Voltage regulator fault diagnosis method based on self-training semi-supervised generative adversarial network

Abstract

The invention discloses a voltage regulator fault diagnosis method based on a self-training semi-supervised generative adversarial network, and relates to the technical field of fault diagnosis, and the method comprises the steps: carrying out the overlapping sampling and two-dimensional conversion of a one-dimensional pressure signal of a voltage regulator, and obtaining a gray image sample; designing an SGAN model and carrying out initial training; adopting a self-training algorithm to use the trained initial classifier to predict category labels of unlabeled samples, expanding samples meeting requirements to a labeled sample set in a repeated marking mode to re-train the SGAN, and a final classifier is stored; and constructing a voltage regulator fault diagnosis model by using the classifier to carry out online diagnosis. According to the method, a classifier containing generation components is used as an initial classifier of a self-training algorithm, so that the basic classification accuracy performance is improved, the feature extraction capability of semi-supervised fault diagnosis on unlabeled samples is improved, the classification category feature extraction capability of a discriminator on the samples is improved by using a SoftMax function, and efficient and intelligent fault diagnosis is realized.

Description

technical field [0001] The invention relates to the technical field of fault diagnosis, in particular to a voltage regulator fault diagnosis method based on a self-training semi-supervised generation confrontation network. Background technique [0002] The pressure regulator is one of the most important parts in the gas transmission pipeline network. Under actual working conditions, internal components such as membranes, springs, valve port pads, valve barrels, etc. are often subject to wear and tear, resulting in a decrease in service life and failure. , an efficient and intelligent fault diagnosis method is particularly important. [0003] At present, due to the development of equipment and instrument technology in the field of voltage regulator fault diagnosis, a large amount of process data has been recorded and stored, and data-driven fault diagnosis methods have become mainstream, such as support vector machine (SVM), neural network, etc. Although these methods have a...

AI Technical Summary

Problems solved by technology

Although these methods have achieved important results in the fault diagnosis of voltage regulators, most of the current fault diagnosis methods for voltage regulators belong to supervised learning, and the training process requires a large amount of labeled data. In actual situations, labeled samples need to be abundant. Expert experience is a time-consuming and labor-intensive task

Therefore, when a large amount of data is unlabeled data, it is difficult for the supervised learning method to play a role in the fault diagnosis of the voltage regulator

Images