Iterative Learning Based Model Identification and State of Charge Estimation of Lithium-Ion Battery

This work focuses on the accurate identification of Lithium-ion battery’s nonlinear parameters by using an iterative learning method. First, the 2nd-order RC model is introduced. Then, when the battery repeatedly implements a discharging trial from SOC 100% to 0%, an iterative learning based recursive least square (IL-RLS) algorithm is presented to accurately identify the nonlinear parameters of the regression model. The essential idea of IL-RLS algorithm is to improve the current parameter estimations by learning the estimation errors of the previous trails. Notably, the IL-RLS algorithm needs to be implemented offline for the long-time repetitive trials, which is the price worth paying to accurately identify the nonlinear parameters. After that, the parameters are identified as the functions of SOC by using the IL-RLS, which are verified by comparing with the result of the classic identification method for current pulses. Finally, by using the classic extended Kalman filter (EKF) as well as the parameters identified by the IL-RLS to estimate the SOC, three dynamic operation conditions are given to show the efficiency of the IL-RLS, where all the SOC estimation errors are less than 2%.