Scalable Charging Optimization of Battery Energy Storage Systems with Deep Reinforcement Learning

This paper presents a scalable data-driven methodology that leverages deep reinforcement learning (DRL) to optimize the charging of battery units within smart energy storage systems (ESS). Battery charging is formulated as an optimization problem for individual battery units. A novel DRL-based architecture based on local data is proposed to derive the optimal policy for each battery unit while ensuring scalability across the entire storage system. This architecture features a shared buffer to aggregate experiences from all agents, enabling the synthesis of centralized training with decentralized execution. The efficacy and scalability of this approach are substantiated through a comprehensive evaluation, demonstrating enhanced performance across various configurations of battery units. The inherent scalability of this methodology facilitates its integration into modular and reconfigurable storage systems, proving the potential for widespread practical applications.