Research On Battery Life Prediction Algorithm Based On Time Series Analysis

In modern society,because of its excellent performance,lithium battery has become an important device to generate electricity in all walks of life.The life information of lithium battery plays an important role in the production,use and recycling of lithium battery.This thesis mainly studies the life prediction of lithium battery.This thesis looks at the reduction of battery life from multiple perspectives.Nowadays,battery life prediction mainly uses the combination of physical model and datadriven,but its accuracy and generalization performance are poor.The first is that after the battery is fully charged,the battery power will continue to decrease in the process of being used by the device,and the power information here is generally called the state of power;the second is that a brand new battery is continuously used by the user to perform the charging and discharging operation.During this period,the remaining maximum capacity of the battery will decrease.Finally,in the process of using the battery,the abnormal state of the battery may be caused by overheating,supercooling or impact.Therefore,this thesis studies the battery life information from three indicators: state of charge,remaining maximum capacity and abnormal state.Battery life is a variable that changes over time.Therefore,this thesis uses the time series analysis method to predict the battery life.First of all,this thesis analyzes the state of charge in the data set,and uses the autoregressive model to predict the state of charge of lithium battery.The results show that the model can improve the timeliness and accuracy of prediction.Secondly,this thesis constructs a gate unit recurrent neural network to predict the maximum remaining capacity of the battery,which not only improves the generalization ability of the model,but also reduces the dependence on the number of training data.Finally,this thesis uses the combination of decision tree and random forest to classify and predict the lithium batteries with abnormal state,and the model can successfully classify the lithium batteries with problems.In summary,this thesis analyzes the life of lithium batteries from three angles,and puts forward an integrated algorithm for predicting the overall life of lithium batteries,which provides effective theoretical basis and experimental reference for producers and users.For producers,accurate and rapid prediction of battery life information can improve the production efficiency and shorten the production cycle.At the same time,in the process of battery use,users can better judge the remaining state of power,so as to plan a better battery use plan.