Novel state of charge estimation of lithium-ion batteries using empirical neural tangent kernel and unscented Kalman filter

Abstract

Accurate state of charge (SoC) estimation is essential for optimizing the performance and safety of rechargeable batteries. Existing methods often struggle with dynamic battery conditions. This study proposes a hybrid approach that combines empirical neural tangent kernel (eNTK) regression with an unscented Kalman filter (UKF) for enhanced SoC estimation. The proposed method identifies the battery model using eNTK regression, splits the obtained regression model to construct prediction and measurement models, and integrates them into a UKF framework. Extensive validation over various driving cycles, temperature, and state of health conditions demonstrate significant enhancements in SoC estimation accuracy, reducing mean square error by an average of 75.35% and the average standard deviation by an average of 47.11%, compared to the EKF using the 2RC model, which incorporates hysteresis characteristics. © 2025 Elsevier Ltd