Dynamic industrial process fault diagnosis method based on GRU depth neural network

Abstract

The invention discloses a dynamic industrial process fault diagnosis method based on a GRU depth neural network. The method divides original data into a plurality of sequence units as the input of theGRU, a GRU network is established through a batch normalization algorithm, extract dynamic characteristics can be effectively extracted from the sequence units, by adopting a softmax regression method, faults are classified according to the dynamic characteristics extracted by the GRU, and the probability interpretation of the classification is provided, so that the diagnosis result is further accurate, the problem of dimension disaster is avoided, and the accuracy of the fault diagnosis of the dynamic industrial process is improved.

Description

technical field [0001] The invention relates to a dynamic industrial process fault diagnosis method based on a GRU deep neural network. Background technique [0002] With the development of modern industrial technology and process control mechanisms, industrial processes have become more and more complex. To improve industrial process safety and product quality, industrial process monitoring and fault diagnosis have received extensive attention in the past decades. Data-driven multivariate statistical process monitoring (MSPM) has been widely used in the monitoring of industrial process operations and production results. Compared with knowledge-based and model-based methods, MSPM methods are easier to set up. Therefore, MSPM models such as Principal Component Analysis (PCA), Independent Component Analysis (ICA) are widely used in industrial process monitoring and fault diagnosis. The framework of traditional fault diagnosis mainly includes two steps: 1) feature extraction...

AI Technical Summary

Problems solved by technology

However, these vector-based augmentation methods may exacerbate the "curse of dimensionality" (a phenomenon in which the computation increases exponentially with the increase of dimensionality in problems involving vector computations) and destabilize feature extraction methods; moreover , the augmented structure is pre-fixed, i.e. the adoption of dynamic information is not adaptively learned from raw process data

Images