An active equalization method for redundant battery based on deep reinforcement learning

Battery equalization is essential in the battery management system. In this paper, an active equalization method based on redundant battery is proposed. The equalization circuit consists of a battery string composed of multiple batteries connected in series and a redundant battery. During the discharging process, one cell in the string is selected by the switch controller to be paralleled with redundant cells for equalization purposes. On this basis, an optimal switch control strategy based on deep reinforcement learning (DRL) is proposed, which takes into ac-count the battery's state of charge (SOC), state of health (SOH), and current distribution during parallel connection. The proposed optimal switching control strategy can achieve equalization with the least number of switching times. Simulation shows that, compared with the greedy algorithm and the rule algorithm, the strategy proposed in this paper can reduce the SOC inconsistency of the battery string to less than 1% with the minimum number of switching times.