Abstract
Electrochemical Impedance Spectroscopy (EIS) is a crucial method for assessing the aging and safety of lithium-ion batteries. However, existing methods for obtaining EIS are time-consuming and costly in terms of hardware. Current data-driven EIS estimation methods face challenges of weak interpretability and low reliability. We propose a reliable EIS estimation pipeline based on an encoder–decoder model. The method is designed based on constructing a set of physics-guided dual-stage deep learning networks using the intrinsic geometric features of EIS for reliability. Additionally, an outlier removal algorithm is designed based on Linear Kronig-Kramers validation for reliability. Experiments on two datasets not only achieved average estimation RMSEs below 1.6 mΩ and 0.62 mΩ, respectively, but also demonstrated the excellent estimation performance and accuracy of the proposed method.
•Design a Lin-KK outlier removal algorithm for battery EIS estimation.•Develop a physics-guided deep neural network for battery EIS estimation.•Link charging curves to EIS using geometric features and SOC.•Design an optimized training process, validated on two datasets.
