Research On The Estimation Of The Health Status Of Pure Electric Vehicle Power Batteries

With the increasing global problems such as energy security,environmental pollution and climate change,the development of automotive industry is gradually transforming to a clean and low-carbon direction with pure electric drive as the main line.Power battery is one of the core components of electric vehicle power system,and its performance index and State of Health（SOH）have a direct and significant impact on the main technical indicators such as power performance and driving range of the vehicle.Therefore,the accurate prediction of the health state of automotive power battery is one of the important issues to be solved in the process of electric vehicle product development and industrialization application.Relying on a school-enterprise cooperation pure electric vehicle product development project of the group,this paper takes the pure electric vehicle power battery as the research object and studies the health state decline law of the pure electric vehicle power battery by analyzing the data collected during the driving of the real vehicle.Based on the battery charging and discharging experiments and data preprocessing,a battery capacity estimation and correction model is built,a battery health state prediction algorithm based on a genetic algorithm improved Gaussian process regression model is proposed,and the model is validated based on the NASA public lithium-ion battery dataset.The research results show that the prediction model established in the paper can achieve more accurate estimation and prediction of the health state of automotive power batteries.The specific research of this paper includes.（1）Firstly,the current research status and development trend of power battery health state prediction are analyzed and summarized.On the basis of analyzing the working principle and health decline mechanism of the battery,the factors affecting the decline of battery capacity are extracted,and two factors,battery use temperature and battery charge/discharge current multiplier,are selected for battery charging and discharging experiments to obtain the decline law between battery capacity and these two factors.（2）According to the estimation and modeling requirements of vehicle power battery capacity,the historical data of a type of pure electric vehicle power battery is analyzed and processed.Firstly,according to the charging state and the characteristics of the battery SOC change during charging,the charging fragment of the vehicle is divided;then the abnormal data is filtered and processed according to the DBSCAN algorithm,and the abnormal data is treated as missing data;then the missing data and the abnormal data filtered out in the previous step are interpolated based on the Newtonian interpolation method.Finally,283667 valid data strips are obtained,and after dividing the fragments,1071 segments of the parking charging fragment are obtained.Lay the foundation for subsequent battery capacity estimation modeling.（3）On the basis of data pre-processing,the battery capacity is modeled and calculated by combining the ampere-time integration method.The box plot method is used to remove outlier points,and based on the previous experiments,the capacity is corrected based on the battery use temperature and charging current multiplier,and then the battery health state is defined based on the capacity ratio method,and the battery health state is estimated and modeled by three fitting methods and wavelet analysis respectively,and finally the wavelet analysis method with less error is selected as the estimation model of the battery health state.（4）According to the characteristics of the vehicle power battery data set,the Gaussian process regression method is selected to model the prediction of the battery health state.To improve the model accuracy,this paper combines the genetic algorithm to optimize the hyperparameter search process of Gaussian process regression kernel function,and validates the model based on NASA public lithium-ion battery dataset,which can meet the performance index of average error rate ≤ 1.5% and maximum error rate ≤ 2%.Finally,the algorithm is validated based on real-vehicle data to demonstrate the feasibility of the algorithm in real-vehicle power battery health state prediction applications.It provides a realistic and feasible solution idea for the power battery health state prediction of pure electric vehicles.