基于物理的锂离子电池电化学模型参数辨识及其双种群优化方法

Abstract Pseudo-two-dimensional (P2D) models are increasingly promising for battery management systems due to their high accuracy, rooted in physical principles. However, their efficacy is hindered by the challenge of accurately identifying multiple parameters, and they often occur non-convergence. Traditional data-driven methods for parameter identification in P2D models, while advanced, are data-intensive and lack essential physical insights, which may lead overfitting. To address these challenges, this study firstly conducts parameter sensitivity analysis to determine the optimal conditions for identifying various parameter types. We then introduce a two-population multi-objective optimization algorithm to efficiently isolate a non-dominated parameter set. This algorithm uniquely incorporates non-convergent populations to enhance the update process of the wolf population, boosting both the effectiveness and reliability of parameter identification. Finally, the solution selection strategy is proposed by utilizing the physical knowledge to accurately identifies 23 parameters of the P2D model. The numerical validation and experimental validation are conducted. The the average percentage error between the identified parameter values and the reference parameter values are compared and verified the effectiveness of two-population multi-objective optimization algorithm and the identification strategy. Experimental validation under different operating conditions demonstrates a significant reduction in the root mean square error . Especially in dynamic operating conditions, the errors are all under 9 mV, affirming the method's precision in battery voltage prediction.