Development of a solar radiation prediction model using Fine Gaussian SVM in machine learning

Abstract

The global transition toward renewable and clean energy sources is imperative for sustainable
development and for reducing the adverse environmental effects associated with fossil fuel use. Among
renewable options, solar energy is one of the most promising due to its abundance, scalability, and
minimal environmental impact. Effective utilization of solar energy in each region requires a
comprehensive understanding of solar radiation components, particularly diffuse radiation, which
significantly influences the performance of solar energy systems. Moreover, accurate estimation of solar
radiation is essential for climate change research, a critical area of global scientific inquiry. Currently,
satellite-based solar radiation forecasting systems offer high spatial and temporal resolution data,
including global, direct, and diffuse radiation components, making them valuable tools for the planning
and optimization of solar power systems. In this study, satellite-based forecasting models were utilized to
estimate diffuse solar radiation for the selected region. To enhance prediction accuracy, a Fine Gaussian
Support Vector Machine (SVM) algorithm was employed to model solar irradiance based on key
meteorological and geographical inputs. Additionally, the Ångström coefficients were determined using
MATLAB to support the development of the predictive model. The performance of the proposed model
was evaluated using various statistical error metrics, including root mean square error (RMSE), mean
absolute error (MAE), and coefficient of determination (R²). The results indicate that the Fine Gaussian
SVM model provides highly accurate predictions of diffuse solar radiation, demonstrating its
effectiveness for solar resource assessment and planning.