Scholarlink

<p>Lithium-ion batteries are widely used in new energy vehicles, energy storage systems, and battery management system applications, making state of health (SOH) estimation under varying conditions crucial for safety, reliability, and lifetime management. However, existing data-driven SOH estimation methods still suffer from feature redundancy, limited cross-condition transferability, and weak output physical consistency in cross-protocol, cross-rate, and cross-temperature scenarios. To address these challenges, this paper proposes a Dual-Criterion Selected Physics-Constrained Relational Temporal Framework (DCS-RTF) for cross-condition SOH estimation with few-shot adaptation. First, 16 candidate health features are constructed from voltage, current, temperature, and incremental capacity (IC) curves, and a Dual-Criterion Six-feature Selection (DCS-6) strategy is developed by combining Pearson relevance/redundancy and SHAP contribution. The framework then integrates relational modeling, TCN-based temporal degradation modeling, and output-level physical-constraint learning to enhance inter-feature dependencies, characterize long-term degradation trajectories, and suppress non-physical SOH fluctuations. In addition, source-domain pretraining, target-domain fine-tuning, and target-domain testing are combined to support cross-condition adaptation. Experiments on the self-built EVE dataset and the public TJU-NCA dataset show that the proposed method achieves the best or best average results in source-domain comparison, cross-condition fine-tuning, and external comparison. Across all possible two-battery combinations, DCS-RTF achieved MAE/RMSE values of 0.584%/0.760%, 0.354%/0.436%, and 0.591%/0.722% in the self-built protocol-transfer, public rate-transfer, and public temperature-transfer scenarios, respectively. The full model further demonstrates improved accuracy, robustness, physical consistency, and generalization capability, indicating that DCS-RTF provides a unified and promising framework for cross-condition battery SOH estimation within the evaluated protocol-, rate-, and temperature-transfer scenarios, with potential value for electrical control-oriented battery management applications.</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

A dual-criterion selected physics-constrained relational temporal framework for cross-condition battery SOH estimation

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Jingwen Yin,
  • Yan Li,
  • Guifeng Zhang

摘要

Lithium-ion batteries are widely used in new energy vehicles, energy storage systems, and battery management system applications, making state of health (SOH) estimation under varying conditions crucial for safety, reliability, and lifetime management. However, existing data-driven SOH estimation methods still suffer from feature redundancy, limited cross-condition transferability, and weak output physical consistency in cross-protocol, cross-rate, and cross-temperature scenarios. To address these challenges, this paper proposes a Dual-Criterion Selected Physics-Constrained Relational Temporal Framework (DCS-RTF) for cross-condition SOH estimation with few-shot adaptation. First, 16 candidate health features are constructed from voltage, current, temperature, and incremental capacity (IC) curves, and a Dual-Criterion Six-feature Selection (DCS-6) strategy is developed by combining Pearson relevance/redundancy and SHAP contribution. The framework then integrates relational modeling, TCN-based temporal degradation modeling, and output-level physical-constraint learning to enhance inter-feature dependencies, characterize long-term degradation trajectories, and suppress non-physical SOH fluctuations. In addition, source-domain pretraining, target-domain fine-tuning, and target-domain testing are combined to support cross-condition adaptation. Experiments on the self-built EVE dataset and the public TJU-NCA dataset show that the proposed method achieves the best or best average results in source-domain comparison, cross-condition fine-tuning, and external comparison. Across all possible two-battery combinations, DCS-RTF achieved MAE/RMSE values of 0.584%/0.760%, 0.354%/0.436%, and 0.591%/0.722% in the self-built protocol-transfer, public rate-transfer, and public temperature-transfer scenarios, respectively. The full model further demonstrates improved accuracy, robustness, physical consistency, and generalization capability, indicating that DCS-RTF provides a unified and promising framework for cross-condition battery SOH estimation within the evaluated protocol-, rate-, and temperature-transfer scenarios, with potential value for electrical control-oriented battery management applications.