Degradation Prognosis for Fast-Charging Batteries via Improved Domain Adaptation

As the market demand for fast-charging lithium-ion batteries grows, the ability to prognose battery degradation based on high-rate charging profiles becomes more valuable, offering a potential solution to alleviate consumer anxiety about the safety and reliability of electric vehicles. However, accurate degradation prognosis remains challenging due to the high variability of charging protocols across cells. To this end, this study proposes a degradation prognosis approach for fast-charging batteries with improved domain adaptation. Incremental capacity analysis of primary phase transition is determined by partial charge voltage within 8 min. Then, a probabilistic neural network is constructed to collaboratively predict two key degradation metrics. A domain-adaptation method that incorporates uncertainty-aware pseudolabeling is introduced, enabling more accurate modeling of battery degradation across varying fast-charging protocols and cell types. Validation using two battery datasets demonstrates that future degradation metrics can be predicted with high accuracy, reduced uncertainty, and lower computational costs. The research outcome also suggests that the data-driven model can be more transparent and reliable by encoding charging voltage into incremental capacity rather than directly predicting battery performance.