Research On State Of Health For Electric Vehicle Power Battery Based On Data Mining

In recent years,worldwide emission regulations of vehicles are becoming strictly and development of new energy vehicles gets more and more attention.Electricity has become primary power source of new energy vehicles as a kind of clean and renewable energy.Associated with plentiful research fields,power battery technology is one of core technologies of electric vehicles,among which research on battery state of health(SOH)has always been focused on.At present,the battery test is generally adopted in research on power battery SOH.However,power batteries have to be removed from electric vehicles and repeated tests for different kinds of batteries are tedious and costly.Considering disadvantages of present methods,running data recoded by battery management system on electric vehicles have been studied based on data mining methods and novel estimation and prediction methods of power battery SOH have been proposed.Contents in this article are as follows.(1)Above all,the background of power battery SOH has been introduced.The definition of SOH is given and relevant researches are reviewed.The structure and working principle of lithium battery is studied and two influencing factors have been analyzed.Modeling procedure based on data mining is introduced and the Python programming language is chosen as data mining tool.Then,the characteristics of common prediction models are analyzed and define the communication between vehicle terminal and data acquisition platform.(2)A series of processing operation has been conducted to running data of the battery electric vehicle(BEV).Correlation analysis is made to determine valuable variables and running data are divided into data fragment for subsequent modeling procedure based on capacity of power battery.Abnormal data are picked and deleted while missing data are added to ensure the data effectiveness and integrity.The running data process provides valuable,effective and completed data for estimation and prediction of power battery state of health.(3)According to the definition of SOH,capacities of power battery are calculated by the Amper-Hour integral methods.Influences of temperature and discharge current are considered and relevant adjustment formulas are given.Due to precision decrease of SOH model,outliers are detected and deleted with the help of box-plot.Then,moving average method is adopted to remove noise and reduce random errors.Three kinds of fitting models have been used to build estimation model of SOH respectively and linear model is chosen to establish estimation model.Hypothesis testing of the estimation model is conducted and the confidence interval of the parameter is given.(4)Back Propagation Neural Network are studied to build power battery prediction model.The verification results of the battery charging and discharging cycle data show that the accuracy of BP neural network model needs to be improved.Then,the GA-BP neural network is used to established the SOH prediction model,and the battery verification results of experimental data shows that the prediction accuracy of the new model is markedly improved.Finally,the improved prediction model is used to predict the SOH.The proposed SOH of electric vehicle power battery estimation and prediction methods are based on data mining technology and can avoid power batteries from being removed for battery test.The proposed methods significantly reduce the costs of power battery research while running data of electric vehicles could be made full use of.It is expected that the research in this article could help with the development of online estimation and prediction of SOH for electric vehicle power battery.