Transferable state-of-charge prediction across temperature domains for lithium-ion and semi-solid batteries

Accurate state-of-charge (SOC) estimation is essential for battery management systems (BMSs) in electric vehicles (EVs) and energy storage systems. The rapid innovation in battery technologies and the complexity of operating environments pose significant challenges to SOC estimation. This paper introduces an application-oriented TabPFN-based SOC estimation framework that adapts a pretrained tabular foundation model to sequential battery data for efficient SOC prediction. A sliding-window feature construction strategy is employed to transform historical current, voltage, and temperature measurements into structured tabular features and capture short-term temporal dependencies. The proposed framework was evaluated on both public and self-made datasets. Notably, the self-made datasets include not only traditional lithium-ion batteries (LIBs) but also advanced semi-solid batteries (SSBs). The results suggest that the TabPFN-based framework has promising applicability for SOC estimation under the tested battery types, temperatures, and driving profiles. Experimental results show that, compared with selected baseline models, this approach achieves high prediction accuracy while reducing task-specific model training and hyperparameter-tuning efforts, suggesting its potential for practical SOC estimation.