论文部分内容阅读
基于主元分析(Principal Component Analysis,PCA)的统计监控模型易受建模数据中离群点的影响;大多工业过程表现出强非线性;且基于PCA的统计性能监控法由于不用过程机理模型的信息从而对故障诊断问题难以在理论上作系统分析,提出基于中心最短距离法CDC(Closest Distance to Center,CDC)/椭球多变量整理法MVT(ellipsoidal Multivariate Trimming,MVT)离群点去除的核主元分析KPCA(Kernel PCA,KPCA)-多支撑向量机MSVMs(Multiple Support Vector Machines,MSVMs)的过程监控方法。该方法首先提出改进尺度的CDC/MVT离群点去除算法以获取正常建模数据;然后利用KPCA来进行故障特征的提取,从而提高非线性统计过程监控的准确性;最后提出MSVMs用来对故障的来源进行分类,以避免求解核主元空间到原始空间的逆映射。将该方法应用到对TE(Tennessee Eastman,TE)过程的监控,表明了所提出方法的有效性,为过程的监控和故障诊断提供了一个新的方法。
Statistical monitoring models based on Principal Component Analysis (PCA) are susceptible to outliers in the modeling data; most industrial processes exhibit strong nonlinearity; and the PCA-based statistical performance monitoring method does not use process model Information to make it difficult to make a systematic analysis on the fault diagnosis problem in theory, and put forward the kernel removing method based on the CDC (Close C Distance to Center) algorithm of the CDS (ellipsoidal multivariate trimming) MVT (MVT) Principal Component Analysis KPCA (KPCA) - Process Monitoring Method for Multiple Support Vector Machines (MSVMs). In this method, we first propose an improved CDC / MVT outlier removal algorithm to get the normal modeling data. Then KPCA is used to extract the fault features to improve the accuracy of the nonlinear statistical process monitoring. Finally, Of the source classification, in order to avoid solving the inverse mapping of the primary space to the original space. The application of this method to the monitoring of TE (Tennessee Eastman, TE) process shows the effectiveness of the proposed method and provides a new method for process monitoring and fault diagnosis.