Comparison of single particle and equivalent circuit analog models for a lithium-ion cell

The physics-based single particle (SP) model was compared to the semi-empirical equivalent circuit analog (ECA) model to predict the cell voltage under constant current charge and discharge for different sets of Li-ion cell data. The parameters of the models were estimated for each set of data using nonlinear least squares regression. In order to enhance the probability of finding the global optima, a combination of the trust region method with a genetic algorithm was applied to minimize the objective function (the sum of squared residuals). Several statistical quantities such as sum of the squared errors, adjusted R 2 , root mean squared error, confidence intervals of the parameters, and prediction bounds were included to compare the models. A significance test ( t test) on the parameters and the analysis of the variances ( F and χ 2 tests) were also performed to discriminate between the goodness of the fit obtained from the two models. The statistical results indicate that the SP model superiorly predicts all sets of data compared to the ECA model, while the computation times of both models are on the same order of magnitude.