Prognostics Of Li-ion Batteries Using Model-based Approach

As main energy storage equipment,lithium battery has good performance in quality and reliability,which has drawn extensive attention of government,businessmen,scholars,environmental organizations,as well as the general public and other sectors of the community.However,the replacement and repair of inoperative batteries will cost losts of money.What's more,lithium battery failure can lead to terrible disasters.To avoid serious accidents and optimize maintenance repair system of lithium battery,there must be breakthroughs in failure detection,condition monitoring,residual life prediction and other related fields.The main work of this paper is listed as follows:1.Aiming at the problem that the capacity is difficult to measure in the practical application,features extracted from voltage,current,time and other parameters,obtained from online sensors,is used to indicate the health state.Subsequently,features extracted from sensor signals are used to train HMMs,which represent different health levels.Then the trained HMMs are used as the state monitor.The data set from CALCE are adopted in the experiment to evaluate the effectiveness of the proposed approach compared with BP neural network.Test results show that this method can achieve the purpose of the study in estimating battery health states.2.BW algorithm is easily trapped in local optimum,because it's a kind of sub-optimal estimation algorithm.Meanwhile,as a kind of global random search algorithm,the convergence speed of PSO is very slow.Therefore,this paper proposes a new training algorithm of CHMM combined BW algorithm and PSO algorithm,which overcomes disadvantages of these two algorithms.The experiment result shows that the new HMMs state monitor trained by the proposed algorithm has higher recognition.3.In this paper,a new integrated RUL prediction algorithm of lithium battery is proposed,which is based on improved ARIMA model and SRCKF algorithm.Firstly,a degradation factor,which is used to characterize the degradation rate of lithium batteries at different stages,is introduced to solve the problem that ARIMA model has poor predictive ability for nonlinear systems.This method improves performance of ARIMA algorithm when it is used to fit the nonlinear degradation trend of lithium battery.Secondly,based on empirical degradation model of the Lithium-ion battery,SRCKF algorithm is used to estimate the battery capacity by multi-step iteration in this paper.Then,to weaken the dependence on the empirical model of SRCKF algorithm,a fusion framework with SRCKF and the improved ARIMA model is introduced to improve the applicability of the algorithm.The experiment results demonstrate that the integrated algorithm improves the forecasting precision of RUL.