Research On Lithium Battery Life Prediction Method Based On Small Data Sets

In today’s growing economy,lithium batteries have been widely used in all walks of life.However,lithium batteries are affected by factors such as high temperature,aging of the diaphragm,improper use and other factors during the recycling process.Therefore,the battery performance gradually decreases,indirectly leading to the decline of the performance of electrical equipment to cause equipment failure,resulting in huge human,material and financial losses.Thus,it is necessary to forecast the remaining useful life of lithium batteries.However,the lithium battery has a long cycle period and it is difficult to obtain more aging data,so there is a small sample problem in RUL prediction of lithium batteries.Therefore,in order to solve the problem of low accuracy of lithium battery RUL prediction under small sample conditions,this article mainly solves the problem by integrating expert knowledge.The main work is as follows:1)Extract indirect health factors.Firstly,according to the internal chemical reaction of the lithium battery,the factors affecting its aging are analyzed;secondly,based on the NASA lithium battery aging data set and the CACLE lithium battery aging data set,six indirect health factors are extracted,including equal voltage boost charge time interval,equal current drop charge time interval,equal voltage drop discharge time interval,equal time charge voltage rise,equal time charge current drop and equal time discharge voltage drop.Finally,the validity of the indirect health factor extracted in this paper is verified by Pearson correlation coefficient and Spearman correlation coefficient.2)Set up a lithium battery RUL prediction model based on an optimization algorithm and a modified BP neural network.Firstly,a lithium battery RUL prediction model based on BP neural network was established.Secondly,because the random initial parameters easily lead to the problem of local minimum in the neural network training process,the best initial weights and threshold were obtained by using genetic algorithm,particle swarm algorithm and simulated annealing algorithm respectively.Finally,the average error was used as the evaluation index,and the optimization effect of different optimization algorithms can be compared.In the field of lithium battery RUL prediction,the genetic algorithm has better optimization effect.3)Establish a lithium battery RUL prediction model incorporating expert knowledge.Firstly,expert knowledge was obtained by drawing the curve between the obtained indirect health factors and capacity,in which there are 6 pieces of monotonic expert-knowledge and 1 uneven expertknowledge.The expert knowledge is mathematically expressed by the way of seeking derivation.Secondly,the expert knowledge was used as the constraint condition,and the 7 constraint conditions were condensed into 1 constraint condition through the KS aggregation function.Thirdly,the augmented Lagrangian multiplier method was seen as an approach to integrate the expert knowledge into the GA-BP neural network training process,to train and fuse GA-BP neural network lithium battery RUL prediction model based on expert knowledge.Compared with traditional GA-BP neural network,the prediction accuracy of the improved algorithm in this paper has been significantly improved.Finally,the generalization ability of the proposed algorithm was carried out through five-fold cross-validation.Verification,the results showed that the algorithm had strong generalization ability.4)Design a set of lithium battery RUL prediction software incorporating small sample data.The software displays the indirect health factors extracted in this article,the RUL prediction results of lithium batteries based on GA-BP neural network,and the RUL prediction results of lithium batteries based on GA-BP neural network fused with expert knowledge.