Scholarlink

<p>State of Health (SOH) estimation for electric vehicle batteries is a core function of the Battery Management System (BMS), and is essential for performance optimization, safety management, residual value assessment, and full-lifecycle battery utilization. This review systematically traces the evolution of SOH estimation methods from experimental measurement and model-based approaches to data-driven methods, and analyzes the advantages and limitations of each technical stage. Building on this evolution, the review critically examines key issues in data-driven SOH estimation, including data heterogeneity, incomplete and noisy field data, inconsistent SOH quantification metrics, feature transferability, model generalization, and evaluation reliability.Particular attention is paid to the lack of true SOH labels in real-world vehicle data. Since complete capacity calibration is rarely available during normal vehicle operation, many field-data-driven models are trained and validated using BMS-estimated capacity, available energy, ampere-hour integration results, or other health indicators as proxy labels. This may introduce circular validation risks, where a model reproduces the proxy-label generation process rather than independently estimating the true electrochemical degradation state of the battery. Therefore, low prediction errors reported on proxy labels should be interpreted with caution unless the label source, baseline method, dataset condition, and validation protocol are clearly specified.Finally, this review summarizes future research directions for reliable and deployable SOH estimation, including standardized data protocols, transparent label-generation mechanisms, physics-constrained learning, uncertainty-aware validation, cross-dataset evaluation, weakly supervised and self-supervised learning, and cloud-edge collaborative deployment. These directions are expected to support the development of more robust, interpretable, and practically applicable SOH estimation methods for electric vehicle battery management.</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

Critical Evolution of Data-Driven Methods for Estimating the State of Health (SOH) of Electric Vehicle Batteries

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Zhigang He,
  • Hongbo Zhu,
  • Hongyu Lou,
  • Yujie Li

摘要

State of Health (SOH) estimation for electric vehicle batteries is a core function of the Battery Management System (BMS), and is essential for performance optimization, safety management, residual value assessment, and full-lifecycle battery utilization. This review systematically traces the evolution of SOH estimation methods from experimental measurement and model-based approaches to data-driven methods, and analyzes the advantages and limitations of each technical stage. Building on this evolution, the review critically examines key issues in data-driven SOH estimation, including data heterogeneity, incomplete and noisy field data, inconsistent SOH quantification metrics, feature transferability, model generalization, and evaluation reliability.Particular attention is paid to the lack of true SOH labels in real-world vehicle data. Since complete capacity calibration is rarely available during normal vehicle operation, many field-data-driven models are trained and validated using BMS-estimated capacity, available energy, ampere-hour integration results, or other health indicators as proxy labels. This may introduce circular validation risks, where a model reproduces the proxy-label generation process rather than independently estimating the true electrochemical degradation state of the battery. Therefore, low prediction errors reported on proxy labels should be interpreted with caution unless the label source, baseline method, dataset condition, and validation protocol are clearly specified.Finally, this review summarizes future research directions for reliable and deployable SOH estimation, including standardized data protocols, transparent label-generation mechanisms, physics-constrained learning, uncertainty-aware validation, cross-dataset evaluation, weakly supervised and self-supervised learning, and cloud-edge collaborative deployment. These directions are expected to support the development of more robust, interpretable, and practically applicable SOH estimation methods for electric vehicle battery management.