Abstract: Lithium metal batteries(LMBs) have emerged as a promising next-generation energy storage equipment due to their ultra-high energy density. However, accurately predicting their state of health(SOH) remains challenging owing to complex degradation mechanisms. An electrochemical impedance spectroscopy(EIS)-based cross-frequency-band multi-source feature extraction methodology is proposed. By leveraging grey relational analysis(GRA) to quantify the nonlinear correlation between EIS characteristics and SOH, five critical frequency-domain features in the mid-to-high-frequency region were screened. Concurrently, a simplified equivalent circuit model was constructed to extract five physical parameter features from the low-frequency region, reflecting solid-phase diffusion and charge-transfer processes. Integrated with a Gaussian process regression(GPR) model, an EIS-SOH estimation framework for LMBs under multi-condition variable coupling was developed. The proposed method demonstrates strong accuracy and robustness for SOH estimation across diverse mechanical preload forces, charge/discharge C-rates, and battery states of charge(SOC). Experimental validation under 12 distinct variable combination scenarios yielded an average root mean square error(RMSE) of 1.65% for SOH prediction, while the RMSE remains below 3% even under extreme operating conditions.

Key words: lithium metal batteries, state of health estimation, electrochemical impedance spectroscopy, multi-source feature extraction, Gaussian process regression