Advancing solar radiation prediction with explainable AI and ensemble learning techniques

The core of sustainable development is empowering future generations with clean, renewable energy. Accurate prediction of solar radiation is imperative for optimizing the efficacy of renewable energy sources and is a critical component of achieving sustainable energy independence. This study examines solar radiation prediction by employing Machine Learning (ML), Deep Learning (DL) algorithms, and Explainable AI (XAI) methods to ensure interpretability. We propose an ensemble ML model named SLRK (Stacking of Linear Regression, Random Forest, and K-nearest Neighbors). Our proposed model outperforms all comparative ML and DL algorithms in terms of regression accuracy and performance metrics, achieving an R 2 score of 98%, MSE of 0.0020, RMSE of 0.0475, and MAE of 0.0201. Other ML models yielded R 2 values from 55% to 97% with MSE values exceeding 0.0043 and RMSE values above 0.0481, and DL models achieved R 2 between 91% and 93% with MSE in the range of 0.0053 to 0.0070 and RMSE between 0.0730 and 0.0837. The t-test confirms that most algorithms demonstrate statistical significance in predicting solar radiation. In addition, XAI techniques, including SHAP, LIME, Shapash, and ELi5, were used to exhibit the impact of temperature, humidity, day of the year, and time of day on solar radiation predictions. These explainability results enhance trust in the model’s predictions and provide practical insights for real-world applications, such as optimal solar panel placement and energy grid optimization. Key challenges remain in improving the model’s robustness under changing climatic conditions and enhancing generalizability across diverse geographic regions, which constitute important directions for future research.