A Method For Online Life Prediction Of Lithium Batteries Based On PCA And RVM Is Presented

With the development and progress of various fields in today’s society,new energy and new environmental protection technology demand and attention.As a representative of new energy,lithium-ion battery has the advantages of environmental protection,energy saving,compact and lightweight,and high battery life,and has been highly valued since its introduction to the market.As an important component of the power system,the working state and remaining life of the lithium battery play a key role in the stable and long-term operation of the whole system.The residual life prediction of lithium-ion batteries based on relevance vector machine（RVM）is presented in this paper.In view of the incomplete charging and discharging characteristics of the battery in the cycle work,the capacity of the battery decreases with the increase of charging and discharging times.On the basis of previous studies on indirect characteristic factors affecting battery capacity,different characteristic factors are considered to have different influences on the prediction results of lithium-ion battery life.A weighted construction method of characteristic factor variables based on principal component analysis（PCA）is proposed to predict the battery capacity attenuation.The main work contents are as follows:（1）To deeply understand and learn the significance of research on health management and life prediction of lithium batteries.Systematically study the basic idea of principal component analysis and the theoretical significance of data processing,as well as the working mechanism of correlation vector machine for online residual life prediction of lithium battery.（2）Based on the existing literature on the performance degradation and failure mechanism of lithium batteries,this paper summarized and analyzed the changes of capacity and performance data of lithium batteries in charge-discharge cycle,and selected three characteristic factors and variables from the monitored data in the charge-discharge process of lithium batteries to represent the health status of the batteries.（3）The online life prediction of lithium-ion batteries based on the traditional relevance vector machine has a single feature,which leads to the unsatisfactory prediction accuracy.A method of weighted linear fusion of predicted values is proposed to get the predicted results which can take multiple variable factors into account by linear fitting of the predicted results of different characteristic variables.（4）According to the basic idea and working principle of PCA,the scoring matrix is determined by taking the three characteristic factors and variables as the research object;The feature coverage degree of different score vectors in the matrix to the original data matrix was analyzed,and then weighted to construct the corresponding fused feature vectors as the input of the relevance vector machine.The prediction model of the online residual life of RVM lithium battery was established,and the results were obtained.Compared with the method of weighted linear fusion of single variable predictive value,the proposed method based on PCA variable weighted construction has less computational cost and wide applicability.At the same time,the prediction results are in better fit with the real capacity value of the battery,which can fundamentally solve the problem better.