摘要
智能网联车路协同系统以及网约出租车的迅速发展,产生了海量的轨迹数据。轨迹数据具有数据量大、准确性高、分布广、易获取等优点,成为交通研究的重要数据来源。排队长度是评价交叉口运行状态的主要参数之一,对交通状态评估和信号优化等具有重要作用。基于轨迹数据对交叉口排队长度进行估计,并结合交叉口历史排队分布对排队长度估计结果的可靠性及精度进行分析。首先建立基于贝叶斯定理的交叉口排队长度估计方法,在联网车辆相同的假设条件下,推导出排队长度与周期内联网车辆停车位置及车道排队长度的概率关系;并利用轨迹车辆排队长度频率分布对实际排队长度频率分布进行近似,解决所存在的未知量问题。然后,结合交叉口历史排队数据,分析在高斯及非高斯情况下交叉口排队长度的置信区间估计问题,并提出用概率分布偏差来描述排队长度,对结果精度进行估计。在仿真分析部分,通过视频识别技术获取交叉口的排队长度数据,并用随机采样方法模拟了交叉口轨迹数据。最后,通过不同时段的交叉口排队长度估算验证所提出的方法,其中凌晨及下午时段的排队长度估算结果的平均M;值分别为0.20及0.61,M;值分别为27.40%及7.47%。结合概率分布分析方法,判断出凌晨时段及下午时段的排队长度分布分别为非高斯分布及高斯分布,计算概率分布偏差分别为10.63%及7.93%,验证了所提出的精度分析方法相比传统分析方法,在小样本场景具有更高的准确性。
With the rapid development of connected vehicle systems and online taxi platforms,massive amounts of trajectory data have been collected.This has become a significant data source for traffic research owing to its high accuracy,wide distribution,and availability.As an important parameter for evaluating the operation status of intersections,queue length is of great significance for traffic status monitoring.This study intends to use trajectory data to estimate the queuing length at intersections and analyze the reliability and accuracy of the queuing length estimation results based on the historical queuing distribution.In this study,a method for estimating the queuing length at intersections based on Bayes’theorem was established.Under the assumption that the probability of each car being a trajectory vehicle is equal,the probability relationship between the queuing length and the parking position of the trajectory vehicle in each cycle was derived.During the estimation process,the queuing length frequency distribution of the trajectory vehicles was used to approximate the actual queuing length frequency distribution,and the unknown problem in the formula was solved.Combining historical queuing data at intersections,this study analyzes the confidence interval estimation problem of intersection queue length under Gaussian and non-Gaussian situations.Furthermore,the definition of probability distribution deviation was proposed to describe the accuracy of the queue-length estimation result.In the simulation experiment,the queuing length data were obtained from the video data of an actual intersection,and the simulated trajectory data were extracted by random sampling.Finally,the proposed method was verified by estimating the queue length at intersections during different periods.In the early morning and afternoon hours,the average values of the queuing length estimation results were 0.20 and 0.61,respectively,and the values were 27.40%and 7.47%,respectively.The actual data collected at the intersection show that the queue length distributions in the early morning and afternoon hours are non-Gaussian and Gaussian,respectively,and the calculated probability distribution deviations are 10.63%and 7.93%,respectively.It is verified that the proposed accuracy analysis method has higher accuracy in small-sample scenarios than traditional analysis methods.
作者
张伟斌
叶竞宇
白孜帅
李熙莹
ZHANG Wei-bin;YE Jing-yu;BAI Zi-shuai;LI Xi-ying(School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210094,Jiangsu,China;School of Intelligent Systems Engineering,Sun Yat-sen University,Guangzhou 510006,Guangdong,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2022年第3期216-225,共10页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2018YFB1601100)
国家自然科学基金项目(71971116).
关键词
交通工程
智能网联
贝叶斯定理
轨迹数据
排队长度估计
排队分布
精度分析
traffic engineering
connected vehicle
Bayes’theorem
trajectory data
queue length estimation
queue distribution
accuracy analysis
作者简介
张伟斌(1975-),男,陕西咸阳人,教授,博士研究生导师,工学博士,E-mail:weibin.zhang@njust.edu.cn;通讯作者:李熙莹(1972-),女,陕西西安人,副教授,博士研究生导师,工学博士,E-mail:stslxy@mail.sysu.edu.cn。
