Adaptive model parameter identification for large capacity Li-ion batteries on separated time scales. (15th December 2016)

• Record Type: Journal Article
• Authors: Dai, Haifeng; Xu, Tianjiao; Zhu, Letao; Wei, Xuezhe; Sun, Zechang
• Is Part Of: Applied energy. Volume 184(2016)
• Page Start: 119
• Page End: 131
• Publication Date: 2016-12-15
• Subjects: Battery dynamics -- Multi time-scaled effect -- Adaptive parameter identification -- Recursive least squares -- Extended Kalman filtering; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2016.10.020
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A novel adaptive parameter identification algorithm for Li-ion batteries is proposed. The algorithm is composed of two modules running on separated time scales. Different battery dynamics are identified with different modules separately. Coupling of the two modules is through the voltage response of slow dynamics. Time scale for the slow identification is dependent on current profiles. Abstract: The accurate identification of battery model parameters is critical to the development of the battery management system (BMS). For large capacity Li-ion batteries, different internal processes happen inside the cell during charging and discharging, which introduce the complex dynamics that occur on different time scales. The multi time-scaled effect of the battery dynamics imposes difficulties on the design of an accurate parameter identification algorithm. As an original contribution, we propose a novel adaptive identification algorithm of the model parameters on separated time scales. The battery dynamics are described with a second-order ECM (equivalent circuit model), where the slow dynamics and fast dynamics are described separately. The parameter identification algorithm is composed of two separated modules, of which one is for the identification of slow dynamics and the other is for the identification of fast dynamics. The two modules are executed on separated time scales. The identification module for slow dynamics is based on extended Kalman filtering (EKF) while the module for fast dynamics is based on recursive least squares (RLS). The coupling of the two modules is through the voltage response of the slow dynamics. To make the algorithm more adaptive, the operation time scale of the slow identification module is not constant, but dependent on current profiles. Validation with experimental results shows that the proposed identification strategy performs better than the traditional RLS based identification methods.