Scholarlink

<p>Accurate and rapid estimation of the State of Health (SOH) is crucial for battery management systems (BMS) in real-world vehicles. In this study, a novel FAP-TCN deep learning network is proposed for estimating the SOH of electric vehicles. This method innovatively proposes the Frequency-Dominant Adaption Periodic Network (FAP) module, which is capable of learning the periodic and frequency-domain characteristic information of battery degradation. At the same time, it incorporates the Temporal Convolutional Network (TCN) to enhance the network’s ability to infer the causal relationships in long time series. To address the issue of insufficient sample data, which leads to a decline in SOH estimation accuracy in real-world vehicles, this study integrates the Model-Agnostic Meta-Learning (MAML) approach and introduces the MAML-FAP-TCN framework to enhance SOH estimation accuracy under small sample conditions. Experimental results demonstrate that the FAP-TCN network achieves good performance on the DVC dataset, with a mean absolute error (MAE) of 1.68% and a root mean square error (RMSE) of 2.08%. Additionally, in small-sample experiments conducted on the NDVNEV dataset, the MAE and RMSE of the MAML-FAP-TCN network are 2.18% and 2.79%, respectively, showing an improvement of approximately 15.8% compared to the FAP-TCN network.</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

Estimation of the state of health of real-world vehicle batteries based on the fusion of frequency domain and periodic signals

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • Juan Wang,
  • Shuyao Hu,
  • Minghu Wu,
  • Yufei Zhang,
  • Minghua Wu,
  • Fan Zhang,
  • Haina Song

摘要

Accurate and rapid estimation of the State of Health (SOH) is crucial for battery management systems (BMS) in real-world vehicles. In this study, a novel FAP-TCN deep learning network is proposed for estimating the SOH of electric vehicles. This method innovatively proposes the Frequency-Dominant Adaption Periodic Network (FAP) module, which is capable of learning the periodic and frequency-domain characteristic information of battery degradation. At the same time, it incorporates the Temporal Convolutional Network (TCN) to enhance the network’s ability to infer the causal relationships in long time series. To address the issue of insufficient sample data, which leads to a decline in SOH estimation accuracy in real-world vehicles, this study integrates the Model-Agnostic Meta-Learning (MAML) approach and introduces the MAML-FAP-TCN framework to enhance SOH estimation accuracy under small sample conditions. Experimental results demonstrate that the FAP-TCN network achieves good performance on the DVC dataset, with a mean absolute error (MAE) of 1.68% and a root mean square error (RMSE) of 2.08%. Additionally, in small-sample experiments conducted on the NDVNEV dataset, the MAE and RMSE of the MAML-FAP-TCN network are 2.18% and 2.79%, respectively, showing an improvement of approximately 15.8% compared to the FAP-TCN network.