Scholarlink

<p>Accurately predicting the State of Health (SOH) of lithium-ion batteries is crucial for ensuring the normal operation of electrical devices. To improve the prediction accuracy of neural networks for the SOH of lithium-ion batteries and their ability to model the complex mechanisms of the battery, this paper proposes a multi-step SOH prediction method based on multivariate feature extraction and fusion with an optimized Transformer. First, multiple features are extracted during the lithium-ion charge–discharge process, and outliers are processed. The features are then divided into Main-feature and Secondary-feature, and multi-feature inputs and fusion are carried out through an attention mechanism. Secondly, to enhance the ability to extract local, global, and temporal features, a new encoder-decoder structure based on Transformer is constructed, integrating dilated causal convolution and LSTM. Finally, experimental results show that the proposed method based on Transformer improvements and feature fusion is effective. When the prediction step length is within 16 steps, the MAE of all battery samples is less than 1.5<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation>, demonstrating significant improvements in prediction accuracy compared to existing methods.</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

Multi-step prediction of lithium-ion battery SOH based on multivariate feature extraction and fusion with optimized transformer

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Hao Xu,
  • Zhanying Li,
  • Yinghao Zhang,
  • Houwen Shen,
  • Yu Gao

摘要

Accurately predicting the State of Health (SOH) of lithium-ion batteries is crucial for ensuring the normal operation of electrical devices. To improve the prediction accuracy of neural networks for the SOH of lithium-ion batteries and their ability to model the complex mechanisms of the battery, this paper proposes a multi-step SOH prediction method based on multivariate feature extraction and fusion with an optimized Transformer. First, multiple features are extracted during the lithium-ion charge–discharge process, and outliers are processed. The features are then divided into Main-feature and Secondary-feature, and multi-feature inputs and fusion are carried out through an attention mechanism. Secondly, to enhance the ability to extract local, global, and temporal features, a new encoder-decoder structure based on Transformer is constructed, integrating dilated causal convolution and LSTM. Finally, experimental results show that the proposed method based on Transformer improvements and feature fusion is effective. When the prediction step length is within 16 steps, the MAE of all battery samples is less than 1.5 \(\%\) , demonstrating significant improvements in prediction accuracy compared to existing methods.