Health Management Of Lithium Batteries Based On Capacity Regeneration Detection

Health management includes the estimation of health state and the accurate prediction of remaining useful life(RUL).As a power supply system,lithium batteries have been widely used in various electronic equipment.In most current researches,capacity is usually as a health indicator(HI)to represent the state of health(SOH)of lithium batteries.But there are two problems existed in the measurement of capacity in lithium batteries.On one hand,the capacity can be obtained after a complete charging or discharging cycle,which will take a long time and only be measured offline.For some batteries,the full charging and discharging will shorten the life.On the other hand,due to the influence of rest time between two cycles,the capacity will produce a sudden regeneration.It may lead to a large prediction error of RUL at the regeneration point.When the capacity is difficult to obtain online,two new health indicators are proposed to characterize the degradation state of lithium batteries.Then the correlation between HI and capacity is analyzed by Pearson correlation coefficient method,which shows the effectiveness of proposed HI.Finally,the online estimation of SOH is realized by the algorithm of Relevance Vector Regression(RVR).For the capacity regeneration of lithium batteries,this paper analyze the reason for the capacity regeneration.Particle Filter(PF)and Mann-Whitney U test are adopted to detect the regeneration points of different batteries.In the RUL prediction process,this paper proposes a hybrid prediction approach combining PF and Auto Regression(AR).The predicted results of AR model are used as real capacity values to update the parameters of PF prediction model.Compared with PF and AR single model and Support Vector Regression(SVR)algorithm,the proposed method has better prediction accuracy.Finally,experiments were carried out on the public data set of NASA,and the results all showed the excellent performance of proposed methods.