SIA OpenIR  > 工业控制网络与系统研究室
基于序列认知的工艺数据异常检测方法研究
Alternative TitleResearch on Abnormal Detection Method of Process Data Based on Sequence Cognition
石贺1,2
Department工业控制网络与系统研究室
Thesis Advisor尚文利
Keyword工业控制系统 异常检测系统 集成机器学习模型 自编码神经网络 长短期记忆神经网络
Pages59页
Degree Discipline控制工程
Degree Name硕士
2019-05-17
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract为构建完善的工业控制网络安全防护体系,保证工业控制系统基础设施安全以及工业产品的高品质要求,本文针对工业控制系统中时序序列的工艺数据异常行为检测这一重要课题进行研究,分析各类异常检测方法的优缺点,提出基于堆叠自编码神经网络降维,长短期记忆神经网络为主检测模型的工艺数据异常检测系统。首先,本文针对工业控制系统进行脆弱性分析,并概述人工智能技术在工业控制系统异常检测领域中的应用。从现场总线层、过程控制层、办公网络融合层三个层面分析现有工业控制系统的安全问题,透过现有的工业信息安全中存在的问题阐述工控系统异常检测的必要性。另外,鉴于人工智能技术近几年的快速发展和在异常检测领域的优异效果,本文重点概述了机器学习技术和深度学习技术在工业控制系统异常检测中的应用形式和各自的适用范围与优缺点。其次,本文着重介绍了对工控系统中工艺数据的预处理与特征工程方法。工艺数据异常检测模型的检测效果依赖于原始的训练样本数据以及对数据进行的特征工程方法,适当的特征偏度转换、平滑数据处理对增强模型的泛化能力非常有帮助,本文把工艺数据特征分为连续型与离散型两种类型分别进行数据预处理与特征工程,然后通过主成分分析法对高维的工艺数据大幅度降维,最后通过集成规则树模型或自编码神经网络模型进行特征选择,为后续的主模型检测提供高质量训练数据。最后,本文通过多种模型试验效果对比,提出一种基于时间序列的异常检测方法。该方法对工艺数据进行相关性分析、向量映射等处理,自编码神经网络降维,然后根据工艺数据在传输序列间的相互关联性,设计基于长短期记忆神经网络的异常检测模型,分别进行多种工艺数据异常检测模型的实验验证分析。实验中选择多种时序模型与非时序模型进行比较,结果表明基于时间序列的异常检测模型能有效提高工艺数据异常检测准确率,并且误报率要低于其他异常检测模型,可以获得较好的异常检测实时性。
Other AbstractTo build perfect industrial control network security protection system, ensure the safety of industrial control system infrastructure, and industrial products of high quality requirements, this article choose in the industrial control system based on the temporal sequence of abnormal behavior detection process data this important subject for research, analysis the advantages and disadvantages of all kinds of anomaly detection model, put forward neural network based on stack from coding dimension reduction both short-term and long-term memory neural network is given priority to detect sequential process data anomaly detection system of the model. First, based on industrial control system vulnerability analysis and summarizes the artificial intelligence technology in the application of industrial control system anomaly detection, from the fieldbus layer, process control, analysis of office network layer three levels of current industrial control system security problem, through the existing industrial information security problem in this paper, the necessity of industrial control system anomaly detection. In addition, due to the hot development of artificial intelligence technology in recent years and the excellent effect in the field of anomaly detection, this paper mainly summarizes the application forms of machine learning technology and deep learning technology in the anomaly detection of industrial control system, as well as their respective scope of application, advantages and disadvantages. Secondly, this paper focuses on the pretreatment of process data and characteristic engineering methods. Abnormal detection model of the effect depends on the original training data and the data processing, the characteristics of the appropriate transformation, smoothing the data processing of skewness is helpful to enhance the generalization ability of the model, in this paper, the characteristics of process data can be divided into continuous and discrete two types respectively for data processing and characteristics of the project, and then introduces the PCA principal component analysis (PCA), since the coding of neural network, integration rule tree feature selection principle of three kinds of feature dimension reduction and dimension reduction effect. Finally, this paper presents a time series-based anomaly detection method by comparing the results of various model tests. The method to process data correlation analysis, vector map processing, such as the encoding neural network dimension reduction, then according to the process data in transmission sequence, the correlation between the LSTM neural network based on short - and long-term memory of anomaly detection model, respectively, for a variety of process data anomaly detection model analysis of the experimental verification. A variety of timing models and non-timing models were selected for comparison in the experiment, and the results showed that the anomaly detection model based on time series could effectively improve the accuracy of process data anomaly detection, and the false alarm rate was lower than that of other anomaly detection models, which could achieve better real-time anomaly detection.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/25202
Collection工业控制网络与系统研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
石贺. 基于序列认知的工艺数据异常检测方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
基于序列认知的工艺数据异常检测方法研究.(1941KB)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
[石贺]'s Articles
Baidu academic
Similar articles in Baidu academic
[石贺]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[石贺]'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.