A Distributed Monitoring and Fault Diagnosis Method for Complex Chemical Production Process

Abstract

The invention discloses a method for distributed monitoring and fault diagnosis of a complex chemical production process. Firstly, the distributed monitoring is implemented by applying the DICA method in the sub-blocks divided based on the flow chart, and then the monitoring results of each sub-block are binarized and combined. The weighted naive Bayesian method CAWNB should be improved to classify fault types. The invention improves the situation of poor monitoring performance caused by too many variables in the system, mutual coupling, autocorrelation and process industry data not satisfying the Gauss assumption; by binarizing the monitoring results of each sub-block and applying the improved The weighted naive Bayesian method CAWNB classifies fault types with high diagnostic accuracy, and can identify fault types in addition to whether abnormalities occur, which improves the performance of distributed fault monitoring and provides reliable reference for operators.

Description

technical field [0001] The invention belongs to the technical field of automatic monitoring, and particularly relates to a distributed monitoring and fault diagnosis method for complex chemical production processes, in which the distributed monitoring is realized by using flow charts to divide sub-blocks and DICA (Dynamic Independent Component Analysis), and further adopting improved simple Bayesian CAWNB (Class-Specific Attribute Weighting Bayes) algorithm to realize the fault diagnosis problem in the chemical production process. Background technique [0002] With the rapid development of complex industrial systems and sensor technology, fault diagnosis is of great significance to ensure that the production process can run safely and with high quality. However, the chemical production process has complex characteristics such as high dimensionality, strong coupling, and autocorrelation, and it is difficult to accurately extract effective features for diagnostic modeling. ...

AI Technical Summary

Problems solved by technology

However, there are two disadvantages of using the PCA method: one is that PCA assumes that the latent variables obey the Gaussian distribution; the other is that the observed variables do not have autocorrelation

However, in applications where attributes such as chemical processes are coupled with each other, the conditional independence assumption of NB is difficult to satisfy, and the classification accuracy will be affected to a certain extent.

Images