(燕山大学 电气工程学院 生物医学工程系, 河北 秦皇岛 066004)
摘 要:阐述了Radviz(radial visualization)技术,即将高维数据样本非线性的投影到二维目标空间。Vizrank优化能够从数以万计的投影图中评价和确定最好的投影方式;能够快速找到容易被领域专家认可的可视化模型,只需少量输入变量(2~7)就能够做到数据的可视化,并且有很好的分类效果。在TEP仿真系统中的应用,表明了Radviz及其优化的可视化故障诊断方法可以将正常与故障状态有效地分开。该可视化故障诊断方法具有简单而不失精确性、易于利用领域专家知识、诊断结果直观形象并容易理解等显著优点。
关键词:故障诊断; Radviz图; Vizrank; 机器学习; Tennessee Eastman过程
Visualized fault diagnosis method based on Radviz and its optimization
XU Yonghong, HONG Wenxue, CHEN Mingming
(Dept. of Biomedical Engineering, College of Electrical Engineering, Yanshan University, Qinhuangdao Hebei 066004, China)
Abstract:This paper investigated radial visualization (Radviz) technique in which highdimensional data samples were mapped nonlinearly into a 2dimensional target space. It proposed principal innovation was a method called Vizrank, which was able to score and identify the best among possibly millions of candidate projections for visualizations. Vizrank was fast and found visualization models that could be easily examined and interpreted by domain experts. Vizrank was always able to find data visualizations with a small number of (two to seven) features and excellent class separation. This paper shows that diagnostic classes in Tennessee Eastman process, which includes 52 features, can be effectively separated with simple two dimensional plots such as Radviz graph. It demonstrates that this visualized fault detection method has the notable merits of simple and without losing accuracy, easy to utilize domain experts’ knowledge, moreover, the diagnosis results is intuitive and more understandable. ......