Research On Analysis And Modeling Of Dynamic Characteristics Of Car-following Behavior Based On Vehicle Trajectory Data

In recent years,the research on car-following behavior has become increasingly popular,and numerous studies use simulation experiments to explore the factors affecting carfollowing behavior.However,the dynamic characteristics analysis of car-following behavior based on vehicle trajectory data and the modeling research considering the impact of multiple factors are slightly insufficient.Based on the selected high-quality vehicle trajectory data,this study proposes a noise reduction and screening method for car-following data,and explores the complex dynamic characteristics of car-following behavior in terms of statistical features and correlation.Then,according to the data analysis results and the shortcomings of the current research,some car-following models reflecting different traffic characteristics or driving characteristics are proposed.Finally,the validity and reliability of the proposed model are discussed through the measured data,stability analysis,and simulation experiments.The main results obtained in this paper are summarized as follows:(1)A car-following model considering the effect of headway and velocity of the preceding vehicle is proposed based on improved optimal velocity function.First,given the shortcomings of the traditional optimal velocity function,a new function considering the boundary conditions of the headway and the velocity of the preceding vehicle is proposed.Then,based on the results of data analysis and taking into account the effects of delay time and acceleration difference,a car-following model that considers the effect of the headway and the velocity of the preceding vehicle is constructed.Finally,the model parameters were calibrated using measured car-following data,and the differences between the proposed model and the control model were verified using the measured data.The results show that the proposed car-following model can better reproduce the car-following behavior and improve the fuel economy and emission of the vehicle.(2)A car-following model considering the car-following critical threshold constraint is proposed based on measured car-following data.First,because of the inconsistency between the car-following headway critical threshold defined by the traditional optimal velocity function and the measured data,a dynamically variable car-following critical threshold constraint function is proposed.Then,based on the optimal velocity criterion,the carfollowing critical threshold constraint function is incorporated into the car-following model.Finally,the vehicle start-stop and emergency braking at intersections are used as simulation scenarios to verify the effectiveness of the proposed model.The results show that by introducing a dynamically variable car-following critical threshold constraint function into the car-following model,the maneuverability and safety level of the vehicle can be significantly improved,and at the same time,the fuel consumption and exhaust emissions of the vehicle can be improved.(3)A car-following model considering the driver’s collision sensitivity is proposed based on the car-following velocity constraint.First,since there are differences in the sensitivity of drivers in different dangerous traffic environments,a driver’s collision sensitivity coefficient is proposed based on the vehicle collision critical time and velocity constraint criteria,which can express the collision sensitivity of the vehicle under acceleration,constant velocity or deceleration.Then,based on the results of data analysis,a car-following model that considers the driver’s collision sensitivity and the velocity of the preceding vehicle is constructed.Finally,the linear stability analysis solution of the model is solved according to the stability theory,and the vehicle start and stop at the intersection and the disturbance interference driving of the ring road are used as simulation scenarios to verify the validity of the model.The results show that as the driver’s collision sensitivity increases,the stability of the model is improved,the phenomenon of stop-and-go vehicles is suppressed,and the safety level of the vehicle during car-following is improved.(4)A hybrid driven car-following model that considers machine learning and dynamics methods is proposed.To determine the best model for predicting the instantaneous velocity of car-following,four data-driven methods were selected for comparative analysis,and a model training and verification scheme was designed according to the characteristics of carfollowing data.The LSTM neural network model is used as a sub-model for predicting the car-following variable.Then,using the full velocity difference model as the host model,and using LSTM to predict the input variables of the host model,a hybrid driven car-following model is constructed,so that it can consider the impact of driver memory and the effect of vehicle information exchange.Finally,simulation experiments and measured data verify the stability and predictive effect of the model,and evaluate the safety level of the model.The results show that the proposed hybrid driven car-following model can not only reduce the velocity and headway fluctuations of the simulated platoon,but also improve the prediction performance of the model,and the safety level of the vehicle is also improved.In summary,the results obtained in this study can provide references for in-depth exploration of factors that affect car-following behavior,and provide necessary theoretical basis and analysis methods for modeling and simulation car-following behavior.