A Foundation-style Lithium-ion Battery Degradation Model with Varying OperatingCondition Adaptation

Accurate prognostics of lithium-ion batteries is essential for the reliability and safe operation of battery system, yet facing challenges brought by nonlinearity of capacity degradation trajectories, noise and sensitivity to dynamic operating conditions such as temperature and charge/discharge profiles. To this end, this paper presents a foundation-style degradation modelling framework that learns shared degradation patterns from historical capacity trajectories and then adapts efficiently to a new cell using its early-cycle observations together with the corresponding operating conditions. The proposed model combines a flexible data-driven degradation backbone with a physics-guided operating-condition adaptation mechanism, enabling robust extrapolation under varying conditions while retaining strong expressive capability. A practical pre-training and adaptation pipeline is established to support capacity forecasting and remaining useful lifetime (RUL) prediction for unseen cells. Experiments on CycleLife-SJTUIE dataset from our lab demonstrate that the proposed approach achieves superior capacity forecasting accuracy compared with several baselines and yields lower RUL prediction errors across multiple cells and evaluation metrics.

Keywords: Degradation modelling, RUL prediction, lithium-ion battery, operating-condition adaptation.