Scholarlink

<p>Lithium-ion batteries are pivotal in the automotive INDUSTRY for their high energy density. Accurate state of charge estimation is essential for optimizing battery performance and longevity. This study utilizes a third-order resistance–capacitance equivalent circuit model with parameters estimated via MATLAB/Simulink Simscape. Four state of charge estimation methods: Coulomb counting, Extended Kalman filter, Unscented Kalman filter, and Extended Kalman-Bucy filter are evaluated. Extended Kalman-Bucy filter demonstrated the highest accuracy (Mean Absolute Error = 0.008%, Root Mean Square Error = 0.01%) but required the longest computation time (32.938&#xa0;s), whereas Coulomb counting was the fastest (6.237&#xa0;s) but least accurate (Mean Absolute Error = 0.0445%, Root Mean Square Error = 0.0548%). To enhance state of charge estimation, a deep neural network is designed to predict equivalent circuit model parameters based on state of charge and temperature. The deep neural network predictions were integrated into the Extended Kalman-Bucy filter using two strategies: Direct Integration and Fusion Integration. The Fusion method demonstrated the best performance (Mean Absolute Error = 0.12%, Root Mean Square Error = 0.15%) but had a higher execution time (605&#xa0;s) compared to Direct Integration (602&#xa0;s) and lookup tables (19&#xa0;s). These findings highlight the potential of deep neural network enhanced filtering techniques to significantly improve state of charge estimation accuracy.</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

Analysis and comparison of SOC estimation techniques for Li-ion batteries

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Mohamed R. Zaki,
  • Mohamed A. El-Beltagy,
  • Ahmed E. Hammad

摘要

Lithium-ion batteries are pivotal in the automotive INDUSTRY for their high energy density. Accurate state of charge estimation is essential for optimizing battery performance and longevity. This study utilizes a third-order resistance–capacitance equivalent circuit model with parameters estimated via MATLAB/Simulink Simscape. Four state of charge estimation methods: Coulomb counting, Extended Kalman filter, Unscented Kalman filter, and Extended Kalman-Bucy filter are evaluated. Extended Kalman-Bucy filter demonstrated the highest accuracy (Mean Absolute Error = 0.008%, Root Mean Square Error = 0.01%) but required the longest computation time (32.938 s), whereas Coulomb counting was the fastest (6.237 s) but least accurate (Mean Absolute Error = 0.0445%, Root Mean Square Error = 0.0548%). To enhance state of charge estimation, a deep neural network is designed to predict equivalent circuit model parameters based on state of charge and temperature. The deep neural network predictions were integrated into the Extended Kalman-Bucy filter using two strategies: Direct Integration and Fusion Integration. The Fusion method demonstrated the best performance (Mean Absolute Error = 0.12%, Root Mean Square Error = 0.15%) but had a higher execution time (605 s) compared to Direct Integration (602 s) and lookup tables (19 s). These findings highlight the potential of deep neural network enhanced filtering techniques to significantly improve state of charge estimation accuracy.