【作者】 沈洋；

【导师】 方志耕；

【作者基本信息】 南京航空航天大学 ， 管理科学与工程， 2012， 博士

【摘要】 近年来，ABMS方法在管理科学领域逐渐受到了人们的关注，成为研究复杂问题的一种重要手段。但是目前关于ABMS方法本身的研究还有诸多不完善之处，不仅缺乏简便易行的建模方法，而且一些重要的技术问题也有待解决，这些缺陷影响了ABMS方法的进一步发展和应用。为此，论文基于多主体系统仿真过程控制方法与贝叶斯网络决策建模问题研究展开研究，旨在从完善ABMS方法本身的角度进行一些有价值的工作，即设计适合多主体系统的仿真事件调度方法、研究不确定性信息下Agent决策过程建模方法，以及面向管理专家的ABMS建模方法。研究成果对于丰富ABMS方法理论、提高ABMS方法在复杂管理问题中的应用水平具有重要的理论和现实意义。论文借鉴离散事件仿真理论原理，结合MAS的特点，提出了适用于复杂事件调度逻辑下的MAS仿真过程控制方法——5步扫描法。本着将仿真调度逻辑与Agent行为逻辑及交互逻辑分离的原则，该方法对多主体仿真事件进行了科学分类，在此基础上设计了方法框架和多主体系统仿真过程控制算法，并完成了该方法的软件实现，较好地解决了使用ABMS方法时仿真调度逻辑难于处理的问题。论文将贝叶斯网络和影响图引入ABMS方法之中，提出了基于贝叶斯网络的不确定性信息下多主体决策模型。在分析MAS中决策类型的基础上，针对Agent的独立决策和协作决策分别建立了解决方案，根据MAS的特点设计了重复性决策贝叶斯网络在线学习算法。在易用、通用、全面、可扩展原则指导下对上述模型完成了软件实现，并从便于使用的角度出发，将5步扫描法和多主体贝叶斯网络决策模型的软件代码封装为SmartAgent程序包。针对管理专家难于将自己的设计思想转化为多主体仿真模型的困难，论文以UML为基础，借鉴软件工程方法，设计了面向管理专家的ABMS建模方法——MEOLW，该方法通过“问题领域模型—逻辑模型—仿真模型”三个基本阶段，实现了ABMS方法中管理专家与IT专家工作的交流和衔接，为不同领域专家协作开发更复杂的多主体仿真系统提供了一条途径。最后，通过对我国房地产市场多主体系统的建模和仿真实验，阐述了本文提出的技术、建模方法和程序包的用法，该房地产市场多主体仿真系统准确地重现了我国房地产市场的运行势态，证明了本文所开发的技术和方法的有效性，也为研究我国房地产市场运作规律提供了一些参考。更多还原

【Abstract】 In recent years, more and more researchers in the field of management science have paidattention to agent-based modeling and simulation(ABMS) method and selected ABMS method astheir important research tool. But there are still many imperfections in the research of ABMS method——the lack of simple and effective modeling methodology, and some important technical issues tobe resolved, these defects have restricted the further development and applications of ABMS method.For this, the thesis concentrates on the modeling and simulation study of multi-agent system (MAS)with complex event scheduling logic and Bayesian network decision-making in order to make somevaluable works to improve the ABMS method, that involve designing simulation event schedulingapproach for MAS, researching the decision-making model of agent under uncertainty, anddeveloping an management experts-oriented ABMS modeling methodology. The research results haveimportant theoretical and practical significance to enrich the ABMS theory, and improve the level ofABMS application on complex management problems.The simulation process control method (named in the thesis as5-step scanning method) forMAS with complex event scheduling logic is proposed which combines the principle of discrete eventsimulation theory with the characteristics of MAS. Based on the strategy of separating simulationscheduling logic, agent behavior logic, and agent interaction logic, a reasonable classification ofmulti-agent simulation events is given,the framework and the process control algorithm for MASsimulation are designed, and the software developing of the method is achieved. The method caneffectively resolve the difficulty of dealing with simulation scheduling logic in ABMS application.Bayesian networks and influence diagrams are introduced in ABMS method, and the multi-agentBayesian network decision-making model under uncertainty is developed. On the basis of analysis ofthe MAS decision type, solutions for the independent decision-making and collaborativedecision-making of the agent are established, and repetitive decision-making Bayesian online learningalgorithm is designed according to the characteristics of the MAS. Based on the principle ofeasy-to-use, universal, comprehensive, and scalable, the software developing of above models areimplemented, the SmartAgent software toolkit is developed to improve the ease of use, in which the5-step scanning method and multi-agent Bayesian network decision-making model are packaged.By reference of software engineering theories, a UML based management experts-orientedABMS modeling methodology——MEOLW is proposed. Through "problem domain model-logical model-simulation model " these three basic stages, the exchanges and matches between managementexperts and IT experts are accomplished in MEOLW methodology, MEOLW provides a way fordifferent type experts to collaboratively develop more complex multi-agent simulation system.Finally, in the case studies, a multi-agent system model of China’s real estate market ecological isdeveloped and the simulation experiments are made. The trend of China’s real estate market isaccurately reproduced in the simulation experiments, that proves the effectiveness of the techniquesand methods developed in the thesis. The model proposed in the thesis also provide some referencefor the study of China’s real estate market.更多还原