论文部分内容阅读
故障识别在化工过程监控中有着至关重要的作用。准确的故障识别能够帮助操作员及时发现并排除故障,避免生产事故的发生。本文运用传统PCA方法对TE过程中5种典型故障数据进行降维,并将所有故障数据投影到正常工况样本的PCA主元空间中,由正常数据样本计算出T~2统计量的闽值,根据Hotelling T~2统计原理,对所有故障数据进行检测,将检测到的故障样本通过SVM的多分类方法进行故障分类。通过TE过程仿真平台的实验表明,PCA SVM方法与PCA KNN、C SVM方法相比较,算法简单,容易实现,计算速度较快,并且突破了很多文献中只有2类或3类故障识别的局限,同时可以达到较高的多分类准确率。
Fault identification in the chemical process monitoring has a crucial role. Accurate fault identification can help operators to detect and rectify faults in time to avoid production accidents. In this paper, the traditional PCA method is used to reduce the five kinds of typical fault data in TE process, and all the fault data are projected into the PCA principal component space of the normal working samples. The normal value of the T ~ 2 statistic According to the Hotelling T ~ 2 statistical principle, all the fault data are detected, and the fault samples detected are classified by the multi-classification method of the SVM. Experiments on the TE process simulation platform show that compared with the PCA KNN and C SVM methods, the PCA SVM method is simple, easy to implement and fast in calculation, and breaks through the limitation of only two or three types of fault recognition in many literatures. At the same time can achieve a high multi-classification accuracy.