Scholarlink

<p>This study develops a remaining useful life (RUL) estimation algorithm for electric vehicle lithium-ion batteries based on multi-scale CNN feature integration and multiple multi-head attention LSTM (MCNN-MHALSTM). The proposed method considers the uncertainty of estimation results and realizes multi-feature fusion and temporal correlation learning via the designed network architecture. Firstly, using the strategy of skip connections, an independent multi-scale CNN is utilized to extract temporal dynamic features from three key battery states: capacity, internal resistance, and temperature Then, Multi-feature integration techniques are utilized to concatenate them and input them into the multiple multi-head attention LSTM (MHALSTM) network to accomplish feature correlation extraction. Secondly, the Long Short-Term Memory (LSTM) algorithm is introduced to estimate the RUL of electric vehicle batteries, and two multi-head attention layers with the same configuration are applied to improve the LSTM algorithm. While achieving high-order feature extraction of battery data, the RUL estimation task of electric vehicle batteries is divided into multiple small tasks, and the problem of feature redundancy mitigation is addressed through resource allocation. Finally, the effectiveness of the MCNN-MHALSTM electric vehicle battery RUL estimation model is verified using the NASA lithium-ion battery capacity degradation dataset.</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 and uncertainty quantification of remaining useful life (RUL) of electric vehicle batteries using a temporal deep model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月22日

查看全文
  • Zhuo Kong,
  • Shuang Li,
  • Chunyu Wang

摘要

This study develops a remaining useful life (RUL) estimation algorithm for electric vehicle lithium-ion batteries based on multi-scale CNN feature integration and multiple multi-head attention LSTM (MCNN-MHALSTM). The proposed method considers the uncertainty of estimation results and realizes multi-feature fusion and temporal correlation learning via the designed network architecture. Firstly, using the strategy of skip connections, an independent multi-scale CNN is utilized to extract temporal dynamic features from three key battery states: capacity, internal resistance, and temperature Then, Multi-feature integration techniques are utilized to concatenate them and input them into the multiple multi-head attention LSTM (MHALSTM) network to accomplish feature correlation extraction. Secondly, the Long Short-Term Memory (LSTM) algorithm is introduced to estimate the RUL of electric vehicle batteries, and two multi-head attention layers with the same configuration are applied to improve the LSTM algorithm. While achieving high-order feature extraction of battery data, the RUL estimation task of electric vehicle batteries is divided into multiple small tasks, and the problem of feature redundancy mitigation is addressed through resource allocation. Finally, the effectiveness of the MCNN-MHALSTM electric vehicle battery RUL estimation model is verified using the NASA lithium-ion battery capacity degradation dataset.