CAJ | 학술논문

针对化工过程监测数据复杂、非线性等特点，本文将一种新的降维算法-核熵成分分析算法应用到化工过程监控。与其他的多元统计分析方法相比，核熵成分分析算法可以保证数据降维过程中的信息损失最小从而建立更加可靠的统计模型，进而提高故障检测的检出率。与核主成分分析相似，核熵成分分析也是将数据映射到一个高维空间，在高维空间中进行主元分析，不同之处是KECA在选取主元时采用了信息保有量较大的主元，使得数据在降维后的信息损失量更少。本文使用某石化企业的润滑油重质过程的数据测试算法监控效果，核熵成分分析算法的故障检出率为98.2%，比核主成分分析算法(69.706%)要高。实验结果显示，核熵成分分析算法的化工过程监控效果优于核主成分分析算法。
침대화공과정감측수거복잡、비선성등특점，본문장일충신적강유산법-핵적성분분석산법응용도화공과정감공。여기타적다원통계분석방법상비，핵적성분분석산법가이보증수거강유과정중적신식손실최소종이건립경가가고적통계모형，진이제고고장검측적검출솔。여핵주성분분석상사，핵적성분분석야시장수거영사도일개고유공간，재고유공간중진행주원분석，불동지처시KECA재선취주원시채용료신식보유량교대적주원，사득수거재강유후적신식손실량경소。본문사용모석화기업적윤활유중질과정적수거측시산법감공효과，핵적성분분석산법적고장검출솔위98.2%，비핵주성분분석산법(69.706%)요고。실험결과현시，핵적성분분석산법적화공과정감공효과우우핵주성분분석산법。
To handle the complex and nonlinear problem for chemical process monitoring, a new technique based on kernel entropy component analysis is applied. Comparing with other statistical process monitoring method, kernel entropy component analysis minimize information loss during dimension reduction process and construct a more valuable regression model, so its fault detect performance are better. Like kernel principle component analysis, kernel entropy component analysis also mapping data from the input space to a higher dimension feature space, and performing conventional principle component analysis in the feature space. The different is kernel entropy component analysis choose the principle components which contain more information, so there less information loss during dimensionality reduction operation. We use industry data from a lubricating oil process to evaluate performance of algorithm, the fault detection rate of kernel entropy component analysis is 98.2%, higher than kernel principle component analysis (69.7%). Experiment result shows that kernel entropy component analysis has a superior process monitoring performance compared to kernel principle component analysis.