Degradation-Aware Energy Management in Residential Microgrids: A Reinforcement Learning Framework

This paper presents a degradation-aware reinforcement learning (RL) framework for real-time energy management in residential microgrids, focusing on optimizing lithium-ion battery usage while balancing economic benefits and battery longevity. We employ the Soft ActorCritic (SAC) algorithm, implemented via Stable Baselines3, to learn non-linear dispatch policies for a 5.2 kWh LiCoO2 battery pack, with degradation modeled using a simplified energy-throughput approach calibrated with NASA dataset measurements. The framework is tested across diverse household profiles over 1-year and 10-year simulations. Results show that RL-SAC outperforms a Model Predictive Control (MPC) baseline, extending battery life and reducing energy purchases in both simulations. These findings highlight RL-SAC’s potential for practical deployment in microgrids, offering a scalable solution for sustainable energy management.