【作者】 李静；

【导师】 安实；

【作者基本信息】 哈尔滨工业大学 ， 管理科学与工程， 2010， 博士

【摘要】 先进出行者信息系统(Advanced Traveler Information Systems, ATIS)作为智能交通运输系统(Intelligent Transportation System, ITS)的一个重要子系统,旨在通过提供实时的交通状况信息,影响出行者的出行行为决策,从而到达对路网出行环境的改善。因而,ATIS环境下驾驶员的行为分析是ATIS服务商实施有效服务策略、确定相应市场规模等决策问题的重要理论依据。ATIS环境下的交通系统由于其内在的动态性、离散性、复杂性和不确定性,使得从严格的数学建模角度去刻画驾驶员行为变得十分困难。而面向Multi-Agent的建模、仿真方法的产生,为描述复杂的交通系统中驾驶员的行为提供了崭新的计算范式。该方法试图通过建立交通中行为个体(主要是驾驶员—车辆单元)的Agent模型,限定Agent "生存”的ATIS交通环境,运用面向Agent (Agent-Oriented)仿真技术,可以在Agent与Agent之间和Agent与环境之间的相互竞争、合作、交互中,再现一个复杂而有序的系统的形成过程,从而考虑通过改变Agent所处的生存环境(主要指信息环境),影响Agent的决策,进而充分了解系统的形成机理,达到对系统的有效控制。本文首先提出了基于双层Agent架构的智能体模型,用于刻画ATIS环境下的驾驶员—车辆单元Agent,包括战略层行为模型和战术层行为模型,其中战略层行为模型负责实时信息下的出发时刻选择、出行路径选择等“头脑性”决策行为；而战术层负责实际的车辆运行中加速、减速、跟驰、换车道等基本仿真行为。在此基础上,为形式化表达驾驶员—车辆单元的战略层行为模型,本文引入了基于BDI的Agent模型,用于描述驾驶员决策过程中的内部推理逻辑,同时针对战术层行为,本文建立了基于“期望车距”的车辆跟驰模型与基于反应型Agent技术的车辆换车道模型。在此基础上,本文利用己建立的驾驶员战术层行为模型,基于微观交通仿真软件Paramics的可扩展编程接口,采用Borland公司通用开发环境Delphi,开发了交互式仿真机DPIS (Delphi-Paramics InteractiveSimulator)。该仿真机能够采用C/S (Client/Server)架构,通过服务器模拟路网交通运行及ATIS信息环境,使得客户端被调查者能够根据模拟的交通信息环境,进行实时信息下的行为决策,从而达到对实时信息下驾驶员行为数据的采集与分析。为实例化前续建立的BDI战略层行为模型,本文采用DPIS仿真机对多名通勤者进行了实时信息下的行为数据采集,并在此基础上,采用“有限理性”的行为分析框架,基于多项Probit (Multinomial Probit Model, MNP)统计分析模型,对实时信息下通勤者的逐日出行动态行为进行了深入分析,并基于离散选择模型和模糊逻辑相结合的方法,对实时信息下驾驶员的“服从”行为进行了刻画,从而达到了对驾驶员—车辆单元Agent战略层行为模型的实例化表达。最后,本文基于所开发的DPIS仿真机和上述建立的驾驶员—车辆单元Agent模型(包含了战略层行为模型和战术层行为模型),讨论了实时信息下驾驶员行为模型在若干ATIS设计与评价情景下的应用,主要包括ATIS信息提供比例的设计与对某种ATIS信息提供策略下的路网通行性能的逐日动态评价。以此为例,阐明了本文所建立的实时信息下驾驶员行为模型在ATIS项目的实施和评价中具体的应用。本文的研究成果可以为ATIS服务商运营管理的决策制定提供理论依据,同时丰富了ATIS环境下驾驶员行为分析的理论和方法,为以后ATIS领域的理论研究和实践探索提供了崭新的思路。更多还原

【Abstract】 As an important subsystem of Intelligent Transportation System (ITS),Advanced Traveler Information System (ATIS) is designed to provide real-timeinformation on traffic conditions to affect driver’s travel decision-makingbehaviors, in further to improve the performance of the whole transportationnetwork. Therefore, analyzing driver’s behaviors under ATIS environment is theimportant theoretical basis of both implementing effective service strategy for theATIS service providers and determining the corresponding market scales.Considering the inherent dynamic, discrete, complex and uncertain nature,transportation system under ATIS is difficult to model by traditionalmathematical methods. However, the modeling and simulating methods based onMulti-Agent provide a suitable and practical way for describing the drivers’behaviors under ATIS. Through developing travel individual behavior Agentmodel (mainly for driver-vehicle unit), and limiting the ATIS traffic environmentfor Agent “survival”, the paper use the Agent-Oriented simulation technology toreproduce a complex and orderly formation process of the system in thecompetition, cooperation, interaction among Agents. Moreover by changing theAgents survival environment (mainly changing the information environment), itinfluences the decision-making process of Agent, thereby fully understands theformation mechanism of the system to achieve the effective control of system.Firstly, an Agent model based on two-layer Agent architecture is presentedto describe the driver-vehicle unit under ATIS. The two-layer model includes thestrategic layer and tactical layer behavior models. The former is responsible forthe “head” of the decision-making behaviors, such as departure time choice,route choice and other behaviors under real-time information; while the latter isresponsible for the actual vehicle running in accelerating, decelerating, followingand lane-changing. This paper introduces the Agent-based BDI model to describethe internal reasoning logic of driver’s decision-making process. Meanwhile, forthe tactical layer, the paper develops the car-following model based on “desireedgap” and lane changing model based on reactive Agent technology.Secondly, the paper uses the driver behavior model established in tactical layer develop a interactive simulator DPIS (Delphi-Paramics InteractiveSimulator), based on the scalable programming interface of microscopic trafficsimulation software “Paramics” and the general developing environment“Delphi” developed by Borland. DPIS can simulate road network traffic andATIS information environment through servers by using the C/S (Client/Server)architecture. Respondents in the client point could make behavior decisionsunder real-time information according to the simulated environment, so as toachieve the driver behavioral data collection and analysis under real-timeinformation.Thirdly, in order to instantiate the BDI strategic layer behavior modelestablished above, the paper collectes the behavior data of several commuters byusing DPIS. Based on the data, the paper builds the behavior analysis frameworkby introducing “bounded rationality” theory and builds the statistical analysismodel based on Multinomial Probit Model (MNP) to analyze the departure timeswitching and route switching behaviors of commuters under ATIS. At the sametime, by combining the method of the discrete choice model and fuzzy logictheory, the paper characterizes the driver’s “obedience” behavior under real-timeinformation, so as to achieve the instantiation expression of driver-vehicle unitAgent strategic behavior model.Finally, based on the DPIS simulator developed and driver-vehicle unitAgent model built above (inluding strategic layer behavior model and tacticallayer behavior model), the paper discusses the application of driver behaviormodel under real-time information and several ATIS design and evaluationscenarios, maily focuses on the design of ATIS information supplying ratio andthe day-to-day dynamic evaluation of road network travel performance under acertain ATIS information supply strategies. In this way, the paper instructs thespecific application of driver behavior models built above under real-timeinformation in the implementation and evaluation of ATIS projects.The results provide theoretical basis for decision-making of ATIS operatorsin operation and management, meanwhile enrich the theories and methods ofdriver behavior analysis under ATIS environment, and provide a new train ofthought for future theoretical research and practical exploration.更多还原