Adaptive Fusion SOH Assessment of Lithium-Ion Batteries Based on EIS Multi-Band Features

Chao Li, Shunli Han, Luo Zhao, Zunheng Yang, Fei Li

Electrochemical impedance spectroscopy (EIS) can quantitatively reflect internal aging mechanisms of lithium-ion batteries, serving as an effective non-destructive tool for State of Health (SOH) evaluation. Single prediction models either fail to balance precision for fresh and aged cells or lack adaptability under variable temperature and SOC conditions. This paper proposes an adaptive dual-model fusion SOH estimation method combining weighted Mahalanobis-distance KNN and power-law capacity fading model. Four-dimensional features extracted from medium-high frequency EIS, real-time SOC and ambient temperature are standardized via Z-score transformation. Dynamic weight allocation is implemented according to battery health intervals: higher weight is assigned to KNN for cells with SOH ≥ 95%, while power-law model dominates for severely degraded cells with SOH < 95%. Multi-temperature full-cycle aging tests on 314 Ah LiFePO4 cells verify the proposed method. The full-life SOH prediction error is controlled below 5%, which outperforms single KNN and pure power-law fitting. This fusion algorithm can be embedded into embedded EIS measurement hardware, supporting online health diagnosis for energy storage and vehicle power batteries.