Adaptive State-of-Health Estimation for Lithium-Ion Battery With Partially Unlabeled and Incomplete Charge Curves

Abstract

State-of-health (SOH) assessment of lithium-ion batteries (LIBs) is essential for electric vehicles (EVs). The existing methods rely on exact capacity labeling for incomplete curves for model training. However, these capacity values cannot be obtained until the charge/discharge process is complete during real operations. Furthermore, the existing models cannot be efficiently updated with newly collected data, causing degenerated performance due to the heterogeneity among different batteries. To overcome these deficiencies, we propose a sequential variational Gaussian mixture regression (SVGMR) model, where the charge curve and capacity are jointly modeled with a Gaussian mixture model (GMM). Due to the generative nature of this model, the information provided by the unlabeled data can also be exploited using the conditional distribution based on observed data to improve the SOH estimation accuracy. In addition, a sequential updating algorithm is developed for online adjustment, which can efficiently assimilate newly collected data of the target battery to further boost the estimation. During the in-field application, the proposed technique can provide SOH estimation with uncertainty based on a random partial segment of the voltage curve. The effectiveness and superiority of the proposed method are validated with case studies.