Explainable AI framework for operational mode classification in hybrid renewable energy systems

The increasing penetration of renewable energy sources requires reliable operational decision support for hybrid renewable energy systems operating under variable conditions. This study presents a physics-guided machine-learning framework for real-time operating-mode classification, distinguishing between grid-import and hydrogen-production states based on the system’s energy balance. Ensemble models, including Random Forest and XGBoost, are trained using physically meaningful inputs such as solar power, wind power, load demand, and system losses. The framework is evaluated under nominal conditions and under physically motivated disturbances, including measurement noise, sensor bias, data loss, and marginal operating conditions near the surplus–deficit transition. Results show that ensemble-based models provide stable, consistent mode identification under uncertainty while maintaining alignment with physical energy-flow behavior. Scenario-based analysis reveals a marginal energy balance regime in which renewable generation approaches demand and losses, leading to increased decision sensitivity and switching instability in deterministic rule-based control. In this regime, the proposed surrogate reduces unnecessary mode transitions and improves decision stability without compromising physical fidelity. Explainability analysis using SHAP confirms that the model’s decisions follow intuitive control logic, driven primarily by renewable availability. Overall, the study demonstrates that physics-guided machine learning enhances robustness, stability, and interpretability of operational mode decisions under realistic uncertainty.