【作者】 彭艳斌；

【导师】 高济；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2008， 博士

【摘要】 借助虚拟组织(VO)的灵活性和动态性,敏捷企业可以灵活地通过与其他企业进行协作,抓住商业机会,达成自身目标,实现效益最大化。随着面向服务的体系结构(SOA)日益成熟并得到IT业的广泛支持,服务协同方式正成为建立高性能VO的主要方式。论文综述了目前建立VO的主流技术,分析了运用当前技术来建立高性能VO所存在的主要问题:1)如何协调VO全局目标和组织成员的个体私有目标,使得即能够保证VO全局目标的实现,又给予组织成员一定的自治空间,解决“可信”危机问题;2)如何引入自治计算技术,建立自主管理系统,解决日益严重的系统管理复杂性问题。针对上述问题,本文提议研究基于准则和策略的自治式多agent服务协同体系(Autonomic Multi-agent Service Cooperation System based on Norm and Policy, AMSSNP),通过建立包含社交行为准则的系统体系结构和实施方法,去解决“可信”危机问题;通过建立策略驱动的自主管理方法,解决管理复杂性问题,建立自治计算系统,从而实现可信自治的高性能VO。本文给出了新型多agent系统理论模型和架构,并进一步阐述了高性能VO所具有的自组织功能,主要工作与贡献包括如下四个方面:1)研究了单agent思维状态问题,提出了动机扩展的agent思维状态模型——NPCD-Agent。单agent思维状态模型是多agent理论的基础。NPCD-Agent模型将动机分为社交行为准则类义务、策略类义务、契约准则类义务和内部愿望,充分考虑了agent行为决策所需要的动机因素,通过给予动机算子新的语义解释以及给出动机冲突消解模型,提高了目标产生的合理程度。实例分析表明,该模型使得agent行为决策过程更加理性。2)研究了多agent理论问题,提出了基于准则和策略的多agent协同理论。该理论模型基于量化一阶多模态逻辑,作为基于准则和策略的新型多agent系统的理论基础,建立逻辑框架及其逻辑语言NPMCT,对多agent协作求解过程进行了形式化描述。模型中引入策略的定义,并在此基础上,给出了新的能力定义,区分动机的表现形式,给出了任务分解分配过程的描述,阐述了理性协商过程以及契约执行和异常处理过程。实例分析表明,该理论模型全面描述多agent协作过程,并且在协作中从理论上消除了“可信”危机问题。3)研究了多agent系统实现问题,提出基于准则和策略的多agent系统架构。该架构作为基于准则和策略的新型多agent系统的实现模型,分为宏观调控层、微观行为层、宏微连接层和资源层。宏观调控层制定VO组织结构和相应的协同行为准则,作为宏观调控Aces交互行为的约束;微观行为层建立事件触发策略驱动的Ace模型,从正面去接受协同行为准则的宏观调空约束;最后通过宏微连接层,从负面去支持协同行为准则的实施,促进了前两个层次的衔接。实例分析表明,该架构使得自治个体agent能够遵守社交行为准则约束,进而多agent的协同行为能够实现VO全局目标,在实践上解决了“可信”危机问题。4)研究了多agent系统的运行机制问题,提出了基于协同网和适用情景的Web服务自动组合方法。Web服务的自动组合能依据商业需求实现服务的快速组装,是实现高性能VO的自组织能力的关键技术,同时也实现了VO作为自治计算系统所具有的自配置特性。本文提出的基于协同网和适用情景的Web服务自动组合方法采用拆分组合的思想。它首先为用户期望的每一个目标服务输出生成完备协同网集,然后通过协同网的组合生成满足用户需求的大粒度组合服务,最后通过综合适用情景对组合服务进行选择,返回最佳组合服务。实例分析表明,该方法具有搜索空间小、速度快和成功率高等特点,能较好的满足用户需求。更多还原

【Abstract】 Agile enterprise will achieve its goal with the help of virtual organization. Companying with the widely supporting from IT, SOA has became the main implementation for virtual organization. This thesis reviewed the domain technologies for building of virtual organization, and then analyzed the main problems about building high performance virtual organization: 1)how to balance the global goal of virtual organization with the private goal of individual, so as to guarantee the global goal together with the private autonomous space of individual, to solve“trustable crisis”problem; 2)how to build autonomic computing system with self-management ability to solve the problem of management complexity problemAiming at the above problems, this paper proposed a model of autonomic multi-agent service cooperation system based on norm and policy, to solve the“trustable crisis”problem and management complexity problem, and then to implement high performance virtual organization. The major work and contributions are as follows:First, we propose an agent mental state model-NPCD-Agent extended by motivation. This model considers all of the motivation ingredients which are necessary in our system, and divide it into four parts: social behavior norm obligation, policy obligation, contract norm obligation and internal desire. Through the new semantic interpretation of motivation operator and conflict solving model of motivation, the goal generated arithmetic is improved. The analysis of demonstration shows that NPCD-Agent model makes the decision making process of agent more rational.Second, we propose a logic framework for multi-agent system based on norm and policy. The model is formalized by expressing it as a theory in a quantified first order multi-modal logic. This theory model describes the process of multi-agent cooperation, from its beginning, with the entire agent recognizing the potential for cooperation with respect to one of its goal, through to team execution. The analysis of demonstration shows that the theory model can describe the whole process of multi-agent cooperation, and solve the believable crisis problem in theory.Third, we propose an implemented framework of multi-agent system based on norm and policy. The implemented framework is divided into four parts: macro regulation layer, micro behavior layer, macro-micro connection layer and resource layer. In macro regulation layer, we define electronic institution of domain and social behavior norm of it to regulate the behavior of agent; in micro behavior layer, we implement policy driven autonomic agent model to map macro regulation into micro behavior; macro-micro connection is used for connecting the above two layers. In this multi-agent system, social behavior norm can be implemented and then the problem of trustable crisis is solved.Last, we propose an automatic service composition based on collaboration network and compatible scene. This approach makes use of the collaboration network and compatible scene through three steps to automate composition of web services. It first builds a complete-collaboration-network for each output that a user asks for; then it combine the complete-collaboration-network of every output into the composed service; after that it calculates integrative compatible scene similarity degree for selecting the final composed service which is returned to the user. The characteristics of this method include a smaller search space, good efficiency and high ratio of hit. It can support the user’s requirement with not only single target but also multiple targets. A set of experiments show that this approach has a good performance and can serve users well even that the service registration repository has a large number of registered services.更多还原