CASIA OpenIR  > 毕业生  > 博士学位论文
面向智能制造实时监控的不确定复杂事件处理研究
其他题名Research on Complex Event Processing with Uncertainty for Real-time Monitoring of Intelligent Manufacturing
毛娜
学位类型工学博士
导师易建强 ; 谭杰
2016-05-24
学位授予单位中国科学院大学
学位授予地点北京
学位专业计算机应用技术
关键词实时监控 Cep Nfa 不确定性 否定操作符 检测优化 多规则耦合 原型系统
摘要
随着社会经济的发展,制造业不断转型升级,如今智能制造被视为新一轮生产模式的核心,成为全球制造业的发展趋势。对生产过程进行实时监控可以大幅提高生产企业的生产效率,提升企业的竞争力。智能制造生产模式下,企业面向产品全生命周期,实现泛在感知条件下的信息化制造,大量感知设备的部署使得企业可以获得大量不确定的实时感知数据流,这对企业信息系统的处理能力提出了巨大的挑战。复杂事件处理(CEP,Complex Event Processing)可以实时处理这些数据并发现数据之间隐含的有用信息,如异常行为或状态检测,使得企业能够对这些关键信息作出反应,提高企业的响应能力。但是,实际生产过程中获得的数据具有强不确定性,复杂事件检测是CEP的核心,对具有不确定特性的数据进行复杂事件检测是一个值得深入研究的问题。
针对智能制造实时监控应用中的不确定感知数据流的处理,本文以SEQ模式下的复杂事件检测为例,对基于自动机模型的不确定复杂事件检测方法进行了深入的研究。以引入不确定机制的复杂事件处理模型为切入点,从事件检测模型和算法两方面探索了不确定复杂事件检测方法及其优化方法,实现支持不确定性的复杂事件检测原型系统,并对多规则耦合情况下的不确定复杂事件检测展开研究。本文的主要工作和贡献如下:
1.  针对智能制造生产模式下不同的实时监控应用需求,建立了不确定复杂事件处理模型。首先,研究在实时监控系统中利用复杂事件处理技术的通用模式,提出基于不确定CEP的分层智能制造实时监控系统框架结构;然后,考虑感知数据流的不确定特性,定义了事件和规则的不确定约束,并在概率维上扩展了SEQ顺序运算符及相应的自动机检测模型,为不确定复杂事件检测提供了支撑;最后,针对多个规则同时进行检测的情况,给出了规则集合的相关概念定义,提出了规则集特征矩阵等规则集特征分析方法。
2.  针对实际监控应用中感知数据流的不确定性,提出基于NFA(Nondeter-ministic Finite Automata)的不确定复杂事件检测方法,提高对不确定数据流的复杂事件检测能力。首先,从事件检测模型出发,分析了现有基于自动机模型的事件检测模型的不足,提出一种新的基于NFA结构事件检测模型rNFA(NFA with run buffers);然后,分析基本事件之间的关系对复杂事件不确定性的影响,以概率理论为基础,建立复杂事件的不确定性计算模型,设计复杂事件的概率计算算法;最后,特别考虑是否含有否定操作符的情况,提出了两种不确定复杂事件检测算法。实验方法表明不确定复杂事件检测方法对于不确定数据流具有较好的事件检测性能。
3.  为了减少复杂事件匹配成本,提高事件匹配效率,提出基于NFA的不确定复杂事件检测方法优化方案。首先,针对SEQ匹配模式的不确定复杂事件检测方法,设计基于事件过滤的不确定事件检测优化算法,分别依据事件类型和概率阈值进行过滤,减少需要处理的事件个数,限制事件检测中间结果集;然后,针对含否定操作符的SEQ匹配模式的不确定复杂事件检测方法,设计基于否定前置的不确定事件检测优化方法,包括基于否定前置的事件检测模型和以此为基础设计的事件检测优化算法,在事件模式匹配的过程中直接考虑否定操作符,及时去除不满足查询要求的事件,减少复杂事件检测过程中的中间结果规模。实验方法表明不确定复杂事件检测优化方法具有较好的优化效果。
4.  针对实际监控应用中存在多个规则,且其中涉及的事件检测规则及基本事件存在耦合关系的情况,提出基于共享检测的多规则不确定复杂事件检测方法,满足对多个复杂事件进行同时检测的需求,避免事件模式间耦合部分的重复检测,优化复杂事件检测过程。首先,扩展了自动机模型,形成基于规则集特征的多规则复杂事件检测模型;然后,基于规则子序列特征矩阵设计匹配特征值计算方法,为检测出正确的复杂事件提供了基础;最后,在以上研究的基础上根据规则集的特征矩阵形成基于共享检测的不确定复杂事件检测算法。理论分析证明了基于共享检测的不确定复杂事件检测方法的有效性及优化效果。
5.  利用研究的不确定复杂事件检测技术,在事件处理引擎SASE 的基础上,引入不确定性因素,基于Java平台实现支持不确定性的铁路货车制造实时监控原型系统。本文以企业铁路货车转向架生产线北线上的生产为例,基于NFA的不确定复杂事件检测方法对输入的事件流和业务逻辑进行处理,对该段生产线上的生产过程进行监控,自动检测了工序间衔接不当和装配过程物料不足的情况,验证了原型系统的有效性和可行性。
其他摘要The manufacturing industry is in constant transformation and upgrading. Now intelligent manufacturing is becoming the trend of the global manufacturing industry, in which enterprises implement the information-based manufacturing in ubiquitous perception for product lifecycle. Real-time monitoring of production processes can greatly improve production efficiency and enhance the competitiveness of manufacturing enterprises. Because of the wide deployment of sensors, manufacturing enterprises get a lot of uncertain real-time data streams, which are challenging for the enterprise information system to deal with. Complex event processing (CEP) is an important technology in processing these data streams and finding the implied useful messages, such as abnormal detection of action, in a real-time way to help manufacturing enterprises improve the response capability. However, the sensor data streams from the real-time monitoring scenarios are highly uncertain. Complex event detection is the core of CEP, so complex event detection on uncertain data is a problem which is worth of further study.
To process uncertain sensor data streams coming from the real-time monitoring scenarios of intelligent manufacturing, this thesis mainly focuses on NFA-based Complex Event Detection method with Uncertainty (UCED method) taking event detection with the SEQ operator for instances. Introducing the uncertainty mechanism to event processing model, we explore UCED methods and their optimization. Besides, we realize a prototype system which can support UCED. At the same time, we study UCED methods for coupling multiple rules. Following are main contributions of this paper.
1.    For the satisfaction of different real-time monitoring applications, a complex event processing model is presented. First of all, a universal hierarchical framework in real-time monitoring systems of intelligent manufacturing based on Uncertain CEP is established. Next, the uncertain constraints of events and rules are defined, besides the SEQ operator and the corresponding event detection model based on NFA are extended in the probability dimension, which are used as the basis of UCED. Finally, the definition of relevant concepts of rule sets and characteristic analysis method for the rule sets based on eigenmatrix are described, so as to detect complex events simultaneously.
2.    Considering high uncertainty in sensor data streams from real-time monitoring applications, a NFA-based UCED method is proposed to improve the ability of complex event detection on uncertain data streams. Firstly, a new event detection model named rNFA is presented, since there have been deficiencies in the existing event detection models based on NFA structure. Secondly, as the relationship between simple events will affect the uncertainty of derived complex events, a probability calculation algorithm for complex events is designed. Finally, the UCED algorithms considering whether there is the negation operator in rules are described. Experimental results show that the proposed NFA-based UCED method has good performance on uncertain data streams.
3.    To reduce the cost and improve the efficiency of complex event matching, an optimized scheme of the described NFA-based UCED method above is proposed. On one hand, as for the UCED method in the SEQ pattern without the negation operators, the UCED optimized algorithm to filter events based on event types and the threshold of probability is presented, which can reduce the number of processed events and intermediate result sets during event detection. On the other hand, as for the UCED method in the SEQ pattern with the negation operators, the UCED optimized method pushing the negation operators to event matching process is presented, which includes the model and the algorithm for UCED with the negation operators. It can decrease the size of intermediate results through removing the events that are not suitable for rules in time. Experimental results show that the proposed optimized methods have better optimization effect compared with the UCED methods above.
4.    As for the existence of multiple rules in real-time monitoring applications in which there are coupling relationships in the related rules and simple events, a UCED method for coupling multiple rules is proposed to detect multiple complex events at the same time and optimize the process of complex event detection by avoiding the repeated detection of the coupling parts of the rules. First of all, the rNFA model is extended and the UCED model for multi-rules based on rule set characteristics is presented. Moreover, the calculation algorithm of the matching feature value for sub-sequence of the rules based on the eigenmatrix is described to be the foundation of detecting correct complex events. Lastly, the NFA-shared UCED algorithm based on the eigenmatrix of the rule set is presented. Theoretical analysis proves the validity and better performance of the proposed UCED method for coupling multiple rules.
5.  Using the UCED methods studied before, a prototype system for real-time monitoring in manufacturing of railway freight cars that can support uncertainty based on SASE is implemented. Taking the production in the north line of bogies of the railway freight cars as an example, the input event streams and business logic based on NFA structure are processed to monitor the production process, which detects the improper connection between operations and the shortage of material in assembly process automatically and verifies the effectiveness and feasibility of the prototype system.
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/11386
专题毕业生_博士学位论文
作者单位中国科学院自动化研究所
推荐引用方式
GB/T 7714
毛娜. 面向智能制造实时监控的不确定复杂事件处理研究[D]. 北京. 中国科学院大学,2016.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
博士论文-毛娜.pdf(4458KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[毛娜]的文章
百度学术
百度学术中相似的文章
[毛娜]的文章
必应学术
必应学术中相似的文章
[毛娜]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。