基于Agent的人工社会应急管理政策分析方法

发布时间：2018-05-24 21:54

  本文选题：人工社会 + Agent　； 参考：《国防科学技术大学》2015年博士论文

【摘要】：复杂系统通常表现为系统中存在大量自主交互的个体以及由此产生的全局行为具有动态性、不确定性、难以重现性或不可重复性等特点,如社会系统、软件密集型的社会技术系统等。复杂系统中出现的某些突发事件(如疾病传染、网络攻击等)可能会诱发系统宏观层面的一些“坏”涌现,例如疾病暴发、系统瘫痪等,从而可能会对系统的利益相关者以及系统中的个体造成严重危害与负面影响。因而,如何提供有效的方法来对复杂系统中的突发事件进行应急管理是目前社会科学、复杂性科学、管理科学、控制科学以及计算机科学等诸多学科所关注的一个重要研究课题。社会计算借助于计算机科学及社会科学等多个领域的思想和技术来研究和分析复杂系统中的社会和计算行为,它已成为突发事件应急管理研究的一个重要方法和手段。本文以突发事件应急管理为具体应用背景,采用社会计算的方法来支持针对社会系统中的突发事件的应急管理政策的评估、分析、优化或选择,通过在计算机世界构建社会系统的计算模型(即人工社会),采用多Agent系统的思想对人工社会进行抽象、建模和仿真,从而开展突发事件应急管理实验。论文通过在人工社会中的可重复性实验展示应急管理政策的有效性,并且可将其应用于现实社会系统的突发事件应急管理中。论文所取得的主要研究成果及创新点概述如下。1.为了支持突发事件应急管理政策的评估、分析与选择,提出了一种轻量级的社会计算方法——PZE(即政策选择、基于僵尸城的人工社会建模以及政策评估)。针对突发事件应急管理所带来的挑战,同时为了解决现有社会计算方法的臃肿性、缺乏灵活性等问题,论文围绕应急管理政策的评估与选择,借助于迭代开发、闭环控制等思想,并引入情景/应对(scenario/response)的应急管理模式,提出了针对突发事件应急管理政策选择、评估与优化的轻量级社会计算方法——PZE。由于PZE方法集成了软件工程、控制学等领域的一些成熟的思想,具有简单、灵活、易使用、低成本等轻量级方法的特性,其主要包括以下三个部分:(P)针对突发事件场景来选择合适的应急管理政策,即政策选择;(Z)基于僵尸城(Zombie-city)模型来构建人工社会模型;(E)借助于人工社会模型开展应急管理实验,评估应急管理政策的有效性,并给出评估结果,而这些评估结果又作为政策选择的反馈,为政策的选择提供有效的支持。2.为了给突发事件及其应急管理提供一般性的分析模型,提出了以Agent、病毒(Virus)和交互(Interaction)为核心的面向应急管理的AVI模型及其扩展模型——僵尸城模型(Zombie-city)。由于社会系统中个体间的局部非线性交互,突发事件(如传染病、谣言传播)会在社会系统中不断地演化和发展,进而产生全局性的宏观行为,因而突发事件演变的本质实际上是传播问题。为了揭示和分析突发事件带来的传播问题,论文提出了一个包含Agent、病毒以及交互三个核心概念的AVI模型,以表示复杂系统中参与传播的个体、传播的载体和内容、传播的媒体和途径。在此基础上,论文进一步提出了针对突发事件应急管理的人工社会模型——僵尸城模型(Zombie-city),该模型除了包含以上三个核心概念之外,还包括了角色、环境(物理环境和社会环境)、规则。角色表示一类Agent的抽象,Agent通过扮演角色来获取不同的行为与属性,并能通过动态地扮演角色来适应环境与自身变化。物理环境表示Agent所生活的空间,社会环境主要指Agent之间的社会关系,Agent行为会受到环境的影响。此外,规则定义了Agent、病毒、交互、角色及环境等的属性或行为规范,其也能用于描述应急管理政策等。该模型能更为系统地刻画面向突发事件应急管理的人工社会,为人工社会建模提供了参考与指导。3.为支持对基于僵尸城模型的人工社会进行建模和分析,提出了一种图形化的人工社会建模语言——ASML(Artificial Society Modeling Language)。为了支持人工社会建模与分析,论文借助于社会组织学思想(包括组织、角色等)以及角色动态扮演机制等,基于Zombie-city模型提出了人工社会建模语言ASML,包括建模语言的元模型、模型、可视化建模图元以及形式化语义等。ASML从不同视点、层次与建模机制提供了五类模型,包括人口统计学模型、环境模型、角色模型、组织结构模型以及交互模型,以便于系统地构建人工社会模型。总体来说,ASML建模语言具有高层抽象、自然建模、多视点、可视化、模型易于理解、表达能力强等特点。此外,为了支撑基于建模语言ASML的人工社会建模,论文还设计并开发了人工社会建模支撑工具ASMLTools,其包括模型的可视化编辑、模型存储、模型载入与导出、模型转换等功能。4.为了支持针对突发事件的应急管理政策的评估,进而支持政策的优化及选择,提出了定量与定性相结合的应急管理政策评估方法。应急管理政策的评估对于政策能否有效地应对和管理突发事件至关重要,为政策的选择或优化提供反馈,论文提出了一种定性和定量相结合的方法。在定量评估方法方面,论文提供了一个一般性的方法框架,基于情景提取人工社会的宏观属性,同时根据具体的应用提取相关的评估参数,并利用数值方法将宏观属性与评估参数相结合定量地评估政策的效果。此外,涌现程度也侧面反映了应急管理政策的效果,论文通过对涌现进行分类,提出了基于信息熵理论的三种定量度量涌现程度的方法,并通过仿真实验展示了三种度量方法的有效性,三种度量也可用于定量地分析应急管理政策的效果。在定性分析方面,论文同样也提出了通过基于定理与规则的形式化推理的应急管理政策定性分析方法。首先,通过基于Zombie-city模型构建人工社会形式化模型并将政策转换为人工社会中的规则,然后利用形式化推理(情景推理)定性地评估和分析政策的有效性。为此,论文还给出了Zombie-city的形式化规约以及人工社会中情景的形式化描述。定性评估方法可用于辅助定量评估方法,其能进一步确认和说明应急管理政策的定量评估结果,通过定量与定性相结合的政策评估方法能更为显式和准确地展示政策的有效性。论文以H1N1在某大学校园中的传播以及互联网环境下社交媒体中的广告信息传播作为案例,开展相关的分析和实验,以展示如何基于人工社会模型Zombie-city及建模语言ASML进行人工社会建模,以及如何利用PZE方法开展应急管理工作及其有效性。
[Abstract]:Complex systems are usually manifested in the existence of a large number of independent individuals in the system and the resulting global behavior, such as dynamic, uncertain, reproducible or unrepeatable, such as social systems, software intensive social technology systems, and some other emergencies in complex systems (such as disease contagion, network attack). It may induce some "bad" emergence at the macro level of the system, such as the outbreak of disease, the paralysis of the system and so on, which may cause serious harm and negative influence to the stakeholders of the system and the individual in the system. Therefore, how to provide an effective method to manage the emergency in the complex system is the present. An important research subject in many disciplines, such as social science, complexity science, management science, control science, and computer science. Social computing, with the help of ideas and techniques in many fields, such as computer science and Social Sciences, studies and analyzes social and computational behavior in complex systems. It has become an emergency emergency. An important method and means of management research. This paper, taking emergency management as the specific application background, adopts the method of social calculation to support the evaluation, analysis, optimization or selection of emergency management policy for sudden events in the social system, and constructs a computing model of the social system in the world of the computer, that is, the artificial society. The idea of multi Agent system is used to abstract, model and simulate the artificial society, so as to carry out the emergency management experiment of emergency. The paper shows the effectiveness of the emergency management policy through the repeatability experiment in the artificial society, and can apply it to the emergency management of the real social system. In order to support the assessment, analysis and selection of emergency management policy of emergency, a lightweight social computing method, PZE (policy selection, artificial social modeling based on zombie city and policy assessment), is proposed by.1.. In order to solve the problem of overstaffing and lack of flexibility of the existing social computing methods, the paper focuses on the evaluation and selection of emergency management policy, with the aid of iterative development, closed loop control and so on, and introduces the emergency management mode of the situation / response (scenario/response), and puts forward the light of emergency management policy selection, evaluation and optimization. The PZE. method of social computing - the PZE method integrates some mature ideas in the fields of software engineering, control and other fields, with the characteristics of lightweight methods such as simple, flexible, easy to use, low cost and so on. It mainly includes the following three parts: (P) the choice of appropriate emergency management policy for emergency scene, that is, policy choice; (Z) Based on the Zombie-city (zombie city) model, an artificial social model is constructed; (E) an emergency management experiment is carried out with the aid of an artificial social model to assess the effectiveness of the emergency management policy and to give the results of the assessment. The results also serve as a feedback to the policy choices and provide effective support for the choice of the policy to give.2. to the emergency and the emergency. Emergency management provides a general analysis model, and proposes an emergency management oriented AVI model with Agent, virus (Virus) and interaction (Interaction) as its core, and its extended model, zombie city model (Zombie-city). Due to the local nonlinear interaction among individuals in the social system, unexpected events (such as infectious diseases, rumor spread) will be in society. In order to reveal and analyze the propagation problems caused by sudden events, a AVI model, which includes three core concepts of Agent, virus and interaction, is proposed to show the participation of the complex system. On the basis of this, the paper further proposes an artificial social model for emergency management of emergencies, the Zombie-city, which includes the three core concepts, including the role, the environment (physical environment and the social environment), and the rules. The role represents the abstract of a class of Agent. Agent plays the role to obtain different behaviors and attributes, and can adapt to the environment and self change by playing the role dynamically. The physical environment indicates the space in which Agent lives, the social environment mainly refers to the social relations between the Agent, the Agent behavior will be affected by the environment. In addition, the rules It defines the attributes or behavior specifications of Agent, virus, interaction, role and environment, and it can also be used to describe emergency management policy. This model can engrave the artificial society of emergency management of sudden events more systematically, and provide reference and guidance for artificial social modeling to support the construction of artificial society based on the zombie city model. Model and analysis, a graphical artificial social modeling language, ASML (Artificial Society Modeling Language), is proposed. In order to support the modeling and analysis of artificial society, the thesis is based on the social histology (including organization, role and so on) and the role dynamic playing machine. Based on the Zombie-city model, the artificial social modeling is proposed. Language ASML, including metamodel of modeling language, model, visual modeling element and formal semantic.ASML, provides five kinds of models from different viewpoint, level and modeling mechanism, including demography model, environment model, role model, organization structure model and intersection model, so as to build an artificial social model in a systematic way. For example, ASML modeling language has high level of abstraction, natural modeling, multi view, visualization, easy to understand model and strong expressive ability. In addition, in order to support artificial social modeling based on modeling language ASML, the thesis also designs and develops the artificial social modeling support tool ASMLTools, which includes model editing, model storage and model. In order to support the assessment of emergency management policy for emergencies and support the optimization and selection of the policy,.4. has put forward the evaluation method of the emergency management policy combining quantitative and qualitative. The assessment of emergency management policy is crucial to the effective response and management of the emergency. In order to provide feedback for policy selection or optimization, a qualitative and quantitative method is proposed in this paper. In terms of quantitative evaluation, the paper provides a general method framework for extracting macro attributes of artificial society based on scenarios, extracting relevant evaluation parameters according to specific use and using numerical methods. Macro attributes and evaluation parameters are combined to quantify the effect of policy. In addition, the degree of emergence also reflects the effect of emergency management policy. By classifying the emergence of the emergence, the paper presents three quantitative measurement methods based on the information entropy theory, and shows the effectiveness of the three measures through simulation experiments. The three measures can also be used to quantitatively analyze the effect of emergency management policy. In the qualitative analysis, the thesis also puts forward the qualitative analysis method of emergency management policy based on formal reasoning based on theorems and rules. First, the artificial social formalized model is constructed based on the Zombie-city model and the policy is converted into an artificial society. In this paper, the formal specification of Zombie-city and the formal description of the situation in the artificial society are given. The qualitative assessment method can be used to assist the quantitative evaluation method, which can further confirm and explain the emergency management policy. The quantitative and qualitative evaluation method can show the effectiveness of policy more clearly and accurately. The paper takes H1N1 in a university campus and the communication of advertising information in social media in the Internet environment as a case to carry out the related analysis and experiment to show how to be based on people. The social and social model Zombie-city and modeling language ASML are used for artificial social modeling, and how to use PZE to carry out emergency management work and its effectiveness.
【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：D035
，

本文编号：1930718