Adaptive control and explainable decision-making in residential energy management using a SHAP-guided multi-objective framework

This paper proposes a novel SHAP-guided (SHapley Additive exPlanations) multi-objective energy management framework that explicitly transforms SHAP-derived feature attributions into dynamic control-rule weights for real-time battery dispatch. The increasing deployment of residential photovoltaic (PV) generation and battery energy storage systems necessitates intelligent strategies that balance multiple, often conflicting, objectives while remaining transparent and interpretable. The proposed architecture integrates an adaptive PV forecasting module with a rule-based controller whose priorities are continuously updated using SHAP values, enabling context-aware and explainable decision-making. Unlike conventional black-box optimization approaches, the framework quantifies the contribution of key variables, such as time-of-use tariffs, state of charge, PV generation, and household demand, to each control action. The method simultaneously minimizes electricity cost, mitigates peak demand, limits battery degradation, and enhances energy autonomy. Comprehensive simulations under deficit, balanced, and surplus PV conditions, using real residential load data and Irish time-of-use tariffs, demonstrate that the proposed approach outperforms Mixed-Integer Linear Programming (MILP), Model Predictive Control (MPC), and a heuristic baseline. It achieves the lowest weekly operating cost and highest savings, eliminates peak-period grid imports, and maintains high self-consumption and self-sufficiency with reduced battery cycling. These results show that SHAP-driven control not only improves holistic system performance but also provides transparent, interpretable, and actionable energy management decisions.