化工学报 ›› 2016, Vol. 67 ›› Issue (3): 925-930.DOI: 10.11949/j.issn.0438-1157.20151963

• 研究论文 • 上一篇    下一篇

基于聚类选择k近邻的LLE算法及故障检测

薄翠梅, 韩晓春, 易辉, 李俊   

  1. 南京工业大学电气工程与控制科学学院, 江苏 南京 211816
  • 收稿日期:2015-12-24 修回日期:2016-01-06 出版日期:2016-03-05 发布日期:2016-01-12
  • 通讯作者: 薄翠梅
  • 基金资助:

    国家自然科学基金项目(61203020,61503181);江苏省自然科学基金项目(BK20141461,BK20140953)。

Neighborhood selection of LLE based on cluster for fault detection

BO Cuimei, HAN Xiaochun, YI Hui, LI Jun   

  1. College of Electrical Engineering and Control Sciences, Nanjing Tech University, Nanjing 211816, Jiangsu, China
  • Received:2015-12-24 Revised:2016-01-06 Online:2016-03-05 Published:2016-01-12
  • Contact: 67
  • Supported by:

    supported by the National Natural Science Foundation of China (61203020, 61503181) and the Natural Science Foundation of Jiangsu Province (BK20141461, BK20140953).

摘要:

针对化工过程在多种运行模式下多种流形结构具有不同最优近邻数问题,提出了基于聚类选择k近邻的局部线性嵌入(LLE)过程监控方法。使用LLE算法提取高维数据的低维子流形,通过局部线性回归得到高维数据空间到特征空间的映射矩阵;选择Silhouette指标作为聚类有效性指标评估嵌入空间样本信息的相似性,进而确定最优近邻数,根据映射矩阵构建故障监控统计量及其控制限,进行故障检测。最后将所提算法与其他经典算法应用于TE化工过程对比分析,验证了算法的有效性。

关键词: 局部线性嵌入, 最近邻数, 子流形, 故障检测, 聚类指标

Abstract:

In the process of chemical engineering, multiple manifold structures has different optimal number of nearest neighborhood under various operating modes. Locally linear embedding (LLE) algorithm based on clustering to select the nearest neighborhood is proposed for nonlinear monitoring. LLE algorithm was performed for dimensionality reduction and extract the available information in high-dimensional data. The mapping matrix from data space to feature space was obtained by local linear regression. The Silhouette index was selected as the clustering validity index to estimate the similarity between the embedded sample information, and further determine the optimal number of neighbors. Process monitoring statistics and its control limits were built based on the mapping matrix. Finally, the feasibility and efficiency of the proposed method were illustrated through the Tennessee Eastman process.

Key words: locally linear embedding, the number of nearest neighbor, sub-manifold, fault detection, clustering index

中图分类号: