Scholarlink

<p>Accurate State of Charge (SOC) estimation under multi-temperature and highly dynamic operating conditions remains a critical challenge for lithium-ion battery management. This paper presents an Improved Dung Beetle Optimization algorithm-Temporal Convolutional Network-Bidirectional Gated Recurrent Unit (DB-TCN-BiGRU) framework for high-precision SOC estimation of lithium-ion batteries. The TCN captures electrochemical dynamics across fast polarization and slow diffusion time scales, while the BiGRU models hysteresis-related bidirectional dependencies in voltage-current responses. The multi-strategy Dung Beetle Optimization algorithm (DBO) adaptively selects network hyperparameters to ensure stable convergence under heterogeneous temperature conditions, and the Kalman layer suppresses noise-induced fluctuations to maintain physically consistent SOC trajectories. Experiments on a 143 Ah lithium-ion cell across three temperatures and three dynamic profiles demonstrate that the proposed framework achieves MAE between 0.31% and 0.77%, RMSE &lt; 1.0%, and R² &gt; 0.999, significantly outperforming conventional TCN-BiGRU and BiGRU baselines. These results confirm the framework’s high robustness, strong generalization capability, and suitability for real-time battery management systems.</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

Improved dung beetle-temporal convolutional network-bidirectional gated recurrent unit for high-precision state of charge estimation of lithium-ion batteries

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Xinyu Yan,
  • Tao Xu,
  • Shunli Wang,
  • Liangwei Cheng,
  • Carlos Fernandez,
  • Frede Blaabjerg

摘要

Accurate State of Charge (SOC) estimation under multi-temperature and highly dynamic operating conditions remains a critical challenge for lithium-ion battery management. This paper presents an Improved Dung Beetle Optimization algorithm-Temporal Convolutional Network-Bidirectional Gated Recurrent Unit (DB-TCN-BiGRU) framework for high-precision SOC estimation of lithium-ion batteries. The TCN captures electrochemical dynamics across fast polarization and slow diffusion time scales, while the BiGRU models hysteresis-related bidirectional dependencies in voltage-current responses. The multi-strategy Dung Beetle Optimization algorithm (DBO) adaptively selects network hyperparameters to ensure stable convergence under heterogeneous temperature conditions, and the Kalman layer suppresses noise-induced fluctuations to maintain physically consistent SOC trajectories. Experiments on a 143 Ah lithium-ion cell across three temperatures and three dynamic profiles demonstrate that the proposed framework achieves MAE between 0.31% and 0.77%, RMSE < 1.0%, and R² > 0.999, significantly outperforming conventional TCN-BiGRU and BiGRU baselines. These results confirm the framework’s high robustness, strong generalization capability, and suitability for real-time battery management systems.