中国海洋大学学报（自然科学版）2025，Vol.55Issue(12)：104-114,11.DOI:10.16441/j.cnki.hdxb.20240356
基于时-频域深度学习的水力压裂微震监测信号消噪方法
Denoising Method of Hydraulic Fracturing Microseismic Monitoring Signals Based on Time-Frequency Domain Deep Learning
摘要
Abstract
Surface microseismic monitoring is widely used due to easy construction and low cost,but its signal is susceptible to noise interference.In this paper,we propose a deep learning method based on Time-Frequency(T-F)domain,which constructs a U-Net training set by synchronizing the squeezing transform,and combines the migration learning strategy to eliminate the noise in the low signal-to-noise ratio data.The actual monitoring data contains a limited number of effective microseismic signals,mak-ing it difficult to construct a sufficiently large label training set.This article adopts a strategy of training using simulated datasets first,and then using actual data for transfer learning.Firstly,we perform syn-chrosqueezing transform on clean simulated microseismic signals and simulated microseismic signals added with real noise to obtain their high-resolution time-frequency spectra respectively,and use them to construct the time-frequency domain training set for the U-net network.After the network training is completed,a small-scale training set composed of actual microseismic monitoring signals is used to con-duct transfer learning for the network and obtain the final denoising network.Using high-resolution time-frequency spectra from multiple microseismic monitoring data to construct a training set can better utilize the correlation of effective microseismic signals between monitoring data channels,resulting in better network training outcomes.The experimental results of simulation and actual data show that the denoising network proposed in this paper can significantly suppress coherent noise and random noise in microseismic monitoring data,greatly improving the signal-to-noise ratio of the data.关键词
微震监测/深度学习/U-Net神经网络/同步挤压变换/迁移学习Key words
microseismic monitoring/deep learning/U-Net neural network/synchrosqueezing trans-form/transfer learning分类
地质学引用本文复制引用
赵欣新,林俊武,黄忠来,秦亮..基于时-频域深度学习的水力压裂微震监测信号消噪方法[J].中国海洋大学学报（自然科学版）,2025,55(12):104-114,11.基金项目
山东省自然科学基金项目(ZR2020MD047) （ZR2020MD047）
福建省自然科学基金项目(2022J011170)资助 Supported by the Natural Science Foundation of Shandong Province(ZR2020MD047) （2022J011170）
the Natural Science Foundation of Fujian Pro-vince(2022J011170) （2022J011170）
