Scholarlink

<p>The state of health (SOH) of batteries is a crucial evaluation parameter in battery management systems. However, it is often subject to unforeseen noise interference in practical operations. This study proposes a novel battery SOH estimation framework, designing a variable parameter model to reconstruct the voltage-incremental capacity (IC) curves during the constant-voltage charging phase and utilizing an extreme learning mechanism to reconstruct the current curves during the constant-current charging phase. Experimental results indicate that even under varying degrees of amplified noise, the framework can still produce smooth voltage, current, and IC curves, with the root mean square error (RMSE) of SOH estimation reaching 2.1%. Furthermore, to address the impacts of noisy input features and frequent fluctuations in ambient temperature on battery SOH estimation, a hybrid model combining a Temporal Convolutional Network with Bidirectional Gated Recurrent Units (TCN-BIGRU-AE) is proposed. Experimental results demonstrate that the adopted 25-step prediction method achieves an RMSE of 2.6% for battery SOH estimation, indicating that the proposed model exhibits good performance and strong robustness in estimating battery SOH.</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

A Deep Learning Hybrid Model for Battery State of Health Estimation Based on Reconstructed Current and Voltage Vurves

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Shixin Song,
  • Cewei Zhang,
  • Chuanxue Song,
  • Wenxian Duan,
  • Feng Xiao,
  • Liqiang Jin,
  • Chunyang Qi

摘要

The state of health (SOH) of batteries is a crucial evaluation parameter in battery management systems. However, it is often subject to unforeseen noise interference in practical operations. This study proposes a novel battery SOH estimation framework, designing a variable parameter model to reconstruct the voltage-incremental capacity (IC) curves during the constant-voltage charging phase and utilizing an extreme learning mechanism to reconstruct the current curves during the constant-current charging phase. Experimental results indicate that even under varying degrees of amplified noise, the framework can still produce smooth voltage, current, and IC curves, with the root mean square error (RMSE) of SOH estimation reaching 2.1%. Furthermore, to address the impacts of noisy input features and frequent fluctuations in ambient temperature on battery SOH estimation, a hybrid model combining a Temporal Convolutional Network with Bidirectional Gated Recurrent Units (TCN-BIGRU-AE) is proposed. Experimental results demonstrate that the adopted 25-step prediction method achieves an RMSE of 2.6% for battery SOH estimation, indicating that the proposed model exhibits good performance and strong robustness in estimating battery SOH.