High-fidelity physics-informed adaptive unscented Kalman filter for lithium titanate battery state-of-charge estimation under extreme sub-zero environments

Abstract

Accurate State-of-Charge estimation for Lithium titanate batteries under extreme sub-zero conditions is critical yet challenging for polar-region electric vehicles, owing to severe electrochemical polarization and highly flat open-circuit voltage plateaus. Conventional filtering algorithms frequently suffer from tracking divergence or numerical degradation caused by uncompensated parameter lookup anomalies at sub-zero temperatures. To resolve this dilemma, this paper proposes a high-fidelity physics-informed Adaptive Unscented Kalman Filter framework. First, a physically smoothed and Arrhenius-corrected parameter lookup base is established to immunize the model against localized parameter collapses. An online recursive least squares parameter loop is then coupled to dynamically track parameter variations induced by polarization. Crucially, an innovation-residual-driven covariance optimization with a dynamic current trigger threshold is integrated to adaptively regulate the measurement noise matrix. Extensive validation under multi-temperature extreme conditions (30 °C, 0 °C, and −25 °C) demonstrates that the proposed framework dramatically suppresses tracking drift in open-circuit voltage (OCV) flat zones. Notably, under the extreme −25 °C environment, the proposed method constrains the maximum absolute State-of-Charge (SOC) error within 1.50% and reduces the Root Mean Square Error to 0.84%, outperforming the baseline Extended Kalman Filter (3.47% maximum error) and standard Unscented Kalman Filter (2.50% maximum error). The results prove that the framework delivers superior robustness for sub-zero battery management systems.