Scholarlink

<p>Accurate estimation of the State of Health (SOH) for lithium-ion batteries is crucial for the safety and reliability of battery management systems. However, traditional methods often suffer from error accumulation and poor adaptability, while data-driven approaches like Long Short-Term Memory (LSTM) networks are hindered by the challenge of hyperparameter optimization. To address these issues, this paper proposes a swarm intelligence-optimized deep neural network for intelligent SOH diagnosis. A multidimensional indirect health indicator (IHI) extraction mechanism is constructed, deriving three highly correlated features—voltage differential slope, constant-current charging time, and temperature change rate—from the charging process. A residual-corrected LSTM network is designed, and the Sparrow Search Algorithm (SSA) is employed to globally optimize seven critical hyperparameters. Experimental results on the NASA and Oxford datasets show that the proposed method substantially alleviates the burden of hyperparameter tuning and improves SOH prediction accuracy compared with a manually tuned LSTM. Compared to the traditional LSTM model, the mean absolute error (MAE) and root mean square error (RMSE) are reduced by approximately 30.4% and 17.6%, respectively, achieving an MAE of 0.3060% and an RMSE of 0.4641% on the primary test set. The method significantly enhances the accuracy and robustness of SOH estimation, providing a viable technical solution for high-precision intelligent diagnostics in practical applications.</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

Swarm intelligence–optimized deep neural network for intelligent diagnosis of lithium‑ion battery state of health

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月30日

查看全文
  • Chong Li,
  • Bangjie Yan,
  • Chunxiang Zhu,
  • Xingkai Zhou,
  • Yinghao Jia

摘要

Accurate estimation of the State of Health (SOH) for lithium-ion batteries is crucial for the safety and reliability of battery management systems. However, traditional methods often suffer from error accumulation and poor adaptability, while data-driven approaches like Long Short-Term Memory (LSTM) networks are hindered by the challenge of hyperparameter optimization. To address these issues, this paper proposes a swarm intelligence-optimized deep neural network for intelligent SOH diagnosis. A multidimensional indirect health indicator (IHI) extraction mechanism is constructed, deriving three highly correlated features—voltage differential slope, constant-current charging time, and temperature change rate—from the charging process. A residual-corrected LSTM network is designed, and the Sparrow Search Algorithm (SSA) is employed to globally optimize seven critical hyperparameters. Experimental results on the NASA and Oxford datasets show that the proposed method substantially alleviates the burden of hyperparameter tuning and improves SOH prediction accuracy compared with a manually tuned LSTM. Compared to the traditional LSTM model, the mean absolute error (MAE) and root mean square error (RMSE) are reduced by approximately 30.4% and 17.6%, respectively, achieving an MAE of 0.3060% and an RMSE of 0.4641% on the primary test set. The method significantly enhances the accuracy and robustness of SOH estimation, providing a viable technical solution for high-precision intelligent diagnostics in practical applications.