Study On SOH Prediction Of Lithium-ion Batteries Based On Electrochemical Impedance Spectroscop

In order to achieve the goal of "30.60" carbon peak and carbon neutrality,China has begun to gradually build a new power system based on new energy.As the core of the new energy industry,energy storage technology and the development of its industry have attracted much attention.However,in order to further improve its dynamic performance and endurance performance,it is of great significance to study a series of problems such as aging modeling,fault diagnosis,failure analysis,online state prediction and early warning of energy storage system.This paper mainly studies the prediction of health state of lithium ion battery based on electrochemical impedance spectroscopy.At present,there are two methods to predict the health status of lithium-ion batteries based on electrochemical impedance spectroscopy: equivalent circuit model method and data-driven method.Firstly,the results of the two methods published so far are summarized and the existing problems are pointed out.The two methods are improved respectively.For the method based on equivalent circuit model,Kramers-Kronig(K-K)relations were first used to verify the impedance data used,so as to ensure the authenticity and validity of the impedance data.Aiming at the problem of unclear definition of the equivalent circuit of lithium ion battery,an innovative method of predicting the equivalent circuit model using impedance data was proposed.Firstly,a variety of equivalent circuit models were proposed.The least square method was used to fit the impedance data with different equivalent circuit models,and the equivalent circuit model with the highest fitting accuracy was selected to ensure the high accuracy of the equivalent circuit.Secondly,the parameters of different units of the equivalent circuit changing with the aging of the battery were extracted,and the correlation between them and the health state of the lithium-ion battery was analyzed.The parameters with high correlation were selected as the final prediction indicators.The multi-equivalent circuit model method proposed in this study can not only ensure the accuracy of the model,but also make the prediction results more real and reliable.In view of the shortage of manual feature extraction in the current data-driven model,two strategies of convolutional neural network and variational autoencoder are proposed to automatically extract health features.It not only ensures the comprehensiveness of feature extraction but also greatly simplifies the complexity of manual feature extraction,which is conducive to practical application.Combined with bidirectional short and short time memory neural network and bidirectional gated cyclic neural network as the final prediction model.The bidirectional neural network model was used to predict the sequence regression,which fully considered the time dependence of impedance data and health status.In addition,due to the large number of parameters in the constructed neural network model,it is necessary to manually adjust the model parameters in the training process to achieve a higher prediction accuracy.In order to further increase the adaptability and robustness of the model,the traditional particle swarm optimization algorithm is improved.The improved particle swarm optimization algorithm is used to automatically optimize the initial learning rate and the number of hidden layer element nodes of the neural network.The simulation results verify the effectiveness of the two model strategies.By comparing the two models,it is concluded that the convolutional neural network combined with the bidirectional short and long time memory neural network is more accurate.The improved equivalent circuit model method and data-driven method are compared with the existing Gaussian process regression model respectively,and the prediction accuracy is improved by 27% and 35% respectively,which shows that the improved method is superior to the health state estimation of lithium-ion batteries.