基于ReliefF的主元挑选算法在过程监控中的应用

发布时间：2018-04-12 14:49

本文选题：过程系统 + 过程控制　；参考：《化工学报》2017年04期

【摘要】：传统的主成分分析(principal component analysis,PCA)算法选取包含大部分方差信息的成分作为主元,并将其应用到过程监控中。但是故障信息不一定会投影到方差较大的成分上,使用方差贡献度挑选主元会导致严重的信息丢失和监控效果的恶化。因此使用ReliefF-PCA算法,其中ReliefF算法从故障角度出发,挑选出在区分正常样本和故障样本上权重更高,效果相对更好的成分作为主元。这样挑选出的主元避免了传统PCA算法在主元挑选过程中出现的主观性、盲目性以及重要信息的丢失。ReliefF-PCA算法在过程监控中主要有两个优势,第1,监控效果更好;第2,对原始数据降维效果更好。随后,基于ReliefF-PCA算法,提出一种加权的故障变量贡献图方法。最后,通过Tennessee Eastman(TE)仿真实验测试,ReliefF-PCA算法达到了预期效果。
[Abstract]:The traditional principal component analysis (PCA) algorithm selects components containing most variance information as principal components and applies them to process monitoring.However, the fault information may not be projected to the components with large variance, and the selection of principal components with variance contribution will lead to serious information loss and deterioration of monitoring effect.Therefore, using ReliefF-PCA algorithm, where ReliefF algorithm from the fault point of view, pick out the normal samples and fault samples in the higher weight, the effect is relatively better as the principal component.In this way, the selected principal components avoid the subjectivity, blindness and the loss of important information of the traditional PCA algorithm in the process of principal component selection. There are two main advantages in the process monitoring. The first, the monitoring effect is better;Second, the dimension reduction effect on the original data is better.Then, based on ReliefF-PCA algorithm, a weighted method of fault variable contribution graph is proposed.Finally, the simulation experiment of Tennessee Eastman test shows that the reliefF-PCA algorithm achieves the desired results.
【作者单位】：华东理工大学化工过程先进控制和优化技术教育部重点实验室;
【基金】：国家自然科学基金项目(61374140,61673173) 中央高校基本科研业务费专项资金(222201714031) 中央高校基本科研业务费重点科研基地创新基金(222201717006)~~
【分类号】：TP277

【参考文献】