引用本文:李锦冰,韩冰,冯守渤,张佳冬,李宇,钟凯,韩敏.基于分块核主成分分析和支持向量机的故障检测[J].控制理论与应用,2020,37(4):847~854.[点击复制]
LI Jin-bing,HAN Bing,FENG Shou-bo,ZHANG Jia-dong,LI Yu,ZHONG Kai,HAN Min.Fault detection based on block kernel principal component analysis and support vector machine[J].Control Theory and Technology,2020,37(4):847~854.[点击复制]
基于分块核主成分分析和支持向量机的故障检测
Fault detection based on block kernel principal component analysis and support vector machine
摘要点击 1546  全文点击 686  投稿时间：2018-11-24  修订日期：2019-05-29
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2019.80923
  2020,37(4):847-854
中文关键词  故障检测  分块核主成分分析  最小二乘支持向量机  特征提取
英文关键词  fault detection  block kernel principal component analysis  least square support vector machine  feature extraction
基金项目  国家自然科学基金项目（61773087）
作者单位E-mail
李锦冰 大连理工大学电子信息与电气工程学部 ljs@mail.dlut.edu.cn 
韩冰 上海船舶科学运输研究所  
冯守渤 大连理工大学电子信息与电气工程学部  
张佳冬 大连理工大学电子信息与电气工程学部  
李宇 大连理工大学电子信息与电气工程学部  
钟凯 大连理工大学电子信息与电气工程学部  
韩敏* 大连理工大学电子信息与电气工程学部 minhan@dlut.edu.cn 
中文摘要
      针对工业系统监测数据为非线性, 且难以辨识复杂工作过程中故障位置的问题, 提出一种基于分块核主成分分析(BKPCA) 和最小二乘支持向量机(LS-SVM) 的集成故障检测方法. 首先对系统监测变量进行分块, 使用KPCA 对每个分块在特征空间中建立T2 和平方预测误差(SPE) 统计量来实时监测系统健康状态, 并使用LS-SVM 对上述过程检测出来的故障数据进行再次判断. 随后计算出现故障后计算每一分块的故障贡献率, 进而确定发生故障的分块. 由于采用了并行分块算法, 可以较简单的确定故障发生位置, 提高的计算效率, 同时LS-SVM 方法的应用也可以提升故障检测的精度. 使用田纳西-伊斯曼化工(TE) 过程数据对本文所提方法进行仿真验证, 试验结果表明所提方法取得了较好效果.
英文摘要
      The measurement data of industrial system is nonlinear and difficult to extract the characteristic information. In the complex large-scale industrial process, an integrated fault detection method based on the block kernel principal component analysis (BKPCA) and least squares support vector machine (LS-SVM) is proposed．Firstly, the measurement variables are partitioned. And KPCA is used to establish the T2 and squared prediction error (SPE) monitoring statistics in the feature space for each block to monitor the health status in real time. The LS-SVM is used to rejudge the faulty data detected by above process. Calculating the contribution rate of each block after the fault occurs, and then the faulty block can be determined. Due to the parallel block algorithm, the location of the fault can be simply found, and the computational efficiency is improved. What is more, the application of LS-SVM can also improve the accuracy of fault detection. The Tennessee-Eastman (TE) process data is used to verify the method proposed in this paper. The results show the effectiveness of proposed method.