A Hybrid SWVMD-iTransformer Framework for Online SOH Estimation of Lithium-Ion Batteries

Accurate estimation of the State of Health (SOH) is crucial for lithium-ion battery management. However, the capacity regeneration phenomenon introduces strong nonstationary fluctuations, presenting a significant challenge for conventional methods in capturing both global degradation trends and local fluctuations simultaneously. To address these challenges, a novel hybrid framework that combines sliding window variational mode decomposition (SWVMD) with an iTransformer is proposed for reliable online estimation. The time interval during equal voltage increase (TEVI) is extracted as an online-accessible health feature and processed via SWVMD to generate multiscale degradation components while preventing information leakage. These components are then fed into the iTransformer model, which leverages feature-dimensional attention to capture the synergistic contributions between degradation trends and local fluctuations. The robustness of the TEVI feature is validated under incomplete charging and noisy conditions, and the sensitivity of the key parameters is systematically analyzed. Extensive experiments on CALCE and NASA data sets demonstrate that the proposed method achieves superior estimation accuracy, with RMSE and MAE below 1.5 and 1.2%, respectively, and a maximum error under 0.05. Furthermore, computational efficiency analysis confirms its potential for onboard implementation, with an average inference time under 3 ms per sample. This study provides a practical solution for online SOH estimation using operationally accessible features, while addressing information leakage.