Solar Irradiance Estimation Using Deep Artificial Neural Network: A Case Study of Yola, Nigeria

Abstract

Accurate estimation of solar irradiance is essential for the design, optimization, and operation of solar energy systems, particularly in regions with high renewable energy potential such as Yola, Nigeria. Traditional empirical and regression-based models often struggle with nonlinear relationships between meteorological parameters and solar irradiance, leading to reduced accuracy. To address this limitation, this study developed a Deep Artificial Neural Network (DNN) model to estimate daily global solar irradiance using ten years (2015–2025) of meteorological data, including sunshine duration, temperature, humidity, wind speed, and cloud cover. The data were preprocessed, normalized and used to train a multi-layered DNN in MATLAB R2024b. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Bias Error (MBE), and Mean Percentage Error (MPE). Simulation results revealed that the DNN achieved high predictive accuracy, with RMSE of 15 W/m², MBE of 2 W/m², and MPE of 5.0%. For comparison, conventional Angstrom–Prescott empirical model applied to the same dataset gave RMSE values ranging from 28 W/m², confirming the superiority of the DNN. The DNN model also demonstrated strong correlation with measured solar irradiance (R² = 0.984), highlighting its robustness in capturing nonlinear meteorological interactions. The proposed DNN-based approach provides a more reliable tool for solar irradiance estimation in Yola and could have potential applicability to other regions in sub-Saharan Africa.