HYBRID MODELING OF ELECTRIC VEHICLE BATTERY DEGRADATION USING PHYSICS-INFORMED MACHINE LEARNING

Accurate prediction of battery degradation is crucial for ensuring the reliability, safety, and performance of electric vehicles (EVs). While data-driven machine learning (ML) models offer high prediction accuracy, they often lack physical interpretability, limiting their application in critical systems. On the other hand, purely physics-based models provide deeper understanding but struggle to generalize across diverse operating conditions. This paper proposes a hybrid modeling approach that combines physics-informed machine learning (PIML) with empirical battery aging data to achieve both accuracy and interpretability. The model incorporates domain knowledge—such as electrochemical degradation mechanisms, capacity fade laws, and thermal effects—into a learning framework based on recurrent neural networks (RNNs) and gradient boosting. Experimental results on real-world EV battery datasets demonstrate that the hybrid model outperforms standalone physics-based and ML models in both prediction precision and consistency. This approach opens new avenues for predictive maintenance, extended battery lifespan, and optimized battery usage strategies.