Estimation Of The State Of Health Of Lithium-ion Batteries Based On Data-driven Methods

Lithium ion batteries face an irreversible decline with long time use and storage.Accurate estimation of state of health(SOH)is helpful for lithium-ion batteries to ensure safety and obtain high service life and it also provides useful information for batteries state of charge(SOC)estimation,state of function(SOF)estimation,and equilibrium management.The data driven method does not consider internal reaction mechanism of battery and finds rule to estimate SOH from charging and discharging data,which has become the focus of research.The paper finds feature vectors to describe the rule of battery aging and models are built to estimate SOH,based on the batteries life experiment and data-driven method.Firstly,the decay mechanism of the battery is analyzed based on Incremental Capacity Analysis(ICA).The influence of discharge ratio and SOC range for batteries is studied.The influence of storage condition for battery aging is studied.Secondly,a novel method of SOH estimation based on constant-voltage charging current curve is proposed and random forest(RF)model is built.RF model not only can handle the high dimension of large scale data sets,but also has out of bag(OOB)samples to improve the generation performance of the model.Results show that the RF model has smaller predictive error.Thirdly,in order to reduce the pressure of the battery management system(BMS)stores data,the paper proposed a novel method based on partial constant-current charging data for evaluating SOH and the support vector machine(SVM)model is built.The grid search method based on cross validation(CV)is applied to optimize parameters of the model.The best voltage range is ensured by comparing the performance in predicting SOH.Results show that SVM model can achieve accurate SOH estimation.Lastly,the local discharging voltage curve is used to describe the change of SOH and least squares support vector machine(LS-SVM)model is built.Due to lack of sparsity,the Fixed Size LS-SVM is proposed to track SOH change.Fixed Size LS-SVM selects appropriate support vectors based on Nystrom method and Renyi entropy.Results show that LS-SVM and Fixed Size LS-SVM modes can give more satisfying and predicting results.Compared with LS-SVM and SVM,Fixed Size LS-SVM predicts SOH based on a little support vectors(SVs),which has more practical values.