返回
类型 综合研究 预答辩日期 2017-11-20
开始(开题)日期 2015-10-26 论文结束日期 2017-06-28
地点 东南大学交通学院三楼会议室 论文选题来源 其他项目    论文字数 9 (万字)
题目 高速公路车辆自主性换道行为建模研究
主题词 高速公路,人工驾驶车辆自主性换道决策模型,网联自动驾驶车辆自主性换道决策模型,深度学习,博弈论
摘要 交通拥堵广泛存在于世界上大多数城市。用于改善交通的方法或算法在被实用之前往往需要校准或验证。然而,由于成本高昂和安全问题,除了非常有限的交通管理方案,大多数不适合直接进行现场测试。此外,随着智能交通系统(ITS)技术的发展,在开始阶段使用现场测试方法考察交通管理策略是不切实际的。因此,需要微观交通仿真工具来执行模拟环境中的交通管理方案规划和评估工作。高速公路作为当前最主要的车辆驾驶环境之一,由此可见,在复杂运输条件下建立更加现实和实用的高速公路微观交通流模型是非常需要的。换道是车辆最基础的微观驾驶行为之一。因此,本文围绕高速公路车辆的自主性换道决策行为建模进行研究。 针对现有高速公路车辆换道研究较少聚焦于自主性换道场景,论文以高速公路车辆的自主性换道行为为研究对象。以NGSIM项目中的US101和I80观测路段车辆轨迹数据为基础,提出了高速公路人工驾驶车辆自主性换道过程的关键节点识别方法,更进一步分析了高速公路人工驾驶车辆在自主性换道过程中的行为特征。 针对现有人工驾驶车辆换道决策模型忽略整合换道准备过程,并且由于公式化模型的过强主观经验性限制模型使用场景的问题,论文提出一种适用性更强的以数据驱动为基础的基于深度学习的高速公路人工驾驶车辆自主性换道决策模型。模型适用场景假设高速公路车辆全部为人工驾驶车辆。该模型将高速公路人工驾驶车辆的自主性换道决策过程分解成目标车道选择、目标间隙选择、目标间隙可接受性判别和换道准备四个子过程。目标车道选择问题可以转化为三分类问题,即可理解为目标车辆根据过去连续多个时刻自身和其周围车辆的行驶状态从当前车道、左侧车道和右侧车道中选择下一个时刻的目标车道。因此,提出采用深度循环神经网络学习器集成方法对目标车道选择问题建模。目标间隙选择问题的实质是从目标车辆从选定的目标车道上的当前相邻间隙及其后边间隙中选择目标间隙,可以理解成二分类问题。目标间隙可接受性判别问题的实质是判别选定的目标间隙是否可以立即进行换道插入动作,同样可以转化成二分类问题。目标间隙选择问题和目标间隙可接受性判别问题均采用深度前馈神经网络集成方法建模。换道准备过程的核心是对目标车辆纵向加速度的预测,论文采用深度前馈神经网络学习器集成方法建模。 基于网联自动驾驶车辆在信息感知、决策制定和驾驶操作等方面与传统的人工驾驶车辆的差异性,人工驾驶车辆换道决策模型不能较好地描述网联自动驾驶车辆的换道决策行为。本文结合网联自动驾驶车辆的特征和功能优势提出一种适用于高速公路网联自动驾驶车辆的分布式自主性换道决策模型。该模型框架由期望换道决策模型和协同换道决策模型构成。对通过引入网联自动驾驶车辆实时车-车信息交互的特征对传统换道模型MOBIL进行改进,在此基础上提出基于MOBIL改进的期望换道决策模型。根据期望换道决策可能存在的相互直接干扰或影响,将网联自动驾驶车辆协同换道的情形划分为四类。针对四类协同换道情形,利用双矩阵法分别构建基于博弈论的网联自动驾驶车辆协同换道决策模型。最后,结合IDM改进模型利用Matlab软件设计数值仿真实验对分布式高速公路网联自主车自主性换道决策模型进行仿真评价。为分析本文提出的网联自动驾驶车辆自主性换道决策模型对交通流的影响,本文设计了四组不同的跟驰+换道模型组合进行比较。
英文题目 RESEARCH ON MODELING DISCRETIONARY LANE-CHANGING BEHAVIORE OF VEHICLES IN FREEWAY
英文主题词 freeway,discretionary lane-changing decision-making model of manned vehicle,discretionary lane-changing decision-making model of connected and autonomous vehicle,deep learning,game theory
英文摘要 Traffic congestion is widespread in most cities in the world. Methods or algorithms used to improve traffic often need to be calibrated or validated before being practiced. However, due to the high cost and safety issues, in addition to very limited traffic management programs, most are not suitable for direct field testing. In addition, with the development of Intelligent Transportation Systems (ITS) technology, it is impractical to use field testing methods to examine traffic management strategies at the beginning. Therefore, micro-traffic simulation tools are needed to perform traffic management planning and evaluation in a simulated environment. Considering that Freeway is one of the most important vehicle driving environment, we can see that it is very necessary to establish a more realistic and practical freeway micro-traffic flow model in the complex transport conditions. Lane change is one of the most basic micro driving behaviors of vehicle. Therefore, this paper focuses on the modeling of the discretionary lane-changing decision-making behavior of freeway vehicles. In view of the existing freeway vehicle lane-changing research ignores the discretionary lane-changing behavior, the paper aims at research on the freeway vehicle discretionary lane-changing behavior. Based on US101 and I80 dataset in the NGSIM project, the key point recognition method of the discretionary lane-changing process of manual vehicle on the freeway is put forward, and the behavioral characteristics of manual vehicle during the discretionary lane-changing process in freeway is further analysed. Aiming at the problem that the existing lane-changing decision model of manual vehicle neglects integrating lane-changing preparation process and empirical formula model is too strong subjective, the thesis proposes a data-driven freeway manual vehicle discretionary lane-changing decision-making model based on deep learning. Model asummption is that all freeway vehicles are manned vehicles. The model divides the discretionary lane-changing decision-making process of freeway manned vehicle into the four sub-processes: targe lane selection, target gap selection, target gap acceptance judgment and lane-changing preparation. The target lane selection problem can be transformed into a three-class classification problem, that is, the target vehicle select the target vehicle lane from the current lane, the left lane and the right lane according to the driving state of itself and its surrounding vehicles in the past several continuous time instant. Therefore, deep feedfoward neural network learner ensemble method is proposed to model target lane selection problem. The essence of the target gap selection problem is that the target vehicle selects the target gap from the current adjacent gap on the selected target lane and its adjacent rear clearance. The essence of the target gap acceptance judgment problem is to determine whether the selected target gap is suitable for target vehicle to insert immediately, and can be converted into a two-class classification problem. Target gap selection problem and target gap acceptability judgment problem are modeled by deep feedforward neural network ensemble method. The core of the lane-changing preparation process is to predict the longitudinal acceleration of the target vehicle and the thesis also uses the deep feedforward neural network learner ensemble method to model it. Based on the differences between information sensing, decision making and driving operation, the lane change model based on manned vehicle can not describe the driving behavior of the connected and autonomous vehicle accurately. In this thesis, a distributed discretionary lane-changing decision-making model for autonomous vehicles is proposed based on the characteristics and functional advantages of connected and autonomous vehicles. The model framework is composed of the expected lane-changing decision model and the cooperative lane-changing decision model. By introducing the real time V2V information interaction characteristics of connected and autonomous vehicle into the traditional lane chenge model—MOBIL, we proposed a based improved MOBIL expected lane change decision model. According to the potential immediate interactive disturbation or influence, cooperative lane change conditions of connected and autonomous vehicle are catergorized into four classes. Aimming at the four coorperative lane change conditons, we adopted to construct based game theory discretionary lane-changing decision-making model of connected and autonomous vehicle. Finally, the numerical simulation experiment is designed by using Matlab software to simulate the discretionary lane-changing decision-making model of freeway connected and autonomous vehicle. In order to compare the influence of the expected lane change decision model and the cooperative lane change decision model on the traffic flow, this paper designs four groups of different car-following+lane-changing model combinations to compare.
学术讨论
主办单位时间地点报告人报告主题
东南大学 2016.10.23 交通学院213室 聂建强 高速公路车辆自主性换道过程目标车道选择模型
东南大学 2015.11.11 交通学院3楼会议室 杨超教授 基于手机数据的个体活动特性研究
东南大学 2015.12.11 交通学院3楼会议室 曲晓波博士 On the fundamental diagram for freeway traffic
东南大学 2016.04.29 交通学院3楼会议室 张学孔教授 智慧出行绿色交通
东南大学 2014.04.16 中山院 刘攀教授 交通安全科研发展的历史与展望
东南大学 2014.11.28 东南院102室 徐铖铖博士 研究、撰写和发表高质量SCI论文的方法和经验
东南大学 2014.09.20 交通学院213室 聂建强 基于手机切换定位技术的道路匹配
东南大学 2015.09.10 交通学院213室 聂建强 普通环境下高速公路车辆自主性换道过程
东南大学 2017.03.08 交通学院214室 聂建强 Decentralized Cooperative Lane Change Decision-Making for Connected and Autonomous Vehicle
     
学术会议
会议名称时间地点本人报告本人报告题目
IEEE Intelligent Transportation Systems Society 2016.11 Rio, Brazil Modeling of decision-making behavior for discretionary lane-changing execution
Transportation Research Board 2017.01 Washington D.C, America Modelling of Vehicle Interaction Behavior during Discretionary Lane-changing Preparation Process on Freeways
     
代表作
论文名称
Decentralized Cooperative Lane-Changing Decision-Making for Connected Autonomous Vehicles
Modeling of decision-making behavior for discretionary lane-changing execution
A Novel Approach to Road Matching Based on Cell Phone Handover
 
答辩委员会组成信息
姓名职称导师类别工作单位是否主席备注
季锦章 正高 其他 中设设计集团股份有限公司
赵佳军 正高 研究员级高级工程师 其他 江苏高速公路联网营运管理有限公司
李文权 正高 教授 博导 东南大学
杨敏 正高 教授 博导 东南大学
何杰 正高 博导 东南大学
      
答辩秘书信息
姓名职称工作单位备注
曲栩 其他 讲师 东南大学