Variational Bayesian and Multi-Kernel Correntropy-Based Unscented Kalman Filter for Battery SOC Estimation

Accurately estimating the state of charge (SOC) is crucial for achieving optimal energy management and ensuring safety in Lithium-ion batteries. However, the precision of SOC estimation is significantly affected by the complexity and variability of the operating environment, including model inaccuracies, communication interference, and variations in load and temperature. To tackle these challenges, we develop a new SOC estimation framework that combines electrochemical modeling with an adaptive and robust Kalman filter. Specifically, a state-space model grounded in electrochemical mechanism is constructed to improve modeling fidelity. Building upon this, we propose a novel filtering algorithm, termed the variational Bayesian and multi-kernel correntropy-based unscented Kalman filter (VBMMKC-UKF), which innovatively integrates variational Bayesian (VB) approximation with the multi-kernel correntropy (MKC) criterion. The VB strategy introduces adaptivity by approximating and updating the measurement noise covariance, while the MKC criterion enables robustness by effectively handling non-Gaussian noise. Additionally, we further incorporate a newly designed smoothing matrix into the filter framework to construct a pseudo-measurement, effectively suppressing outliers and enhancing estimation accuracy. Extensive simulations on real-world datasets with varying dynamics and noise conditions clearly demonstrate the proposed filter’s effectiveness and robustness in SOC estimation.