltaic systems, assessment of regional climate and crop growth modeling. However, this estimate depends on expensive devices, namely pyranometer and pyranometer. Considering the difficulty of acquiring these devices, predicting such values through mathematical and computational models is a convenient approach where costs can be reduced. In particular, machine learning methods have been successfully and widely applied for this task. However, the choice of the correct machine learning model, its parameters sets, and the variables used influence obtained results. This work presents a methodology that optimizes the aforementioned points to efficiently predict solar radiation in the state of Minas Gerais, Brazil. Currently, no work presents a computational model for the entire state. For this, data from 51 cities in Minas Gerais are used, obtained by the automatic weather stations of the National Institute of Meteorology. Two machine learning models, Artificial Neural Network and Multivariate Adaptive Regression Spline, were optimized through a Simple Genetic Algorithm, and the results compared to those available in the literature. The best results were found at the Guanhães station, with R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document} of 0.867 and RMSE of 1.68 MJ m-2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-2}$$\end{document} day-1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-1}$$\end{document}, and at the Muriaé station, with R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document} of 0.864 and RMSE of 1.64 MJ m-2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-2}$$\end{document} day-1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-1}$$\end{document}. The models had their metrics compared to each other through the methodology of performance profiles, where the Multivariate Adaptive Regression Spline model proved to be more efficient. The results demonstrate that computational models perform better than the empirical models currently used.