信息增强随机邻域嵌入在故障检测中的应用  

Information-enhanced Stochastic Neighbor Embedding for Fault Detection

作　　者：卢春红[1] 王杰华[1] LU Chun-hong;WANG Jie-hua(School of Information Science and Technology,Nantong University,Nantong 226019,China)

机构地区：[1]南通大学信息科学技术学院,江苏南通226019

出　　处：《控制工程》2021年第5期938-943,共6页Control Engineering of China

基　　金：国家自然科学基金资助项目(62003179);江苏省高校自然科学研究面上项目(17KJB530008)。

摘　　要：通过使高维空间的数据相似度按概率分布,在邻域嵌入后仍然保持相同的概率分布来实现非线性降维,随机邻域嵌入方法已成功应用于过程监测。然而,这种方法仅仅关注了数据成对样本之间的局部相似关系,忽视了远距离样本之间的非近邻关系。针对这个问题,提出了信息增强的随机邻域嵌入方法,利用基于几何测距的随机邻域概率分布度量样本之间的局部相似关系,同时关注了远距离样本之间的非局部结构关系,实现了非线性过程的特征提取。提出的信息增强随机邻域嵌入方法与传统的随机邻域嵌入方法相比,保持了更完整的结构信息,能更灵活地表征过程系统的状态,更具竞争力。所提方法在TE化工过程中进行仿真验证,并与现有的几种方法进行对比,结果表明了该方法的优越性。Stochastic neighbor embedding is a nonlinear dimensionality reduction method,which makes the data similarity in the high-dimensional space according to the probability distribution and retains the same probability distribution after the neighborhood embedding.It has been successfully applied in process monitoring.However,it only considers local similarity relationship between pairwise samples of data,and neglects non-neighbor relationship between samples faraway.To solve this problem,information-enhanced stochastic neighbor embedding is proposed in this paper.It uses stochastic neighbor probability distribution based on geometric distance to measure local similarity relationship between samples,learns nonlocal structure relationship between samples faraway,and realizes feature extraction of nonlinear process.Compared with the traditional stochastic neighbor embedding,the proposed method preserves more complete structure information,represents current status of process systems more flexibly,and is more competitive.The proposed method is simulated and validated in the Tennessee Eastman chemical process and compared with several existing methods.The results indicate its superior performance.

关 键 词：过程监测 信息增强 随机邻域嵌入 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]