A novel parameter adaptive method for state of charge estimation of aged lithium batteries

Abstract

The state of charge (SOC) of lithium battery is an important evaluation index for battery management system (BMS) in electric vehicles. In this study, a novel parameter adaptive method adapted to battery aging has been proposed for SOC estimation. In order to accurately estimate the SOC, the Thevenin model was firstly to established. Model parameters were then identified based on the recursive least square (RLS) method. In order to improve the accuracy and the self-update ability of the RLS method, an improved RLS method with an activation zone (RLS-AZ) was developed. The activation zone refers to an interval according to the error between the predicted value and measured value of the terminal voltage. When the error is in the interval, the covariance matrix of RLS is reset to the pre-defined value. In order to consider the influence of battery aging on the evolution of open circuit voltage (OCV), a back-propagation neural network (BPNN) between OCV and the SOC and capacity was established. Based on the identified parameters above, the unscented Kalman filter (UKF) was applied to estimate the SOC under degradation. The robustness and accuracy of the proposed method was verified, and the results show that the error of terminal voltage estimated by improved RLS method is reduced to one order of magnitude compared with the original RLS method. The SOC estimation error for the aged battery is less than 0.8% even if given an inaccurate initial value for estimator.