Research On Deep Learning-Based Noise Suppression Method For Microseismic Data

Hydraulic fracturing technology is an important tool in unconventional oil and gas extraction.Microseismic monitoring technology is commonly used in the field industry to collect the microseismic signals generated during hydraulic fracturing.However,the field microseismic data collected by geophones has strong noise interference and weak energy,which significantly increases the difficulty of studying and interpreting microseismic data.Therefore,how to suppress the noise in microseismic data is one of the core issues in microseismic data processing.Currently,deep learning is widely used in various fields due to its powerful feature characterization capability.However,research on deep learning techniques in the field of microseismic data noise reduction is still in the early stage.For this reason,this paper provides an in-depth study of deep learning-based noise reduction methods for microseismic data,aiming to develop advanced denoising models for microseismic data and further improve the accuracy and efficiency of microseismic data processing.The main works of this paper are illustrated as follows:1)To address the problem of unsatisfactory noise reduction(such as waveform distortion)for the noise-reduced microseismic signals,this paper proposes a microseismic denoising method based on the U-Net network.Firstly,the Ricker wavelet is used to synthesize the clean microseismic data,and different levels of noise are added to construct the noisy microseismic data,then the noisy data is sliced and expanded to feed the model.Next,a dilated convolution with varying expansion coefficients is introduced in the U-Net network to obtain more signal features,which aims to reduce waveform distortion and retain more effective signals.Then,an envelope entropy is employed as the loss function of the proposed model to implement the unsupervised processing for field microseismic signal noise reduction.Finally,the proposed network is trained to verify its effectiveness and superiority.Experimental results show that the proposed denoising method based on the U-Net network can effectively remove the random noises in the microseismic signals,improving the events of images and recovering the microseismic signals’ amplitudes.2)To tackle the problem of high-frequency feature loss and low-frequency noise retention caused by ignoring noise in the same frequency band as the effective signal in traditional denoising models,this paper proposes a multi-channel self-coding microseismic denoising method based on complementary ensemble empirical mode decomposition.Firstly,an empirical modal decomposition method is deployed to decompose the microseismic data.Then,the decomposed modal components are used as inputs for different channels of multi-channel networks to obtain the denoised signal.Finally,a novel mutual correlation loss function is utilized to achieve unsupervised processing of microseismic data noise reduction,which solves the problem of difficulty in obtaining pure field microseismic data.The experimental results demonstrate that the proposed method can effectively retain effective signals,completely remove the noises,reduce the number of model parameters,and improve the model’s computational efficiency.3)In this paper,two proposed denoising methods for microseismic data are applied to the processing of field microseismic data.The experimental results verify that the two proposed denoising methods can effectively process the field microseismic data,provide a new idea for microseismic data noise reduction to a certain extent,which is also of great significance for microseismic data processing,interpretation and subsequent guidance of hydraulic fracturing operations.