Prediction Of Health State And Remaining Useful Life Of Lithium-Ion Battery Based On EIS-PSO-CNN-SAM

With the development of battery technology,lithium-ion batteries have been widely used in various fields However,during the actual cycle of the battery,the chemical processes inside the battery are not completely reversible,which can lead to deterioration of battery performance,even cause catastrophic consequences.In addition,existing prediction methods of battery state of health(SOH)and remaining useful life(RUL)based on electrochemical impedance spectroscopy(EIS)data mainly rely on qualitative interpretation of parameters in Nyquist diagram,while direct determination of battery SOH and RUL from impedance values or spectrum has not been thoroughly investigated.Based on this,a prediction method based on EIS data is proposed in this paper,which can accurately predict the SOH and RUL of lithium-ion batteries without relying on the battery history cycle data,which is of great significance for ensuring the safe and efficient operation of batteries.Firstly,in order to investigate the relationship between the impedance of the battery and the SOH and RUL,this paper uses the battery impedance and phase data obtained at different health states and temperatures based on the electrochemical impedance spectroscopy method to study the visualization and quantification of the frequency response of the battery impedance.Meanwhile,the mapping relationship between SOH and RUL and battery impedance was studied in the full impedance spectrum at different temperatures.It is found that the negative imaginary part of impedance increases significantly during the aging process of the battery,and it is less affected by temperature.However,the real part of impedance is greatly affected by temperature.This fully demonstrates the negative imaginary part of the impedance spectrum has a good ability to fit and characterize the SOH and RUL of the battery.Secondly,features were extracted from the negative imaginary part data of the battery impedance,and principal component analysis algorithm(PCA)was used to process and screen the negative imaginary part data to achieve the effect of data purification.Meanwhile,the fully adaptive noise set empirical mode decomposition method(CEEMDAN)was introduced to decompose multiple sub-sequences of the battery capacity data.The noise subsequence is removed by quantitative analysis and the quality of sample data is improved.Finally,the convolutional neural network(CNN)is used to model the decomposed subsequences,and the weight of neurons in the hidden layer of the network is adjusted based on the self-attention mechanism(SAM),so as to further optimize the feature extraction ability of CNN.At the same time,the PSO-CNN-SAM model is constructed by using the particle swarm optimization(PSO)to optimize the superparameters.In addition,Gaussian process regression(GPR),support vector regression(GS-SVR)and PSO-CNN-LSTM-SAM were designed as competitive models for comparative study.The results show that PSO-CNN-SAM model shows high accuracy and stability in predicting SOH and RUL.Compared with the published related methods,the R~2 value of the predicted results of the PSO-CNN-SAM model increased by 14%,which further verified the accuracy of the model.