Bayesian parameter identification in electrochemical model for lithium-ion batteries

Abstract

Electrochemical models can characterize the internal behavior of cells and are powerful and effective tools for the design and management of batteries. This study proposes a comprehensive framework of Bayesian parameter identification to determine the parameter distributions in the electrochemical model and to estimate the global sensitivity of the parameters for lithium-ion batteries. Bayesian inference in parameter identification can simultaneously determine accurate parameter estimates and parameter identifiability, given specific voltage measurements. Among the several parameters in the pseudo-two-dimensional model, 15 parameters are selected to estimate the posterior distributions. The reconstructed voltage curves through the estimated parameter distributions are consistent with the reference voltages, with relative errors of less than 0.7% at various discharge rates. Changes in the parameter distributions and identifiability were investigated for different discharge rates through the estimated joint and marginal distributions of the parameters. Moreover, based on variance-based global sensitivity analysis, the identifiability of the electrochemical parameters according to the discharge rates is quantitatively analyzed. We demonstrate that the Bayesian parameter identification simultaneously obtains the parameter distributions and identifiability, considering the correlation between various parameters. The proposed framework can help to analyze the behaviors of batteries according to specific operating conditions and materials.