Physics-guided multi-source collaborative state of charge estimation for lithium iron phosphate batteries

Lithium iron phosphate (LFP) batteries are widely deployed in electric vehicles and energy storage systems. However, accurate state of charge (SOC) estimation is limited by the weak observability of the voltage plateau region and thermo-mechanically coupled drift of surface force. To address these challenges, a physics-guided multi-source collaborative method (P-MSCM) is proposed. Incremental force analysis (IFA) is employed to extract intrinsic mechanical features, which are further combined with voltage gradients to construct a multi-source physical feature set. Shannon entropy is utilized to dynamically quantify the signal confidence of both electro-mechanical dimensions. These real-time evaluations are then transformed into Bayesian prior constraints to guide the variational Bayesian adaptive extended Kalman filter (VB-AEKF), achieving optimal adaptive measurement noise covariance reconstruction. Validations across a wide temperature range (15 °C–45 °C), multiple aging states (state of health≈100% and 91%), and dynamic load profiles demonstrate the superiority of the proposed method. The P-MSCM reduces the root mean square error by up to 84.1% compared to conventional single electrical models, maintaining dynamic estimation errors within ±2.0% even under severe multi-physics coupled interferences. These results confirm that the proposed method achieves high accuracy and strong robustness under complex operating conditions.