CASIA OpenIR  > 毕业生  > 硕士学位论文
基于SVDD和CLIPS的生产过程异常检测和诊断
Alternative TitleResearch of Process Abnormal Detection and Diagnosis Based on SVDD and CLIPS
沈大伟
Subtype工学硕士
Thesis Advisor庄诚
2012-05-28
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline控制理论与控制工程
KeywordSvdd 过程异常检测 过程异常诊断 规则提取 专家系统 Clips Svdd Process Abnormal Detection Process Abnormal Diagnosis Rules Extraction Expert System Clips
Abstract随着生产过程自动化程度的不断提高,大量的过程数据被不断的采集和保存下来,操作人员的工作也逐渐从常规控制转移到对过程的监控和管理。被采集和保存下来的数据反映了过程的运行状态和趋势,然而面对海量的数据,即使有经验的操作人员也很难准确的区分正常和异常工况,特别是在故障发生的早期。解决上述问题的方法可以分为两个步骤,首先由过程数据构造少数监控变量,根据监控变量的统计特征来监视过程的运行状态;然后当检测到过程异常时,再诊断过程异常数据,找到导致过程异常的原因,指导过程操作。在本论文中,采用SVDD(Support Vector Data Description)方法对过程数据进行建模实现过程异常检测,采用专家系统的方法对过程异常进行诊断。主要工作内容如下: 论文针对SVDD算法在对过程数据进行建模的过程中时间复杂度过高的问题,提出了一种基于过程趋势信息的非均匀数据采样方法,该方法有效的降低了SVDD训练时所需的数据集大小,同时为了有效的提取过程数据中的关键样本,本文提出将每次训练时所产生的支持向量和测试时的错分样本相结合构造最终的模型训练数据集,在不降低SVDD模型的分类性能的条件下,有效的减少了训练的时间复杂度。 针对SVDD算法难以定位过程异常的问题,采用决策树算法分析SVDD标定的过程异常数据和正常过程数据的差别,辅助操作人员定位过程异常的位置,同时将决策树与SVDD异常检测结合生成过程异常规则,通过人工分析,提取诊断规则,建立专家系统诊断知识库。 针对在实际用专家系统的方法进行过程异常诊断时,专家系统的知识库需要频繁更新的问题,本文开发了一个专家系统编程语言CLIPS的接口程序,程序将输入的产生式规则自动转化为可执行的CLIPS代码。同时针对CLIPS不支持反向链规则的问题,提出一种基于消息处理机制的反向推理方法。本文中分别在TE过程仿真平台和某合成氨粗煤气变换工段上验证了本文所提方法的有效性和实用性。
Other AbstractAs the increasing use of advanced control system in the production process, the work of operators have changed to process monitor and management from regular controls. On the other hand, with the advances in the sensor and computer technology, massive process data have been collected and stored which contains the information about the process states and trends. Facing the massive data, the operators could not easily distinguish the normal and abnormal operating mode, even for the experienced operators, especially in the early stages of a abnormal event. An approach to deal with the problem has two steps: First, monitoring a statistical variable from the process data, which reflects the overall states of the process. Then, if abnormal state is detected, a diagnosis process will be excuted to find the reason of the abnormal state. In this thesis, a one-class classifier SVDD(support vector data description) will be used to model the normal process data and then to classify the normal and abnormal process states. Expert system will be used to to diagnosis the process abnormallities. The main contents of this thesis are as follows: For the problem of the high computational complexity of SVDD when training with massive process data, this thesis puts forward a non-uniform sampling method to deal with the problem. This method could efficiently reduce the size of the training set. On the other hand, in order to efficiently extract essential samples from the process data, the method collects support vectors from the training set and the false outliers from the testing set. At last, all these collected samples are combined to form a training set. All these data preprocessing has lead to significant decrease in the time complexity of SVDD training while keeping equivalent classification performance with the model trained on the whole process data. For the problem that SVDD can not locate the process abnormallity directly,this thesis uses the decision tree algorithm to analyse the differences between the normal and abnormal process data to asssit the operators to locate the process abnormallities. At the same time, we can extract diagnosis rules by analysing the results from decision tree analysis, which can be used to update knowledge database of expert system. For the problem that domain experts could not update the knowledge database directly when using the expert system to diagnosis process abnormal data. This thesis has developed a interface ro...
shelfnumXWLW1715
Other Identifier200828014628011
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7617
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
沈大伟. 基于SVDD和CLIPS的生产过程异常检测和诊断[D]. 中国科学院自动化研究所. 中国科学院研究生院,2012.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20082801462801(2794KB) 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
[沈大伟]'s Articles
Baidu academic
Similar articles in Baidu academic
[沈大伟]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[沈大伟]'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.