基于传递熵的MPCA间歇过程监测方法

传统统计分析方法忽略了变量间作用关系,而传递熵可以有效地表达变量间作用关系,因此提出了一种基于传递熵的MPCA间歇过程监测方法.利用传递熵表达变量间的作用关系,在计算传递熵时采用非参数核密度估计法,利用该方法不依赖于数据先验分布知识的特点来处理非高斯分布的过程数据,通过构建传递熵矩阵,结合滑动窗,实现对间歇过程变量间信息传递的动态表达,最后对传递熵矩阵进行多向主元分析方法(MPCA)建模,实现间歇过程监测.通过青霉素发酵的仿真,结果表明与传统多变量统计过程控制(MSPC)方法作对比,本文监测方法能更及时准确地监测到过程异常.

MPCA Online Monitoring Based on Transfer Entropy for Batch Process

The traditional statistical analysis methods ignore the relations between variables. Transfer entropy could express relations between variables effectively. So this paper proposes an MPCA online monitoring method based on entropy transfer for batch process. The transfer entropy is adopted to describe the complex relations between process variables. The non-parametric kernel density estimation method which does not depend on the prior distribution of data is utilized to calculate transfer entropy to deal with the non-Gauss distribution of the process data. By constructing the transfer entropy matrix combined with the sliding window to achieve the expression of dynamic information transfer between process variables, the MPCA model is then established based on these matrices for detecting faults of batch process. The simulation results show that, compared with the traditional MSPC method, the proposed method can timely identify the faults with better accuracy.