A hierarchical adaptive extended Kalman filter algorithm for lithium-ion battery state of charge estimation. (June 2023)

• Record Type: Journal Article
• Authors: Wang, Dongqing; Yang, Yan; Gu, Tianyu
• Is Part Of: Journal of energy storage. Volume 62(2023)
• Publication Date: 2023-06
• Subjects: Lithium-ion battery -- State of charge (SOC) -- Extended Kalman filter (EKF) -- Sage-Husa estimator -- Hierarchical identification principle; Energy storage -- Periodicals; Energy storage -- Research -- Periodicals; 621.3126
• DOI: 10.1016/j.est.2023.106831
• Languages: English
• ISSNs: 2352-152X

Abstract

Abstract: State of charge (SOC) is a key state in the battery management system (BMS). For a second-order equivalent circuit model (ECM), a hierarchical adaptive extended Kalman filter (HAEKF) algorithm is investigated for SOC estimation. Firstly, by a Sage-Husa estimator updating process noise online, an adaptive EKF algorithm is adopted to improve SOC estimation. Then, by the hierarchical identification principle, the circuit state equation model is decomposed into two fictitious state equation submodels with different sampling rates to solve the fast/slow dynamic problem existing in the two different resistance-capacitance (RC) networks, respectively, in which states are alternatively estimated by the HAEKF algorithm. In the two dual-rate submodels, the faster sampling rate for the fast dynamic submodel ensures the timely state updates, the slower sampled rate for the slow dynamic submodel reduces the computational burden and error accumulation. Finally, under the urban dynamometer driving schedule (UDDS) test condition, experimental results verify that the HAEKF algorithm has high SOC estimation accuracy, lower computational cost, and strong robustness under left biased measurement noise variance. Highlights: The state equation model is decomposed into two submodels with dual-rate sampling. The states in two submodels are alternatively estimated by the HAEKF algorithm. The Sage-Husa estimator is adopted to online update the process noises. The investigated algorithm has high estimation accuracy, lower computational load. The investigated algorithm is robust under left biased measurement noise variance.