Scholarlink

<p>Accurate estimation of the State of Health (SOH) of batteries is crucial for the safe operation of Battery Management System (BMS). Traditional SOH estimation methods require complex mechanistic modeling and depend on expert knowledge. Existing deep learning-based estimation methods fail to make full use of battery degradation stage information and have limitations in accuracy and generalizability. To make up for the deficiencies of previous research work, we propose a novel network model called multi-scale convolutional long short-term memory network combined with feature segment localization (MLSTM-FSL) in this paper. The MLSTM-FSL model mainly consists of a multi-scale convolutional long short-term memory (MSC-LSTM) module and a feature segment localization (FSL) module. The MSC-LSTM module uses multi-scale convolutional and LSTM networks to learn the information of different scales in the battery multi-feature time series and obtain temporal dependencies. Meanwhile, the auxiliary FSL module mines the battery capacity degradation stage information, which significantly improves the prediction accuracy of the model. Experimental results on the NASA dataset and the self-constructed battery fault dataset based on the NASA dataset and Toyota dataset show that the MLSTM-FSL model has a significant advantage in terms of performance, and exhibits a strong generalizability for the fault detection task.</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

SOH prediction and fault early warning of lithium-ion batteries based on MLSTM-FSL model

  • 期刊论文

  • 发表日期：

    2025年3月8日

查看全文
  • Xiaohua Wang,
  • Mengting Li,
  • Min Hu,
  • Nanbing Ni

摘要

Accurate estimation of the State of Health (SOH) of batteries is crucial for the safe operation of Battery Management System (BMS). Traditional SOH estimation methods require complex mechanistic modeling and depend on expert knowledge. Existing deep learning-based estimation methods fail to make full use of battery degradation stage information and have limitations in accuracy and generalizability. To make up for the deficiencies of previous research work, we propose a novel network model called multi-scale convolutional long short-term memory network combined with feature segment localization (MLSTM-FSL) in this paper. The MLSTM-FSL model mainly consists of a multi-scale convolutional long short-term memory (MSC-LSTM) module and a feature segment localization (FSL) module. The MSC-LSTM module uses multi-scale convolutional and LSTM networks to learn the information of different scales in the battery multi-feature time series and obtain temporal dependencies. Meanwhile, the auxiliary FSL module mines the battery capacity degradation stage information, which significantly improves the prediction accuracy of the model. Experimental results on the NASA dataset and the self-constructed battery fault dataset based on the NASA dataset and Toyota dataset show that the MLSTM-FSL model has a significant advantage in terms of performance, and exhibits a strong generalizability for the fault detection task.