Chemical process fault monitoring method based on active learning

Abstract

The invention provides a chemical process fault monitoring method based on active learning, comprising: acquiring attributes of chemical process signals in real time; establishing an initial training signal set and an unlabeled observation signal set according to historical signals of a chemical process; updating the training signal set and the unlabeled observation signal set; establishing a naive Bayes classifier model according to the training signal set and forecasting class labels of unlabeled observation signals; correcting the naive Bayes classifier model by using an active learning method to obtain a final naive Bayes classifier model; and monitoring chemical process faults by using the real-time acquired attributes of the chemical process signals as inputs of the final naive Bayes classifier model. According to the method, a naive Bayes classifier is established on the basis of a finite number of labeled data and a large number of unlabeled observation data, and useful samples in the unlabeled observation signals are searched by using an active learning method to retrain the naive Bayes classifier, so that new fault types can be discovered and the classification precision can be improved.

Description

technical field [0001] The invention relates to the technical field of chemical process fault monitoring, in particular to an active learning-based chemical process fault monitoring method. Background technique [0002] In recent years, due to the development of distributed control systems, a large amount of industrial process data has been effectively collected and stored, and these data contain a large amount of valuable industrial process information, making the data-driven industrial process monitoring method widely used. focus on. In data-driven fault monitoring methods, Naive Bayesian classifier is a commonly used method, but before applying Naive Bayesian classifier to fault monitoring method, there are two potential problems to be solved: 1) Naive Bayesian classifier Yaesian classifiers require labeled normal and faulty data to model, which means that naive Bayesian classifiers cannot discover unknown faults. 2) Naive Bayesian classifiers require a large amount of ...

AI Technical Summary

Problems solved by technology

In industrial practice, it is difficult to collect all fault type data, generally only a limited amount of labeled data and a large number of unlabeled observation data, which limits the application of naive Bayesian classifiers in industrial practice

Images