一种基于多向局部线性嵌入的故障检测方法  

Fault Detection Method Based on Multiway Locally Linear Embedding

作　　者：郭金玉[1] 袁堂明[1] 李元[1] 

机构地区：[1]沈阳化工大学信息工程学院,沈阳110142

出　　处：《小型微型计算机系统》2015年第9期2149-2152,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61034006;61174119)资助;辽宁省博士启动资金项目(20131089)资助

摘　　要：为了改善传统算法因工况改变导致的故障误诊断,在保持数据结构的基础上,降低高维数据的维度,提出一种基于多向局部线性嵌入(MLLE)的故障检测方法.根据重构误差最小的原则,确定样本局部最优权值矩阵,通过提取相关矩阵的特征向量将样本嵌入到低维的数据中.在低维的数据中运用主元分析进行建模.对新的样本数据,在原始数据中找到k个近邻样本并确定相应的权值.在局部的低维数据中重构低维空间的新样本,计算其统计量SPE和T2.根据统计量是否超过控制限进行故障检测.将该方法应用到青霉素发酵过程的故障检测中,并与传统算法进行对比.仿真结果表明,该方法能够有效地识别正常工况改变与过程故障引起的统计量变化,误报率明显下降.In order to improve the misdiagnosis of fault caused by conditions change,reduce the dimension of high-dimensional data based on preserving the data structure,a fault detection method based on multiw ay locally linear embedding（ M LLE） w as proposed.The locally optimum w eight matrix w as determined based on the principle of the minimum reconstruction error. The samples w ere embedded in low-dimensional data by extracting the eigenvectors of the correlation matrix. Principal component analysis w as used to the low dimensional data for model building. For a new sample,w e find the k-nearest neighbor in the original sample data and determine the appropriate w eights. The low-dimensional new data w as reconstructed in a local low-dimensional space,and the tw o statistics index,that is,SPE and T2 w ere calculated. The faults w ere detected according to the statistics index exceeded the control limit. The method is applied to the fault detection of penicillin fermentation process in comparison w ith traditional algorithms. Simulation results show that this method can effectively identify the changes in statistics index caused by normal operating conditions change and process fault,so that the false alarm rate decreases obviously.

关 键 词：批次过程 故障检测 多向主元分析 局部线性嵌入 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]