In a variety of scientific and technical disciplines, machine learning has proved to be a powerful tool for solving complex, nonlinear problems. Its application to the modeling of global solar radiation (GSR) is a particularly promising approach to overcoming the limitations of traditional statistical and physical approaches. Recognized as the proportion of solar energy that reaches the Earth’s surface per unit area, GSR involves complex atmospheric and geophysical features, making its prediction a highly dynamic and nonlinear challenge. In this work, we explore the implementation of artificial neural networks (ANNs) as an enhanced machine learning approach for GSR prediction, with a particular focus on the mathematical models, learning algorithms, and optimization techniques that define their structure. An overview is given of the mathematical foundations of RNA, involving the use of backpropagation and stochastic gradient descent for parameter optimization. Additionally, the relevance of ANNs for capturing temporal, spatial, and nonlinear interdependencies between meteorological variables and GSR is explored, particularly in the context of the Rabat region. Based on the results, a comparative analysis of two different ANN architectures is reported, highlighting their ability to improve model accuracy and reliability over traditional methods.