基于RISOMAP的非线性过程故障检测方法  

Non-linear process fault detection method based on RISOMAP

作　　者：张妮[1] 田学民[1] 蔡连芳 

机构地区：[1]中国石油大学(华东)信息与控制工程学院,山东青岛266580

出　　处：《化工学报》2013年第6期2125-2130,共6页CIESC Journal

基　　金：国家自然科学基金项目(61273160);山东省自然科学基金项目(ZR2011FM014);中央高校基本科研业务费专项资金(10CX04046A);山东省博士基金项目(BS2012ZZ011)~~

摘　　要：化工过程监控数据存在非线性特点,且过程常常运行于多个模态,针对该类问题,提出基于相对等距离映射(relative isometric mapping,RISOMAP)的过程故障检测方法,该方法采用相对测地距离构造高维空间的距离关系阵,运用多维尺度变换(MDS)计算其低维嵌入输出,从高维数据中提取子流形信息和残差信息分别构造监控统计量进行故障检测,同时运用核ridge回归在线计算测试数据的低维输出,核矩阵通过综合相似度进行更新。数值算例和TE过程的仿真结果表明,RISOMAP方法可以更为有效地实施故障检测,故障检测的灵敏度较高,同时也为基于流形学习的多模态过程故障检测的实施提供了一条思路。Industrial processes are often operating under different modes, while there are nonlinear correlations between data monitored. Aiming at these problems, a fault detection method based on relative isometric mapping （RISOMAP） was proposed. Relative geodesic distance was used to establish distance matrix in the high dimensional space, and multi dimensional scaling （MDS） was used to calculate output in the low dimensional embedded space. Information of sub-manifold and error could be obtained, and then monitoring statistics were built for fault detection. Meanwhile, kernel ridge regression was used to obtain the lower dimensional output of test data. Besides, kernel matrix was updated through integrated similarity. The simulations of visualization case and TE process illustrated that in contrast to fault detection methods based on kernel principal component analysis （KPCA） and ISOMAP, the proposed method could detect process fault more effectively and quickly. It also provided an idea to implement fault detection without prior knowledge in the multimode process.

关 键 词：相对测地距离 子流形 核ridge回归 故障检测 非线性过程 多模态过程 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]