A dual-objective data-driven framework combining Bayesian optimization and improved differential evolution for rapid and accurate parameter identification of lithium-ion battery P2D models

The electrochemical pseudo-two-dimensional model for lithium-ion batteries offers high accuracy and strong physical interpretability, making it widely used in battery diagnostics, lifetime prediction, and fast-charging control. However, the P2D model involves a large number of parameters that are difficult to determine accurately through experiments. Furthermore, due to the model's highly nonlinear nature, the parameter identification problem is often ill-posed, with multiple parameter combinations capable of fitting the same experimental data. In this study, we propose a novel parameter identification framework that combines Bayesian Optimization with an improved Differential Evolution algorithm. For the first time, both constant-current discharge data and state-of-charge versus open-circuit voltage data are simultaneously used as dual-objective inputs for optimization. The identification results are evaluated based on the fitting errors of both terminal voltage and open-circuit voltage, which enhances the accuracy of parameter estimation and significantly reduces the identification time. All 21 electrochemical parameters can be identified within 2.5 h. The proposed method is validated using a series of tests at 25 °C, including multi-rate charge/discharge experiments, Dynamic Stress Test, Federal Urban Driving Schedule, Hybrid Pulse Power Characterization, and open-circuit voltage measurements. The identified pseudo-two-dimensional model demonstrates excellent agreement with experimental data, with a voltage error below 5 mV and a SOC error below 0.5 %. The mean absolute error of the model-predicted voltage under dynamic operating conditions is less than 12.7 mV, further demonstrating the accuracy and effectiveness of the proposed identification method.