Scholarlink

Parameter estimation of battery module in energy storage stations is fundamental for battery management and fault diagnosis. This paper proposes a battery module model based on Thevenin equivalent circuits and a physics-informed recurrent neural network (nECM-RNN). The model embeds the physical information of an n-order Thevenin circuit into the recurrent neural network, ensuring that the forward propagation process adheres to the circuit structure. By utilizing the information transmission of the hidden layers, the model can achieve online estimation of battery module parameters. Numerical experiments show that for a battery module, the 104-order nECM-RNN model achieves a parameter estimation average error of less than 5%, a port voltage estimation error of less than 1%, and a DC internal resistance estimation error of approximately 0.2%. The nECM-RNN leverages the advantages of both data-driven and model-driven approaches, and by adjusting the order n, it can be applied to battery modules of different scales. This provides a scalable method for the online estimation of battery module parameters in energy storage stations, offering a flexible solution for battery module management.

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

Study on Modeling Energy Storage Battery Module Based on the Thevenin Equivalent Circuit and Physics-Informed Neural Network

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chuanqi Lin,
  • Guogang Zhang,
  • Chenchen Zhao,
  • Lingna Liu

摘要

Parameter estimation of battery module in energy storage stations is fundamental for battery management and fault diagnosis. This paper proposes a battery module model based on Thevenin equivalent circuits and a physics-informed recurrent neural network (nECM-RNN). The model embeds the physical information of an n-order Thevenin circuit into the recurrent neural network, ensuring that the forward propagation process adheres to the circuit structure. By utilizing the information transmission of the hidden layers, the model can achieve online estimation of battery module parameters. Numerical experiments show that for a battery module, the 104-order nECM-RNN model achieves a parameter estimation average error of less than 5%, a port voltage estimation error of less than 1%, and a DC internal resistance estimation error of approximately 0.2%. The nECM-RNN leverages the advantages of both data-driven and model-driven approaches, and by adjusting the order n, it can be applied to battery modules of different scales. This provides a scalable method for the online estimation of battery module parameters in energy storage stations, offering a flexible solution for battery module management.