基于EIS与混合专家模型的锂离子电池SOH估计方法OA

State of health(SOH)constitutes a crucial metric for quantifying performance degradation in lithium-ion batteries,with accurate estimation being paramount for ensuring energy storage sys-tem reliability.To overcome the limitation of conventional voltage-current approaches requiring ex-tensive charge/discharge cycle monitoring in practical scenarios,a portable impedance measurement system was leveraged to acquire electrochemical impedance spectroscopy(EIS)data across 20 opera-tional conditions during battery aging experiments.Eight health indicators(HIs)correlated with SOH were established and validated via Pearson correlation analysis.Capitalizing on the rich infor-mation content of EIS,a specialized mixture of experts(MoE)framework was proposed wherein a two-dimensional convolutional neural network(2D CNN)-based gating network predicts state of charge(SOC)labels using real/imaginary EIS components and voltage data at 40 discrete frequen-cies,while twenty genetic algorithm(GA)-optimized GRU expert sub-networks estimate SOH based on condition-specific HIs.This arcHItecture decouples SOC effects on EIS through gating-controlled expert activation,utilizing SOC labels for branch selection.By implementing 10%SOC interval parti-tioning,the model comprehensively spans equilibrium and non-equilibrium battery states across di-verse operating profiles.Requiring under 2 minutes for single-estimation data acquisition,the solu-tion demonstrates efficient temporal performance.Experimental validation yields R²=0.949 8,RMSE=0.003 2,and MAPE=0.26%on test sets,confirming satisfactory SOH estimation accuracy and demon-strating its potential for practical deployment.