Abstract:

Traditional data-driven process monitoring methods mainly ignore the consideration of process mechanism. The predictive residuals-based process monitoring methods realize the approximation of the local process mechanism by a data-driven regression model, and use the residual of the regression model as a feature to better monitor the abnormal deviation of the process. But their approximation of the local process mechanism ignores the specific process information. As a form of process information, process topology can be used to extract the process and causal relationship between variables. Based on this, a statistical process monitoring method based on process topology information is proposed in this paper. Considering specific process information, a modified index for variable selection is proposed by extracting the process and causal relationship between variables from the flowchart. On this basis, a process monitoring model based on prediction residuals is established to monitor the process deviation. The proposed method combines specific process information with a data-driven regression model to achieve a more accurate approximation of the local process mechanism. The proposed method is applied to a continuous reforming unit, the result shows that the method proposed in this paper can detect fault earlier compared to process monitoring method based on prediction residual or PCA (principal component analysis) model.

Key words: process systems, principal component analysis (PCA), partial least square (PLS), residual, early fault identification

摘要：

传统数据驱动的过程监测方法主要基于历史数据和统计学知识建立，往往忽视了对过程机理的考虑。基于预测残差的过程监测方法则通过数据驱动的回归模型实现对局部过程机理的近似，在预测残差的基础上建立监测模型实现了对过程偏离更好的识别。但其建立回归模型实现对局部过程机理的近似时主要基于数据，很少考虑具体流程信息。作为流程信息的一种表现形式，流程拓扑结构常被用来提取变量间的进程与因果关系，如果在建立回归模型时结合流程的拓扑结构，则可使得所建立的回归模型中包含一定的流程信息，使其对局部机理的近似更为准确。基于此，本文提出一种基于流程拓扑信息的统计过程监测方法。该方法利用流程的拓扑结构，提取变量间的进程与因果关系，建立回归模型实现对局部过程机理的近似。在此基础上建立基于预测残差的过程监测模型，实现对过程偏离的监测。该方法被应用于某连续重整装置的过程监测中，结果表明其监测效果要优于基于主元分析和基于预测残差的过程监测方法。

关键词: 过程系统, 主元分析, 偏最小二乘法, 残差, 故障早期识别