Scholarlink

<p>Lithium-ion battery research is vital for advancing modern electronics, electric vehicles, renewable energy storage, and sustainable energy systems; however, the nonlinear and dynamic characteristics of lithium-ion batteries have significantly increased the difficulty of accurately estimating the state of charge. This study proposes a dual-scale deep learning model for estimating SOC. To begin with, the input voltage and current are decomposed by complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and then denoised by detrended fluctuation analysis and adaptive wavelet threshold denoising. Following this, the SOC is decomposed into high- and low-frequency components by performing K-means clustering on the sample entropy of the CEEMDAN decomposition results. Subsequently, a bidirectional long short-term memory network is applied to estimate the high-frequency components, while an AdaBoost hybrid kernel extreme learning machine is employed to estimate the low-frequency components. Experiments based on data from the McMaster University show that the proposed model achieves higher estimation accuracy compared to other models. Specifically, the root mean square error is reduced from 1.92% to 0.88%, and the mean absolute error is reduced from 1.5% to 0.73% at 25&#xa0;°C.</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 dual-scale deep learning model for estimating lithium-ion battery SOC by data denoising

  • 期刊论文

  • 发表日期：

    2025年10月2日

查看全文
  • Sai Wang,
  • Jie Ding,
  • Dezhi Shen,
  • Huibo Chen

摘要

Lithium-ion battery research is vital for advancing modern electronics, electric vehicles, renewable energy storage, and sustainable energy systems; however, the nonlinear and dynamic characteristics of lithium-ion batteries have significantly increased the difficulty of accurately estimating the state of charge. This study proposes a dual-scale deep learning model for estimating SOC. To begin with, the input voltage and current are decomposed by complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and then denoised by detrended fluctuation analysis and adaptive wavelet threshold denoising. Following this, the SOC is decomposed into high- and low-frequency components by performing K-means clustering on the sample entropy of the CEEMDAN decomposition results. Subsequently, a bidirectional long short-term memory network is applied to estimate the high-frequency components, while an AdaBoost hybrid kernel extreme learning machine is employed to estimate the low-frequency components. Experiments based on data from the McMaster University show that the proposed model achieves higher estimation accuracy compared to other models. Specifically, the root mean square error is reduced from 1.92% to 0.88%, and the mean absolute error is reduced from 1.5% to 0.73% at 25 °C.