Prediction Of Remaining Useful Life Of Lithium-ion Battery Based On New Health Indicator

With the increasingly serious problems of environmental pollution and energy shortage,and with the requirements of ‘Made in China 2025' policy for the development of energy-efficient and new energy vehicles,the development of Electric Vehicles(EVs)has become the inevitable trend of future traffic.Lithium-ion(Li-ion)battery is the key power supply equipment for EVs,and in order to avoid the breakdown of EVs due to the decline of useful life of Li-ion battery to failure threshold,scientific and accurate prediction of Remaining Useful Life(RUL)of Li-ion battery has become one of the key technologies in Battery Management System(BMS).In this thesis,the following work has been done for the prediction of RUL of Li-ion battery:First of all,the degradation mechanism of Li-ion battery is studied.Based on the internal structure of Li-ion battery,the internal factors causing aging are analyzed,and five external factors affecting RUL of Li-ion battery are given and analyzed respectively.At last,two direct Health Indicators(HI)of Li-ion battery are introduced and analyzed.Secondly,a new HI is extracted from the Li-ion battery charging data to predict RUL of Li-ion battery.In order to solve the problem that it is difficult to directly measure the actual capacity when predicting RUL of Li-ion battery online,the Li-ion battery data provided by the NASA PCoE Center and the University of Maryland CALCE Center are analyzed in this thesis,and a framework for HI extraction and optimization based on Li-ion battery charging data is proposed.At the same time,Pearson correlation analysis and Spearman rank correlation analysis are used to analyze the correlation between the capacity and the extracted HI,and the Box-Cox transform method is used to optimize the extracted HI.Finally,the accurate estimation of the battery capacity is accomplished based on the newly extracted HI.Finally,a Multi-kernel Support Vector Machine(MSVM)model based on Particle Swarm Optimization(PSO)is proposed to successfully predict the RUL of Li-ion battery.In order to solve the problem of low prediction precision and weak generalization ability caused by the traditional single kernel function used in Support Vector Machine(SVM)model,a novel MSVM based on polynomial kernel and radial basis kernel convex combination is proposed in this thesis.Moreover,the PSO algorithm is used to optimize the parameters of the MSVM model,and the prediction accuracy of the model is improved.At last,the model is validated based on the experimental data of NASA PCoE center and CALCE center of the University of Maryland.The results show that the prediction accuracy and generalization of the PSO-MSVM model are improved well.The new HI and PSO-MSVM model presented in this thesis can provide reliable data for battery safety management and other safety systems,and can successfully realize the accurate prediction of RUL of Li-ion battery.This research is of certain reference value for BMS development of EVs.