Research On Modeling And Predicting Of Remaining Driving Range Of Battery Electric Vehicles Based On Data

Nowadays,the world automobile industry is facing a great change that has not happened in a century.The new energy vehicles represented by battery electric vehicles are developing rapidly.However,with no major breakthrough in power battery technology and low charging pile density,the inability to accurately predict the vehicle’s remaining driving mileage leads to the phenomenon of "mileage anxiety" and the conservative vehicles energy management strategies chosen by manufacturers.All these seriously affect the experience of battery electric vehicles.Therefore,accurate prediction of the remaining driving range of battery electric vehicles is of great significance for the promotion of battery electric vehicles and the development of refined management strategies.In this paper,aiming at stable and accurate prediction of the remaining driving range of battery electric vehicles,the modeling and predicting of the remaining driving range of battery electric vehicles were carried out with the operation data of battery electric vehicles as the research object under the guidance of statistical learning method theory.A set of method system of modeling and predicting of the remaining driving range of battery electric vehicles was formed,which was "data exploration and preprocessing-feature construction-modeling and tuning-model fusion-model prediction-prediction results evaluation".The main research of this paper is as follows:Ⅰ.Processing of battery electric vehicles of operation data.All fields of the battery electric vehicles operation data in the database were explored,and the meaning and data type of each field were described.The mileage-related operation data of battery electric vehicles were extracted from the database.The processing scheme of outlier screening and missing value filling and remaining driving range field construction in the early stage of data was designed.The original operation data was pre-processed in the integrated development environment programmatically according to this scheme.Ⅱ.Research and selection of statistical machine learning algorithms.The problem of predicting the remaining driving range of battery electric vehicles was abstracted to the machine learning modeling level,which was the regression problem.After the research of the algorithm,it was determined to model and predicted the remaining driving range of battery electric vehicles based on Ridge Regression Algorithm,Random Forest Algorithm and XGBoost Algorithm.Ⅲ.Analysis of factors affecting the remaining driving range of battery electric vehicles.Based on the energy consumption of battery electric vehicles and the remaining available energy of power battery pack,several factors influencing the remaining driving range of battery electric vehicles were analyzed and summarized.Ⅳ.Data feature construction.Based on the prior knowledge and the application symbol transformations based on genetic programming,the new data features were constructed,and the original data dimensions were expanded.Experiments showed that the evaluation indexes of the model can be improved by using the dimension expanded data for modeling.Ⅴ.Model fusion.Stacking,Bagging and Weighted model fusion were described in detail.In the integrated development environment,different model fusion methods were used to predict the remaining driving range of battery electric vehicles.The experimental results showed that the Bagging model fusion method was helpful to improve the modeling and prediction effect of the remaining driving range.Based on the method system of modeling and predicting,and evaluation method system of the remaining driving range of battery electric vehicles proposed in this paper,the maximum RAE of the predicted remaining driving range was less than 3.5%,and the MAE was about2 km,and the RMSE was close to 3km.Compared with the existing research,the accuracy of the modeling and predicting results of the battery electric vehicles remaining driving range in this paper has been greatly improved.