Transfer from Lithium to Sodium: Promoting Battery Lifetime Prognosis Application

Abstract

Lithium-ion batteries (LIBs) have achieved substantial progress; however, the limited availability of lithium resources poses a significant challenge to their continued scalability. Sodium-ion batteries (SIBs) offer a promising, cost-effective alternative, yet their widespread adoption is hindered by limited research—particularly in performance degradation and health management. This knowledge gap reinforces a development bottleneck, in contrast to the extensive diagnostic and prognostic studies available for LIBs. To address this, we construct a comprehensive SIB cycling life dataset and propose a transfer learning-based prognostic framework. Our approach utilizes knowledge from LIBs to speed up prognostics development for SIBs, while enhancing modeling flexibility and prediction accuracy. Central to this method is a Dual-Dynamic Mode Decomposition (Dual-DMD) model that captures both shared degradation behaviors and battery-specific deviations. By extracting universal features from LIB data and modeling SIB-specific characteristics, effective knowledge transfer is achieved. An online transfer factor optimization mechanism further mitigates distributional discrepancies between LIBs and SIBs. Degradation trajectories are then predicted using an adaptive unscented Kalman filter (AUKF). The proposed Dual-DMD framework achieves high predictive accuracy, with average mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean square error (RMSE) of 5.2%, 4.1%, and 4.8%, respectively, across all SIB test samples.