Prediction of state-of-health and remaining useful life for lithium-ion batteries using short-term relaxation voltage

Accurate estimation of state-of-health (SOH) and remaining useful life (RUL) is critical for battery management systems . However, existing methods often depend on complex feature engineering or extensive historical cycling data, posing challenges for real-time online monitoring. To address this limitation, we innovatively propose a dual-task collaborative Gaussian process regression framework based on short-term relaxation voltage (RV). By extracting RV data after a single charging cycle, online SOH estimation is achieved. Additionally, time-series voltage features are integrated to establish an RUL prediction model . Validation is conducted on 41 batteries with varying temperatures, C-rates, and chemistries. When using merely 10-minute RV data, the SOH prediction achieves MAE and RMSE below 1.3 % and 1.6 % respectively, while RUL estimation maintains MAE and RMSE within 50 cycles. Furthermore, incorporating voltage decay rate as an additional input further reduces errors. Finally, the transfer learning model enhanced with a linear transformation layer achieves an MAE below 2.8 % for SOH prediction and 65 cycles for RUL prediction. This work presents a lightweight, interpretable, and high-accuracy paradigm for battery health assessment under variable conditions. • This method enable accurate prediction of SOH and RUL within ten minutes. • The method requires no feature engineering or historical cycling information. • The generalization capability was validated through testing on 41 battery cells. • Incorporating voltage decay rate as an additional input further reduces errors. • A transfer learning model enhanced with a linear transformation layer is proposed.