A neural network-based model for addressing global solar radiation data scarcity in tropical and equatorial regions
Applied Journal of Physical Science · 30.04.2026 · 7 (2) · 34-48
DOI: 10.31248/ajps2026.138 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Journal of Physical Science, online erschienen am 30.04.2026. Verfasst von J. E. Olubusade und 5 weiteren Autor:innen.
Abstract (Originalsprache)
Energy is as vital as life itself. Among various energy forms, solar is chief, and its potential to sustain life and other activities on Earth is significant. To achieve optimal efficiency in the design, sizing, calibration, manufacture, or deployment of any solar application, it is crucial to accurately measure the Global Solar Radiation (GSR) at the surface. GSR can be quantified through direct and indirect methods. Data obtained via the direct method is considered more accurate but scarce. Conversely, data acquired through indirect methods is accessible for any location of interest, but its precision is often doubtful. This study aims to develop a GSR model to address data scarcity, mitigate instances of erroneous estimations, and the intricacy of existing models that attempt to predict GSR. A multi-layer neural network, consisting of 49 neurons in the hidden layer, was selected from multiple training trials from which a new Soft Computing Model (SCM) emerged. The network was trained, tested, and validated using a 25-year dataset on monthly averages, comprising solar flux, relative humidity, and temperature change as the input nodes. A regression coefficient of 0.9832 was obtained during the neural network training phase, indicating a strong agreement between predicted and measured values. To assess the predictive performance of the trained network, a new dataset was introduced for testing, yielding a regression coefficient of 0.9737, a mean squared error of 0.0015 and a mean absolute error of 0.0210 when compared with measured data, representing the best performance among all training iterations. Comparative evaluation with existing models and deployment across different locations confirmed that the SCM consistently performed well. The deployment results further confirmed that the model is well-optimised for tropical and equatorial climates, while recalibration would be required to ensure reliable performance in temperate and high-latitude regions.
Zitieren
APA
Olubusade, J. E., Oyedum, O. D., Aibinu, M. A., Ezenwora, J. A., Eichie, J. O., & Folorunso, A. (2026). A neural network-based model for addressing global solar radiation data scarcity in tropical and equatorial regions. Applied Journal of Physical Science, 7(2), 34-48. https://doi.org/10.31248/ajps2026.138
Vancouver
Olubusade JE, Oyedum OD, Aibinu MA, Ezenwora JA, Eichie JO, Folorunso A. A neural network-based model for addressing global solar radiation data scarcity in tropical and equatorial regions. Applied Journal of Physical Science. 2026;7(2):34-48. doi:10.31248/ajps2026.138
BibTeX
@article{Olubusade2026,
author = {Olubusade, J. E. and Oyedum, O. D. and Aibinu, M. A. and Ezenwora, J. A. and Eichie, J. O. and Folorunso, . A.},
title = {A neural network-based model for addressing global solar radiation data scarcity in tropical and equatorial regions},
journal = {Applied Journal of Physical Science},
year = {2026},
volume = {7},
number = {2},
pages = {34--48},
doi = {10.31248/ajps2026.138}
}
EUVIMED