Parameter Optimization of Battery Energy Storage System Considering Degradation Using Reinforcement Learning

Accurate and sustainable operation of battery energy storage systems (BESS) is critical for supporting renewable energy integration, ensuring both short-term reliability and long-term asset preservation. This study proposed a reinforcement learning (RL)-based scheduling framework designed to minimize power mismatch while mitigating degradation in lithium-ion batteries. The framework dynamically adapted to fluctuations in photovoltaic generation and residential load, enabling real-time decision-making. The performance was evaluated over a 30-day horizon using three indicators: average power mismatch, cumulative capacity loss, and system stability index (SSI). Results demonstrated that the proposed method achieved near-perfect load balance with an average mismatch of only 0.002 kW, while cumulative degradation remained limited to 0.22% and SSI was maintained at 0.96, reflecting high operational stability. The estimated daily degradation rate of 0.0073% corresponded to an annual capacity loss of approximately 2.7%, significantly lower than the 5–6% typically observed in uncontrolled cycling scenarios. Comparative analysis with simulated annealing (SA) and multi-objective genetic algorithm (MOGA) highlighted the balanced performance of the RL method. While MOGA eliminated mismatch at the expense of excessive degradation (0.60%) and simulated annealing reduced degradation but suffers from high mismatch (0.012 kW), the RL framework delivered the most balanced trade-off across all metrics. These findings confirm the potential of RL as a practical and sustainable strategy for PV–BESS integration, providing both technical resilience and extended battery lifetime.