可变窗自适应核主元分析的化工过程故障诊断算法  

Fault diagnosis algorithm of chemical industry process based on variable window adaptive KPCA

作　　者：赵小强[1] 杨武[1] 薛永飞[1] 

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050

出　　处：《东南大学学报（自然科学版）》2013年第A01期93-97,共5页Journal of Southeast University：Natural Science Edition

基　　金：甘肃省自然科学基金资助项目(1112RJZA028);甘肃省教育厅硕士生导师资助项目(1003ZTC085)

摘　　要：针对核主元分析法(KPCA)处理非线性时变化工过程故障诊断易产生误报的问题,提出一种基于可变窗技术的快速自适应核主元分析(AKPCA)故障诊断算法.该算法在过程监控中当大量的数据块来临时,分别对KPCA模型进行迭代更新和复旧处理,通过计算SPE和T2统计量来调节滑动窗的尺寸,从而实现KPCA模型的进一步更新,克服了传统的自适应KPCA方法只能够处理某个时刻的一个观察点数据的缺点,能够有效地排除异常样本影响,从而提高时变过程性能检测的准确性.将该方法应用于酮苯脱蜡过程的监控中.与KPCA、滑动窗KPCA(MWKP-CA)的监测性能进行比较,仿真结果表明该方法能够很好地降低过程故障的误报率,对非线性时变的化工过程具有更可靠的检测性能.As the kernel principal component analysis（KPCA） method is easy to generate false alarm for nonlinear and time-varying chemical process,a variable w indow fast adaptive kernel principal component analysis algorithm is proposed.When large amounts of data blocks are provided,the proposed algorithm first updates and dow ndates the KPCA model separately,then adjusts the size of the moving w indow by calculating SPE and T2 statistics to further update the KPCA model.This algorithm overcomes the w eakness of the traditonal adaptive KPCA that it can only handle the data of one observation at one moment,and it can effectively eliminate the impact of abnormal samples,thus improving the accuracy of fault detection.The algorithm is applied to monitor the ketone-benzol dew axing process.Compared w ith the KPCA and the moving w indow KPCA（MWKPCA） methods,simulation results show that the proposed method can decrease the false alarm rate and has more reliable performance for nonlinear and time-varying chemical process.

关 键 词：故障诊断 核主元分析 可变窗 酮苯脱蜡 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]