Research And Application On Remaining Useful Life Prediction Of Lithium-ion Battery

It is widely used in electric vehicles,aerospace and electric energy storage because lithium-ion batteries have the advantages of high specific energy value,long cycle life and no memory effect.With the repeated charging and discharging of lithium-ion batteries,battery capacity will gradually decrease and the cycle life will decay,which brings serious safety and reliability problems.Therefore,it is of great practical significance to carry out research on health assessment and RUL prediction of lithium-ion batteries,which can effectively reduce the safety risks of all kinds of systems using lithium-ion batteries.At present,the degradation mechanism of lithium-ion battery is not completely clear,so it is extremely difficult to establish an accurate degradation model for prediction.However,the historical data of lithium-ion battery contain abundant degradation information.To this end,this paper carries out RUL prediction of lithium-ion battery,starting with data and mechanism,in order to solve the key theory and technical problems,and then carries out system development and application on this basis.The main work and achievements of this paper are as follows:(1)Traditional kernel support vector regression(KSVR)algorithm has unique advantages in solving small sample and nonlinear prediction problems.It is a typical nonlinear prediction problem to predict RUL based on the limited historical battery capacity data.Therefore,a prediction method of RUL for lithium-ion battery based on KSVR and Improved particle swarm optimization(NPPSO-KSVR)was proposed.This method firstly constructs KSVR model based on historical capacity data of lithium-ion batteries,then uses improved particle swarm optimization(NPPSO)to optimize penalty factor C and nuclear parameters,and finally conducts training and prediction on typical sample sets selected from the batteries of NASA and the university of Maryland.The experimental results show that the capacity prediction and RUL performance of NPPSO-KSVR method is better than that of the traditional method,and has strong universality.(2)Particle filter algorithm has strong modeling ability for nonlinear andparametric non-gaussian problems and particle degradation problem.The probability distribution of lithium-ion battery capacity noise does not always satisfy the gaussian distribution.Therefore,aiming at two kinds of lithium-ion application scenarios for accuracy demands high or time sensitive,the paper puts forward the simple resampling method and adaptive resampling method to improve the particle degradation problems,a simple resampling method based particle filter(SPF)and adaptive resampling method based particle filter(APF)were designed for lithium-ion battery capacity and RUL prediction Based on the degradation mechanism of lithium battery.The experimental results show that the capacity prediction performance of APF method is obviously better than that of the traditional method,but the computational complexity is slightly higher,so it can be applied to scenarios with high precision requirements.The capacity prediction performance of SPF method is slightly better than that of traditional method and the computer is less complex.it can be applied to scenes sensitive to the time requirement.(3)In order to make full use of the historical data and degradation mechanism information of lithium ion battery and combine the advantages of NPPSO-KSVR and PF algorithms,two prediction methods of RUL for lithium-ion battery,NPPSO-KSVR-SPF and NPPSO-KSVR-APF,were proposed based on the fusion of data and mechanism.This method uses the NPPSO-KSVR algorithm to predict the capacity after the k time by nonlinear quasi combination of the historical capacity data Time k and before,and then takes the predicted capacity as the actual value of SPF and APF to construct the PF model.The experimental results show that the prediction performance of the two methods RUL is better than that of the traditional method and NPPSO-KSVR method.The algorithm takes less time and can meet the requirements of precision and time.(4)In order to test the effectiveness of the algorithm,an battery capacity and RUL prediction system was developed with kingview,SQL Server data management system,Matlab and SCM.The system first collects battery performance parameters through the tec-06 battery capacity detector And EBC-X battery tester,then transmits them to the upper computer through the Ethernet module,finally carries out algorithm realization,analysis,display and data storage in the database based on the RUL prediction platform of kingview and Matlab.The measured data in tieneng battery group show that the NPPSO-KSVR-SPF and NPPSO-KSVR-APF methods RUL have high prediction accuracy and short time consumption,and can meet the requirements of actual engineering scenarios.