AI-Powered Global Solar Radiation Forecasting: Fusing Machine Learning with Satellite Vision

Authors- K. Harika, Telugunta Sahana, Surampudi Yamini Satya Sai Priya, Jafari Kulsum, Thadisetti Siva Venkata Sai Kumar, Shaik Sameeruddin

Abstract-Accurately predicting Daily Global Solar Radiation (DGSR) is crucial for applications in renewable energy, agriculture, and climate research. This study explores the potential of Machine Learning (ML) algorithms combined with satellite imagery to enhance DGSR forecasting. Traditional ML-based approaches rely on meteorological parameters such as temperature, wind speed, atmospheric pressure, and sunshine duration, along with radiometric factors like aerosol optical thickness and water vapor. In this work, we investigate the integration of normalized reflectance from satellite images across multiple spectral channels to refine solar radiation predictions. Two ML models, Artificial Neural Networks (ANN) and Support Vector Machines (SVM), were employed as regression techniques. The findings highlight that the selection and quantity of input parameters significantly impact the accuracy of DGSR predictions. Furthermore, ANN demonstrated superior performance over SVM, achieving an RMSE of 212.21 W h/m², NRMSE of 3.46%, MAPE of 2.85%, MBE of -7.26 W h/m², and an R-value of 0.99. In contrast, the SVM model produced an RMSE of 441.95 W h/m², NRMSE of 6.6%, MAPE of 5.62%, MBE of 69.46 W h/m², and an R-value of 0.98. These results underscore the effectiveness of leveraging satellite imagery alongside ML techniques to improve the accuracy of DGSR forecasting.

DOI: /10.61463/ijset.vol.13.issue2.225