A Novel Active Cell Balancing Approach Based on Reinforcement Learning for SoC Balancing of Four Lithium-Ion Battery Cells

Battery lifetime and performance are critical concerns for electric vehicles (EV’s) and energy storage systems (ESS). However, factors such as environmental variations and manufacturing defects often result in charge imbalance among battery cells, leading to reduced energy capacity and power performance of the entire battery pack. To address this issue, cell equalization becomes necessary. In recent years, the advancements in deep reinforcement learning (DRL) have made it a viable tool for battery management systems (BMS) in EV’s. In this research paper, a new method is introduced for active cell balancing of a battery pack consisting of four series connected lithium-ion (Li-ion) batteries. The approach utilizes deep reinforcement learning (DRL) within a MATLAB simulation. A deep Q-learning (DQL) algorithm is used for the training of DRL agent, and a DC-DC Zeta converter is used to transfer the excess charge from the overcharged cell to the undercharged cell in the battery pack. Additionally, the proposed equalization topology focuses on selecting the specific cells that requires balancing, aiming to enhance the speed of the equalization process. The simulation results demonstrated that SoC convergence among four Li-ion cells (with an SoC difference of as little as 0.5%) occurs within 500 s using the proposed novel integration.