A study on energy consumption analysis and prediction of electric bus at intersections considering driving behavior.

When passing through an intersection section, the relationship between driving behavior and energy consumption of pure Electric Buses (E-Bus) is unclear. In this study, natural driving data on two Bus Rapid Transit (BRT) routes were collected to quantify and analyze the implied relationship between driving behavior and energy consumption when entering an intersection point. Furthermore, it is proposed that predicting energy consumption on the basis of distinguishing whether an intersection is stopping or non-stopping would be a more accurate scenario. Concerning the working method, firstly, statistical analysis is used to observe the difference between the energy consumption of the stopping and non-stopping samples; secondly, correlation analysis and linear regression are used to analyze the significant parameters related to whether to stop or not, and energy consumption; finally, machine learning method is used to establish the classification model of whether to stop or not at an intersection as well as a prediction model the energy consumption of the intersection.The results show that the model accuracy of XGBoost-KNN is higher than that of KNN and XGBoost in predicting whether to stop or not, which is 84.4%. For predicting energy consumption, the GBDT has the lowest prediction accuracy; as for XGBoost and SVM, which have a higher prediction accuracy, distinguishing whether to stop or not helps to enhance the model's prediction accuracy. Furthermore, after distinguishing whether to stop or not, SVM outperformed XGBoost in R2, MAE, and RMSE. Research results provide a new perspective for studying the relationship between the driving behavior and energy consumption of pure electric buses at intersections. Meanwhile, they also offer the possibility for further research on the applicability of energy consumption when expanding from the BRT to more complex mixed traffic environments.