摘要
在化工过程监控领域,Vine Copula模型为描述高维复杂变量之间相依关系提供了一种新的思想,在不降维的基础上直接刻画变量之间复杂的相关关系。传统的Copula函数模型选择方法是基于赤池信息准则(Akaikeinformation criterion,AIC),但是在利用AIC准则时不仅要计算Copula的密度函数,而且边缘分布的拟合效果也直接影响了AIC的取值。本文提出了基于核密度估计的R-Vine Copula (kernel estimation-based R-vine Copula, KRVC)选择方法,并将其应用在化工过程监控领域。通过核密度选择原理得到R-Vine模型,然后利用高密度区域(HDR)与密度分位数表等理论,构建非高斯态广义局部概率指标(GLP)。该方法在TE(TennesseeEastman)过程中以及醋酸脱水过程中的应用验证了KRVC方法在过程监控中的良好性能。
Vine Copula model provides a new approach for describing the interdependence between high-dimensional complex variables during chemical process monitoring, which can directly characterize correlation without dimensional reduction. Traditional Copula function model selection methods are based on AIC (Akaile information criterion). However, the Copula density function needs to be calculated and the fitness of the edge distribution directly affects AIC values. Therefore, a kernel estimation-based R-Vine Copula (KRVC) selection method was proposed and applied in chemical process monitoring. The R-Vine model was obtained by selection criterion based on kernel density estimation. The generalized local probability (GLP) of the non-Gaussian state was constructed using the highest density region (HDR) theory and density quantile table. The monitoring results of the TE (Tennessee Eastman) process shows that the proposed KRVC approach achieves good performance in chemical process fault monitoring.
作者
周南
李绍军
ZHOU Nan;LI Shao-jun(Key Laboratory of Advanced Control and Optimization for Chemical Processes,East China University of Science and Technology,Ministry of Education,Shanghai 200237,China)
出处
《高校化学工程学报》
EI
CAS
CSCD
北大核心
2019年第2期443-452,共10页
Journal of Chemical Engineering of Chinese Universities
基金
国家自然科学基金(21676086)
上海市自然科学基金(14ZR1410500)
