Study On Dynamic Characteristics Of Car Following Based On Experiment And Modeling

As the basis of microscopic traffic flow theory,car-following theory has always been a hot topic in the field of transportation.Its research methods are mainly divided into measured experimental methods and modeling simulation methods.The experiment of controllable variables can eliminate the interference of mixed factors on the road,and obtain more accurate experimental results.For example,the data source of this paper—car following experiment of 25 cars on straight sections by Jiang et al.It is proved that the relationship between speed and headway is not unique,instead occupying a 2D space,which is a good confirmation of the viewpoint of the three-phase traffic flow theory.As an important tool for time-frequency analysis,wavelet transform plays a key role in data processing in recent years.This paper uses wavelet transform to quantitatively analyze the speed,headway and acceleration phase duration of the vehicle in the car-following experiment,and obtain the corresponding cumulative the distribution,and fitting results show that the probability density distributions of the three statistics show log-normal,log-normal,and first-order decay exponential trends,respectively.At the same time,the stability of driving behavior will decrease as the platoon’s position is pushed back.Based on a large amount of empirical and experimental data,the car-following model has also been continuously improved,mainly reflected in the adjustment of the governing equations and the calibration of model parameters.This article discusses two classic car-following models: inertial model and intelligent driver model: Considering the actual driving condition,the acceleration terms are supplemented to the inertia model,and the simulation results can well reproduce the experimental rules such as the speed standard variance and the spatiotemporal patterns.Considering the heterogeneity of drivers,the parameters of the intelligent driver model were randomized,and the results showed that the phase duration distribution of the model was more similar to the experiment.