【作者】 李祯；

【导师】 杨放春；

【作者基本信息】 北京邮电大学 ， 计算机科学与技术， 2008， 博士

【摘要】 随着Internet技术和WWW(World Wide Web)技术的发展,近年来互联网行业呈现一派繁荣的景象。互联网无论在人民生活中还是商业领域中都扮演着举足轻重的角色,特别是越来越多的在线商业应用被构建和部署,用来满足客户多方位的交易需求。然而隶属于不同商业机构的应用在接口设计上各不相同且接口与内部逻辑之间耦合紧密,这给多应用间的互通带来了很大的困难。另外,Web应用的种类众多、数目庞大,对应用的查找和访问都需要消耗大量的人工成本。基于以上现状,对于新一代互联网服务的需求大致可归结为以下两点:1.建立松耦合、跨平台、高互操作性和高可扩展性的服务模型;2.实质性地提高服务行为的自动化程度,解放客户和开发人员的工作。语义Web服务(Semantic Web Service)技术致力于满足人们对新一代互联网服务的期望和需求。语义Web服务技术是以Web服务(Web Service)技术为基础,以语义网(Semantic Web)技术为补充的新一代互联网服务解决方案。其中Web服务作为一种新兴的分布式计算技术和一种开放的业务提供方式,为Web应用提供了松耦合的设计模式和无缝互操作的技术支持。对Web服务的研究主要集中在服务发现、服务组合和服务质量保障上。语义网技术将网络信息和资源以知识的形式描述和存储,使得机器能够对这些知识进行自动地处理和集成。因此语义Web技术允诺与在线商业应用相关的操作可以自动执行,在极大程度上提高了应用的灵活性和互操作能力。在语义Web服务的相关研究中,服务组合问题是最为重要和困难的。任何一个服务只能满足相对单一的客户需求。服务提供商们迫切的希望现有的服务能够被自动地、快速地组合成增值的复杂服务用来满足更为多样化的需求。在一个服务组合场景中,客户首先向服务组合引擎提供他对于组合服务的功能性和非功能性(在本文中主要指QoS(Quality of Service))需求,然后组合引擎基于工作流技术生成服务组合的执行路径,随后组合引擎将访问服务发现主体,通过语义匹配的方法为执行路径中的每个任务找到满足功能性需求的服务实例。在这之后,相对于每个任务会出现众多的功能相似的候选服务,这些候选服务的每一个可能的组合被称作一个服务组合执行计划。最后最优的执行计划通过一个QoS感知的服务组合过程得出。从概念上讲,目前对于QoS感知的服务组合的研究内容包括:QoS聚合(根据单个服务的QoS计算整个计划的总体QoS值)、和最终决策(在候选计划中找到满足客户QoS限制的综合最优方案)。本论文的研究重点在于QoS感知的语义Web服务组合过程,然而就作者的知识而言,这方面的研究工作还处于起步阶段,仍存在一些问题:1.不能支持含有不确定QoS信息的组合决策;2.不支持不确定的客户偏好和群决策场景;3.组合决策算法在任务数量多和候选服务规模大的情况下运行效率低;4.组合算法的执行自动化程度低。为了解决上述难题,为服务组合提供有效地决策支持,本论文分为3个部分分析与研究了描述混合QoS模型(支持实数型、区间型和语言型QoS数据)的QoS本体以及该本体的管理框架和进化策略,混合QoS模型感知的服务组合决策算法和服务组合系统的实现框架。本论文取得了以下几个方面的科研成果:1).混合QoS本体及本体管理框架的创建。以往的QoS模型均以明确的实数表示,这种描述方式会造成大量信息的丧失并会误导最终的决策过程,为此本论文提出了一套混合QoS本体—OWL-S-QoS用来对QoS信息进行更为高效、准确的语义描述,该QoS模型混合了确定和不确定的表示方法(区间型表示法和语言型表示法)。另外本论文还提出了一种主客观相结合的QoS声誉属性模型--WSrep,用来增强现有声誉模型抵御恶意打分攻击的能力。此外本论文还讨论了该本体的管理框架和进化策略,这主要包括定义了与该本体相关的添加、修改和删除操作,提出了本体(连锁)进化算法和基于本体摘要匹配的本体版本控制算法用来保证知识更新的实时性和一致性。仿真试验证实:WSrep相比与主观声誉模型,具有更高的抵御恶意打分攻击的能力;基于摘要的本体版本控制算法相比传统算法具有更高的运行效率。2).提出一种基于不确定多属性决策理论的服务组合算法(UncertainMulti-attribute decision making-based service Composition algorithm,UMC)解决混合QoS模型感知的语义Web服务组合难题。该算法主要是从全局优化的角度在考虑客户QoS限制的基础上解决最优执行计划的选择问题,与以往的组合算法不同,UMC不仅能够支持混合QoS模型,还能有效地适应不确定的客户偏好和群决策场景。另外,UMC还具有良好的运行时效率,推动了服务组合算法在大规模问题空间情况下的实用进程。UMC分为两个主要组成部分:UMC-Core和UMC-DH(Distributed and Heuristic framework for UMC),其中UMC-Core用于对混合QoS进行综合评估,共包括三个主要的步骤:去模糊化(将语言型表示的数据转变为实数),决策矩阵规格化,候选方案综合评估。UMC-DH是针对解决组合算法效率难题而为UMC设计的分布式与启发式相结合的执行框架,主要包括分布式求解局部最优解、并行结构k启发聚合和全局回溯控制策略。本文通过仿真试验证实,UMC算法具有较高的决策灵敏性和稳定性,另外相比于其他类似算法UMC算法具有执行速度快、最优解近似度高和最优解成功率高的优势。3).组合决策算法的执行框架设计。本论文研究了UMC的执行框架,该框架基于智能多主体系统技术,其中主要包括服务组合引擎设计、服务注册中心设计、客户主体设计和知识管理主体设计。另外,还给出了多主体之间的两组交互模型:服务组合交互模型和本体进化交互模型。这是一个自动化程度高且可扩展的服务组合系统架构,为整个系统的实现提供了强有力的理论基础和技术支持。更多还原

【Abstract】 As the rapid growth of Internet and WWW (World Wide Web) technologies, the internet industries are booming in recent years. WWW acts as one of the most important roles in people’s lifes and business domains. Now more and more online business applications are constructed and deployed to satisfy the various customer requirments. However, the solutions owned by different corperations usually have heterogeneous interfaces, which are often tightly coupled with their inner service logics. It brings the communication difficulties among applications. Further various and numerous web applications are challenging the searching and accessing approaches. In this case, there are two basic requirments for the next generation online service model: 1. Building a loosely coupled, platform-independent, highly interoperable and scalable service paradigm. 2. Evidently improving the automatization of service activities for releasing the works of customers and developers.Semantic web service technologies aim to fulfil the expectations and requirments on the next generation internet service. Semantic web service integrates the benefits of web service and semantic web ideologies. Conceptually, web service as a novel distributed technology and an open service delivering mode brings the loosely coupled design pattern and seamless interoperability to web applications. Researches on web service focus on service discovery, service composition and QoS guarantee. Further the semantic web is a roadmap for waving cerrent web to the next generation web. On such a web, information and resources are given as knowledges. Software agents can process and integrate useful info automatically. Consequently, semantic web service promises online service logics can be executed automatically to improve the flexibility and interoperability of applications.Sevice composition is one of the most difficult aspects in the semantic web service researches. Actually, any single service only can satisfy relative simplex requirement. Service providers and customers call for existing services can be composed expediently to deliver stronger functionalities. In a service composition context, the composition engine first determines the composition pattern based on workflow design technology after user submits his functional and non-functional (mainly mean QoS in this thesis) requirements. Then service implementations are discovered by semantic functional matchmaking. After that, a large number of functional overlapping services may be yielded for each task. Each possible combination of these service candidates along the execution path is called an execution plan. Finally, a QoS (Quality of Service) -aware composition process is needed to choose the optimal plan for the composite service. Conceptually, QoS-aware composition includes QoS aggregation and final decision making. The overall QoS of each execution plan is calculated in the aggregation step and the optimal decision is made by the decision maker based on the customer QoS Constraints.Our work focuses on the QoS-aware semantic web service composition. Within our knowledges, researches on this field are underway. Several scientific problems remain: 1. The composition algorithms can not be adeptive with uncertain QoS values. 2. Can not cope with uncertain customer preferences and group decision making settings. 3. Run-time performances of existing composition algorithms decrease severely when tasks and candidate services increasing. 4. The automatization of algorithm execution is low. For solving the above difficulties, this thesis introduces a more descriptive hybrid (integrates certain and uncertain descriptions) QoS ontology, its management framework and evolution strategies. Further we introduce a hybrid QoS model-aware semantic web service composition algorithm. In addition, the implementing strategies of the composition system are discussed. Our research work has several significant contributions:1) Constructing the hybrid QoS ontology and management framework. Most current QoS models accept certain (real number-based) descriptions. This kind of model makes losing much QoS info and misleading the decision process. This thesis introduces a hybrid QoS ontology—OWL-S-QoS to describe QoS attributes more efficiently and accurately. This QoS model integrates certain and uncertain (interval type and linguistic type) expression types. Further, this thesis also presents an enhanced reputation model-WSrep to reduce the influence of maliciouse rating attack. In addition, the management framework and evolution strategies for OWL-S-QoS are also discussed, which includes defining add, modify and delete operations on concepts, a ontology chaining evolution algorithm for assuring the effectiveness and consistency of knowledge updating. The evolution algorithm includes a ontology abstract matchmaking based versioning algorithm and a hybrid QoS aggregation evolution algorithm. The simulation results show the WSrep has the strong ability to defense the malicious rating attack. Further the abstract matchmaking based ontology versioning algorithm has better performance than existing approaches.2) Presenting an Uncertain Multi-attribute decision making-based service Composion algorithm (UMC) to solve the difficulty of hybrid QoS-aware semantic web service composition. UMC aims to choose the best execution plan from a global viewpoint based on customer QoS constraints. UMC is different from other similar approaches in several ways: 1. UMC can be adapted with hybrid QoS model. 2. UMC can be used with uncertain customer preference and in group decision making scenario. 3. UMC has better run-time performance than others in large problem scales. UMC has two logical parts: UMC-Core and UMC-DH (Distributed and Heuristic framework for UMC). UMC-Core is used for assessing candidates with hybrid QoS values and includes three sequential steps: linguistic QoS defuzzification, heterogeneous QoS values normalization and synthetically evaluation. UMC-DH is designed for improving the run-time efficiency of UMC, including distributed local optimal solution finding, k-heuristic QoS aggregation for parallel structures, and global backtracking control strategy. The simulation results prove that the UMC has high decision sensitivity and stability. Further, it has better run-time performance, approximation ratio and success ratio than other similar approaches.3) Designing the UMC implementation strategies. This thesis discusses a multi-agent technology-based UMC framework for implementing the service composition approaches. It includes service composition engine (QComposer) designing, service register designing, customer agent designing and knowledge manager designing. In addition, two interaction models of involved intelligent agents are discussed: service composition interaction model and ontology evolution interaction model. This framework is a high automatic and scalable one, which brings the basic theories and technologies for implementing the service composition system.更多还原