优化动态核主元分析的工业过程故障监测方法  

Industrial Process Fault Monitoring Method Based on Optimized Dynamic Kernel Principal Component Analysis

作　　者：杨芳 王亚君[1] 沈亚慧 YANG Fang;WANG Yajun;SHEN Yahui(College of Electronic and Information Engineering,Liaoning University of Technology,Jinzhou 121001,Liaoning,China)

机构地区：[1]辽宁工业大学电子与信息工程学院,辽宁锦州121001

出　　处：《实验室研究与探索》2024年第1期6-10,19,共6页Research and Exploration In Laboratory

基　　金：国家自然科学基金项目(61503169,61802161);辽宁省自然科学基金项目(2020-MS-291)。

摘　　要：针对现代生产工业过程中数据的非线性多模态特征,提出了一种基于人工大猩猩部队优化动态核主元分析(GTO-DKPCA)的故障监测方法。利用自回归移动平均时间序列模型和核主成分分析(KPCA)方法构建DKPCA模型,对过程各阶段的批次数据进行DKPCA处理。通过正常数据和故障数据特征构建自适应度函数,利用人工大猩猩部队优化算法对DKPCA核参数进行优化,以发现最优的非线性特征;通过计算各时间点的霍特林统计量T2和平方预测误差(SPE)统计量进行故障监测。青霉素发酵过程故障监测结果表明,GTO-DKPCA方法比多向核主元分析(MKPCA)和多动态核主元分析(BDKPCA)有更好的监测效果,适应性和准确性更高。A fault monitoring method based on artificial gorilla force optimization algorithm optimized dynamic kernel principal component analysis(GTO-DKPCA)is proposed for nonlinear multimodal data in modern industrial production processes.Using the autoregressive moving average time series model and kernel principal component analysis(KPCA)method,a DKPCA model is constructed to process batch data at each stage of the process.We construct an adaptive degree function based on normal and fault data features,and use the artificial gorilla force optimization algorithm to optimize the DKPCA kernel parameters to discover the optimal nonlinear features.Fault monitoring is carried out by calculating the Hotelling statistic T2 and the squared prediction error SPE statistic at each time point.The results of penicillin fermentation process indicate that the GTO-DKPCA method has better monitoring effect,adaptability,and accuracy than multi directional kernel principal component analysis(MKPCA)and multi dynamic kernel principal component analysis(BDKPCA).

关 键 词：动态核主元分析 人工大猩猩部队优化算法 故障监测 青霉素发酵 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]