多线性主元分析的应用与研究  

RESEARCH AND APPLICATION OF MULTILINEAR PRINCIPAL COMPONENT ANALYSIS

作　　者：孔晓光[1] 郭金玉[1] 林爱军 

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《计算机应用与软件》2013年第12期84-86,共3页Computer Applications and Software

基　　金：国家自然科学基金项目(61174119)

摘　　要：为了减少计算量和信息丢失,提出一种运用多线性主元分析(Multilinear PCA)进行间歇过程故障诊断的新方法。首先运用Multilinear PCA直接对间歇过程三维数据进行降维,得到低维的投影向量。然后所有批次向投影向量上投影得到得分向量,计算SPE统计指标控制限,建立Multilinear PCA模型。建立Multilinear PCA模型后,计算新批次的得分向量和SPE(Squared Prediction Error)统计指标,根据统计指标是否超限监视生产过程的运行。最后,在检测出故障之后,采用SPE贡献图诊断故障原因。仿真实例表明:与多向主元分析法MPCA相比(Muhiway Principal Component Analysis)。Multilinear PCA提高了过程性能监视和故障诊断的准确性,较早地发现过程异常。In order to decrease the computation complexity and information loss, in this paper we propose a new method to diagnose the in- termittent process fault with multilinear principal component analysis （ multilinear PCA）. First, we apply the multilinear PCA directlY to the di- mension reduction of three-dimensional data of intermittent process and get the projection vectors with low dimensions. Then all the batches are projected onto the projection vectors to get the scoring vectors, and we calculate the control limits of SPE statistics indexes and build the multi- linear PCA model. Thirdly,we calculate the scoring vectors and SPE statistics indexes of the new batch, and monitor the production operation according to whether the statistical index exceeding the control limit. Finally,we adopt SPE contribution chart to diagnose the fault cause when the faults are detected. Simulation examples show that compared with muhiway principal component analysis （ MPCA）, the muhilinear PCA improves the accuracy of process performance monitoring and fault diagnosis, and finds the abnormal process earlier.

关 键 词：间歇过程故障诊断 多向主元分析 多线性主元分析 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]