Ensemble Deep Learning for Global Solar Irradiance Forecasting in Egypt: A Multi-Model Approach

doi:10.1007/s41748-026-01083-w

Ensemble Deep Learning for Global Solar Irradiance Forecasting in Egypt: A Multi-Model Approach

2026-03-03

0

PRE

AI

DOI：10.1007/s41748-026-01083-w

Original

Original request for help

Save

Abstract

En 中文

Accurate hourly forecasting of global horizontal irradiance (GHI) is essential for the reliable integration of solar energy into Egypt’s expanding renewable-power sector. This study develops a multi-model deep learning forecasting framework using ERA5 reanalysis data from 56 spatially distributed locations across Egypt for the period 2010–2022. Nineteen meteorological and temporal predictors were preprocessed through standardized scaling, Pearson correlation-based filtering, and ablation-guided feature selection, resulting in a refined set of ten influential variables. Three recurrent deep learning architectures, including the recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU), were trained and evaluated using a unified pipeline and a multi-metric performance assessment. Model accuracy was quantified using the coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), normalized RMSE (nRMSE), Nash-Sutcliffe efficiency (NSE), and Kling-Gupta efficiency (KGE). Among the evaluated models, the GRU demonstratedthe highest predictive accuracy and stability, achieving R2 = 0.9988, RMSE = 10.43 W/m2, MAE = 2.77 W/m2, nRMSE = 2.09%, NSE = 0.9988, and KGE = 0.9987, on the independent test set. Error distribution analyses and dataset-specific evaluations confirm the robustness and generalization capability of the GRU across Egypt’s diverse climatic zones. The proposed framework provides a scalable and computationally efficient solution for national-scale hourly solar irradiance forecasting, supporting photovoltaic planning, grid management, and long-term energy system development in Egypt. This graphical abstract illustrates the complete workflow of a multi-model deep learning framework for solar irradiance forecasting over Egypt. The process begins with ERA5-based meteorological datasets collected from 56 spatially distributed locations over a 12-year period. The data then undergo preprocessing, including normalization and Pearson-correlation-based feature selection, which reduces the predictor set to ten significant variables. These refined inputs are subsequently used to train and evaluate three deep learning architectures: Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Model performance is assessed using standard statistical metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the Coefficient of Determination (R2). Among the tested models, GRU demonstrates the highest predictive accuracy, achieving an R2 value of 0.9958. The optimized forecasting framework provides reliable hourly solar irradiance estimates across Egypt, supporting applications in solar energy planning, renewable energy integration, and grid management. Overall, the graphical abstract summarizes a scalable and robust forecasting solution suited to Egypt’s climatic diversity. Developed a deep learning framework for forecasting GHI across Egypt using 12 years of ERA5 data. Evaluated RNN, LSTM, and GRU models, with GRU showing the highest predictive accuracy. The framework provides scalable and reliable forecasts for solar energy planning and grid management.

Keywords:

Solar irradiance forecasting

Global Horizontal Irradiance (GHI)

Deep learning

Gated Recurrent Unit (GRU)

Feature selection

ERA5 reanalysis

Egypt

Ensemble Deep Learning for Global Solar Irradiance Forecasting in Egypt: A Multi-Model Approach

Abstract

Journal

Organization