• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2026, Vol. 62 ›› Issue (18): 256-267.doi: 10.3901/JME.260773

• 特邀专栏:大模型驱动的高端装备智能运维 • 上一篇    

扫码分享

保障数据隐私的锂电池多用户协同智能健康监测通用基础模型

张微1, 常希鹏1, 李响2, 杨绍杰3   

  1. 1. 沈阳航空航天大学航空宇航学院 沈阳 110136;
    2. 西安交通大学现代设计与转子轴承系统教育部重点实验室 西安 710049;
    3. 加州大学圣地亚哥分校机械与航空航天工程系 拉霍亚 92093 美国
  • 收稿日期:2026-01-05 修回日期:2026-06-22 发布日期:2026-11-20
  • 作者简介:张微,女,1988年出生,教授。主要研究方向为工业人工智能,机械装备智能运维等。E-mail:zw_7126257@163.com;常希鹏,男,2002年出生,硕士研究生。主要研究方向为锂电池寿命监测,动力电池大数据状态估计。E-mail:cxp124320@163.com;李响(通信作者),男,1990年出生,教授。主要研究方向为工业人工智能、机器视觉、神经形态感知与计算等。E-mail:lixiang@xjtu.edu.cn
  • 基金资助:
    国家自然科学基金(52475125)资助项目。

Privacy-preserving General Foundation Model for Multi-user Collaborative Intelligent Health Monitoring of Lithium-ion Batteries

ZHANG Wei1, CHANG Xipeng1, LI Xiang2, YANG Shaojie3   

  1. 1. College of Aerospace Engineering, Shenyang Aerospace University, Shenyang 110136;
    2. Key Laboratory of Modern Design and Rotor-Bearing Systems, Ministry of Education, Xi'an Jiaotong University, Xi'an 710049;
    3. Department of Mechanical and Aerospace Engineering, University of California at San Diego, La Jolla, CA 92093 USA
  • Received:2026-01-05 Revised:2026-06-22 Published:2026-11-20

摘要: 锂电池健康监测对储能、交通、运载等工业领域至关重要,近年来数据驱动方法在电池健康监测领域取得显著成果,现有高精度的健康监测方法依赖于大规模监测数据的集中训练,但在实际工程中,单一用户难以收集大规模数据,且传统模型普遍适用于有限电池型号与运行工况,难以泛化普适,因此智能监测方法的工程应用面临巨大挑战。在工程场景下,不同用户往往拥有不同电池型号、工况的数据,将不同用户的数据集中建模能够解决单一用户数据稀缺与模型泛化性差的问题。然而,电池数据往往包含着用户的隐私,让不同用户共享其私有数据将泄漏隐私,因而难以在工程中应用。提出保障数据隐私的锂电池多用户协同智能健康监测通用基础模型,各用户在本地完成模型训练,多用户间仅交互模型信息;提出Mamba状态空间模型作为电池监测数据的统一序列编码器;建立局部特征空间尺度对齐方法,缓解不同数据源间的尺度偏移,解决数据分布偏差难题;在12个数据集上的实验结果表明,该模型能够在隐私约束下有效整合多用户不同电池型号、工况以及退化模式数据中的健康状态退化知识,学习具有良好泛化能力的统一电池健康表征,稳定实现多用户协同的高精度智能健康监测。

关键词: 锂离子电池, 隐私保护, 多用户协同, 通用基础模型, 健康状态监测

Abstract: Lithium battery health monitoring is critical for industrial fields such as energy storage, transportation, and logistics. In recent years, data-driven methods have achieved remarkable progress in battery health monitoring. However, existing high-accuracy monitoring methods rely on centralized training with large-scale monitoring data. In practical engineering applications, a single user is often unable to collect sufficient data, and conventional models are typically designed for limited battery types and operating conditions, making it difficult for them to generalize across diverse scenarios. As a result, the engineering deployment of intelligent monitoring methods faces significant challenges. In real-world settings, different users often possess data from different battery types and operating conditions. Integrating data from multiple users can alleviate the problems of data scarcity at a single user and poor model generalization. Nevertheless, battery data often contain user privacy information, and sharing private data among users may lead to privacy leakage, making such approaches difficult to apply in engineering practice. To address this issue, a privacy-preserving multi-user collaborative intelligent health monitoring foundation model for lithium batteries is proposed, in which each user trains the model locally and only model information is exchanged among users. A Mamba state-space model is introduced as a unified sequence encoder for battery monitoring data. In addition, a local feature-space scale alignment method is established to alleviate scale shifts among different data sources and mitigate data distribution bias. Experimental results on 12 datasets show that, under privacy constraints, the proposed model can effectively integrate degradation knowledge from multiple users with different battery types, operating conditions, and degradation modes, learn a unified battery health representation with strong generalization capability, and stably achieve high-accuracy collaborative intelligent health monitoring across multiple users.

Key words: lithium-ion batteries, privacy preservation, multi-user collaboration, general foundation model, health state monitoring

中图分类号: