【作者】 高明；

【导师】 姜继忱；

【作者基本信息】 东北财经大学 ， 技术经济与管理， 2013， 博士

【摘要】 Gartner在2009年业务流程管理(Business Process Management, BPM)的调研报告中指出,BPM已经逐步被纳入企业的核心战略目标,面临着前所未有的发展机遇。但随着知识经济环境下的企业和组织的知识协同和创新活动日益增多,对传统的工作流管理理论、方法和工具提出了严峻的挑战。知识经济背景下的企业创新活动表现为复杂流程下面向内容、知识资源的多角色、跨组织的知识协同。面向知识协同的跨组织多实体工作流是传统工作流管理系统与知识管理系统的有机结合,不仅支撑和管理着企业、组织的各个实体之间的复杂业务协作关系,还能够管理和控制各类知识资源在企业、组织的各个实体间产生、流动和更新的知识协同流程和知识生命周期的全过程。多数工作流模型和管理系统用来支持成熟的产品制造过程或固定的业务目标的实现路径管理,注重流程活动间的串行、并行等过程结构的自动化、半自动化控制。在面向知识和协同的应用,传统的工作流模型难于应付复杂的跨组织、多角色、知识密集型任务的建模和管理。要解决的主要问题包括协作过程中角色、任务及资源的动态分配协调、知识资源建模和工作流过程模型的结合,以及知识资源和知识协同的生命周期管理等。早期的E-Learning系统基于传统的教育观点,注重课程资源的建设和学习内容在计算机和互联网上的多媒体呈现方式,面向学习内容、学生培养和考评,忽视了学习活动的过程管理和实时监控,对学习活动中多实体协作和知识协同等还不够重视。工作流技术被引入E-Learning,利用企业、组织已有的IT基础设施,将学习流程整合到企业的工作流程中。产品化的工作流管理软件和传统的工作流建模技术,流程结构固定,形式化描述精确,易于实现和管理,但不可避免的带来了流程“僵化”,难以灵活变更的缺点,在知识分享和动态任务创建时,很难实现上下文感知的自适应学习和知识推送。解决此类问题应在工作流建模阶段把学习资源、学习过程的建模和管理,学习过程和学习资源的协作、协同结合在一起,引入新技术和新模型,在协作的学习流程活动中实现对知识资源的重用和积累,实现满足企业、组织成员的个性化知识需求的、情景感知的知识过滤、推送或推荐机制以支持创造性的知识协同学习。利用语义网和本体技术对知识资源进行语义建模和标注,构造统一的、抽象的、高层的知识本体库,借助语义网查询和推理技术,可以更好的依据动态学习环境中的上下文信息在本体知识库中提取面向用户个性化需求的知识资源推送给学习者。面向内容协作、知识协同的复杂应用中,企业、组织的某些具体业务会随着用户需求和业务目标而变化,企业、组织内外的协作关系存在着高度的变化性。对于此类业务协作,流程设计者无法预先对某个流程的所有环节精确的给定一个固定结构的静态流程模板；而且随着用户需求和业务目标的变化,原有的静态流程模板也可能无法适应新的环境条件。面向工作流的自动服务规划,能够针对工作流中随需变化的结构和模式,结合人工智能规划技术和算法,自动生成满足用户需求和业务目标的服务节点执行序列,按需构造目标驱动的子流程实例；传统工作流的静态流程模板作为上层的、宏观的控制机制来统一协调高层视角中固定不变的协作活动。规划图、HTN及其扩展或组合在自动服务规划领域应用的最为广泛。但是上述方法侧重于研究服务组合问题的规划求解实现和相关规划算法的优化,没有考虑到在服务规划中引入用户需求和业务目标的高层的、具有指导意义的抽象模型,还是基于较低层、细节的任务层面。如何在宏观意义上形成满足用户需求和业务目标的组合服务仅仅依靠人工智能领域的规划器是不够的,还需要针对服务组合的特点和相关应用领域做一些工作。跨组织多实体协同的工作流中服务节点的组合问题不仅涉及流程服务节点间的逻辑顺序关系(即结构特征),也涉及到服务节点存在的众多候选服务的选择和优化问题。工作流被建模和验证后,运行阶段被实例化,工作流引擎需要根据业务约束和资源约束对各个服务节点的具体服务对象(任务执行者)进行绑定,如果存在满足功能需求的候选服务对象(任务执行者)时,如何选择功能相同但服务质量(Quality of Service, QoS)不同的候选服务对象(任务执行者)是服务组合选择和优化的关键问题。本文的主要创新是：第一,提出了一个基于分层颜色Petri网的面向资源协同的工作流系统(Resource-oriented Collaborative Workflow,ROCWF)的过程模型和资源控制模型(Task Resource Multi-role Collaboration Model,TRRC),给出了基于分层颜色Petri网的过程模型和支持角色、任务和资源关联、感知的资源控制模型的联合建模方法,为减小建模和验证的复杂性,将流程验证工作分为主干流程验证和任务子流程验证,给出了相应的颜色集定义和验证规则,提出了角色、资源感知的动态任务调度算法和相应的分层颜色Petri网模型,对多线索任务执行中的隐式并行进行了讨论、界定、建模和验证。引入了虚拟资源抽象层,将结构化、半结构化和非结构化资源以及元数据资源纳入统一的管理视角,设计和实现了元数据驱动的工作流的流程实例级和任务实例级的资源协作和管理机制。给出了借助Java反射机制和XStream设计并实现的一个基于企业服务总线和动态服务代理设计模式的服务接口,简化了系统内外异步服务的集成。采用语义网和本体技术对E-Learning系统内的学习资源、学习者的认知模型和偏好进行本体建模和语义标注,提出并实现了一个基于本体和工作流的个性化领域内容推荐方法：针对网络学习提出和设计了两种常用的学习认知类型和对应的具体认知过程模型,分为偏好导向的学习流程和主题导向的学习流程；为学习资源库定义了学习领域内容模型,进行了学习内容的学习偏好、学习主题和学习认知过程的标注。在认知过程、偏好和主题三个维度的综合配合下,利用用户上下文和流程上下文信息进行本体知识库的语义查询,生成个性化的推荐学习内容。第二,设计和实现了一个目标驱动、面向内容协作的分层任务网络规划方法的扩展模型(GCCHTN),用于支持面向知识协同工作流中的自动服务规划。在问题定义阶段融入目标、内容和任务的分层描述和基于语义关系的形式化模型,基于通用和可复用的原则设计和实现了一组用于目标、内容和任务分解映射的HTN操作算子,在此基础上围绕如何实现目标向任务、任务实现目标,内容向任务、任务生成内容等关系设计了相应的算法。运行时能够动态的、按需的构造HTN域的复合任务网络。而且根据环境参数、需求条件的不同,动态生成复合任务,克服了传统HTN和SHOP2需要显式、固化定义算法所需Method和Operator的缺点。规划领域方法库的核心主控方法“任务实作处理方法”提供了扩展点,实现了自定义领域方法的静态关联和动态关联机制。第三,在面向知识协同的跨组织多实体工作流的QoS计算模型中引入了信息质量和协同度指标。使用三角模糊数来描述具有不确定性的QoS属性,将一种改进的模糊层次分析法方法应用到QoS的计算中,来构造集成用户偏好的模糊QoS权重,减少人工任务量,提高了评价的准确率。提出了一种基于服务对象的服务提供者的组合流程协同度的计算方法,解决服务对象间协同度考量的不可能性。根据服务节点出现概率,结合服务提供者协同效应权重矩阵加权得到绑定流程实例的协同效应矩阵,最后求得绑定流程实例整体的协同度。在自动服务组合优化算法的研究中,提出了结合单点均匀变异策略、交叉变异后2-竞标赛选拔策略、最大维数的多点交叉和短期记忆注入的改进的遗传算法。在此基础上,提出了结合均匀分布变异策略、短期记忆注入配合克隆变异全收纳策略和最大维数多点交叉策略的改进的克隆选择算法；并将其用于改善免疫记忆克隆规划算法。在改进的免疫记忆克隆规划算法的基础上提出了一个基于混合克隆和选择性适度协同策略的并行短期记忆协同克隆选择算法(ParaCoSIMCSA)。ParaCoSIMCSA算法不仅可配置性强,并行效率高,简单易用,而且能结合多克隆和单克隆的优点,有较强的突破局部最优点吸引的能力,因此该算法有较强的适应性和稳定性。更多还原

【Abstract】 Business Process Management (BPM) has gradually been incorporated into the core strategic objectives of enterprises according to a research report of Gartner in2009. BPM is facing an unprecedented development opportunity. However, with the increasing knowledge collaboration and innovation activities of enterprises and organizations in the knowledge-based economy environment, traditional workflow management theories, methods and tools are facing enormous challenges.Innovation activities of enterprises in the context of the knowledge-based economy involve complex content-oriented, knowledge-intensive collaboration processes of multi-role, cross-organizational knowledge-based cooperation. Cross-organizational knowledge-intensive workflows are combination of the traditional workflow management system and knowledge management system. They not only support and manage the complex business collaboration relations, but also control and coordinate the whole lifecycle of knowledge including its creation, transferring and updating.Most workflow model and management system is designed to support management of mature products manufacturing process with fixed business objectives and unchangeable realization path, mainly focusing on the automation of activities of sequential, parallel and semi-automatic structure. Traditional workflow model is difficult to cope with complex cross-organizational, multi-role, and knowledge-intensive tasks’modeling and management. The main issues to be resolved include dynamic allocation of roles, tasks and resources in a collaborative process model which combines knowledge resource modeling and workflow process modeling, as well as knowledge-based collaborative lifecycle management.Traditional E-Learning system focuses on the construction of course resources and its representation techniques based on multimedia technologies on the internet, supporting teaching, learning and evaluation activities. These systems ignore the process management and monitoring of learning activities and multi-entity collaboration and knowledge-oriented cooperation. Workflow technologies are introduced into development of E-Learning system, make use of existing IT infrastructure of enterprises and organizations, and integrate the learning process into the enterprise workflows. Commercial workflow management software and traditional workflow modeling techniques are building upon the static structure of process, and its formal description is accurate and easy to implement and manage, but inevitably make process "rigid" which is difficult to be changed flexibly. It is hard to realize context-aware adaptive learning and knowledge recommendation mechanism in the stage of knowledge sharing and dynamic creation of tasks.To solve these issues the combination of modeling, management and collaboration of learning resources and learning processes is a must by using new technologies and new modeling techniques which can help to support reuse and accumulation of knowledge resources, to meet the knowledge needs of members of businesses and organizations, to providing personalized, context-aware knowledge filtering, pushing or recommendation mechanisms and to support creative knowledge based collaborative learning.Semantic web and ontology based modeling technology can be used to annotate and semantically model knowledge resources, build a unified, abstract and high level ontology based knowledge base. By using semantic query and reasoning technology, personalized user-oriented knowledge can be extracted from the ontology based knowledge base according to context information in dynamic learning environment.In content-oriented knowledge-intensive collaborative complex applications, specific business operations of enterprises and organizations will change oftenly because the user needs and business goals inside and outside the collaborative relationship of enterprises and organizations are highly viable. Process designers can not pre-define all aspects of a process model given a static process template of fixed structure. With changing user needs and business objectives, original static process template may not be applicable in a new situation.Automatic service planning in workflow is to automatically generate services execution sequence by using artificial intelligence planning techniques suited for changeable structure and patterns in workflow, create object-driven sub process instances on demand and traditional workflow can be used to control fixed activities at high level.Planning graph, HTN and their extensions or combinations are widely used in the field of automatic service planning. However, these methods focus on realization and application of service composition planing and optimization of the planning algorithm at low level, do not take into account the user needs and business objectives in the service planning at high level. Composition services which meet user needs and business objectives can not only rely on the field of artificial intelligence planning but also need more and specific research on related areas and technologies.In cross-organizational multi-entity collaboration workflows, service nodes composition problem not only involve the logical sequence relations (structural characteristics) of the flow among service nodes, but also relate to existing service node selection and optimization of the huge number of candidate services. Workflow instance to be instantiated by the workflow engine after modeling and verification needs to be bind to concrete service node object (task execution entity) based on business constraints and resource constraints to meet the functional requirements. But how to choose and compose these candidate service objects with same functionality but different QoS (Quality of Service) to make optimized compostion process instance is the key problem to research on.Innovation of this thesis mainly including:Firstly, a hierarchical colored Petri-net based Resource-oriented Collaborative Workflow model (ROCWF), its resource control model (Task Resource Multi-role Collaboration Model, TRRC) and the joint modeling method are proposed which support roles, tasks and resources association and perception. In order to reduce the complexity of modeling and verification, process validation task is divided into backbone process validation and tasks verification of sub-process, corresponding color-set definitions and validation rules are given. A dynamic task scheduling algorithm and corresponding hierarchical colored Petri-net model supporting role and resource awareness are designed. The thesis also analyzed and discussed the implicit parallel process execution in multi-thread task workflow. A virtual resource abstraction layer is introduced to put structured, semi-structured and unstructured resources and metadata resources into a unified management perspective, then a metadata-driven workflow’s process instance and task instance level resources collaboration and management mechanisms are designed and implemented. By using Java reflection mechanism and XStream, an Enterprise Service Bus and dynamic service agent design pattern based service interface is implemented to simplify asynchronous service integration inside or outside of the system.Semantic web and ontology modeling and semantic annotation of learning resources is used within the E-Learning system, proposed a recommendation method based on the learner’s cognitive model and preferences to provide personalized content in workflow. Two common learning cognitive types and the corresponding specific cognitive process models are proposed including preference-oriented learning processes and topic-oriented learning process. In learning resources library, the learning content domain model is defined including semantic annotations of learning content, learning preferences, learning topics, and learning cognitive processes. User context and process context is used to construct semantic query on knowledge base and generate personalized recommended learning content in three dimensions of the cognitive processes, preferences and topics.Secondly, a goal-driven, content-oriented collaboration extended model and method (GCCHTN) of hierarchical task network planning is proposed and implemented which can be used to support automatic service planning in knowledge-oriented collaborative workflows. The HTN operators are designed based on generic and reusable principles to cope with the mapping relationships of the hierarchical objectives, contents and tasks according to their formal model of the problem definition and semantic relations. The algorithm is designed to conduct content and task decomposition based on principles about how to achieve the goal of the task, perform the task to achieve the goal, what is the relationship of the content to the task and how the task generate the content designed algorithm. At runtime it supports dynamic, on-demand HTN task network decomposition. Composite task can be dynamically generated according to different environmental parameters and requirements, so as to overcome the shortcomings of traditional HTN and SHOP2which need to explicitly pre-define Method and Operator. The core of planning method is "task implementation method", which provides extension points used to embed custom static and dynamic specific methods and operators.Finally, in cross-organizational knowledge-oriented collaborative multi-entity workflow, this thesis introduces the quality of information and collaborative degree in its QoS computation model. Triangular fuzzy numbers are used to describe the uncertainty of QoS attributes, and an improved fuzzy analytic hierarchy process is applied to the calculation of QoS to construct fuzzy QoS weights integrated with user preferences, these strategies can reduce the amount of human task and improve the evaluation accuracy. Based on service providers’relations of the service objects, a calculation method of the composition process collaboration degree is proposed to avoid difficulty and infeasibility of traditional computation method. The collaboration degree of binding compositon process instance can be obtained by combining the service providers collaboration weight matrix of the binding process instance with the service node emergence probability.In the research of optimization algorithm of automatic service composition, this thesis proposed an improved genetic algorithm which is composed of single point uniform distribution mutation strategy, crossover, mutation and2-tournament selection strategy, maximum dimension multi-point crossover strategy, and short-term memory injection strategy. An improved clonal selection algorithm is proposed as well by using uniform distribution mutation strategy, short-term memory injection and maximum dimension multi-point crossover strategy. The improved clonal selection algorithm is used to improve the immune memory clonal programming algorithm.Based on the improved immune memory clonal programming algorithm, a hybrid clonal strategy and selective parallel cooperation based short-term memory injection clonal selection algorithm (ParaCoSIMCSA) is proposed. ParaCoSIMCSA algorithm not only can be configured easily, with high parallel efficiency and ease of use, but also combines the advantages of polyclonal and monoclonal, with better ability to jump out of local minima, so it appears to be a stable and robust discrete optimization algorithm.更多还原