Scholarlink

<p>To improve the accuracy of internal short circuit (ISC) fault diagnosis for lithium-ion batteries (LIBs), the study proposed a fusion model based on the incremental capacity curve (ICC). In the fusion model, the gated recurrent unit (GRU) model is utilized to extract ISC time-order features from the raw data of the ICC. The convolutional neural network (CNN) model is applied to extract spatial features from 2D images of the ICC transformed by the Gramian angular summation field (GSAF) encoding method. With the multi-head self-attention (MHSA) mechanism processing the fused features of two models, the ISC degree of the LIB is diagnosed. To evaluate the performance of the fusion model on ISC fault diagnosis for LIBs, three different models, CNN, GRU, and CNN and GRU two-branch parallel fusion self-attention mechanism (CNN-GRU-SA), are used for comparison with the fusion model. The results show that the fusion model has the highest ISC fault diagnostic accuracy at different initial states of charge (SOCs). Accuracy rates of the model for ISC fault diagnostics are all above 90% at different initial SOCs. The fusion model also has the best other evaluation indicators, with average values of recall, precision, and F1 score all exceeding 90%.The proposed method is particularly suitable for early ISC fault diagnosis in battery management systems of electric vehicles, enhancing safety and reliability.</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

Study of Lithium-Ion Battery Internal Short Circuit Diagnosis Based on Multi-model Fusion Network Model

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Zhongbo Zhang,
  • Yongjia Zeng,
  • Caihang Liao,
  • Wei Yu,
  • Wenbo Zhu,
  • Haibing Li,
  • Qin Liu

摘要

To improve the accuracy of internal short circuit (ISC) fault diagnosis for lithium-ion batteries (LIBs), the study proposed a fusion model based on the incremental capacity curve (ICC). In the fusion model, the gated recurrent unit (GRU) model is utilized to extract ISC time-order features from the raw data of the ICC. The convolutional neural network (CNN) model is applied to extract spatial features from 2D images of the ICC transformed by the Gramian angular summation field (GSAF) encoding method. With the multi-head self-attention (MHSA) mechanism processing the fused features of two models, the ISC degree of the LIB is diagnosed. To evaluate the performance of the fusion model on ISC fault diagnosis for LIBs, three different models, CNN, GRU, and CNN and GRU two-branch parallel fusion self-attention mechanism (CNN-GRU-SA), are used for comparison with the fusion model. The results show that the fusion model has the highest ISC fault diagnostic accuracy at different initial states of charge (SOCs). Accuracy rates of the model for ISC fault diagnostics are all above 90% at different initial SOCs. The fusion model also has the best other evaluation indicators, with average values of recall, precision, and F1 score all exceeding 90%.The proposed method is particularly suitable for early ISC fault diagnosis in battery management systems of electric vehicles, enhancing safety and reliability.