【作者】 梅珊；

【导师】 朱一凡；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2010， 博士

【摘要】 病毒感染对人类健康的影响显而易见,严重时可以威胁生命;特别是病毒传播所引起的传染病则可能引起全社会的恐慌和动荡,同时须耗费大量的人力、物力和财力用于对传染病的预防、控制、诊断和治疗。除了自然发生的传染病,恐怖分子也可能将投放病毒生物剂作为其开展生物恐怖袭击的手段,生物恐怖危害的预警与应急决策已成为当前国际上关于生物恐怖防御和传染病防治等的重要研究领域。由于研究病毒传播引起的传染病扩散问题不可在真实世界进行实验,因此计算机建模和仿真是研究此类问题的一种有效途径。本论文研究目的在于提出描述病毒传播特点和动态性的一种有效方法,通过计算机上生成的虚拟复杂网络来描述和刻画真实世界个体之间交互的复杂性以及病毒传播动态过程,然后支持仿真。在此基础上,我们可基于对仿真结果的分析预测传染病的发展趋势,并根据特定病毒的防治要求为公共卫生措施的制定提供决策支持。在深入研究常用的传统系统动力学建模、基于agent的建模和基于复杂网络的建模这三种病毒传播建模方法的基础上,提出一种支持个体层和群体层建模的一体化方法——复杂agent网络方法。该方法无缝集成多agent系统方法和复杂网络两种方法,利用两者的优点并规避其不足,能为个体层和群体层的一体化建模提供方法指导。在复杂agent网络方法指导下分析病毒传播过程,构建了基于病毒传播过程动态性的传染病流行规模统计模型,作为在仿真过程中度量传染病流行规模(也即疫情严重程度)的统计量并计算得到系统的状态输出。然后,从两个层次分析影响病毒传播过程及疫情发展的因素:在群体层次主要研究个体复杂交互规律的宏观描述、传染病扩散阻断措施和人口统计学方面的影响;在个体层次研究宿主个体agent建模,描述个体与病程发展相关的基本属性和行为,并重点研究了通过个体交互活动引起(空气传播和性传播)病毒传染的概率模型。影响个体规避传染决策的因素包括情感和认知两种,且这两种因素之间互相影响,共同影响决策。对情感和认知之间的影响以及二者在决策过程中的作用进行研究,构建了个体规避传染的决策框架;在框架指导下,综合考虑个体的初级、二级和高级情感以及个体对疫情认知的知识表示之间的影响关系,提出一种支持自主学习的个体情感和认知FCM表示模型,并通过非线性Hebbian自主学习对专家初始设定的因果影响关系权值进行调整;在计算模糊认知图中各种因素值的基础上,可根据个体对不同传染病传播的具体反应情况,设定个体规避传染的决策规则。作为支持病毒传播过程建模的一个关键算法,基于配置模型的scale-free病毒传播复杂网络生成算法可快速、灵活和鲁棒地生成给定网络尺寸和幂律度分布指数的复杂网络,以支持病毒传播社会网络的构建。基于该算法生成复杂网络的过程中,边的创建能满足病毒传播过程中个体之间交互形成的局部判定条件,同时对边赋予其与交互关系相关的属性值,更符合真实世界情况;此外,避免了其他一般算法所要求的随机增删结点和边的操作,从而能够保留所有个体病程发展全过程的状态信息,也即保证了个体agent的自治性和独立性。基于前述研究,设计并实现了一个病毒传播仿真原型系统,并在此基础上构建了封闭高校内甲型H1N1传播仿真和荷兰阿姆斯特丹男同性恋HIV流行病仿真两个应用实例。在设计不同想定并对仿真结果进行分析之后,分别给出了控制这两种不同类型传染病扩散的公共卫生决策建议。这两个应用实例验证了论文所提出的复杂agent网络方法在支持病毒传播过程建模和仿真方面的有效性。论文研究成果可有力支持病毒传播建模和仿真,为模拟真实世界病毒传播问题提供了一般性参考过程和建模框架,并为传染病扩散过程的模拟和预测提供了量化分析支撑,这对控制无论是自然发生的传染病还是由生物恐怖袭击引发的传染病,都具有重要的理论意义和现实意义。更多还原

【Abstract】 Virus infections obviously harm human health and even threat human life. The prevalenceof infectious diseases caused by viruses can arouse social panic and disturbance, andin the meantime cost large amount of resources for prevention, containment, diagnosesand treatment. Apart from the study of natural infectious diseases, the study of bioterroristicattacks and their subsequent infectious diseases becomes an important field, inwhich the precaution of bioterrorism and crisis decision making are of great internationalconcerns.Modeling and Simulation is an effective approach to mimic and study virus propagation,due to infectious diseases’non-experimentation among human in the real world. Theaim of this thesis is to present a method for describing the characteristics and dynamicsof virus propagation and characterizing the complex interactions among individuals bygenerating virtual complex networks so as to support stochastic simulations. Afterwards,we can predict the trends of infectious diseases by analyzing simulation results, and thensupport public health policy making based on features of viruses concerned.The paper presents a novel Complex Agent Networks method, which is a comprehensivesolution to the modeling at individual and population scales, based on deep investigationson three common epidemic modeling methods, i.e., traditional equation-basedmodeling, agent-based modeling and complex-network-based modeling. Our ComplexAgent Networks method can seamlessly integrate Multi-agent Systems and complex networks,utilizing the merits of both methods while avoiding their demerits. This methodcan generally guide the overall modeling at individual and population scales.Based on the utilization of the Complex Agent Networks method to analyze the dynamicvirus propagation procedure, we construct a statistical model for assessing the epidemiologicalimpact of virus propagation in simulations. Then we analyze the influentialfactors of virus propagation on different scales. On population scale, the macro characterizationof the complex interactions between hosts, intervention policies of infectiousdiseases and demographical influences are studied. On individual scale, hosts’individualagent modeling, individual properties and behavior related to infection progression andparticularly the probabilistic model of virus infection caused by interactions are studied.The two interacting factors that influence individual decision-making for avoiding infections are emotion and cognition. A decision framework is constructed based on thestudy of the relationship between the two factors and their roles in the decision-making.Guided by this framework, a Fussy-Cognitive-Map-based representation model of individualemotions and cognition which supports unsupervised learning is presented, withprimary, secondary and senior emotions and cognitive knowledge of infectious diseasesconsidered. Initial causal influence weights between each pair of concepts can be tuned byapplying Nonlinear Hebbian Learning. We can set individual decision rules for avoidinginfections mapping to the iterated values of concepts in a fussy cognitive map, accordingto people’s reactions in reality when they are faced with infectious disease spreading.Then a key algorithm, named configuration model based scale-free complex networkgenerations, is presented to support rapid flexible and robust generations of complex networkswith given size and power law degree distributions. The social networks for viruspropagation can be generated based on the algorithm, with local principles of forming interactionsamong individuals considered, and interaction properties assigned to edges. Inaddition, random addition and deletion of vertices and edges required by other algorithmscan be avoided in our algorithm, and thus the infection progression information of eachindividual can be retained to guarantee agents’autonomy and independence.Next, a virus propagation simulation prototype is designed and implemented basedon aforementioned study and two applications are built to illustrate the prototype. One isfor simulating the human swine influenza A [H1N1] propagation in a closed universityarea, the other is for simulating the HIV epidemic among men who sex with men in Amsterdam,the Netherlands. After performing different scenario simulations, suggestions forpublic health decision making are given to hold back these two different infectious diseases.These two applications are advantageous to validate the Complex Agent Networksmethod presented previously.This study can effectively support the modeling and simulation of virus propagation,and provides a general reference procedure and framework for mimicking virus propagationin the real world. Quantitative analyses of the spreading of infectious diseases,conducted based on this study, helps theoretically and practically to investigate the interdictionsof both naturally occurring and bioterroristically aroused infectious diseases.更多还原