SIA OpenIR  > 工业控制网络与系统研究室
基于时序数据挖掘的化工生产工艺异常诊断研究
Alternative TitleResearch on Abnormal Diagnosis of Chemical Production Process Based on Time Series Data Mining
李俊朋
Department工业控制网络与系统研究室
Thesis Advisor刘意杨
Keyword时间序列 线性分段 核主成分 异常检测
Pages66页
Degree Discipline控制工程
Degree Name专业学位硕士
2020-05-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract化工工业,也属于流程工业,主要是通过能源、设备和其他资源来混合或分离各种成份并引起化学反应,从而达到增值的目的。由于工艺或者控制等原因,在化工生产过程中会存在一些异常,造成产品质量的不稳定,因此,需要对化工生产进行异常判别和诊断,进而稳定产品质量。在实际的化工反应中,由于反应机理和参数等问题,难以直接建模。但在化工生产过程中存储大量随时间变化的参数数据,这些数据在一定程度上反应了生产状况信息和产品结果信息,因此,可利用这些随时间变化的工艺参数挖掘生产过程的信息,进行异常研究。本文以甘肃某地车间内的含能炸药黑索今的生产线为背景进行研究,对含能炸药黑索今的化工生产流程进行分析并研究影响生产的因素,本文以甘肃某地车间内黑索今的生产工艺参数数据为基础,对化工生产过程中产生的工艺异常进行判别和诊断,本文主要工作如下:1)分析研究了甘肃某地车间内的含能材料黑索今的生产背景和生产工艺特点,根据黑索今的实际化工生产线的特点提出了基于时序数据挖掘的化工生产工艺异常的研究思路。并针对黑索今生产线上采集到的历史工艺参数数据存在的问题进行缺失值填充处理和数据转换等。2)为能够提取工艺参数的波动和变化趋势特征,在工艺参数预处理的基础上,实现对化工工艺参数序列的分段线性表示,针对目前时间序列数据挖掘中,时间序列线性分段表示多局限于对相邻的极值点或波动幅度大的特征点进行的评价进行来筛选、确定分段点的选择,容易陷入局部最优,不能准确对时间序列趋势进行表示的问题,本文提出了基于转折点和趋势段的时间序列趋势特征提取算法,通过两组实验表明,该方法不仅具有良好的抗噪能力和趋势提取能力,在相同压缩率的情况下具有更好的拟合精度,可为后续数据挖掘提供更好的特征。3)本文化工生产过程本身具有整体运行不稳定的情况,核主元分析基于单一时刻进行异常检测的方式忽略了参数的时间特性,对本课题的化工生产工况不适用,因此本文提出了基于时间段的时间序列趋势特征核主元分析检测方法,在时间序列分段的基础之上,构建反应工况运行趋势的特征,弥补核主元分析检测方式的缺点,实现了黑索今生产过程中异常的检测,并实现了对异常原因的诊断。
Other AbstractThe chemical industry, which is also a process industry, mainly uses energy, equipment and other resources to mix or separate various components and cause chemical reactions to achieve value-added purposes. Due to process or control and other reasons, there will be some abnormalities in the chemical production process, resulting in unstable product quality. Therefore, abnormal identification and diagnosis of chemical production are needed to stabilize product quality. In actual chemical reactions, it is difficult to directly model due to problems such as reaction mechanism and parameters. However, the chemical production process stores a large amount of parameter data that changes with time. These data reflect the production status information and product result information to a certain extent. Therefore, these time series parameter data can be used to mine production process information and conduct abnormal research. Based on the research on the production line of energetic explosive in a workshop in Gansu Province, this paper analyzes the chemical production process of energetic explosive RDX and studies the factors that affect the production. Based on the historical production data of RDX in the workshop, this paper distinguishes and diagnoses the abnormal production process data in the chemical process. The main work of this article is as follows: 1) Analyze and study the production background and production process characteristics of the energetic material RDX in a workshop in Gansu. According to the characteristics of RDX's actual chemical production line, a research idea of abnormal chemical production process based on time series data mining is proposed. At the same time, according to the problems existing in the historical process parameter data collected by RDX's production line, it performs missing value filling processing and data conversion. 2) After data preprocessing, in order to be able to extract the trend characteristics of the time series, realize the piecewise linear representation of the process time series parameters in chemical production. For the current time series linear representation is mostly limited the evaluation of adjacent extreme points or feature points with large fluctuations is used to screen and determine the selection of segment points, which is easy to fall into the local optimum and cannot accurately represent the time series trend. This paper proposes a problem based on turning points and Time series trend feature extraction algorithm for trend segments. Through two sets of experiments, this method not only has good anti-noise ability and trend extraction ability, but also has better fitting accuracy under the same compression rate, which can be used for subsequent data mining. Provide better features. 3) In this paper, the chemical production process itself has an unstable overall operation. The abnormal detection of nuclear principal component analysis is often based on a single time sampling. This detection method ignores the time characteristics of the parameters and is not applicable to the chemical production conditions of this subject. Therefore, this paper A detection method based on the time series trend characteristics is proposed. Based on the PLA of time series, the characteristics of the reaction trend of the operating conditions are constructed to make up for the shortcomings of the nuclear principal component analysis and detection method, and the process of abnormal production can be identified, and Realize the diagnosis of abnormal causes.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27147
Collection工业控制网络与系统研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
李俊朋. 基于时序数据挖掘的化工生产工艺异常诊断研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
基于时序数据挖掘的化工生产工艺异常诊断研(2284KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李俊朋]'s Articles
Baidu academic
Similar articles in Baidu academic
[李俊朋]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李俊朋]'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.