Identification of Contact Relationship of Electrical Engineering Distribution Network with Two-Dimensional Wavelet Threshold Deep Neural Network

Abstract

With the improvement of electrification in power systems, accurate and rapid fault location helps to repair faults, which is of great significance to the stability of distribution network operation. As an important part of power distribution in the power system, the electrical engineering distribution network is directly connected to the power transmission system and power end users. Its safety and reliability are related not only to the power sales interests of power companies, but also to the power users’ rights and interests in power consumption. In this paper, an improved threshold based on the peak-sum ratio (PSR) is proposed, the time-frequency features in the disturbance signal are extracted through the continuous transformation of the two-dimensional wavelet threshold deep neural network to generate the disturbance time-frequency map of the electrical engineering distribution network, and then the deep learning model is used to analyze the model. After the classification performance is continuously optimized, the signal disturbance identification of electrical engineering distribution network is realized. By calculating the PSR of the distribution network, the correction factor can adaptively adjust the general threshold according to the noise distribution characteristics of different disturbance signals. By analyzing the data one by one, it can be seen that the improved threshold function has obvious advantages in the input signal-to-noise ratio of 10–12 dB, 16–18 dB, and 21–26 dB. At 13 dB, 14 dB, 20 dB, 27 dB, and 28 dB, the SNR difference of the distribution network is very small, and at 15 dB, 19 dB, 29 dB, and 30 dB, it is slightly inferior, but its denoising effect is generally better. The example results have shown that the recognition accuracy of the two-dimensional wavelet threshold denoising method in a noise-free environment has been effectively improved, and it has a certain anti-noise performance. The method proposed in this study has few feature extraction steps, is easy to implement, and is suitable for more types of disturbances.