An explainable state of health estimation method for sodium-ion batteries based on Kolmogorov-Arnold networks

Sodium-ion batteries (SIBs), due to their low cost, abundant raw materials, and environmental benefits, are emerging as a viable alternative to lithium-ion batteries. Nevertheless, SIB-specific degradation mechanisms, together with limited data availability, continue to constrain state-of-health (SOH) estimation accuracy, cross-condition generalization robustness, and model interpretability in energy-storage applications. Therefore, this paper proposes a multilayer interpretable framework based on a Kolmogorov–Arnold Network (KAN) for precise SOH estimation. The study first constructs an SIB aging dataset through experiments and extracts 30 basic features. Subsequently, secondary feature selection and interpretability analysis are performed using the Integrated Gradients method to explore the relationship between features and the battery's physical degradation mechanisms. Based on this, a Temporal Residual KAN-LSTM model is designed, integrating the KAN module with residual connections along the time step direction. Experimental results showed that the model, with only 2780 parameters, achieves the maximum error of less than 1.55 %, demonstrating outstanding prediction efficiency. Furthermore, symbolic formulas and feature importance analysis reveal the key decision-making logic of the model, providing strong support for enhancing model interpretability. This study not only offers an efficient solution for SIB management but also sets a new benchmark for developing battery SOH models that combine high interpretability with predictive capability.