Research On Remaining Useful Life Prediction Of Lithium-ion Batteries For Electric Vehicle

Research on remaining useful life(RUL)prediction of lithium-ion batteries is an indispensable part of fault prediction and health management.In order to ensure the safe driving of electric vehicles and maximize the use of energy,it is particularly important to accurately and timely predict the RUL of lithium-ion batteries.Due to the complexity of the internal structure of lithium-ion batteries,the mechanism model is difficult to include the degradation trend of lithium-ion batteries,and data-driven came into being.This research studies the problems existing in the construction of health indicator(HI)and the selection of prediction algorithms in data-driven.The main contents are as follows:1.For the problem of poor universality and low generalization of HI,this research constructs a novel high-quality HI for RUL prediction of lithium-ion batteries.The relationship between the experimental data of various lithium-ion batteries cycle life and RUL degradation is analyzed.A novel HI,Average Voltage Drop(AVD),is built based on the discharge voltage.The correlation between AVD and capacity is verified by Pearson correlation coefficient and Spearman’s rank correlation coefficient.The minimum value of correlation coefficient of various lithium-ion batteries is 0.9783,indicating that AVD is strongly correlated with capacity.2.For the problem of low correlation between prediction algorithms and HI,in this thesis,RUL indirect prediction model is constructed by AVD combined with Relevance Vector Machine(RVM)and Long Short-Term Memory(LSTM)respectively.Finding the relatively optimal parameters of LSTM and kernel function parameter of RVM are involved,so as to establish the relatively optimal prediction model and improve the prediction accuracy.The experimental results show that,first,the RVM prediction model can accurately and indirectly predict the RUL of various lithium-ion batteries,and the maximum value of Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)are 0.0169 and 0.0134 respectively,proving the effectiveness,universality and generalization of AVD.Second,the LSTM prediction model can track the capacity decline trend for a long time,and the maximum value of RMSE and MAE are 0.0911 and0.0784 respectively.The predictive performance of LSTM and RVM on various lithiumion batteries is compared and analyzed,indicating that RVM is more suitable for AVD than LSTM.3.For the problem that RVM indirect prediction model can not achieve RUL online and single point prediction,this thesis proposes a RUL direct prediction method for lithium-ion batteries based on RVM-RVM,which is verified on various lithium-ion batteries.The experimental results show that this method can effectively achieve RUL direct prediction.