中国机械工程 ›› 2026, Vol. 37 ›› Issue (10): 2502-2510.DOI: 10.3969/j.issn.1004-132X.2026.10.013

• 系统能量管理与回收利用 • 上一篇    下一篇

考虑锂电池老化与热效应的电动汽车SAC能量管理策略

王春(), 申渝秦(), 曾权宏, 唐滔   

  1. 四川轻化工大学机械工程学院, 宜宾, 644000
  • 收稿日期:2025-08-10 出版日期:2026-10-25 发布日期:2026-11-10
  • 通讯作者: 申渝秦
  • 作者简介:王春,男,1984年生,教授。研究方向为先进储能系统人工智能管理与控制。发表论文30余篇。E-mail:susewangchun@163.com
    申渝秦*(通信作者),女,1999年生,硕士研究生。研究方向为车用复合电源能量管理。E-mail:15191026912@163.com。
  • 基金资助:
    四川省自然科学基金(2024NSFSC0145);四川省自然科学基金(2026NSFSC1314);过程装备与控制工程四川省高校重点实验室开放基金(GK202405);四川轻化工大学科研创新团队计划(SUSE652A004)

SAC-based Energy Management Strategy for Electric Vehicles Considering Lithium Battery Aging and Thermal Effects

WANG Chun(), SHEN Yuqin(), ZENG Quanhong, TANG Tao   

  1. School of Mechanical Engineering,Sichuan University of Science and Engineering,Yibin,Sichuan,644000
  • Received:2025-08-10 Online:2026-10-25 Published:2026-11-10
  • Contact: SHEN Yuqin

摘要:

针对电动汽车混合动力储能系统在复杂动态工况下的多目标协调带来的实时决策与鲁棒性挑战,提出了一种基于软动作-评价(SAC)的能量管理策略。该策略融合专家先验约束以抑制非理性探索,加速在高维连续动作空间中的学习并增强稳定性。显式引入锂离子电池的热效应与老化模型,在充放电与功率分配上实现多目标协同优化,兼顾系统效率与电池寿命。在混合工况的10 ℃、25 ℃、40 ℃条件下,所提策略的能量损失性能分别达到动态规划方法的97.0%、98.2%和97.3%,展现出在复杂动态环境中更优的能效表现与可靠性。研究表明,结合物理特性建模与强化学习的混合方法为混合动力储能系统的智能调度提供了一条有效路径。

关键词: 电动汽车, 混合动力储能系统, 能量管理, 深度强化学习

Abstract:

A soft actor-critic (SAC)-based energy management strategy was proposed to address the real-time decision-making and robustness challenges arising from multi-objective coordination of HESS for electric vehicles under complex dynamic operating conditions. Expert prior constraints were integrated into the proposed strategy to curb erratic exploration, accelerate learning in high-dimensional continuous action spaces, and improve system stability. Meanwhile, thermal dynamics and aging models of lithium-ion batteries were explicitly incorporated, and multi-objective collaborative optimization of charging-discharging processes and power split was realized to balance system efficiency and battery lifetime. Under the mixed driving cycle at 10°C, 25°C, and 40°C, the energy-loss performance of the proposed strategy reaches 97.0%, 98.2%, and 97.3% of that achieved by the dynamic programming method, which demonstrates superior energy performance and reliability under varying thermal conditions.The results suggest that integrating physics-informed models with reinforcement learning is an effective route for adaptive HESS scheduling.

Key words: electric vehicles, hybrid energy storage system (HESS), energy management, deep reinforcement learning

中图分类号: