Converter transformer multi-scale fusion feature extraction method based on vibration signals

Abstract

The invention relates to a converter transformer multi-scale fusion feature extraction method based on vibration signals, and belongs to the field of electric power. The method comprises an image generation module and a multi-scale information fusion module. Firstly, a time domain characteristic spectrum and a frequency domain characteristic spectrum are obtained by calculating a Markov transition field (MTF) matrix of a vibration signal sequence and a refined frequency sequence respectively, then a time-frequency energy characteristic spectrum of a signal is obtained through continuous wavelet transform, and then multi-scale characteristic spectrums are fused through a channel attention module and a space attention module. The problems that a traditional method is low in information utilization rate and a deep learning method is difficult to build can be solved.

Description

technical field [0001] The invention belongs to the field of electric power and relates to a multi-scale fusion feature extraction method of a converter transformer based on vibration signals. Background technique [0002] The vibration generation principle of the converter transformer and the power transformer is basically the same. The vibration mainly comes from the magnetostrictive effect of the iron core and the forced vibration of the winding. In addition, vibration is also affected by many factors such as voltage and current levels, power factor, and three-phase unbalanced conditions. Therefore, the vibration signal collected from the box contains a large amount of effective information that can reflect the state of the converter rheological machinery. In recent years, the evaluation method of transformer operating state based on vibration signal has been widely concerned at home and abroad, but the state identification method for converter vibration signal is rarely...

AI Technical Summary

Problems solved by technology

The traditional method is good for feature extraction of simple vibration signals, but there are still the following problems: 1) The effect of processing high complexity, non-periodic and non-stationary signals is not good; 2) The observation angle is single, and the resolution of time domain and frequency domain cannot be balanced. There is a loss of information;

[0004] 3) The calculation is cumbersome and depends on certain prior knowledge, which is not conducive to big data processing

[0006] To sum up, the traditional method has the problems of high complexity, large amount of data, and low information utilization rate, while the deep learning method based on vibration signals is difficult to build, the training efficiency is low, and it is impossible to accurately identify the operating state of the converter.

Images