Research on Feature Extraction Method of Converter Transformer Vibration Signal Based on Markov Transition Field

Abstract

In view of the high complexity of the vibration signal of the converter transformer and the large amount of data, the construction of the feature extraction model of the converter transformer based on the vibration signal is difficult and the accuracy is not high. This paper proposes a feature extraction model of commutation vibratiob signals based on Markov transition field and residual convolutional neural network. This paper first divides the interval according to the signal amplitude and calculates its Markov transition field matrix, and then obtains the two-dimensional representation of the one-dimensional vibration signal by calculating the Markov transition field matrix through the transition matrix. Finally, the feature extraction of the vibration map is performed through the residual convolutional neural network. Analysis of the actual measured data at the converter station shows that the average working condition recognition accuracy of the model in this paper reaches 93.1%. It is better than classic time series processing networks such as long and short-term memory networks and one-dimensional convolutional neural networks. It solves the problem of difficulty in building and training long vector deep learning networks by constructing two-dimensional representations of one-dimensional vectors. It is based on commutation The research on fault detection methods of variable vibration signals provides the basis.