Non-destructive and rapid parameter identification of a simplified electrochemical model for lithium-ion batteries via multi-step and physical-informed methods

Accurate parameter identification of simplified electrochemical models for lithium-ion batteries (LIBs) is crucial for battery management and control. However, existing methods often struggle with parameter coupling, computational efficiency, and physical consistency. This paper presents a non-destructive and rapid parameter identification methodology through an integration of multi-step (MS) and physical-informed (PI) approaches. First, Fisher information matrix-based identifiability analysis reveals parameter coupling relationships, enabling model reconstruction through parameter aggregation. Hierarchical clustering analysis then categorizes parameters into high and low sensitivity groups, establishing the foundation for MS identification. The proposed MS strategy uniquely addresses low-sensitivity parameter challenges through sequential optimization, while the PI method incorporates electrochemical constraints to ensure physically consistent results. An improved particle swarm optimization (IPSO) algorithm also significantly advances population diversity and search capabilities. Numerical validation demonstrates exceptional performance of the proposed identification framework, achieving a 29.73 % reduction in mean absolute percentage error of parameters compared to the baseline framework, with most parameters maintaining relative errors below 5 %. The proposed IPSO algorithm also has the best convergence characteristics and parameter identification results. Experimental validation under the dynamic stress test condition yields a mean absolute error of 10.12 mV and a root-mean-square error of 14.38 mV, with complete identification requiring only 13.94 seconds. The methodology's generalizability and practicality are comprehensively validated across diverse operating conditions, external datasets, multiple cathode materials, and even incomplete datasets. The proposed model and method hold considerable promise for extensive applications in adaptive battery control, performance evaluation, and health diagnosis systems.