A Reliable, Physics-informed Model for Assessing Battery State of Health Across Diverse Battery Types

키워드

확장요약문

Reliable State of Health (SoH) assessment of lithium-ion batteries must remain accurate across aging states, operating conditions, and battery types, yet data-driven models are mostly validated under a single condition. This study proposes a physics-informed rate-capability hybrid (PIRCH) that combines data-driven learning with electrochemical modeling. As illustrated in Fig. 1, the hybrid multiplies a data-driven aging term at a reference discharge rate by a rate-capability factor from a calibrated Doyle–Fuller–Newman model, whose transport coefficient is linked to the measured internal resistance. Across 25 cylindrical and prismatic cells from three datasets, recurrent and attention-based networks beat a persistence baseline in at most 18% of free-running forecasting cases. On measured 1.5 C data under leave-one-cell-out validation, the hybrid achieves the lowest mean absolute error of all tested models, 0.194 Ah, six times lower than without rate correction. In simulated extrapolation to 2–3 C, R2 falls from 0.9999 to 0.08 when the physics parameterization is mismatched, confirming that reliability rests on cell-specific calibration.