Research on the performance of variable-parameter models for lithium iron phosphate batteries in rail transit systems

Complex multi-physics coupling processes occur within lithium batteries, and high-precision modeling can accurately describe the dynamic characteristics of these batteries, thus facilitating predictions regarding their state performance and degradation trends. Lithium iron phosphate batteries have become the dominant battery type in the rail transit sector due to their advantages of high safety and long cycle time. Based on a second-order resistor-capacitor equivalent circuit model, this paper proposes improved variable resistor-capacitor parameter models that can change the main parameters according to the state of charge (SOC) and current value of batteries, thus affecting the output of the models. Constant-current charging and discharging experiments, as well as pulse power characteristic experiments were conducted on lithium batteries to identify the required model parameters using the nonlinear least squares algorithm. The identified parameters were organized into tables, and the missing values were filled using the surface fitting tool and the <italic>k</italic>-nearest-neighbor nonparametric interpolation algorithm. This enabled the parameter values to be adapted to battery operating conditions across different SOC and current levels. The performance of the proposed variable-parameter models was verified, with the results indicating that the root mean squared error (RMSE) of the models is about 8 mV under fixed operating conditions. Furthermore, the RMSE of the <italic>k</italic>-nearest-neighbor interpolation model is about 20% of that of the surface fitting variable-parameter model under non-fixed operating conditions. Finally, the models’ accuracy was tested using the real data from a new energy locomotive. The comparison results show that the accuracy of the <italic>k</italic>-nearest-neighbor interpolation variable-parameter model surpasses that of the surface fitting model. This paper provides a new modeling approach for second-order variable-parameter models tailored for lithium iron phosphate (LiFePO₄) power batteries.