Interpretable deep learning with uncertainty quantification for lithium-ion battery SOH estimation

Abstract

The rapid proliferation of lithium-ion batteries in electric vehicles and grid-scale energy storage systems has underscored the critical need for advanced battery management systems, particularly for accurate state of health (SOH) monitoring for massive cells. This paper has proposed an interpretable deep learning framework for SOH prediction, in which a long short-term memory (LSTM) network optimized by the sparrow search algorithm (SSA) serves as the core predictor. To enhance the transparency and reliability of the model, Deep SHAP (SHapley Additive explanations) is employed to interpret the contribution of each health feature, and uncertainty is quantified through confidence intervals derived from stochastic seed variation. To validate the proposed model, experiments were conducted on 12 batteries from the NASA and CALCE public datasets, as well as a proprietary dataset at PolyU. The experimental results show that the proposed model significantly outperforms others, with RMSE, MAE, and MAPE all below 5%. This work supports the practical application of interpretable SOH estimation in battery management systems to improve the safety and reliability of energy storage system operations.