Hybrid feature selection framework for reliable solar irradiance prediction using ground-based meteorological data

Accurate solar irradiance (SI) prediction is essential for optimizing solar energy (SE) generation and improving the reliability of renewable energy systems. However, the complex and nonlinear interactions among meteorological variables often limit the performance and interpretability of conventional forecasting models. The main challenge addressed in this study is the reliable identification of physically meaningful predictors under nonlinear, non-normal, multicollinear, and sensor-sensitive meteorological conditions. This study develops a reliability-aware feature evaluation framework that integrates established statistical, nonlinear, and explainable methods to assess ground-based meteorological predictors under non-normal, multicollinear, and sensor-sensitive conditions. Linear relationships between predictors were first examined using Pearson, Spearman, and Kendall correlation analyses together with principal component analysis (PCA) to identify dominant variance structures. Nonlinear feature sensitivity was then evaluated using a bilayer neural network, while SHAP and LIME techniques were applied to provide global and local interpretability of model predictions. The results indicate that average battery voltage (ABV) is the most influential predictor of SI (RMSE = 0.203), followed by relative humidity (RMSE = 0.239) and wind direction (RMSE = 0.244). PCA results show that the first three principal components explain 62.88% of the total variance in the dataset. Explainability analysis further confirms that ABV and RH contribute most significantly to SI prediction, while rainfall exhibits minimal global influence due to its sparse occurrence. The framework improves understanding of nonlinear meteorological interactions and supports reliability-aware predictor evaluation for SI modelling. The findings provide valuable insights for SE planning, smart grid integration, and sustainable renewable energy management.