Short-term and long-term solar irradiance forecasting with advanced machine learning techniques in Zafarana, Egypt

Abstract The increasing demand for renewable energy sources has positioned solar energy as a pivotal component in the global transition towards sustainable power generation. As the demand increases for solar energy production, the need for technical specifications, resource cost increases, and output power prediction increases. Thus, recent studies in machine learning (ML) and deep learning (DL) techniques have opened new ways for improving solar irradiance predictions by leveraging historical data. This paper proposes an integrated framework for forecasting solar irradiance, combining feature selection techniques with machine learning models to address region-specific challenges in Zafarana, Egypt, aimed at improving predictive accuracy using historical data sourced from the NASA Power Project for both short-term and long-term horizons. The framework begins with feature selection techniques, including One-Way ANOVA, Boruta, and Random Forest, to identify key variables influencing solar irradiance. This is followed by the implementation of ML and DL models, including Linear Regression (LR), Decision Tree (DT), Gradient Boosting (GB), Random Forest (RF), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model. The analysis reveals that RF and GB achieved high accuracy, with R² scores of 0.9948 and 0.9724, respectively, for one-day forecasts and 0.978 and 0.954, respectively, for one-month forecasts. The results indicate that the proposed machine learning approaches significantly outperform traditional forecasting methods, demonstrating their potential for optimizing solar energy management.