SIA OpenIR  > 信息服务与智能控制技术研究室
工业过程感知序列预处理及融合方法研究
其他题名Research on Preprocessing and Fusion Methods for Industrial Process Sensing Series
苏卫星1,2
导师朱云龙
分类号TP311.13
关键词时间序列 数据预处理 多源信息融合 异常检测 突变检测
索取号TP311.13/S91/2014
页数102页
学位专业机械电子工程
学位名称博士
2014-11-24
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门信息服务与智能控制技术研究室
摘要近年来,随着物联网技术的发展及其在工业领域的推广应用,大量的各种类型的传感器、自动识别系统、工业无线网络系统等新型的信息获取与传输技术被引入到工业生产现场,构建起具有泛在特征的工业物联环境,使得面向工业生产过程控制与管理的泛在信息的获取与传输成为可能。在工业过程感知信息极大丰富的基础上,信息的分析、处理与融合利用成为提升人们对各种工业活动的信息感知和管理控制能力的关键所在。为此,本文从工业过程感知序列预处理与融合利用入手,开展相关关键技术及方法研究,为工业感知信息融合处理提供具有一定实用性的技术方法和解决思路。具体研究内容包括:1、为满足大数据量感知序列异常值检测需求,提出了一种计算简单快速的基于边缘化后验比检验的异常值在线检测方法。该方法将基于“偏差”的检测思想与统计学理论相结合,首先利用基于数据的鲁棒建模方法对待检测数据进行拟合并得到拟合残差,然后利用统计学知识分析拟合残差以最终确定数据的异常情况。为了实现算法的在线检测要求,引进了两窗口结构并对其加以改进;通过合理的选取先验分布以及对未知参数进行边缘化处理的方式,有效地减少了算法中的参数个数,提高了算法的可用性。实验证明该方法具有良好的检测准确性和实用性。2、针对工业过程控制系统异常值检测需求,在改进传统小波异常数据检测方法基础上,提出基于小波与RBF的异常数据在线检测方法。该方法利用鲁棒RBF网络在线拟合被检测据,采用可在线应用的迭代小波变换对残差信号进行小波分解,利用隐马尔可夫模型分析小波系数自动在线检测数据异常,避免了检测阈值的设定。为保障方法能够在线应用,提出一种利用原始数据训练RBF网络模型的训练算法,使得网络训练不依赖于干净数据。仿真和比较实验证明,该方法对于工业过程中的具有较大振荡特征非稳态序列有较好的检测效果。3、针对传统突变点检测算法具有大延时的问题以及实际数据中同时含有突变点、异常点的实际情况,提出一种基于小波变换有效分数向量的异常点、突变点检测算法。该方法通过引入有效分数向量作为检测统计量,有效避免了传统检测统计量随着数据增多而无限增大的缺点;提出利用小波分析统计量的办法,有效地克服了传统突变点检测算法中存在大延时的缺陷;利用李氏指数及小波变换的关系,实现了在一个检测框架内同时在线检测异常点以及突变点,使得该检测算法更符合突变点及异常点同时存在的实际情况。仿真实验和性能比较结果证明了所提方法的有效性和实用性。4、对于同构多传感器系统在无先验知识、无系统模型条件下对同一未知目标进行在线测量的数据融合问题,提出一种基于改进模糊聚类的多源同构感知序列在线融合方法。该方法采用引入噪声类的鲁棒模糊聚类方法分析同时刻多源数据,避免了传统模糊聚类融合方法中对聚类数设定的依赖,同时能够有效去除系统偏移较大的数据源和异常信号对融合结果的不良影响;通过引入隶属度函数影响因子,增加历史融合结果对当前融合的指导作用,有效降低偶然错误对融合结果的影响,提高了融合精度。实验结果证明了该方法较传统方法在系统适应性和融合精度方面的优势。5、针对多源异构感知序列的融合问题,提出一种能够对异构观测序列进行融合处理的融合框架和方法。该方法首先对多源感知序列的同一时刻的观测值分别进行模糊化处理,并在样本规则库的作用下形成对目标对象可能状态支持程度的证据。最终,在随机集表示的D-S证据理论融合框架下,完成多证据的融合,进而达到异构信息融合的目的。最后利用应用案例的求解来验证该方法在解决多源异构信息融合上的可行性。
其他摘要In recent years, with the development of IOT technology and its application in the industrial field, a large number of various types of sensors, automatic identification systems, industrial wireless network and other new data acquisition and transmission technology are applied into the industrial workshops, which building an industrial IOT environment and making the ubiquitous information sensing for industrial process control and management possible. With the great wealth of the sensing information in industrial process, analyzing, processing and fusing this information becomes the key to raise capabilities of people on the managing and controlling the various industrial activities. So this paper researches the preprocessing and fusing methods for industrial process sensing series in order to provide some practical ideas and solutions for industrial sensing information processing and the research includes:. 1. In order to meet the massive time series outlier detection requirements, a simple on-line outlier detection method is presented which is based on marginalized posterior ratio test. The main detection principle is that row data are fitted by a data-based robust modeling method and thus get fitting residuals, then the residuals are analyzed with statistics-based method to judge whether the data are outliers or not. During the detection process, double-window structure is introduced and improved to fit on-line detection use, well selected prior distribution and marginalization processing for unknown parameters also makes the number of parameters decrease sharply and thus improves the availability of the method. The results of simulation in the end of the paper prove that this method has a higher accuracy and availability for on-line outlier detection for time series. 2. In order to meet the needs of industrial process control systems outlier detection, a new outlier detection algorithm combining wavelet technique with a robust Radial Basis Function (RBF) network is proposed by improving the shortcoming of the conventional WT method for the data in PCS. The method fits the data by a robust RBFN and can do online wavelet decomposition of the residual signals by an iteration wavelet. A Hidden Markov Model (HMM) is adopted as an analysis tool to accomplish on-line automatic detection without pre-selecting threshold. In order to fits for the application of online detection, a robust Radial Basis Function Network (RBFN) training algorithm is proposed, which can train RBFN online using the original data as training set without the need of clean data. Simulation and comparison results verifies that the method gets a better detection results for unsteady sequence of industrial processes with a larger oscillation characteristics. 3. Aiming at the conventional change-points detection method existing the shortages on time delay and inapplicability for the time series mingled with outliers in the practical applications, an outlier and change-point detection algorithm for time series, which is based on the wavelet transform of the efficient score vector, is proposed in this paper. The algorithm introduces the efficient score vector to solve the problem of the conventional detection method that statistics often increasing infinitely with the number of data added during the process of detection, and proposes a strategy that analyzing the statistics by using wavelet in order to avoid the serious time delay. In order to distinguish the outlier and change-point during the detection process, we propose a detecting framework based on the relationship between Lipschitz exponent and the wavelet coefficients, by which both outlier and change-point can be detected out meanwhile. The advantage of this is the detection effect is not subject to the influence of the outlier. It means that the method can deal with the time series containing both outliers and change-points under actual operating conditions and is more suitable for the real application. Eventually, the effectiveness and practicality of the proposed detection method have been proved through simulation results. 4. In order to deal with the problem of multi-sources data fusion for a homogeneous multi-sensor system during the online measurement of an unknown target without prior knowledge or system model, this paper proposes a new fusion method based on advanced fuzzy clustering. The method utilizes the robust fuzzy clustering, which includes a noise cluster besides the normal clusters, to process the multi-sources data with the some timestamp. The advantage of this is that it doesn't need to select the number of the clusters, which is much more important for traditional FCM clustering based data fusion method, and can avoid the ill effects of both the data sources with large system drift and the abnormal signals to the fusion result. Furthermore, an impact factor for membership function is added to guide the current fusion process by historical fusion results, which helps to reduce the probability of falling into local minimum during the iterative calculation and improve the fusion accuracy. The simulation results show that, comparing with the weighted averaging fusion method and traditional FCM clustering based fusion method, the proposed method has more advantages in the adaptability and accuracy. 5. For the fusion of multi-source heterogeneous sensing series, we propose both framework and method for fusing the heterogeneous observed sequence. With the method the observations of the same time from multi-source sensing series are fuzzy processing first and then an evidence of supporting the state of the target object is getten according to a sample rule. Eventually, the fusion of multi-evidences is made under the framework of D-S evidence theory merge rule denoted by random set and thus achieving the purpose of heterogeneous information fusion. At the end of this chapter the solving processes of an application case veritys somehow the feasibility of the proposed method for multi-source heterogeneous information fusion.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/16749
专题信息服务与智能控制技术研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
苏卫星. 工业过程感知序列预处理及融合方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2014.
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
工业过程感知序列预处理及融合方法研究.p(5270KB) 开放获取CC BY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
[苏卫星]的文章
百度学术
百度学术中相似的文章
[苏卫星]的文章
必应学术
必应学术中相似的文章
[苏卫星]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。