Revealing the degradation patterns of lithium-ion batteries from impedance spectroscopy using variational auto-encoders

The aging life estimation of lithium-ion batteries (LIBs) is of great significance to the use, maintenance and economic analysis of energy storage systems. The estimation method of aging life based on electrochemical impedance spectroscopy (EIS) has received more attention due to its high accuracy. Due to the high dimensionality of impedance data, it cannot directly reflect the decline trend, this paper proposes a method of combining variational auto-encoders (VAE) and bidirectional gated recurrent unit (BiGRU) for automatic feature extraction from EIS data and battery aging life estimation. Reducing the dimensionality of impedance data through VAE can greatly reduce computational complexity, and the high-quality features extracted are mapped to battery capacity using BiGRU, which can improve estimation accuracy. The validation results show that the method can adapt to various working conditions, and has good estimation accuracy and robustness. MAE and RMSE are less than 1.27 mAh and 1.43 mAh, respectively. We demonstrate that the latent variables extracted from EIS data using unsupervised learning method reliably represent the degradation patterns of LIBs.