Multi-physics coupling model parameter identification of lithium-ion battery based on data driven method and genetic algorithm

The increasing demand for electric vehicles necessitates accurate battery modeling to ensure performance, safety, and longevity. This study develops a comprehensive coupled mechanism model for lithium-ion batteries that integrates electrochemical, aging, and thermal phenomena. To address the challenge of identifying numerous unknown parameters within the model, a data-driven approach is employed. First, Latin Hypercube Sampling is employed to generate a diverse set of parameter combinations. Subsequently, the coupled mechanism model is simulated using these combinations to produce a dataset of macroscopic responses. This dataset is utilized to train an artificial neural network, creating a meta-model that significantly accelerates the optimization process. Following this, sensitivity analysis is conducted to identify the most influential parameters. Finally, a genetic algorithm is used to optimize these parameters, minimizing the discrepancy between model predictions and experimental data. Results reveal that among 33 model parameters, 9 high-sensitivity parameters and 10 medium-sensitivity parameters are identified as significantly influencing the model output. By refining these parameters, the model achieved mean absolute errors of 0.0147, 0.2132, and 0.0163 for voltage, temperature, and capacity simulations, respectively. These results demonstrate the high accuracy and effectiveness of the proposed approach, offering a robust and efficient method for lithium-ion battery modeling.