Industrial process fault method based on robust semi-supervised discriminant analysis

Abstract

The invention relates to an industrial process fault classification method based on robust semi-supervised discriminant analysis. In the off-line modeling stage, firstly, historical training samples are randomly marked, and all the historical training samples are standardized by means of the mean value and the standard deviation of the marked historical training samples under the normal working condition, so that the influence of different dimensions on modeling is eliminated. Furthermore, for each known working condition, a sample identification criterion based on a deviation degree threshold value is established by using the marked historical training sample, and the historical training sample from an unknown fault category is identified. A fault classification model based on semi-supervised discriminant analysis is established in combination with marked historical training sample information and unmarked historical training sample information derived from a known fault category. In the online use stage of the model, the established sample recognition criterion is utilized to recognize online samples from unknown fault categories, and the robustness of an existing semi-supervised discriminant analysis method in an unknown fault scene can be remarkably improved.

Description

technical field [0001] The invention relates to a method for classifying industrial process faults, in particular to a method for classifying industrial process faults based on robust semi-supervised discriminant analysis. Background technique [0002] Modern industrial processes are increasingly large-scale and complex, resulting in a significant increase in the possibility of faults; when faults propagate in large-scale and complex industrial processes, it will lead to low product quality, high production energy consumption, equipment damage, personnel Casualties, environmental pollution and other serious consequences. For example, on July 19, 2019, the air separation unit of the Henan Gas Group Yima Gasification Plant in Sanmenxia City, Henan Province experienced a "sand explosion" due to a liquid leakage failure in the cold box, which eventually led to an explosion, killing 15 people and seriously injuring 16. . On September 8, 2020, the flue gas desulfurization fan of...

AI Technical Summary

Problems solved by technology

[0002] Modern industrial processes are increasingly large-scale and complex, resulting in a significant increase in the possibility of faults; when faults propagate in large-scale and complex industrial processes, it will lead to low product quality, high production energy consumption, equipment damage, personnel Casualties, environmental pollution and other serious consequences

For example, on July 19, 2019, the air separation unit of the Henan Gas Group Yima Gasification Plant in Sanmenxia City, Henan Province experienced a "sand explosion" due to a liquid leakage failure in the cold box, which eventually led to an explosion, killing 15 people and seriously injuring 16.

On September 8, 2020, the flue gas desulfurization fan of Xiangfen Hongyuan Coking Co., Ltd., Xiangfen County, Linfen City, Shanxi Province suddenly shut down, causing the flue to burst and killing 2 people

[0006] However, for existing semi-supervised discriminant analysis-based fault classification methods for industrial processes, see figure 1 , there are still the following problems that have not been effectively resolved:

[0007] (1) When using unlabeled historical training sample information, unlabeled historical training samples from unknown fault categories cannot be identified; these samples are essentially equivalent to outlier (or noise) data, which will seriously affect the model classification performance, see figure 2

[0008] (2) The established fault classification model cannot identify online samples from unknown fault categories, resulting in poor practicability of the model and urgent improvement in classification performance

Images