Power transformer winding fault diagnosis method based on GSMallat-NIN-CNN network

Abstract

The invention discloses a power transformer winding fault diagnosis method based on a GSMallat-NIN-CNN network, and the method comprises the steps: carrying out the measurement of a vibration condition of a transformer winding through a multi-channel sensor, and obtaining multi-source vibration data of a transformer; converting the multi-source vibration data obtained by measurement into grayscaleimages by using GST gray level transformation; decomposing each grayscale image into a high-frequency component sub-image and a low-frequency component sub-image layer by layer by adopting a Mallat algorithm, and performing image fusion on a high-frequency component sub-image and a low-frequency component sub-image; reconstructing the fused grayscale images, and encoding the vibration grayscale images according to the fault state of the transformer winding; establishing a transformer fault diagnosis model based on a GSMallat-NIN-CNN network; randomly initializing network parameters, dividinga training set and a test set, and training and optimizing the network through the training set; and storing the trained network, and testing the network through the test set. According to the method,t noise in a multi-source signal is effectively suppressed, the integrity of feature information is improved, the calculated amount is reduced, and the fault diagnosis accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of fault diagnosis of power transformer windings, in particular to a method for diagnosing faults of power transformer windings based on a GSMallat-NIN-CNN network. Background technique [0002] As an indispensable part of the power grid, the power transformer's working status is closely related to the safe and reliable operation of the power system. Therefore, whether the hidden fault of the transformer can be quickly detected and dealt with in time is the focus of the power industry. [0003] Domestic statistics show that winding faults are the main faults of transformers, accounting for 70% of the total faults. The traditional transformer fault diagnosis methods are all offline detection methods, the test data is not real-time, and the fault location and category cannot be effectively determined. The vibration method collects the surface vibration signal of the transformer oil tank in real time, analyzes...

AI Technical Summary

Problems solved by technology

Traditional transformer fault diagnosis methods are all offline detection methods, the test data is not real-time, and the fault location and category cannot be effectively determined

The vibration method collects the surface vibration signal of the transformer oil tank in real time, and analyzes the change of the winding force to judge the operating state of the transformer. The commonly used extraction methods include Fourier transform (FFT), wavelet transform (WT) and Hilbert-Huang transform (HHT). results, but limitations

Two-dimensional image recognition is an emerging fault diagnosis method in recent years. This method uses the spatial scale invariance to perform grayscale transformation through polar coordinates, converts one-dimensional vibration signals into two-dimensional grayscale images, and reduces the amount of calculation. However, related literature The noise processing link of the transformer is relatively weak, which reduces the diagnostic accuracy of the transformer

Images