Non-destructive and adaptive negative electrode impedance estimation of lithium-ion batteries using ensemble learning

The negative electrode (NE) impedance of lithium-ion batteries is a key indicator that reflects their internal electrochemical dynamics. Traditional invasive methods relying on reference electrodes (REs) fail to satisfy the demands of non-destructive, online monitoring in engineering applications. To overcome this limitation, this manuscript proposes a data-driven method based on ensemble learning to achieve non-destructive and adaptive estimation of NE impedance. The research integrates an improved dual-RE experimental design with ensemble learning algorithms. A total of 1050 electrochemical impedance spectroscopy (EIS) datasets are systematically acquired from two battery types within a temperature range of 0–45 °C and a state of charge range of 20 %–80 %. Features are extracted through equivalent circuit model analysis and distribution of relaxation times representation, and a precise mapping model is established to connect battery impedance with NE impedance. The proposed model achieves a coefficient of determination (R2) above 98.5 % for estimating NE polarization resistance. The predicted NE EIS curves yield a mean absolute percentage error (MAPE) below 8.1 %, while performance under unseen conditions maintains MAPE within 9.25 %, demonstrating great generalization ability. Moreover, based on predicted impedance features, a linear internal temperature estimation model is constructed. This approach reduces the mean absolute error by 14.5 % compared with conventional methods and exhibits strong adaptability across different battery capacities. This study provides a novel technical pathway for electrode-level parameter estimation, highlights the essential role of NE impedance in accurate state perception, and contributes to advancing intelligent battery management system.