Uncertainty-Aware State Estimation : A Multi-Modal Signal Fusion Framework

Rasheed, Muhammad Babar and Khan, Inam Ullah and Ma, Xiandong (2026) Uncertainty-Aware State Estimation : A Multi-Modal Signal Fusion Framework. IEEE Transactions on Energy Conversion. ISSN 0885-8969

Text (main_IEEEPES_FINAL)
main_IEEEPES_FINAL.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (643kB)

Abstract

Despite significant advances in data-driven battery modeling, no existing framework simultaneously guarantees estimation accuracy, well-calibrated predictive uncertainty, and computational tractability for real-time battery management systems. This letter addresses this gap by proposing an uncertainty-aware multi-modal signal fusion framework that hierarchically extracts complementary features from voltage, current, and temperature measurements and adaptively combines them through linear synthesis, polynomial basis expansion, and kernel mapping. The fusion mechanism is underpinned by formal convergence guarantees, with weights converging at O(logk/k), ensuring provable robustness alongside computational efficiency. Experimental validation across three real-world electric vehicle datasets yields SOC RMSE of 0.07% and voltage RMSE of 0.1 p.u. (Vp.u.=Vmeasured/Vnom, corresponding to 0.36-0.42 V absolute error), and coverage deviation within 0.2 percentage points of the nominal 95% prediction interval, all within millisecond-scale cycle times. These results establish a new state-of-the-art in the joint optimisation of estimation accuracy, uncertainty calibration, and real-time BMS deployment.