Scholarlink

<p>Assessing the state of charge (SOC) is vital for ensuring the accurate and efficient operation of lithium-ion batteries in electric vehicles and smart devices. To keep these batteries reliable, safe, and have an appropriate service life across applications such as electric vehicles and portable electronics, precise SOC estimation by the battery management system (BMS) is essential. This study uses transfer learning with a long short-term memory (LSTM) network to analyze how training and testing variables affect the accuracy of SOC estimation. It also emphasizes the use of an adaptive feedback correction-gain extended Kalman filter (AFGEKF) and an EKF, both of which utilize independently collected operational data and LSTM-estimated SOCs to improve performance. Through iterative improvements, this approach enhances denoising and SOC accuracy across different operating conditions. The comprehensive results show that the optimal mean absolute error, mean squared error, and mean absolute percentage error are 0.4544%, 0.7326%, and 0.9371% for the LSTM model; 0.3069%, 0.4093%, and 0.3577% for the LSTM-EKF model; and 0.14687%, 0.3169%, and 0.2492% for the proposed LSTM-AFGEKF model at 0&#xa0;°C, 25&#xa0;°C, and 45&#xa0;°C with a ternary battery. The study demonstrates that the LSTM training and testing hyperparameters significantly influence SOC estimation accuracy. Additionally, the proposed LSTM-AFGEKF model’s ability to deliver precise SOC estimates makes it a reliable, high-performance model.</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 novel long short-term memory-adaptive feedback-correction gain extended Kalman filter for the high-precision state-of-charge estimation of lithium-ion batteries

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Emmanuel Appiah,
  • Shunli Wang,
  • Paul Takyi-Aninakwa,
  • Etse Dablu Bobobee,
  • Ebenezer Koukoyi

摘要

Assessing the state of charge (SOC) is vital for ensuring the accurate and efficient operation of lithium-ion batteries in electric vehicles and smart devices. To keep these batteries reliable, safe, and have an appropriate service life across applications such as electric vehicles and portable electronics, precise SOC estimation by the battery management system (BMS) is essential. This study uses transfer learning with a long short-term memory (LSTM) network to analyze how training and testing variables affect the accuracy of SOC estimation. It also emphasizes the use of an adaptive feedback correction-gain extended Kalman filter (AFGEKF) and an EKF, both of which utilize independently collected operational data and LSTM-estimated SOCs to improve performance. Through iterative improvements, this approach enhances denoising and SOC accuracy across different operating conditions. The comprehensive results show that the optimal mean absolute error, mean squared error, and mean absolute percentage error are 0.4544%, 0.7326%, and 0.9371% for the LSTM model; 0.3069%, 0.4093%, and 0.3577% for the LSTM-EKF model; and 0.14687%, 0.3169%, and 0.2492% for the proposed LSTM-AFGEKF model at 0 °C, 25 °C, and 45 °C with a ternary battery. The study demonstrates that the LSTM training and testing hyperparameters significantly influence SOC estimation accuracy. Additionally, the proposed LSTM-AFGEKF model’s ability to deliver precise SOC estimates makes it a reliable, high-performance model.