【作者】 李志伟；

【导师】 冉斌； 孙兴焕；

【作者基本信息】 东南大学 ， 交通运输工程（专业学位）， 2017， 硕士

【摘要】 近些年来,随着我国机动车拥有量的快速提升,道路交通通行需求也快速增加。同时,高速公路大规模建设的减缓以及高速公路节假日免费通行政策使得交通供给与交通需求矛盾格外尖锐。现阶段的发展已经不容许我们继续扩大道路建设,改善高速公路运营与管理水平是唯一缓解供给与需求失衡问题的方法,而高速公路交通状态的全面感知是一切运营管理的基础,其中路段密度对高速公路管理者尤为重要。此外,机器学习技术、视频处理技术、传感器技术的飞速发展,也带来了车辆的变革。众多互联网公司与传统车企都投入重资加速智能汽车的研发,许多实用技术已经应用在车辆上,如自适应巡航系统、自主停车系统等等,智能汽车实际在道路行驶不会久远。在此情况下,未来的道路交通环境将会很长一段时间为智能汽车与普通汽车混合的局面。本文立足于智能网联汽车与普通汽车混合的交通环境,对高速公路路段密度估计方法进行研究,进而估计道路交通状态。首先,为了实现智能网联车辆与普通车辆混合的交通环境。论文对现有的交通模型进行了全面的了解,结合分析了智能网联车辆应该具有的驾驶特性,选定了合适的模型模拟智能网联车辆的驾驶行为,并利用NG-SIM数据库对模型的参数进行了标定。同时,论文以此为基础提出了智能网联汽车的控制策略,也对各个策略下的模型参数进行了说明。其次,研究智能网联车辆与普通车辆混合车流宏观交通流基本图。为确定混合交通流的宏观交通参数,论文利用Vissim的二次开发功能,结合前文提出的智能网联车辆模拟模型以及NG-SIM数据标定软件参数,实现了智能网联车辆与普通车辆混合环境的仿真。在此基础上多次实验获取了混合交通流的宏观交通参数。然后,通过数值仿真方法,确定智能网联车辆占比达40%时,智能网联车辆平均速度能够准确表征路段区间平均车速,并以此将混合交通流分为两类。此后,论文提出不同条件下的两种路段密度估计方法:利用固定检测器数据的kalman滤波估计方法,基于BP神经网络融合kalman估计值与智能网联车辆数据的融合方法,并讨论了各自适用场景。最后,论文设计仿真场景,对提出的路段密度估计方法进行了验证,并对两种方法使用环境进行研究。研究得出神经网络融合方法与kalman估计方法孰优孰劣与道路交通流和智能网联车辆占比有关。总体为自由流时段应用kalman滤波估计方法,拥挤流及高智能网联车辆占比且交通流发生变化环境,应用神经网络方法较优。更多还原

【Abstract】 With the rapid growing of vehicle ownership in China recently,traffic demand of road traffic is increasing along with it.At the same time,the slowing grows of highway construction,as well as the highway toll free policy in red-letter day makes the contradiction of traffic supply and traffic demand sharper.At the present stage of development has not allow us to continue to expand the road construction,and then improving the level of highway operation and management is the only way to alleviate the problem.The basis of all operations management is the overall perception of highway traffic state,the road density is particularly the important one.In addition,the speedy development of machine learning technology,video processing technology,and sensor technology brought about changes in vehicles.Many Internet Company and traditional car companies turn their eyes on the intelligent vehicle research and development,and many practical techniques have been applied in the vehicle,such as adaptive cruise system,autonomous parking system and so on.The actual intelligent vehicle in the road will not be long.In this case,the future road traffic environment will be smart cars and manual cars mixed situation for a long time.This paper is established in the connected and autonomous vehicles and manual vehicles mixed environment,and research the method of highway density estimation.First,in order to realize the connected and autonomous vehicles and manual vehicles mixed environment,this thesis presents a comprehensive understanding of the existing traffic models.Combined with the analysis of the characteristics of connected and autonomous vehicle driving behavior,thesis has choose a suitable model to simulate the connected and autonomous vehicles.The parameters of this model were calibrated by the NG-SIM database,and then control strategy of the connected and autonomous vehicle are proposed,model parameters of each strategy are also described.Secondly,the fundamental diagrams of traffic flow is studied.In order to determine the macroscopic traffic parameters of the mixed traffic flow,second development function of the Vissim is selected.And then creating the simulation of the connected and autonomous vehicles and manual vehicles mixed environment,which the vehicle simulation model and parameters are mentioned above.On the basis of these experiments,the macroscopic traffic parameters of mixed traffic flow are obtained.Then,while the penetrance of the connected and autonomous vehicles reaches 40%,that the connected and autonomous vehicles’ average speed can accurately characterize the link average speed is found through the numerical simulation.And the mixed traffic flow is divided into two categories.Since then,two road density estimation methods,kalman filtering method and fusion estimation method based on BP neural network,are proposed.Their application scenarios are also discusses in the latter.Finally,the simulation scenario is designed to validate the proposed method.The thesis concludes that the superiority of those two methods is related to traffic flow and the penetrance of the connected and autonomous vehicles.While the traffic is free flow,Kalman filter estimation method is chosen.And while the traffic is confused or the traffic flow has phase transition,BP neural network fusion method is better.更多还原