Scholarlink

<p>With increasing concerns about charging and range anxiety in electric vehicles (EVs), developing safe and fast charging control strategies is particularly important for ensuring the safety of EVs and improving user’s charging experience. This paper proposes a novel safe and fast charging control strategy for lithium-ion batteries based on deep reinforcement learning, capable of adapting to dynamic changes in the charging environment. Firstly, a unilateral sampling soft actor-critic (USSAC) algorithm is proposed and integrated with an electrochemical-thermal coupling model to train an agent capable of optimizing charging speeds while adhering to multiphysical constraints. The trained agent can then provide a safe and fast charging control strategy and be updated online. Secondly, an adaptive negative pulse regulation method that autonomously adds negative pulses base on multiphysical constraints is proposed to further enhance the safety of the charging process. Finally, the proposed charging strategy is simulated and experimentally verified under different environment conditions. The experimental results show that, compared to the commonly used fast charging strategies, the proposed USSAC-based charging strategy can dynamically provide the optimal charging current in real-time according to the battery environment and its own status, effectively mitigating the risks of overcharge, lithium plating, and thermal hazards.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Safe and fast charging control strategy for lithium-ion batteries based on deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Hengwei Xie,
  • Shengzhe Liu,
  • Ruifeng An,
  • Xiaojun Tan

摘要

With increasing concerns about charging and range anxiety in electric vehicles (EVs), developing safe and fast charging control strategies is particularly important for ensuring the safety of EVs and improving user’s charging experience. This paper proposes a novel safe and fast charging control strategy for lithium-ion batteries based on deep reinforcement learning, capable of adapting to dynamic changes in the charging environment. Firstly, a unilateral sampling soft actor-critic (USSAC) algorithm is proposed and integrated with an electrochemical-thermal coupling model to train an agent capable of optimizing charging speeds while adhering to multiphysical constraints. The trained agent can then provide a safe and fast charging control strategy and be updated online. Secondly, an adaptive negative pulse regulation method that autonomously adds negative pulses base on multiphysical constraints is proposed to further enhance the safety of the charging process. Finally, the proposed charging strategy is simulated and experimentally verified under different environment conditions. The experimental results show that, compared to the commonly used fast charging strategies, the proposed USSAC-based charging strategy can dynamically provide the optimal charging current in real-time according to the battery environment and its own status, effectively mitigating the risks of overcharge, lithium plating, and thermal hazards.