| With the continuous development of lithium-ion batteries,the issue of lithium-ion battery safety is of great concern.Usually when the capacity decays to 70% of the rated capacity,the life is considered to be terminated,so in order to guarantee the safety of use,the remaining life of lithium-ion batteries must be predicted.The current prediction methods are mainly model-driven and data-driven.Model-driven requires professionals to complete,and the model parameters are complicated and not easy to estimate,which leads to the limitation of the use of this method.Data-driven methods require a large amount of data,and with the accumulation of lithium-ion battery data from major institutions and the development of machine learning in recent years,it has become possible to predict the remaining life of lithium-ion batteries using machine learning algorithms.Support vector machine has high generalization ability based on the principle of minimizing structural risk,and also has good performance in dealing with small sample data and nonlinear regression problems,so this paper will use support vector machine for prediction of remaining life.The data are used from the lithium-ion battery dataset published by NASA,and the remaining life is predicted by extracting the features that can show the capacity decline of lithium-ion batteries.Firstly,by observing the charge/discharge current,voltage and temperature change curves of lithium-ion batteries,nine features that can reflect the capacity decline are extracted,and then the features are normalized and used as the input of the model.Secondly,this paper proposes feature-weighted SVR based on the standard SVR,and the method to determine the feature weights is gray correlation,and two models are used to predict the battery No.B0006.The data empirically prove that the featureweighted SVR prediction effect is better than the standard SVR prediction effect,which has some practical value. |
