An Interpretable TCN– Transformer Framework for Lithium‐Ion Battery State of Health Estimation Using SHAP Analysis

ABSTRACT Accurate state of health (SOH) estimation of Li‐ion batteries is essential for ensuring safety, reliability, and prolonging battery lifespan in energy storage systems and electric vehicles. This study proposes a hybrid temporal convolutional network (TCN)–transformer framework that effectively captures both short‐term temporal dynamics and long‐term degradation trends in battery degradation. To enhance model interpretability, SHAP (SHAPley Additive exPlanations) analysis is incorporated, allowing a detailed quantification of each input feature's contribution, such as internal resistance, capacity, Constant current charge time, and constant voltage charge time, to SOH estimation. Experimental validation using the dataset from University of Maryland demonstrates that the proposed model outperforms traditional methods, including long short‐term memory, recurrent neural network, gated recurrent units, transformer, and TCN, across various evaluation metrics (mean absolute error, mean squared error, root mean squared error, R ). Beyond improved predictive accuracy, SHAP analysis offers valuable insights into battery degradation mechanisms by highlighting nonlinear relationships between key features and SOH estimation results. It reveals critical factors, including electrochemical aging, capacity fading, and voltage instability, that influence battery health. This interpretable and robust framework not only enhances SOH estimation performance but also supports the development of more reliable battery management systems by providing actionable insights into degradation patterns and health indicators.