State of health estimation of lithium‐ion batteries: A multiscale Gaussian process regression modeling approach

Abstract

Accurate state of health (SOH) estimation in lithium‐ion batteries, which plays a significant role not only in state of charge (SOC) estimation but also in remaining useful life (RUL) prognostics is studied. SOC estimation and RUL prognostics often require one‐step‐ahead and long‐term SOH prediction, respectively. A systematic multiscale Gaussian process regression (GPR) modeling method is proposed to tackle accurate SOH estimation problems. Wavelet analysis method is utilized to decouple global degradation, local regeneration and fluctuations in SOH time series. GPR with the inclusion of time index is utilized to fit the extracted global degradation trend, and GPR with the input of lag vector is designed to recursively predict local regeneration and fluctuations. The proposed method is validated through experimental data from lithium‐ion batteries degradation test. Both one‐step‐ahead and multi‐step‐ahead SOH prediction performances are thoroughly evaluated. The satisfactory results illustrate that the proposed method outperform GPR models without trend extraction. It is thus indicated that the proposed multiscale GPR modeling method can not only be greatly helpful to both RUL prognostics and SOC estimation for lithium‐ion batteries, but also provide a general promising approach to tackle complex time series prediction in health management systems. © 2015 American Institute of Chemical Engineers AIChE J, 61: 1589–1600, 2015