Study On Life Prediction Method Of Lithium Ion Battery

Lithium ion batteries are widely used in automotive,aerospace and many other fields because of their long cycle life,large capacity and environmental protection.In the process of multiple cycle charge and discharge of lithium-ion battery,a series of irreversible electrochemical reactions will occur inside the battery,resulting in aging,mainly manifested in the decline of battery capacity and life.Therefore,it is necessary to study the degradation mechanism and life prediction method of battery.On the one hand,it enables us to replace the battery that will fail in time,so as to avoid failure and even immeasurable loss to battery application equipment.On the other hand,predicting the service life of lithium battery can reverse optimize the user’s use behavior,prolong the service life of battery,and promote the development of lithium battery industry in the direction of low-carbon and environmental protection.The prediction methods of lithium-ion battery life include model-driven method and data-driven method.The model-driven method generally only considers the trend of the internal characteristics of the battery,and the generalization ability is limited.Starting from the data-driven method,this paper uses the machine learning related model to predict the life of lithium battery.The main work is as follows:In the first part,based on the trend analysis of the charging and discharging voltage,temperature,current and other characteristics of lithium-ion battery in NASA Ames diagnostics center of excellence(pcoe)laboratory database,11 relevant characteristics related to capacity decline and life decay are extracted,and the Pearson correlation coefficient is used to analyze the correlation degree between the extracted characteristics and capacity,so as to evaluate the rationality of the extracted characteristics.In the second part,the capacity of lithium-ion battery is modeled and predicted based on the integrated learning method and deep learning method,so as to determine the end life of the battery.In the aspect of integrated learning,random forest regression and XGBoost are mainly selected and compared.From the analysis of the prediction results,it is concluded that the stability of random forest regression is higher than XGBoost and the fitting effect is good;In terms of deep learning,we mainly select the traditional BP neural network and the LSTM model integrating cyclic dropout and L2 regular term,add cyclic dropout to the LSTM structure,which acts on the cyclic connection structure of LSTM,and set L2 regular term in the input link for constraints,which can prevent the model from over fitting to a certain extent and make the model more inclusive.Finally,two sets of battery data sets are used to train the deep learning model and the random forest regression model with good performance in integrated learning.Then two sets of new data sets are used to verify the model.The model verification results show that the LSTM model integrating cyclic dropout and L2 regular term has better effect and can predict the life of lithium-ion battery relatively accurately.