核熵成分分析算法及其在化工过程监控中的应用研究  

Chemical process monitoring using kernel entropy component

作　　者：康德礼[1] 李秀喜[1] 

机构地区：[1]华南理工大学化学与化工学院,广东广州510640

出　　处：《计算机与应用化学》2013年第8期855-858,共4页Computers and Applied Chemistry

基　　金：国家科技支撑计划(2012BAK13B02);国家自然科学基金(21176089

摘　　要：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一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。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.

关 键 词：过程监控 化工过程 核熵成分分析 

分 类 号：TQ015.9[化学工程] TP391.9[自动化与计算机技术—计算机应用技术]