Remaining Useful Life Prediction of Lithium-Ion Batteries Using Physics-Informed Gaussian Process Regression Under Multi-Phase Degradation

The widespread adoption of lithium-ion batteries in electric vehicles and energy storage systems has made accurate remaining useful life (RUL) prediction increasingly critical for system reliability and maintenance planning. This paper presents a novel physics-informed Gaussian Process Regression (GPR) framework for battery RUL prediction, specifically addressing the challenges of multi-phase degradation patterns. The proposed approach integrates a physics-informed mean function with an adaptive cross-scale kernel structure to enable robust prediction across different degradation phases. The framework combines a phase-aware mean function that explicitly models transition behaviors, an adaptive kernel design for temporal dependencies, and a covariance modulation mechanism that automatically adapts to local degradation characteristics. Experimental validation on both NASA and TOYOTA battery datasets demonstrates the framework's effectiveness across diverse aging patterns. Experimental validation demonstrates the framework's effectiveness across diverse aging patterns, achieving RMSE reductions of up to 75% on the TOYOTA dataset compared to traditional GPR approaches, while maintaining robust regeneration tracking capabilities on the NASA dataset despite its complex capacity fluctuations.