Development of machine learning models for the estimation of diffuse solar radiation in the humid-subtropical climatic region of India

• DOI: 10.1007/s44292-025-00030-0
• Publication Date: 2025-04-08T21:36:00Z
• Authors: Farooque Azam; Naiem Akhtar; Shahid Husain
• Keywords: Humid subtropical climate

Abstract In this research paper, the investigation focused on developing machine learning models and comparing them with the best empirical models for estimating monthly average diffuse solar radiation. This document shares findings and information gathered from a 3-year study from August 2020 to July 2023, where measurements were taken and analyzed. Two pyranometers, one with a shading ring, were used to measure global and diffuse solar radiation. The study found that the mean values of global, beam, and diffuse solar radiation were 22.39 MJ/m2 day, 14.51 MJ/m2 day, and 7.80 MJ/m2 day, respectively. The average values for the sky-clearness index, diffuse fraction, and diffusion coefficient were also determined as 0.71, 0.36, and 0.25, respectively. To assess the suitability of different models for estimating diffuse solar radiation using only global solar radiation as input, six machine learning models, namely KNN, SVM, RF, GPR, MLP, and XGBoost, and four best empirical models for the region have been evaluated. Various well-established statistical indicators were utilized for a comprehensive evaluation of the models. These statistical metrics were then converted into scaled values to calculate each model’s Global Performance Indicator (GPI). XGBoost model outperformed the others, achieving a GPI value of 6.073.