面向交通控制的多智能体设计的理论与模型研究

VIP免费
3.0 赵德峰 2024-11-19 5 4 5.39MB 129 页 15积分
侵权投诉
i
摘 要
城市交通的飞速发展带来了城市交通拥挤、交通延误、交通事故和交通污染
等一系列问题。解决城市交通问题的最直接的办法就是修建更多的道路和桥梁以
提高路网的通行能力。但是由于资金和城市空间限制，这一方法往往不是十分可
行。智能交通系统（ITS）的发展为解决城市交通问题提供了新思路。智能交通系
统包括城市道路交通信息采集、处理、发布、决策，它运用各种先进的技术和科
学方法，实现城市交通管理的自动化、现代化和智能化。智能交通系统不仅能提
高路网的利用率、交通的安全性，而且关系到土地资源和能源的合理利用、环境
污染和噪声的改善，乃至国民经济的持续发展和社会经济效益的全面提高。城市
交通信号控制系统作为智能交通系统中的一个重要子系统，已经成为当前研究的
热点课题之一。本文通过对城市交通信号控制问题的研究，寻求更有效的城市交
通控制方法，为缓解城市交通的拥堵问题，改善交通状况提供科学依据。
针对城市交通路网的实际情况，本文研究借鉴多智能体技术的概念和方法，建
立了多智能体城市交通信号控制系统（AUTCS）。系统包括三类信号控制智能体：
交叉口交通控制智能体、干线协调交通控制智能体和区域交通控制智能体。正常
交通流量下，各路口的主要控制工作由交叉口Agent给出控制策略；一般堵塞情况
下，由交叉口Agent和上游交叉口Agent协商解决；中等堵塞情况下，交叉口Agent
和干线控制Agent通信，将控制权交给干线控制Agent；严重堵塞情况下，由区域控
制Agent来进行整体控制。由于智能体都包含知识库和推理机制，使得这种具有自
主性、主动性、交互性的智能体交通控制不仅充分考虑了相关车队信息和周边路
口的交通状况，兼顾到路网交通控制整体优化的实现，而且能通过自学习不断增
加自身的控制经验，逐渐达到较优的控制效果。因此，对于大城市、特大城市的
大规模城市交通网，这种多智能体的城市交通信号控制系统，在实时控制效果方
面具有一定的优势。
本论文从单路口交通信号控制入手，将具有感知和反应特点的信号控制智能体
作为交叉口信号控制器，通过对到达车辆的模糊聚类处理形成对路口交通状态的
定量描述，根据交通控制规律和经验建立面向各种交通状态的信号控制规则集，
对交叉口车流进行实时控制。同时以车辆总延迟时间作为交通控制的优化指标，
采用PSO算法对信号控制规则进行优化，在整个信号过程中使用不同的规则组合进
行信号控制，淘汰控制指标较差的控制方案，对信号控制规则集进行持续的改进，
建立了一种具有学习能力的单交叉口交通信号控制模型。本文设计了基于博弈论
的智能体协调模型，当路口车辆较多，单个交叉口控制不能明显改善控制效果时，
交叉口智能体可和相邻交叉口进行协调，通过协调控制来改善控制效果。
本文系统地分析了干线信号协调控制的最佳适用条件，设计了干线控制智能
体，给出了干线智能体信号控制模型和自学习模型。当路口出现中等堵塞时，启
动干线智能体生成线控方案。干线智能体采用模糊聚类对交通状态进行了识别，
针对不同的交通状况采用不同的干线控制策略。非高峰期时，以车辆延误为控制
目标，上行高峰期或下行高峰期时，采用单向绿波控制，双向高峰期时，采用双
向绿波控制。线控模型的公共周期采用模糊算法确定，相位差优化模型采用Q-学
习算法确定。
本文研究还包括当干线智能体控制效果不佳，路口出现严重堵塞时，系统启动
区域智能体进行区域协调控制，实现整个控制子区域交通网络的最优控制。本文
设计了控制子区的动态划分方法，区域智能体以控制子区为优化对象，将控制子
区划分为若干干线，逐条干线进行控制。区域协调模型以平均延误最小、车辆平
均停车次数最小和路口排队长度最小等性能指标中的一项或几项构造目标函数，
进行优化。控制子区内各路口的周期和相位差由区域智能体采用模糊控制算法获
得，发送给干线智能体，再由干线智能体发送给路口智能体执行。
为了验证多智能体的城市交通信号控制系统的可行性和有效性，编制交叉口、
干线和区域控制智能体、结合AIMSUN仿真软件，模拟智能体交通控制，并对一
个由9个路口组成的交通网络进行仿真实验，模拟感应控制和多智能体交通信号控
制方法，分析比较不同控制方法的控制效果。仿真结果表明，在相同的条件下，
多智能体信号控制模型的路网平均延迟时间比感应控制方式平均减少15.7%，车辆
的平均速度提高了10.4%。
关键词：城市交通控制，多智能体系统，分布式控制，模糊算法，粒
子群算法，绿波控制，Q-学习算法，交通控制微观仿真
iii
ABSTRACT
With the rapid development of urban traffic, a series of problems are coming such
as urban traffic congestion, traffic delay, traffic accident and traffic pollution, and so on.
Obviously, building more roads and bridges are direct solution to resolve these urban
traffic problems, but due to the funds and the urban room, those methods are not
feasible. Development of intelligent transportation system (ITS) provides new solution
for urban transport problems. Intelligent transportation systems, including traffic
information collection, processing, publishing, decision-making process, uses a variety
of advanced technology and scientific methods to achieve automation, modernization
and intelligent of urban traffic management. The application of ITS will not only
improves the transportation utilization and safety, but also connects with land resources
and energies exploitation, environment improvement, national economic and social
revenue development. Urban traffic signal control system is one of the most important
subsystems of the ITS, is hot problem to study. By the research urban traffic signal
control, appropriate urban traffic signal control methods are seeking, for solving the
urban traffic congestion and congestion problems.
According to the actuality of traffic network, using the concepts and approaches of
multi-agent system, agent oriented urban traffic signal control system (AUTCS) is
established. It includes three types of agent, intersection control agent, trunk control
agent and area control agent. Under normal traffic flow, intersection agent gives
intersection control strategy; under light congestion, intersection agent and upstream
intersection agent negotiates control strategy; under middle congestion, intersection
agent communicates with trunk agent, hands over control to trunk agent; under serious
congestion, area agent performs overall control. Since agent contains knowledge base
and inference mechanism, the use of such autonomous, proactive, interactive agent in
signal control, not only gives full consideration to the vehicle query information and
traffic conditions around the intersection, taking into account optimization of overall
network traffic, but also greatly reduces the complexity of control algorithms. Therefore,
for large cities, large-scale urban transportation network, this multi-agent-based urban
traffic signal control system, has a considerable advantage in real-time traffic control.
This dissertation starts with the signal control of isolated intersection. It makes the
signal control agent with perception and reaction characters as the intersection signal
controller. It forms quantitative description for traffic states of intersection through
fuzzy classifying the arrived vehicles, constructs the signal control rules according to
the disciplines and the experience of traffic control for all traffic states, and performs
real-time control on intersection vehicles. Vehicle delay as the optimizing goal of traffic
control, Particle Swarm Optimization (PSO) algorithm is applied to optimize signal
control rules. During signal control process, different combinations of control rules is
applied and eliminates several schemes with worse results, thus the signal control rules
can be improved continuously. These measures form a signal control model with
self-learn ability for single intersection. This dissertation also puts forward an agent
coordination model based on game theory. Under the large traffic flow, single
intersection control can not improve the control effect obviously, intersection agent
negotiates with upstream intersection agent, realizes of the overall coordination control
to improve control effect.
This dissertation systematically analyzes the optimal application conditions of
trunk coordination control, designs trunk control agent, and gives control model and
self-learning model of trunk agent. Under middle congestion, trunk agent starts to make
line control strategies. Trunk agent identifies traffic state with fuzzy classifying methods,
applies different control strategies in different traffic conditions. In off-peak periods,
vehicle delay is control target; in up line or down line peak period, one-way green wave
control is applied; in bi-peak period, two-way green wave control is used. Common
signal control cycle in trunk model determined by fuzzy algorithm, phase offset
optimization uses Q-learning algorithm.
When trunk control results less effect, and serious congestion occurs, area control
agent performs overall coordination control. It optimizes whole traffic flow, and realizes
optimal control of regional transport network. The dissertation gives traffic area
dynamic partition method, area agent regards traffic area as optimizing object, divided
the area into a number of trunk lines, and control trunk one by one. Area coordination
model constructs objective function with one or several performance parameters,
including minimum travel delay time, minimum vehicle stop times, and minimum
vehicle query length, and performs optimization. Signal cycle and phase offset of
intersections in control area are obtained by fuzzy control algorithm, and sent to trunk
agent, then sent to intersection agent for implementation.For testing the feasibility and
effect of multi-agent oriented urban traffic signal control system, program of
intersection, trunk and area control algorithm is made, applies to AIMSUN simulation
v
system, and performs simulation in 9-intersection traffic network. AIMSUN simulates
control process of inductive control method and agent-oriented control method, analysis
and compares control effect of different methods. Simulation shows AUTCS signal
control approach reduces 15.7% average vehicle delay than the actuated method and
increases 10.4% average vehicle speed in the same condition.
Key Word: Urban Traffic Control, Multi-Agent System, Distributed
Control, Fuzzy Algorithm, Particle Swarm Optimization, Green Wave
Control, Q-learning, Microscopic Simulation of Traffic Control
I
目 录
中文摘要
ABSTRACT
第一章 绪 论··············································································· 1
§1.1 选题目的与意义·································································1
§1.2 城市交通控制研究的发展·····················································3
§1.2.1 交通信号控制的历史····················································· 3
§1.2.2 国外交通信号控制系统的发展········································· 3
§1.2.3 国内交通信号控制系统的发展········································· 5
§1.2.4 交通信号控制系统发展方向············································ 6
§1.2.5 Agent 技术在交通控制系统的研究现状······························ 6
§1.3 本文研究内容····································································8
§1.4 本文的特色与创新······························································9
第二章 多智能体城市交通信号控制系统建模······································ 11
§2.1 智能体和多智能体系统······················································ 11
§2.1.1 智能体（Agent）的定义················································11
§2.1.2 多智能体系统（Multi-Agent System,MAS）·······················12
§2.1.3 多智能体交通控制系统的可行性和优越性·························12
§2.1.4 多智能体的城市交通信号控制系统的优势·························13
§2.2 AUTCS 的Agent 的设计····················································· 14
§2.2.1 Agent 的结构类型························································ 14
§2.2.2 Agent 的结构设计························································ 15
§2.2.3 Agent 的智能决策功能··················································16
§2.2.4 Agent 的自学习功能····················································· 17
§2.3 Agent 交通控制系统设计···················································· 18
§2.3.1 交通控制系统设计目标················································· 18
§2.3.2 交通控制系统的个体智能体···········································18
§2.3.3 Agent 交通控制系统结构···············································19
§2.4 交通控制 Agent 的通信······················································ 27
§2.4.1 Agent 的通信方式························································ 27
§2.4.2 交通控制 Agent 的通信设计··········································· 28
§2.4.3 交通控制 Agent 的拓扑结构··········································· 29
§2.5 交通控制 Agent 的实现······················································ 29
§2.5.1 CORBA 结构······························································ 29
§2.5.2 基于 CORBA 的Agent 开发步骤····································· 30
§2.5.3 基于 CORBA 的Agent 开发过程····································· 31
§2.6 本章小节········································································ 34
第三章 单交叉口交通控制智能体模型··············································· 35
§3.1 单交叉口控制基本理论······················································ 35
§3.1.1 交叉口控制基本参数···················································· 35
§3.1.2 单交叉口信号控制模型················································· 35
§3.2 交叉口 Agent 的设计························································· 38
§3.2.1 交叉口控制目标·························································· 38
§3.2.2 交叉口 Agent 结构······················································· 38
§3.2.3 交叉口控制 Agent 配置················································· 40
§3.2.4 交叉口 Agent 的智能决策过程········································ 40
§3.2.5 交叉口 Agent 工作流程················································· 42
§3.3 交叉口 Agent 的控制模型··················································· 43
§3.3.1 交叉口描述································································ 43
§3.3.2 交叉口 Agent 信号控制原理··········································· 43
§3.3.3 模糊控制器的设计······················································· 44
§3.3.4 交叉口 Agent 控制策略················································· 45
§3.4 交叉口 Agent 的自学习模型················································ 46
§3.4.1 路口交通状态描述······················································· 46
§3.4.2 交叉口交通信号控制规则集···········································46
§3.4.3 PSO 算法优化信号控制策略···········································47
§3.5 交叉口 Agent 的协调模型··················································· 51
§3.5.1 已有的协调控制模型···················································· 51
§3.5.2 基于博弈论的多智能体协调控制·····································51
§3.6 本章小节········································································ 53
第四章 干线协调控制智能体模型····················································· 55
§4.1 干线控制系统相关理论······················································ 55
§4.1.1 干线控制模型相关参数················································· 55
§4.1.2 干线控制系统的定义···················································· 56
§4.1.3 传统的干线控制控方法················································· 56
§4.1.4 智能干线控制系统······················································· 57
§4.1.5 智能体干线控制模型的优势···········································59
§4.2 干线控制 Agent 设计························································· 59
§4.2.1 干线控制目标····························································· 59
§4.2.2 干线控制 Agent 结构···················································· 60
§4.2.3 干线控制 Agent 的配置················································· 61
§4.2.4 干线控制驱动····························································· 61
§4.2.5 干线控制 Agent 的工作流程··········································· 61
§4.3 干线 Agent 控制模型························································· 62
§4.3.1 干线交通控制原理······················································· 62
§4.3.2 干线交通状况的识别···················································· 63
§4.3.3 周期的确定································································ 65
§4.3.4 绿信比的确定····························································· 66
§4.3.5 相位差优化模型·························································· 66
§4.4 本章小节········································································ 71
第五章 区域控制智能体模型····························································73
§5.1 区域控制相关理论···························································· 73
§5.1.1 区域信号控制的含义···················································· 73
§5.1.2 区域信号控制系统介绍················································· 73
§5.2 交通控制区域的划分························································· 79
III
§5.2.1 交通控制子区定义······················································· 79
§5.2.2 控制子区划分的原则···················································· 79
§5.2.3 周期原则子区自动划分过程···········································80
§5.3 区域控制 Agent 的设计······················································ 81
§5.3.1 区域控制内容····························································· 81
§5.3.2 区域 Agent 控制目标···················································· 82
§5.3.3 区域控制 Agent 的结构················································· 83
§5.3.4 区域控制 Agent 的配置················································· 83
§5.3.5 区域控制 Agent 的工作流程··········································· 84
§5.4 区域控制 Agent 模型························································· 85
§5.4.1 区域控制原理····························································· 85
§5.4.2 区域控制目标函数······················································· 85
§5.4.3 优化周期··································································· 86
§5.4.4 优化相位差································································ 87
§5.4.5 优化绿灯时间····························································· 88
§5.5 AUTCS 模型的优势···························································88
§5.5.1 AUTCS 模型的优点····················································· 88
§5.5.2 AUTCS 模型与 SCATS 模型比较·····································89
§5.5.3 AUTCS 模型与 SCOOT 模型比较····································90
§5.5.4 AUTCS 模型与其它多智能体模型比较····························· 91
§5.6 本章小节········································································ 92
第六章 多智能体城市交通控制模型的仿真验证··································· 93
§6.1 AIMSUN 仿真软件···························································· 93
§6.1.1 AIMSUN 简介·····························································93
§6.1.2 AIMSUN 的功能结构··················································· 93
§6.2 仿真内容········································································ 95
§6.3 设计技术路线·································································· 96
§6.4 仿真环境的搭建······························································· 96
§6.5 仿真分析········································································ 97
§6.5.1 点控制仿真································································ 97
§6.5.2 线控制仿真·······························································102
§6.5.3 面控制仿真·······························································105
§6.5.4 仿真结果分析····························································109
§6.6 本章小节······································································· 110
第七章 结论与展望······································································ 111
§7.1 结论············································································· 111
§7.2 今后的工作···································································· 112
§7.3 展望············································································· 113
参考文献····················································································· 114
在读期间公开发表的论文和承担科研项目及取得成果··························· 122
致谢··························································································· 123
第一章 绪 论
1
第一章 绪论
§1.1 选题目的与意义
经济的高速发展，交通运输也飞速增长，伴随而来的是交通状况的不断恶化，
以及一系列的交通问题：交通拥挤、交通污染严重、交通安全问题等。面对如此
多的交通问题，专家们极力地寻求解决的途径。过去常用的方法是增加交通供给，
多修路、建高架。但是交通供给不是无限的，而且交通供给的代价非常昂贵。因
此交通供给达到一定的水平以后，人们将解决交通问题的视线转向对交通需求的
控制，如发展公共交通，减少城市中小汽车的使用数量、提高泊车收费等，希望
从供、需这一交通矛盾的两方面解决问题。随着现代信息通讯技术和计算机技术
的快速发展，多学科专家从系统工程学的角度出发，提出了基于现代信息通讯技
术和计算机技术的、“人、车、路、环境”四位一体的交通系统，通过对交通信
息的利用，在不增加现有交通供给的条件下，大大提高道路交通的安全性、交通
设施的效率和行车舒适性，减少交通污染。在此基础上逐步形成了智能交通系统
的概念[1]。
智能交通系统（Intelligent Transportation System,简称ITS）是将信息技术、通
信技术、控制技术和系统集成等有效地应用于交通运输系统，从而建立起大范围
内发挥作用的实时、准确、高效的交通运输系统。城市交通是一个复杂的大系统，
城市交通规则和城市交通信号控制仅仅是城市交通网络建设和道理交通管理的重
要环节，单独从车辆方向考虑或单独从道路方面考虑都是片面的，凭借它们尚不
足以解决交通拥挤和交通安全问题。所以把人、车、路、环境综合起来考虑，充
分应用现代科学技术的智能交通系统为解决城市交通问题提供了全新的方法[2]。
近年来，我国的交通控制已有很大的进展，但相对于城市的扩展、路网规模
的扩大和车辆的急剧增多，交通控制系统仍显滞后。现有控制方式在处理大规模
路网的交通控制、路网控制的整体优化以及实时控制等方面存在明显不足，对于
车辆的实时信息处理分析简单，特别是对于车流的急剧变化在控制方案的选择上
很难做出相应调整。这说明交通系统的复杂性难以使用传统方法精确描述和解决。
智能控制理论的发展，为城市交通问题的解决提供了新的思路。多智能体交通控
制系统是基于专家系统的智能交通控制系统，该系统以知识库中的知识为基础进
行推理，生成最佳控制方案；并具有自学习功能，能够较强地适应环境的变化，
可以从控制过程中总结控制经验，生成知识，最终获得较好的控制效果。以智能
体为基本要素、使用分布式问题求解方法的多智能体系统（Multi-Agent Systems，
MAS）是未来城市智能交通系统发展的必然趋势。交通控制系统的分布式拓扑结
构非常适合采用多智能体技术，基于多智能体技术的控制策略将是实现城市交通
面向交通控制的多智能体设计的理论与模型研究
2
智能优化控制的重要途径，有助于区域综合交通网络管理的实现。
本文依托上海市科委国际技术合作基金项目“基于多智能体的道路多交叉口
信号集成控制仿真系统及信息诱导技术研究”，主要研究面向交通控制的多智能体
设计的理论与模型，研究多智能体技术在交通区域控制中的应用。本文研究的意
义如下：
(1) 城市规模的不断扩大，在区域控制中应用智能控制方法已经成为必然趋
势。随着现代城市的规模扩大，在对城市交通控制的研究中，以整个城市或区域
作为研究对象的区域控制方法越来越受到研究人员的重视。传统的区域控制方法
是通过对整个区域的路网及交通流进行数学抽象，在此基础上建立数学模型[3,4]，
运用运筹学以及最优控制理论求取最优控制变量，以期使整个交通流系统的整体
控制达到最优。但由于城市交通流系统是一个典型的非线性、动态时变的、不确
定性的复杂大系统，因此想要对它建立精确的数学模型比较困难，所用的最优算
法在实际应用中未必最优，因此智能控制方法逐渐应用到了交通系统控制研究当
中，应用专家系统等智能化方法来解决交通系统问题。
(2) 多智能体技术在解决城市交通分布控制问题上具有明显的优势。从计算机
科学的角度来看，城市交通系统是个分布式的多任务实时系统，必须有正确的理
论算法和有效的工具才能保证系统在动态运行过程中的正确性和可靠性。智能体
具有感知能力、问题求解能力和外界通信能力，能使逻辑上和物理上分散的系统
并行、协调地实现问题求解，能在过程运行中进行决策或控制，因此能较好地进
行交通控制。
(3) 国内外研究表明，传统的建模方法已经不能很好地刻画复杂系统，需要采
用新的建模理论与仿真方法。而 Agent 建模理论和仿真技术是最具活力、最有效
的方法之一，适合于复杂系统的研究。Agent 建模的基本思想是通过模拟现实世界，
将复杂系统划分为与之相应的 Agent（每个 Agent 具有各自的数据、知识、模型以
及接口等），以自底向上的方式，从研究个体微观行为着手，进而实现系统宏观
行为，这种设计方法，符合交通运输系统的形成规律。MAS 的目标是将复杂系统
分解成小的、彼此相互通讯及协调的、易于管理的系统，从而找到管理和控制这
类系统的有效方法。
综合以上可知，本研究顺应我国交通发展的实际需要，将为我国城市智能交
通系统提供一个切实可行的理论模型，为交通部门提供强有力的交通管理和控制
工具。
摘要：

i摘要城市交通的飞速发展带来了城市交通拥挤、交通延误、交通事故和交通污染等一系列问题。解决城市交通问题的最直接的办法就是修建更多的道路和桥梁以提高路网的通行能力。但是由于资金和城市空间限制，这一方法往往不是十分可行。智能交通系统（ITS）的发展为解决城市交通问题提供了新思路。智能交通系统包括城市道路交通信息采集、处理、发布、决策，它运用各种先进的技术和科学方法，实现城市交通管理的自动化、现代化和智能化。智能交通系统不仅能提高路网的利用率、交通的安全性，而且关系到土地资源和能源的合理利用、环境污染和噪声的改善，乃至国民经济的持续发展和社会经济效益的全面提高。城市交通信号控制系统作为智能交通系统中...

展开>> 收起<<
面向交通控制的多智能体设计的理论与模型研究.pdf

共129页,预览10页

还剩页未读， 继续阅读

相关推荐

更多
立即下载
作者：赵德峰 分类：高等教育资料 价格：15积分 属性：129 页 大小：5.39MB 格式：PDF 时间：2024-11-19

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务

作者详情

相关内容

更多

推荐作者

热门标签

/ 129
评分 收藏
分享
立即下载 VIP免费下载
客服
关注