A Physics-Informed Hybrid Data-Driven Approach With Generative Electrode-Level Features for Lithium-Ion Battery Health Prognostics

Lithium-ion (Li-ion) battery health management is crucial for ensuring the safety and stability of electronic products, particularly in estimating remaining useful life (RUL). To achieve rapid and accurate prognostics and improve model interpretability, this article proposes a physics-informed hybrid data-driven approach with generative electrode-level features for battery health prognostics. Initially, an electrochemical-informed data generative model is developed to reconstruct battery electrode-level state. Subsequently, features are extracted from cell-level aging states and the synthetic aging data to enhance interpretability. Furthermore, a physics-informed hybrid neural network (PIHNN) is introduced to integrate electrode-level aging states with cyclic cell-level features for battery RUL prediction. Validation is performed using four battery datasets, demonstrating the high accuracy, feasibility, and real-time performance of the proposed method and different battery aging modes can be identified by the proposed method effectively.