基于核独立元分析的非线性过程故障检测方法研究

摘要：独立元分析（Independent component analysis,ICA）是盲源分离(Blind source separation, BSS)方法的一种，是能够从多个源信号的线性混合中分离出源信号的技术,是一种基于信号的高阶统计特征的算法。但是ICA在对于处理强非线性和非高斯性特征的变化信号上有一定的局限性,针对同一非线性的过程数据，使用核独立元分析（Kernel Independent component analysis ，KICA）方法能够很好地解决这个问题。核独立元分析通过规范相关性将比较函数通过核函数扩展到一个再生的核希尔伯特空间, 然后再在这个空间对比较函数进行计算。随着科技的快速发展，在现代工业生产控制系统日益复杂的趋势下，核独立元分析(KICA)相比独立元分析(ICA) 是更为有效的用于提取特征的方法，可以更好地判断故障的发生。本文分别使用了独立元算法与核独立元算法通过对工业流程中正常工作下的测量数据进行处理后建立故障检测模型,通过监控所设定的检测统计量有无超出控制限的置信区间来对故障进行检测。再经过田纳西伊斯曼过程(Tennessee Eastman)的数据进行仿真对比后，验证了核独立元算法更具有效性。 69489

毕业论文关键词：核独立元分析;非线性过程;TE过程;故障检测

Research on Nonlinear Process Fault Detection Method Based on Kernel Independent Element Analysis

Abstract: Independent component analysis is a kind of blind source separation method, which is a technique capable of separating the source signal from the linear mixing of multiple source signals. It is an algorithm based on signal high order statistical feature. However, Independent component analysis has some limitations on the change signal for dealing with strongly nonlinear and non-Gaussian features. The kernel independent component analysis method can solve this problem well for the same nonlinear process data. The kernel independent component analysis extends the comparison function through a kernel function to a regenerated kernel Hilbert space by normative correlation, and then computes the comparison function in this space. With the rapid development of science and technology, under the increasingly complex trend of modern industrial production control system, the kernel independent component analysis is more effective than the independent component analysis method for extracting features, and can judge the occurrence of the fault better. In this paper, we use the independent element algorithm and the kernel independent element algorithm to establish the fault detection model by processing the measurement data under the normal working process in the industrial process. By monitoring the set of the measured statistics, the confidence interval beyond the control limit is Fault detection. After comparing the data of Tennessee Eastman process, it is verified that the kernel independent element algorithm is more effective.

Key Words: kernel independent component analysis; nonlinear process; Tennessee Eastman process; fault detection; fault diagnosis

1 引言 1

1.1 研究背景 1

1.2 研究意义 2

1.3 研究内容 3

1.4 全文安排 4

2 基于数据驱动的性能监控与故障诊断算法 5

2.1 主元分析 5

2.2偏最小二乘法 8

2.3 独立元分析 10

2.4 对比分析 16

3 核独立元分析 17

3.1 KICA算法基于核独立元分析的非线性过程故障检测方法研究:http://www.youerw.com/jisuanji/lunwen_78406.html