Battery degradation trajectory early prediction with degradation recognition and physics-guided under different charging strategies

The degradation trajectories of lithium-ion batteries vary significantly depending on charging strategies. Accurate prediction of their long-term performance is crucial for optimizing charging management and extending battery lifespan. However, due to the constraints of real-world operating conditions, it is usually difficult to obtain a large volume of fully labeled battery data, and achieving high-precision early-stage predictions remains challenging. This study proposes an early-stage degradation trajectory prediction method that integrates degradation pattern recognition with physics-guided. First, optimal charging segments are extracted using a sliding window technique, followed by K-means clustering to identify three typical degradation modes. Second, polynomial fitting and Monte Carlo simulation are employed to augment the limited data from three reference batteries. Finally, a transfer learning framework is designed by incorporating two physical constraints into the loss function. The proposed approach successfully achieves accurate degradation trajectory prediction for 124 target batteries. The mean values of RMAEs and MAPEs are only 0.0031 Ah and 0.215 %. This study provides a novel approach for battery health management.