Scholarlink

<p>Lithium-ion batteries are widely used in electric vehicles and energy storage systems due to their high energy density and long cycle life. However, capacity fading and performance degradation are inevitable during long-term operation. Accurate estimation of the state of health (SOH) is essential for ensuring battery safety and extending service life. In this work, a Multi-Scale Entropy Attention Network (MSEA), integrating entropy-based signal features with a hybrid attention mechanism, is proposed. A weighted multi-scale sample entropy, independent of data length, is developed as the input feature. By leveraging the synergy of efficient channel attention, knowledge-assisted networks, and enhanced multi-scale attention, the model achieves cross-channel feature weighting, semantic enhancement, and temporal dependency modeling. Furthermore, the arithmetic optimization algorithm is employed for adaptive hyperparameter search, improving model convergence and generalization. Experimental results on multiple real-world and public datasets demonstrate that MSEA significantly outperforms benchmark models such as TCN, GRU, and GCN in both prediction accuracy and robustness. On the representative test sample Pro03, MSEA reduces the MAE to 0.60%, and MAXE decreases from 4.42% to 1.37%. Moreover, under the condition of 150&#xa0;mV superimposed random noise, MSEA still maintains a low error range of 0.55%–0.62%, confirming its strong noise resilience and stability. Ablation studies further verify that the ECA, KAN, and EMA modules play essential roles in performance enhancement. Overall, the proposed MSEA method demonstrates substantial advantages in accuracy, robustness, and cross-system adaptability, providing a feasible solution for SOH monitoring and intelligent battery management systems.</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 novel SOH estimation method of lithium-ion batteries based on entropy signal features and multi-attention mechanisms

  • 期刊论文

  • 发表日期：

    2025年11月22日

查看全文
  • Yan Li,
  • Jingwen Yin

摘要

Lithium-ion batteries are widely used in electric vehicles and energy storage systems due to their high energy density and long cycle life. However, capacity fading and performance degradation are inevitable during long-term operation. Accurate estimation of the state of health (SOH) is essential for ensuring battery safety and extending service life. In this work, a Multi-Scale Entropy Attention Network (MSEA), integrating entropy-based signal features with a hybrid attention mechanism, is proposed. A weighted multi-scale sample entropy, independent of data length, is developed as the input feature. By leveraging the synergy of efficient channel attention, knowledge-assisted networks, and enhanced multi-scale attention, the model achieves cross-channel feature weighting, semantic enhancement, and temporal dependency modeling. Furthermore, the arithmetic optimization algorithm is employed for adaptive hyperparameter search, improving model convergence and generalization. Experimental results on multiple real-world and public datasets demonstrate that MSEA significantly outperforms benchmark models such as TCN, GRU, and GCN in both prediction accuracy and robustness. On the representative test sample Pro03, MSEA reduces the MAE to 0.60%, and MAXE decreases from 4.42% to 1.37%. Moreover, under the condition of 150 mV superimposed random noise, MSEA still maintains a low error range of 0.55%–0.62%, confirming its strong noise resilience and stability. Ablation studies further verify that the ECA, KAN, and EMA modules play essential roles in performance enhancement. Overall, the proposed MSEA method demonstrates substantial advantages in accuracy, robustness, and cross-system adaptability, providing a feasible solution for SOH monitoring and intelligent battery management systems.