Scholarlink

The growing demand for lithium-ion batteries, projected by Statista to increase sevenfold between 2022 and 2030 and reach 4.7 terawatt-hours, necessitates secure battery management. This includes the accurate estimation of battery states, particularly the State of Charge (SOC). Current research often relies on Recurrent Neural Network models that consider variable conditions, which limits their ability to adapt to real-world scenarios. To overcome this limitation, we propose a novel approach that couples a Doyle–Fuller–Newman (DFN) electrochemical model with a Transformer architecture, utilizing both synthetic data and an online dataset. Our model demonstrates promising results, with an Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) of 1.27% and 0.86%, respectively.

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

Generalizing SOC Estimation: A Hybrid Physics-Guided Deep Learning Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hafsa Khayrane,
  • Asma R’guibi,
  • Mohamed Louzazni

摘要

The growing demand for lithium-ion batteries, projected by Statista to increase sevenfold between 2022 and 2030 and reach 4.7 terawatt-hours, necessitates secure battery management. This includes the accurate estimation of battery states, particularly the State of Charge (SOC). Current research often relies on Recurrent Neural Network models that consider variable conditions, which limits their ability to adapt to real-world scenarios. To overcome this limitation, we propose a novel approach that couples a Doyle–Fuller–Newman (DFN) electrochemical model with a Transformer architecture, utilizing both synthetic data and an online dataset. Our model demonstrates promising results, with an Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) of 1.27% and 0.86%, respectively.