An Ensemble Bias-Correction Method With Adaptive Weights for Dynamic Modeling of Lithium-Ion Batteries

Accurate modeling of the electrical behavior of a lithiumion (Li-ion) battery can provide accurate dynamic characteristics of the battery during charging/discharging and relaxation phases, which is essential to accurate online estimation of the battery state of charge (SoC). This paper proposes an ensemble bias-correction (BC) method with adaptive weights to improve the accuracy of an equivalent circuit model (ECM) in dynamic modeling of Li-ion batteries. The contribution of this paper is twofold: (i) the development of a novel ensemble method based on BC learning to model the dynamic characteristics of Li-ion batteries; and (ii) the creation of an adaptive-weighting scheme to learn online the weights of offline member BC models for building an online ensemble BC model. Repeated pulsing tests with single and multiple C-rates were conducted on seven Li-ion battery cells to evaluate the effectiveness of the proposed ensemble BC method. The analysis results with the use of an ECM demonstrate that the proposed method can reduce, on average, the voltage modeling error of the ECM by at least 50%.