A framework for joint SOC and SOH estimation of lithium-ion battery: Eliminating the dependency on initial states

Model-based methods are prevalently used for estimating battery states, which forms the foundation of battery management systems. However, the efficacy of these methods relies on accurate initial state settings, and inaccuracies can precipitate substantial instability and even divergence, thereby posing serious threats to battery safety. This issue is particularly acute in the joint estimation of state of charge (SOC) and state of health (SOH) due to their interdependence. This study endeavors to obviate this dependency on initial states. Initially, two radial basis function auto-regressive models with exogenous inputs (RBF-ARXMs) are formulated to capture battery nonlinear dynamics and establish correlations between SOC, SOH, and observations. Building on these models, we derive effective target distributions and advocate for applying a Markov chain Monte Carlo method to sample and infer initial SOC and SOH values. Finally, a co-estimation method for SOC and SOH is developed, utilizing the unscented Kalman filter in conjunction with an RBF-ARXM. Validation on the Oxford and NASA degradation datasets demonstrates that the proposed methods achieve reliable initial state inference and SOC-SOH co-estimation throughout the batteries’ life cycle. Additionally, validation on the A123 dynamic driving dataset shows that our methods provide accurate initial SOC inference and SOC estimation under complex conditions.