PiDDM Improves Battery SOH Prediction with Physics

Accurate prediction of a lithium-ion battery's State-of-Health (SOH) is crucial for reliable energy storage systems. However, purely data-driven models often struggle with generalization across different cycling protocols and can produce physically implausible results during long-term predictions, such as non-physical capacity regeneration.To overcome these limitations, a new framework called PiDDM (Physics-Informed Differentiable Degradation Modeling) has been introduced. PiDDM enhances neural network training by embedding empirical Arrhenius degradation kinetics, specifically related to solid electrolyte interphase growth and loss of lithium inventory, directly into the training objective. This ensures that the model's capacity fade predictions are physically consistent under various operating conditions.Evaluated on a public dataset of 55 batteries across six operating protocols, PiDDM achieved the lowest average prediction error among tested models. It substantially reduced mean squared error compared to a multilayer perceptron and a baseline physics-informed neural network. Crucially, during extrapolation (predicting the final 10% of cycle life after training on the first 90%), PiDDM accurately captured accelerated end-of-life degradation, avoiding the nonphysical capacity regeneration seen in baseline models.