Chemical fault diagnosis method based on unbalanced correction convolutional neural network

Abstract

The invention provides a chemical fault diagnosis method based on an unbalanced correction convolutional neural network, and the method comprises the following steps: S1, carrying out the data preprocessing of a TE process; S2, synthesizing a sample; S3, carrying out data dimension reduction; and S4, constructing a CNN incremental learning network. The method has the beneficial effects that the proposed II-CNN framework can be used for synthesizing unbalanced data, and the importance of boundary samples is considered, so that the synthesized samples are more representative; on the basis, dimension reduction is carried out on the data, and the complex learning process is simplified; and finally, for the arrival of a new fault type, updating the structure and parameters of the CNN network by adopting incremental learning. The method is superior to an existing static model method, and has remarkable robustness and reliability in chemical fault diagnosis.

Description

Technical field [0001] The present invention belongs to the field of chemical industry, and more particularly to an unbalanced correction convolutional neurological network increment method for chemical fault diagnosis. Background technique [0002] Chemical process troubleshooting is one of the most important procedures in process control systems, which is critical to ensuring the successful operation of the chemical process and improving the safety of chemical processes. The fault diagnosis model is designed to detect the abnormal state of the production process, find the root cause of the fault, assist in making reliable decisions, and excludes the system failure. The fault diagnosis model can convert historical data into process information according to data collected from multiple sensors, and determine whether a fault has occurred, thereby ensuring the safety, efficiency and economy of complex chemical processes. [0003] The current intelligent fault diagnosis method based...

AI Technical Summary

Problems solved by technology

However, most of these methods have the following flaws: 1) They assume that the data samples under different failure modes are balanced or equal, but this assumption is not always applicable to real chemical processes, and data imbalance will cause the classifier to fail to learn To complete category knowledge, reduce the classification accuracy of the classifier, because the data imbalance will cause the classifier to pay less attention to a few faults; 2) As the production progresses in the actual industrial process, one or several New fault types, with the arrival of new fault categories, these models require a complete retraining process

Images