基于流形学习算法的非高斯过程监控方法研究及在化工过程监控中的应用

发布时间：2018-04-10 18:28

本文选题：非高斯 + 最大方差展开　；参考：《华东理工大学》2015年硕士论文

【摘要】：随着国民经济的快速发展以及人民生活水平的显著提高,整个社会对流程工业的生产制造提出越来越高的要求,其安全与稳定不容小觑。保证生产安全和提高产品质量是流程工业亟待解决的问题,过程监控技术的合理运用是解决这些问题的有效途径。由于近年来计算机控制系统和智能仪表在生产中的应用,可以获取大量数据。通过对过程数据的统计分析进而对系统的运行状态进行监控,已然成为近年来的热门研究领域。多变量统计过程监控(Multivariate Statistical Process Monitoring, MSPM)方法作为一种重要的基于数据驱动的监控方法,受到了学术界和工业界的普遍关注。但是,传统MSPM方法受到过程中的很多条件限制,例如过程数据应尽量满足线性关系、高斯分布。然而实际工业由于种种因素的影响,数据变量间呈现强非线性以及非高斯分布等关系。因此,在前人研究的基础上,对实际工业中存在的问题进行了分析,进行了如下工作： 1.针对传统监控方法在处理数据进行建模时破坏非线性结构,监控效率不高的情况,提出了一种基于否定选择算法(Negative Selection Algorithm, NSA)的过程监控方法。该方法首先采用最大方差展开(Maximum Variance Unfolding, MVU)方法从原始数据中提取低维流形,再利用NSA对低维流形进行建模得到“超球体群”模型,从而实现对过程的监控。TE平台仿真表明提出的方法较其他方法具有更好的检测能力。 2.针对支持向量数据描述(Support Vector Data Description, SVDD)方法在处理大样本时遇到的“维数灾难”的问题,同时利用其在处理小样本数据独有的优势,提出了一种基于LTSA-Greedy-SVDD的过程监控方法。该方法首先引入局部切空间排列(Local Tangent Space Alignment, LTSA)算法提取低维流形,之后引入Greedy方法提取特征建模样本,从而大大地减少了运算时间。TE平台仿真以及应用仿真表明了该方法的有效性。 3.针对过程数据是高斯分布和非高斯分布的混合体等问题,提出了一种基于加权联合指标的过程监控方法。该方法利用非高斯-高斯两步策略提取过程数据的有用信息建立统计模型,之后采用加权策略对两个统计量进行加权得到新的统计指标。基于加权联合指标的监控方法的有效性在数值系统仿真中得到了验证。TE过程仿真和工业应用也都表明了提出的方法的有效性。
[Abstract]:With the rapid development of the national economy and the remarkable improvement of the people's living standard, the whole society has put forward more and more high requirements for the production and manufacture of the process industry, and its safety and stability should not be underestimated.Ensuring production safety and improving product quality are the urgent problems to be solved in process industry, and the rational application of process monitoring technology is an effective way to solve these problems.Because of the application of computer control system and intelligent instrument in production in recent years, a large amount of data can be obtained.It has become a hot research field in recent years to monitor the running state of the system through the statistical analysis of the process data.As an important data-driven monitoring method, multivariate Statistical Process monitoring (MSPMM) has attracted much attention from academia and industry.However, the traditional MSPM method is restricted by many conditions in the process, for example, the process data should satisfy the linear relation as far as possible, Gao Si distribution.However, due to the influence of various factors, the data variables are strongly nonlinear and non-Gao Si distribution.Therefore, on the basis of previous studies, the problems existing in the actual industry are analyzed, and the following work is done:1.A process monitoring method based on negative selection algorithm (NSAs) is proposed to solve the problem that the traditional monitoring method destroys the nonlinear structure and the monitoring efficiency is not high when the data is modeled.In this method, the maximum Variance unfolding (MVU) method is used to extract the low-dimensional manifold from the original data, and then the "hypersphere group" model is obtained by using NSA to model the low-dimensional manifold.The simulation results show that the proposed method has better detection capability than other methods.2.Aiming at the problem of "dimension disaster" encountered by support vector data description Vector Data description (SVDDD) method in dealing with large samples, and taking advantage of its unique advantages in dealing with small sample data, a process monitoring method based on LTSA-Greedy-SVDD is proposed.In this method, the local tangent space arrangement Local Tangent Space alignment (LTSA) algorithm is introduced to extract the low-dimensional manifold, and then the Greedy method is introduced to extract the feature modeling samples.Thus, the computation time is greatly reduced. Te platform simulation and application simulation show the effectiveness of the method.3.Aiming at the problem that the process data is a mixture of Gao Si distribution and non-#china_person1# distribution, a process monitoring method based on weighted joint index is proposed.The statistical model is established by using the useful information of process data extracted by non-Gaussian two-step strategy, and the new statistical index is obtained by weighting the two statistics by weighted strategy.The effectiveness of the monitoring method based on weighted joint index is verified by numerical system simulation. Te process simulation and industrial application also show the effectiveness of the proposed method.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ015.9

【参考文献】