SHapley Additive exPlanations-guided rule-based energy management: bridging machine learning interpretability and adaptive control strategies

• SHAP-guided EMS optimizes battery use with improved charge/discharge strategies. • Hybrid LR-XGBoost model with SHAP analytics drives intelligent decision-making • Feature importance analysis directs rule-based control in solar+storage systems. • Performance within 5% of MILP while requiring less computational resources. • Weekly analysis shows consistent grid cost reduction across PV generation levels. The effective integration of photovoltaic (PV) systems with battery storage is essential for advancing sustainable energy adoption yet translating forecasts into adaptive and interpretable control remains a key challenge. This study introduces a SHapley Additive exPlanations–Guided Energy Management System (SHAP-EMS) that directly embeds model interpretability into real-time control for residential solar-battery systems. A hybrid Linear Regression–eXtreme Gradient Boost (LR-XGBoost) model provides one-hour-ahead PV forecasts, while a SHAP-weighted rule-based controller dynamically adjusts decision priorities based on feature importance, system state, and temporal interactions. Results demonstrate that SHAP-EMS achieved an 18.3% reduction in peak grid imports (63.2% to 44.9%), a 4.5% decrease in total imports compared with Mixed-Integer Linear Programming optimization, and consistently high self-consumption ratios under polycrystalline PV conditions. By efficiently adapting to temperature fluctuations and generation variability, the framework illustrates how SHAP values can be leveraged to transform black-box forecasts into transparent, computationally efficient, and adaptive control strategies, establishing a novel paradigm for explainable energy management.