CASIA OpenIR  > 毕业生  > 博士学位论文
基于事件图的实时感知数据流处理
Alternative TitleReal-time Sensor Data Stream Processing with Event-graph
李娜
Subtype工学博士
Thesis Advisor杨一平
2014-05-25
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
KeywordCep 事件图 感知数据流 分布式处理 事件调度 Cep Event-graph Sensor Data Stream Distributed Processing Event Scheduling
Abstract随着感知设备的广泛部署,感知数据流处理研究具有重要的理论与应用价值。复杂事件处理(CEP,Complex Event Processing)模型作为流数据处理的主要工具之一,由于其简单性和智能性,已经受到学界和工业界相当关注。复杂事件处理一般用于及时发现应用场景中系统感兴趣的语义信息,如异常行为或状态检测。而在利用CEP模型处理感知数据流时,感知数据流的复杂事件检测性能是影响应用系统性能的关键因素。 针对生产制造过程中的异常检测场景的感知数据流处理,本文对基于事件图的复杂事件检测及性能优化进行了深入研究,包括感知应用中的复杂事件模型、并发事件检测调度、并行复杂事件检测,以及规则动态调整的及时处理。本文的主要工作和贡献如下: 1.面向感知的复杂事件模型方面:对事件代数的顺序运算符进行了扩展,添加了计数约束,丰富了事件语义;对标准的ECA规则模型进行了扩展,加入了截止期要素,为复杂事件检测的实时性分析提供了支撑;针对规则集上复杂事件检测的实时性和灵活性,分析了规则集特征及规则集特征对于复杂事件检测响应时间和动态规则集适应性的影响。 2.并发事件检测调度方面:针对复杂事件间的耦合关系带来的并发事件检测问题,分析了耦合事件结点上的不同事件实例调度顺序对并发事件检测响应时间的影响;基于扩展事件图模型,提出了考虑截止期约束的事件调度优化模型;基于事件结点的静态拓扑结构和动态截止期特征,给出基于三种优先级评价指标的事件调度策略,优化了并发事件检测任务执行顺序。实验方法表明基于事件调度的并发事件检测能够减少复杂事件检测响应时间。 3.并行复杂事件检测方面:针对多个事件图的并发事件检测情形,为充分利用处理节点资源,减少不同事件图间事件检测任务的空等待时间,设计了基于规则集划分的并行复杂事件检测方法;考虑规则间的组合关系和基本事件的特征,分别提出基于规则子集矩阵上精确覆盖的全局最优规则集划分算法,以及基于基本事件聚类的局部最优规则集划分算法。实验方法表明基于规则集划分的并行复杂事件检测方法能够减少复杂事件检测响应时间。 4.规则集动态调整的及时处理方面:针对规则动态调整问题,分析了规则集上的规则添加、删除和更新操作对基于事件图的复杂事件检测输出和时间性能的影响,提出面向规则集动态调整的事件图集调整方法。在单处理节点上,针对事件图调整过程中子结点不完整和结点间关系不完整导致的事件检测错误,提出"失效-激活"机制;在多处理节点上,针对规则添加对处理节点上规则分布的影响,提出可及时处理规则添加的规则集划分算法。实验方法表明利用提出的面向规则集动态调整的事件图集调整方法,能够在保证复杂事件检测正常执行的同时不影响复杂事件检测性能。
Other AbstractWith the prevailing of sensors, researches on sensor data stream processing is important from both aspects of theory and application. Complex Event Processing (CEP) is an important technology in processing stream data. With advantages on simplicity and intelligent, CEP has got great attention from both research and industry fields. It is generally adopted to discover semantic information of system interests from application scenarios, such as abnormal detection of action. Using CEP to process sensor data stream, the complex event detection plays an important role in determining sensor applications performance. To process sensor data streams come from the abnormal detection scenario of manufacture process, this thesis mainly focuses on the event-graph model based complex event detection method and its improvements. Following are main contributions of this paper: 1.Complex event model in sensor applications: The event semantic is extended by adding counting constrains to the temporal sequence event algebra. The standard ECA rule model with deadline extension is used as the basis of analyzing real-time complex event detection implementation. Both rule set characteristics and its affection towards the response time and dynamic rules adjustments are studied, so as to improve real-time performance and flexibility for complex event detection. 2.Concurrent event detections scheduling: With the common event nodes between different event-graphs, event-graphs will be merged into an event-graph. However, the common event nodes will produce problems of multiple event detection tasks waiting for the same event instance. As different scheduling order of the common event instance among these concurrent event detections will lead to different complex event detection response time, we proposed the event scheduling method optimization model related to deadline constrain. We get three priority metrics from both aspects of topological particularity and the deadline. Experimental results show that the proposed event scheduling can yield a substantial improvement in complex event detection response time. 3.Parallel complex event processing: Concurrent event detections exist among multiple event graphs. Towards reducing the waiting time among these concurrent event detection tasks, we proposed two algorithms to partition the rule set to different processing nodes and implemented the parallel complex event processing based on these rule set partition algorithms. The global...
shelfnumXWLW2035
Other Identifier201118014629083
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6602
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
李娜. 基于事件图的实时感知数据流处理[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20111801462908(2741KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李娜]'s Articles
Baidu academic
Similar articles in Baidu academic
[李娜]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李娜]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.