Scholarlink

Accurate estimation of state-of-health (SOH) is crucial for optimizing charging and discharging strategies and prolonging the lifespan of lithium-ion batteries (LIBs). This paper propose a deep learning model that integrates a multilayer perceptron (MLP) and a bidirectional long and short-term memory (BiLSTM) neural network with attention mechanism for SOH estimation. This model leverages historical and current health indicators(HIs) data, as well as aging information of the LIBs in the SOH history sequence to improve accuracy. Firstly, the model extracts four HIs that are strongly related to the SOH from the LIB operation process. Next, a channel attention based MLP is proposed to realize the feature extraction of HIs, and a temporal attention based BiLSTM is proposed to realize the temporal feature extraction of historical SOH sequence. Then, the feature information of both HIs and historical SOH are fused to realize accurate estimation of current SOH. Besides, the paper propose a parameter-adjustable asymmetric loss function that deals with overestimation and underestimation differently by adjusting the parameters to reduce overestimation. Finally, the proposed model’s effectiveness is validated using aging data from \(LiCoO_2\) batteries sourced from the University of Maryland’s Center for Advanced Life Cycle Engineering battery dataset. Experimental results demonstrate the method’s capability to achieve accurate SOH estimation, effectively reduce overestimation, and outperform existing data-driven methods.

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

State of Health Estimation for Lithium-Ion Batteries with an Attention-Integrated BiLSTM-MLP Hybrid Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Maolin Yang,
  • Yishun Liu,
  • Bo Li,
  • Chunhua Yang

摘要

Accurate estimation of state-of-health (SOH) is crucial for optimizing charging and discharging strategies and prolonging the lifespan of lithium-ion batteries (LIBs). This paper propose a deep learning model that integrates a multilayer perceptron (MLP) and a bidirectional long and short-term memory (BiLSTM) neural network with attention mechanism for SOH estimation. This model leverages historical and current health indicators(HIs) data, as well as aging information of the LIBs in the SOH history sequence to improve accuracy. Firstly, the model extracts four HIs that are strongly related to the SOH from the LIB operation process. Next, a channel attention based MLP is proposed to realize the feature extraction of HIs, and a temporal attention based BiLSTM is proposed to realize the temporal feature extraction of historical SOH sequence. Then, the feature information of both HIs and historical SOH are fused to realize accurate estimation of current SOH. Besides, the paper propose a parameter-adjustable asymmetric loss function that deals with overestimation and underestimation differently by adjusting the parameters to reduce overestimation. Finally, the proposed model’s effectiveness is validated using aging data from \(LiCoO_2\) batteries sourced from the University of Maryland’s Center for Advanced Life Cycle Engineering battery dataset. Experimental results demonstrate the method’s capability to achieve accurate SOH estimation, effectively reduce overestimation, and outperform existing data-driven methods.