Electrochemical Expert-Free Battery Remaining Capacity Estimation via Impedance Spectroscopy and CNN-Attention-BiLSTM Deep Learning Frameworks

Health assessment is a critical function in lithiumion battery management. Nowadays, methods for estimating battery remaining capacity based on electrochemical impedance spectroscopy (EIS) show promising development prospects. However, most existing studies still require electrochemical expertise and complex mechanism analysis. To address this limitation, this study proposes a fully automated data-driven remaining capacity estimation framework, called the convolutional neural network - attention - bidirectional long short-term memory (CNN-Attention-BiLSTM) model. The raw EIS data is fed directly into the proposed model, and the hyperparameters of the model are automatically globally optimized by the Coyote Optimization Algorithm (COA). The model achieves Root Mean Square Errors (RMSE) <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;1 \text{mAh}$</tex>, demonstrating its high accuracy. Furthermore, this framework identified the most critical frequency (0.05 Hz) by evaluating the contribution of features to the model and demonstrated that the increase in diffusion impedance is the key indicator of battery aging.