基于分子动力学的跟驰特性及其模型

Car-following Characteristics and Its Models Based on Molecular Dynamics

为更好地研究车辆跟驰特性,缓解道路交通拥堵,在车辆跟驰行为受前导车和道路环境等影响的基础上,将单车道道路虚拟为一维管道,道路上的跟驰车辆抽象成相互作用的分子。考虑需求安全距离和期望速度2个影响因素,基于分子动力学构建车辆相互作用势和分子壁面势函数,并建立基于相互作用势函数的分子跟驰模型,给出跟驰车辆的加速度模型。在实际交通环境中建立视频采集试验路段,采集试验路段不同点位的交通流样本,从视频中获得模型所需数据,并将数据分为两部分,一部分用于参数标定,其余用来模型验证。将车辆运行状态分为常态行驶、起动加速和减速停车3种。根据实测交通数据分别对3种车辆运行状态下的经典GM模型和分子跟驰模型进行参数标定,选取3种不同运行状态下的试验数据各3组,代入标定后的分子跟驰模型与经典GM模型计算模型输出加速度,并与实测加速度进行误差分析对比,结果表明,分子跟驰模型输出加速度与实测加速度之间的误差,总体上比经典GM模型要小,而且根据绝对误差方差显示,分子跟驰模型较经典GM模型稳定性更高。选取有代表性的一组跟驰过程进行数据绘图,对比可以看出分子跟驰模型输出加速度与实测数据变化趋势几乎一致,其拟合效果比经典GM模型更好。

In order to better study the car following characteristics and alleviate road traffic congestion, the single lane road is virtualized as 1 D pipeline, and the following cars on the road are abstracted into interacting molecules based on the influence of leading vehicle and road environment on the car following behavior. Considering the demand safe distance and expected speed, the vehicle interaction potential and molecular wall potential functions are constructed based on molecular dynamics. The car following model based on the interaction potential function is established, and the acceleration model of following car is given. The video acquisition test road section is built in the actual traffic environment, and the traffic flow samples of different positions in the test section are collected. The required data of the model are obtained from the video, which are divided into 2 parts. One is used for parameter calibration, and the other is used for model verification. The vehicle running states are divided into normal driving, starting acceleration and deceleration parking. The GM model and the molecular car following model in the 3 driving states are parameter calibrated respectively according to the measured traffic data. The selected 3 groups of experimental data in the 3 running states are plugged into the calibrated molecular following model and the classical GM model to output the acceleration, and its error is compared with that of the measured acceleration. The result shows that （ 1 ）the error between the output acceleration and the measured acceleration is smaller than that from the classical GM model; （2） according to the showing of absolute error variance, the molecular following model is more stable than the classical GM model. A typical set of car following process is selected for data drawing, the comparison shows that the changing same as that of the measured data, trend of output acceleration of the molecular following model is almost the and the fitting effect is better than that of the classic GM model.