摘要
针对电动汽车电池热管理系统中传统控制方法温控精度不足及环境适应性差的难题,提出基于深度强化学习的智能控制体方法。基于电池热电耦合模型与制冷空调系统模型,应用强化学习中的双延迟深度确定性策略(TD3)算法进行控制策略训练,通过双重价值网络与延迟策略更新机制,克服传统强化学习中的过高估计问题。结果表明:在夏季充电的训练工况下,能够将电池包的平均温度控制在25℃左右;在冬季充电的训练工况下,能够将电池包的平均温度控制在20℃左右,电池模组之间的最大温差控制在1℃以内。同时,在控制动作上,智能体控制的压缩机转速的调整更为平缓,与比例-积分-微分控制、开关控制相比,智能体控制在夏季放电时最高节能了32.1%,充电时最高节能了15.8%,冬季放电时最高节能了17.0%,充电时最高节能了26.3%。此外,在环境条件变化时,智能体能够及时调整控制动作,将电池包的温度控制在目标温度附近。该研究利用TD3强化学习算法能够在多变的环境条件下平稳、精准地控制电池热管理系统,证明了强化学习在电池热管理中的可行性与有效性。
To address the challenges of insufficient temperature control accuracy and poor environmental adaptability in traditional control methods for electric vehicle battery thermal management systems,an intelligent agent control method based on deep reinforcement learning is proposed.Based on a battery electro-thermal coupling model and an air conditioning refrigeration system model,the twin delayed deep deterministic policy gradient(TD3)algorithm in reinforcement learning is applied to train the control strategy.By utilizing dual critic networks and a delayed policy update mechanism,the issue of overestimation common in traditional reinforcement learning is overcome.Results show that under summer charging conditions,the average battery pack temperature can be controlled around 25℃,while under winter charging conditions,it can be maintained around 20℃,with the maximum temperature difference between battery modules controlled within 1℃.Moreover,the compressor speed adjustments made by the intelligent agent are smoother.Compared with proportional-integral-derivative control and on-off control,the intelligent agent control achieves energy savings of up to 32.1%during summer discharging,15.8%during summer charging,17.0%during winter discharging,and 26.3%during winter charging.Additionally,when environmental conditions change,the agent can promptly adjust control actions to maintain the battery pack temperature near the target.This study demonstrates that the TD3 reinforcement learning algorithm can achieve stable and precise control of the battery thermal management system under varying environmental conditions,proving the feasibility and effectiveness of reinforcement learning in battery thermal management.
作者
席椿富
赵东鹏
黄驰
邹子豪
黄琨杰
谢翌
XI Chunfu;ZHAO Dongpeng;HUANG Chi;ZOU Zihao;HUANG Kunjie;XIE Yi(China Automotive Engineering Research Institute Co.,Ltd.,Chongqing 401122,China;School of Mechanical Engineering,Tianjin University,Tianjin 300350,China;Key Laboratory of Low-Grade Energy Utilization Technologies and Systems of Ministry of Education,Chongqing University,Chongqing 400044,China;School of Energy and Power Engineering,Chongqing University,Chongqing 400044,China;School of Mechanical and Transportation Engineering,Chongqing University,Chongqing 400044,China)
出处
《西安交通大学学报》
北大核心
2026年第2期24-37,共14页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(52472375)。
关键词
电池热管理
电池热电耦合模型
强化学习
制冷空调系统
battery thermal management
battery thermoelectric coupling model
reinforcement learning
air conditioning modeling
