Research On Life Prediction Technology Of Lithiumion Battery Based On Data Driven

Lithium-ion battery is widely used in new energy vehicles because of its long cycle life,high energy density and environmental protection.With the development of new energy vehicle industry,the society puts forward higher requirements for the recycling of lithium-ion batteries.Accurate prediction of remain useful life(RUL)can improve the reliability and performance of batteries.In this paper,data-driven method is used to evaluate the reliability of lithium-ion battery life from two aspects of indirect prediction and direct prediction,combined with machine learning algorithm.This paper first introduces the development status of life prediction of lithium-ion battery at home and abroad,and briefly analyzes the attenuation principle of lithiumion battery.Aiming at the problem that the uncertainty of battery cannot be evaluated in RUL indirect prediction,an improved sparrow search optimization algorithm combined with RVM is adopted to reduce the prediction error by optimizing the initial value of RVM kernel function.At the same time,aiming at the problem that the prediction accuracy of health factor based on constant voltage discharge time is not high,two improved health factors are proposed,and the correlation between health factor and battery capacity is analyzed by partial correlation coefficient method.The simulation results show that the prediction accuracy of the two improved health factors is significantly higher than that of the common health factors.The prediction error is less than 1%,which verifies the effectiveness of the improved indirect prediction model.Secondly,aiming at the problem of low accuracy in the direct prediction process of RVM algorithm,the method of data dimension promotion is adopted to improve the sample feature size,and the best data dimension promotion parameters are selected by comparison.An improved RVM algorithm based on online learning is proposed,which improves the shortcoming of low long-term prediction error of RVM model and realizes the real-time update of RVM.The experimental results show that the performance of the improved RVM algorithm is improved.The improved RVM algorithm gives the reliability interval of lithium battery life prediction according to the existing data,which provides the basis for the uncertainty of lithium battery life prediction.Finally,in view of the characteristics of incomplete charging and discharging,complex actual working conditions and few available samples,gamma stochastic process is used to simulate the battery capacity decline.Based on this,a Bayesian data fusion method is proposed to fuse the theoretical working condition data and available actual working condition data to build a battery life evaluation model,which realizes the evaluation of small sample actual working condition RUL prediction and reliability evaluation of lithium-ion battery under different conditions.