摘要：在工业生产中，生产安全与产品质量一直是过程工业的两大重要主题。故障诊断技术自70年代以来一直是过程监控领域的一个重要分支。而在故障诊断技术中，比较有发展前景的的是基于数据的多变量统计分析方法，这是由于过程工业本身难以建立精确模型、强耦合的多变量关系等特点决定的。
偏最小二乘法（PLS，Partial Least Squares）分析方法是一种利用统计原理提取过程数据的有用信息建立过程模型的降文技术。偏最小二乘法是一种数学优化技术，它通过最小化误差的平方和找到一组数据的最佳函数匹配。 用最简的方法求得一些绝对不可知的真值，而令误差平方之和为最小。 通常用于曲线拟合。很多其他的优化问题也可通过最小化能量或最大化熵用最小二乘形式表达。 本论文以偏最小二乘法的算法原理出发，系统地探讨了利用PLS对多变量生产过程进行故障检测与诊断的整个思路与应用情况。
TE过程(Tennessee Eastman Process)是基于实际工业过程的仿真案例。它是由美国Tennessee Eastman化学公司的过程控制小组的J.J.Downs和E.F.Vogel提出的，这个案例很适合于研究过程控制技术，因而，在提出后引起了过程控制学术界的关注，被广泛用于过程控制研究。5888
关键字： 偏最小二乘；TE过程；过程监控；故障检测与诊断
Partial least squares algorithm in the application of process monitoring
Abstract:Industrial production safety and product quality has always been a process of two important themes in industrial production. Fault diagnosis technology since the 70 s has been an important branch in the field of process monitoring. In the fault diagnosis technology, more promising is based on the data of multivariate statistical analysis method, this is due to the process industry itself is difficult to establish precise model of multivariable, strong coupling relationship, etc.
PLS is a statistical principle is utilized to extract data useful information to build process model of dimension reduction techniques. Partial least squares is a kind of mathematical optimization techniques, it is found by minimizing the error sum of squares of a set of data the best match function. With the minimalist approach for some absolutely unknowable truth value, and make the square sum of error is minimal. Usually used for curve fitting. Many of the other optimization problems can also be used by minimizing or maximizing entropy energy with the least squares expression form. Based on partial least squares algorithm principle, systematically discusses the multivariate process to make use of PLS and application of fault detection and diagnosis of the whole train of thought.
The Tennessee Eastman Process is based on the real industrial Process simulation cases. It is by the Tennessee Eastman chemical company process control team J.J.D owns and E.F.Vogel, this case is very suitable for process control technology research, therefore, after the proposed caught the attention of the process control in academia, research is widely used in process control.
Keywords: partial least squares; TE process; process monitoring; fault detection and diagnosis
目录
1 绪论 1
1.1 研究背景 1
1.2 过程监控的研究现状 2
1.3 本文工作 8
2 基本方法介绍 9
2.1引言 9
2.2 主元分析法（PCA） 9
2.3 独立元分析法（ICA） 11
2.4 偏最小二乘算法（PLS） 12
3 过程检测模型 15
3.1 TE简介 15
3.2 TE流程 16
3.3 TE控制目标 17
4 仿真实验 20 偏最小二乘算法在过程监控中的应用研究:http://www.751com.cn/zidonghua/lunwen_3154.html