Online Modeling And Simulation Of Car-Following Behavior Based On Data-Driven Methods

Car-following behavior is one of the main research areas of traffic flow theory.Through the research and modeling of car-following behavior,we can quantify and analyze the interaction between the vehicles,reveal the characteristics of the single-lane traffic flow,and provide the basis for traffic simulation.Therefore,in the traffic simulation system,the car-following model is one of its basic models.The higher the accuracy of the car-following model,the higher the realistic and reducing degree of traffic simulation.This is of great significance for the use of traffic simulation for road capacity calculation,service level division,and transportation policy evaluation to improve the traffic environment and the service level of the transportation system.Traditional traffic simulations are mostly offline.With limitations such as hysteresis,it is difficult to provide timely guidance for real-time management and control of traffic.Through the collection and processing of real-time vehicle driving data and traffic control system information,the online traffic simulation system can restore the actual traffic scene as much as possible and quickly predict the future traffic conditions.As the basic model of the online traffic simulation system,the online car-following model can use the real-time data to update itself online,which can better simulate the current car-following behavior of the real traffic scene and ensure the accuracy of online traffic simulation.This thesis focuses on the data-driven online traffic simulation car-following model.The main work includes:(1)Construction of the car-following behavior model based on online support vector regression method.Based on in-depth analysis of the characteristics of car-following behavior,for the limitation of the existing car-following model which cannot accurately reflect the car-following behavior in dynamic traffic environment,this thesis introduces the Online Support Vector Regression(Online-SVR)method,which can be used to nonlinear time-varying systems with a better real-time modeling capability,using data-driven method,to realize online modeling of car-following behavior and to achieve fast and accurate prediction of driving behavior in dynamic traffic environment.(2)Calibration and verification of Online-SVR-based car-following modelBased on the field vehicle trajectory data of the Next Generation Simulation(NGSIM)project,the parameter calibration and error evaluation of the online car-following model proposed in this thesis are carried out,and the prediction accuracy and macro-recurring ability are verified by simulation experiments.(3)Design and implementation of online car-following system of USTCMTS2.0In the traffic simulation software USTCMTS2.0,the function of online simulation is added,and the proposed online car-following model is implemented,which provides the core engine for the online simulation experiment of road traffic.