Research On State-of-Health Estimation Method For Lithium-Ion Batteries Based On Features Of Time-Frequency Domain

Lithium-ion batteries are increasingly used,and accurate state-of-health estimation plays a crucial role in improving the operating performance and safety performance of the battery system.In this paper,a systematic study of the lithium-ion battery state-ofhealth estimation method is conducted,and a complete set of feature extraction and stateof-health estimation method is proposed from the perspective of frequency domain data and time domain data respectively.First,for the feature extraction based on the frequency domain data,this paper proposes an equivalent circuit model parameter identification method based on partial frequency band electrochemical impedance spectrum to deeply explore the battery aging state information contained in the impedance spectrum data,which can avoid the problem of using time domain data for parameter identification without guaranteeing the accurate physicochemical properties of the parameters.At the same time,the use of only partial impedance spectrum data also lays a certain foundation for its online application.In view of the poor accuracy of the commonly used equivalent model in fitting the impedance spectrum data,an improved model structure is proposed in this paper.The validation results on the impedance spectrum data of different temperatures and aging states show that the proposed model can achieve higher fitting accuracy and parameter stability,and the model complexity is within the acceptable range.Secondly,for the feature extraction based on time domain data,this paper proposes to use the improved incremental power analysis method based on voltage-charge power model to achieve simple and efficient multi-type feature extraction based on monitored current,voltage and other time domain data,in order to address the problems that the traditional incremental power analysis method is plagued by noise caused by discrete numerical calculations.In order to enrich the theoretical study on the voltage-charge power model and the selection of model orders,this paper presents a systematic comparison of the models with different structures and orders from the perspectives of curve fitting consistency,feature correlation,model complexity,robustness to battery degradation and potential for battery health state estimation based on experimental data of whole-life accelerated life aging of Li-ion batteries with two different chemical materials.A systematic comparison is conducted to select the most suitable model for health state estimation for both batteries.Finally,a support vector regression algorithm-based model for estimating the health status of lithium-ion batteries is developed in this paper.After analyzing the correlation between the equivalent circuit model parameters and the incremental power curve characteristics and the battery capacity degradation respectively,the parameters with higher correlation are selected as the input to build the support vector regression prediction model,and better estimation accuracy and stability are achieved compared with several other models.Meanwhile,the robustness of the established support vector regression prediction models is evaluated,considering that the inherent inconsistency among cells will affect the estimation accuracy of the offline learning models when applied online.To address the problem of poor generalization performance due to feature differences,the concept of model fine-tuning in migration learning is introduced,and the generalization performance of the model is substantially improved by introducing a small portion of new data for retraining.