基于集成KPCA的非线性工业过程状态监测  

NONLINEAR INDUSTRIAL PROCESS CONDITION MONITORING BASED ON INTEGRATED KPCA

作　　者：郑丹 陈路 童楚东[1] Zheng Dan;Chen Lu;Tong Chudong(Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,Zhejiang,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机应用与软件》2023年第6期16-22,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61773225);浙江省自然科学基金项目(LY20F030004)。

摘　　要：传统的KPCA(Kernel Principal Component Analysis)过程监测方法一般根据经验选取需要的核函数及一定宽度的参数,这样做是非常盲目的。同时单一KPCA模型不能对所有故障都有好的监测效果。为了解决此问题,提出基于集成KPCA的非线性工业过程状态监测方法。通过选取一系列的核函数及其参数构建不同的KPCA模型得到子模型,用贝叶斯方法将众多子模型的监测统计量转化为故障概率,分两步进行融合,得到最终监测结果。实验结果表明,该方法显著地提高了监测性能,同时减小核函数及参数选取对故障监测的影响。The traditional kernel principal component analysis(KPCA)process monitoring method usually selects necessary kernel functions and parameters with a certain width based on experience,which is obviously very blind.Besides,only a single KPCA model cannot have a proper monitoring result for all failures.To solve this problem,a state-monitoring method of the nonlinear industrial process based on integrated KPCA is proposed.Different KPCA models were created by selecting a series of kernel functions and parameters to obtain sub-models,and the Bayesian method was used to transform monitoring statistics of the sub-models into failure probability.The integration was performed in two steps,so as to obtain the final monitoring result.The experimental results show that this method can significantly improve the monitoring performance and reduces the effect of kernel function and parameter selection on failure monitoring.

关 键 词：核主成分分析 集成学习 贝叶斯融合 故障监测 田纳西-伊斯曼过程 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术] TP277