Research On Life Prediction Method Of Lithium Battery Of Electric Vehicle Based On Data-driven

With the increasing acceptance of electric vehicles,the performance and reliability of lithium-ion batteries have also received more and more attention.Lithium-ion batteries are used as power components of electric vehicles,its performance limits the cruising range,acceleration performance and climbing ability of electric vehicles.Cycle life is an important evaluation index for the performance of lithium batteries.Predicting the cycle life of lithium-ion batteries accurately and quickly has a positive effect on the manufacturing of batteries and rational management.However,because the lithium-ion battery system is very complex,the mechanism model of capacity decline is limited to finite side reactions.Therefore,from the perspective of data,the research on the topic of using early cycle data to predict the cycle life of lithium-ion batteries has important theoretical significance and research value.In this paper,18650 batteries commonly used in electric vehicles are used as experimental objects.The main research contents are as follows:(1)When there is only data of a single battery capacity and cycle number,from a statistical point of view,the particle filter algorithm(PF)is used to update the state parameters in the capacity decline empirical formula,and the state parameters are predicted by polynomial fitting of the state parameters,so it can be continuously updated during the process to improve the prediction accuracy.At the same time,the regularized particle filtering algorithm(RPF)is used to improve the discrete sampling in the resampling process to continuous sampling,which solves the problem of particle exhaustion in the traditional particle filtering algorithm.(2)In response to solve the problem that the RPF algorithm relies too much on the capacity decline empirical formula,a battery life prediction method based on particle swarm optimizating long-short-time memory network(PSO-LSTM)is proposed,while using Adam-optimized PSO-LSTM to achieve faster and more accurate life prediction,the prediction results are more suitable for actual conditions than RPF,and the advantages and disadvantages of the two methods are analyzed.(3)When there are many lithium-ion battery capacity decay data and data information under multiple working conditions,the exhaustive method is used to select the combination feature of the first 100 cycles and the battery life,whose pearson correlation coefficient is the largest,as the input of the ridge regression model.At the same time,using the relevant combined features of the first 30 cycles,by comparing the classification accuracy and AUC value of multiple classification models and integrated learning models with single or multiple features,support vector machine(SVM)was selected as the classification model of battery life.Life cycle prediction and classification of lithium batteries are realized by using earlier cycle data information than the improved RPF algorithm and Adam-optimized PSO-LSTM algorithm.(4)Build a battery accelerated life test platform.Accelerated life experiments at 2 C,3C,4C and other operating conditions for 18650 single cells commonly used in electric vehicles are conducted at 30℃.The feasibility of the above algorithm was verified by data cleaning,interpolation,correction and standardization.At the same time,a battery life prediction software interface integrating Adam-optimized PSOLSTM algorithm and Ridge regression model was designed to evaluate the applicability of various methods.