Learning of Battery Model Bias for Effective State of Charge Estimation of Lithium-Ion Batteries

Abstract

State of charge (SOC) estimation of lithium-ion batteries has been extensively studied and the estimation accuracy was mainly investigated through the development of various battery models and dynamic estimation algorithms. All battery models, however, contain inherent model bias due to the simplifications and assumptions, which cannot be effectively addressed through the development of various algorithms such as Kalman filtering (KF) or particle filtering (PF). Consequently, as observed in some study, battery SOC estimation using a typical extended KF in fact is not very accurate where the error could range from 5% to 10% or even more depending on the battery characteristics. This paper proposes bias characterization of the battery model, so that accuracy of the baseline model could be significantly improved and eventually SOC estimation could be much more accurate than the one only using the baseline model. This paper reports great potential for improving battery SOC estimation with the bias characterization and proposes two methods for actual bias modeling. In particular, the polynomial regression model and the Gaussian process regression model are proposed to examine the effects of the two methods on bias modeling and SOC estimation using a typical battery circuit model. Results are demonstrated in both simulation and lab testing using three battery charging/discharging profiles with the cross-validation technique.