Explainable Deep Reinforcement Learning for Resilient and Battery-Aware Microgrid Control

Microgrids are crucial for integrating renewable energy and improving power system resilience, yet daily operation is complicated by intermittent resources, uncertain forecasts, and the need to preserve battery life. This paper presents an eXplainable Deep Reinforcement Learning (XDRL) framework for hourly microgrid energy management under uncertainty, built on Proximal Policy Optimisation (PPO). The reward function jointly considers a Resilience Index (RI) and a life-cycle-aware battery term, so learned policies protect supply while limiting depth of discharge stress. Robustness is promoted through curriculum learning across sequentially noisier training scenarios. Explainability, essential for industrial adoption, is provided through post-hoc Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) analyses that expose how decisions depend on net-energy balance, recent state of charge, and different load demands. The framework is evaluated on a cyclone-prone coastal microgrid in Kothapatnam, India. The test system comprises solar PV, wind generation, and battery storage serving essential, business, and agricultural loads with priority-based dispatch. Under uncertain forecasts, the DRL controller attains a Resilience Index of 0.9956, within 0.33% of a Model Predictive Control (MPC) benchmark (0.9989), while extending expected battery life by about 5% (15.9 vs. 15.1 years) and producing smoother charge/discharge profiles. Monte Carlo simulations confirm robustness (median RI 0.992, 5–95%: 0.984–0.999; median battery life ∼16 years). The XDRL framework integrates uncertainty, multi-objective optimisation, and explainability for resilient microgrid control