面向可靠web服务的智能处理技术研究

发布时间：2018-04-22 17:03

本文选题：服务计算 + 服务推荐　；参考：《北京理工大学》2014年博士论文

【摘要】：面向服务计算是基于互联网的企业分布式应用系统采用的一种先进的计算范式，其高可复用性体现在为了满足更复杂的功能需求，单个独立的服务可以组合起来提供增值服务。随着对网络服务的依赖性越来越大，同时由于网络环境的复杂性，学术界研究者和工业实践者一直从事服务/服务组合可靠性的相关研究。本文以提高服务组合的可靠性为目标，在服务组合形成的三个阶段：服务的选择、服务组合的设计及服务组合的运行监控，深入的分析和研究保证各阶段可靠性的方案，，论文的主要贡献及研究内容如下：（1）在服务的选择阶段，提出一种改进的信任感知的服务推荐方法，即在结合推荐技术和用户间的信任关系的研究中，给出两者的组合方法，同时考虑用户间的不信任信息对推荐效果的影响。其中，提出被信任度的概念，并利用信任评价的相似性来更好地发现相似用户集合，同时给出用户信任网络的重构方法，改进了服务推荐效果。（2）在服务的选择阶段，提出一种新的服务信誉管理方法，不仅考虑用户对服务质量属性的不同偏好，而且还兼顾服务评价的时态性。为了给用户提供更好的参考信息，引入评价者的敏感度概念，通过实验表明提出的综合信誉模型是有效的。（3）在服务组合的设计阶段，为了验证业务流程中错误处理机制和补偿机制的正确性，首先给出BPEL语义扩展模型，然后提出利用图规划技术验证其可行性。（4）在服务组合的运行阶段，为了提高服务组合的错误处理能力，提出基于案例推理的错误容忍框架。在总结各种错误信息的基础上，提出将已有的错误处理实例作为历史案例存放在数据库中，以应对未来发生的错误。详细阐述案例的表示方法，案例的搜索方法和案例的重用方法的选择方案。为了减小错误处理代价，根据错误的影响域，提出一个自适应的区域重配置算法。通过真实的数据实验验证了此框架的可行性。（5）为了提高服务组合的自适应能力，分别提出基于概率上下文无关文法和学习自动机的自适应方法。两种方法分别从不同的角度对服务组合进行建模。前者主要关注服务间的交互信息，概率上下文无关文法对自适应策略的建模具有良好的可扩展性，利用概率信息，提出基于步长的策略选择方案。后者主要关注包含服务组合的运行环境在内的一个整体建模。学习自动机通过自身的学习能力，不断的更新每个被监控服务的性能信息，在错误发生时，可以及时地生成最优（次优）的服务组合实例。两种方法都可以保证组合继续正确的运行。
[Abstract]:Service-oriented computing is an advanced computing paradigm used in Internet-based enterprise distributed application systems. Its high reusability is reflected in the fact that, in order to meet more complex functional requirements, a single independent service can be combined to provide value-added services. With the increasing dependence on network services, and because of the complexity of the network environment, academic researchers and industry practitioners have been engaged in research on the reliability of service / service composition. In order to improve the reliability of service composition, this paper aims at three stages of service composition formation: service selection, service composition design and service composition operation monitoring. The main contributions and research contents are as follows: 1) in the stage of service selection, an improved trust aware service recommendation method is proposed, that is, in the research of the combination of recommendation technology and trust relationship between users, the combination of the two methods is given. At the same time, the effect of distrust information on the effect of recommendation is considered. Among them, the concept of trusted degree is put forward, and the similarity of trust evaluation is used to better discover the similar user set. At the same time, the reconstruction method of user trust network is given, which improves the effect of service recommendation. 2) in the service selection stage, a new service reputation management method is proposed, which not only takes into account the different preferences of users on the attributes of service quality, but also takes into account the temporal nature of service evaluation. In order to provide users with better reference information, the concept of sensitivity of evaluators is introduced, and the experimental results show that the proposed comprehensive reputation model is effective. In the design phase of service composition, in order to verify the correctness of error handling mechanism and compensation mechanism in business process, the semantic extension model of BPEL is given first, and then the feasibility of using graph planning technology to verify the feasibility is proposed. In order to improve the error-handling ability of service composition, a case-based reasoning (CBR) based error tolerance framework is proposed. On the basis of summarizing all kinds of error information, it is proposed that the existing error handling examples be stored in the database as historical cases to deal with future errors. The method of case representation, the method of case searching and the method of case reuse are discussed in detail. In order to reduce the cost of error processing, an adaptive region reconfiguration algorithm is proposed according to the error influence domain. The feasibility of the framework is verified by real data experiments. In order to improve the adaptive ability of service composition, an adaptive method based on probabilistic context-free grammar and learning automata is proposed. The two methods model service composition from different angles. The former focuses on interactive information between services. Probabilistic context-free grammars have good scalability for adaptive policy modeling. Using probabilistic information, a strategy selection scheme based on step size is proposed. The latter focuses on a whole modeling including the runtime environment of service composition. Learning automata continuously updates the performance information of each monitored service through its own learning ability and can generate the best (sub-optimal) service composition instance in time when errors occur. Both methods ensure that the combination continues to work correctly.
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】