Background noise 11.
<div><p>In the mining field, hydraulic fracturing of coal - seam boreholes generates a large number of weak microseismic signals. The accurate identification of these signals is crucial for subsequent positioning and inversion. However, when dealing with such signals, traditional automat...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | , , , , , , |
| منشور في: |
2025
|
| الموضوعات: | |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | <div><p>In the mining field, hydraulic fracturing of coal - seam boreholes generates a large number of weak microseismic signals. The accurate identification of these signals is crucial for subsequent positioning and inversion. However, when dealing with such signals, traditional automatic microseismic waveform identification algorithms have difficulty in accurately identifying weak waveforms and are prone to misjudging background noise. This study innovatively introduces the deep - learning convolutional neural network (CNN), integrating the concepts and methods of computational communication to analyze microseismic signals. 8,341 pieces of background noise data and 5,860 pieces of microseismic data are carefully selected from the data of coal - seam borehole hydraulic fracturing. After adding noise at 12 levels and performing translation with 10 different degrees of displacement, 101,123 pieces of background noise and 102,546 effective waveforms are obtained. Subsequently, by applying the information - propagation dynamics model of computational communication, microseismic signals are regarded as information carriers. A signal - propagation network is constructed, and features such as network degree distribution are extracted. These features, combined with traditional time - domain and frequency - domain features, are converted into time - domain and Fourier images and then input into a two - dimensional CNN model. Experiments show that the time - domain CNN model achieves a precision rate of 100% and a recall rate of 68% in microseismic event identification, significantly outperforming traditional methods such as AIC, STA/LTA, and the Fourier CNN model. Furthermore, the time-frequency fusion CNN model—integrating time-domain waveforms, Fourier frequency-domain features, and time-frequency characteristics (e.g., short-time Fourier transform)—achieves an identical precision rate of 100% and a higher recall rate of 72%, outperforming the single-domain time-domain CNN model. The integration of computational communication concepts (e.g., signal propagation network topological features) and multi-domain features enables the model to capture comprehensive spatiotemporal and dynamic signal characteristics, further validating its superiority in identifying weak microseismic signals with low signal-to-noise ratios (SNR).This indicates that the combination of time - domain images and computational - communication technology is more suitable as the input data for the CNN model. It can effectively distinguish microseismic waveforms from background noise, opening up a new path for the identification of mine microseismic signals and demonstrating the application potential of computational communication in this field.</p></div> |
|---|