Scholarlink

<p>Microseismic (MS) source localization plays a crucial role in rockburst risk mitigation and reservoir stimulation evaluation. As a deep learning approach, convolutional neural networks (CNNs) have shown promising capabilities in extracting local spatial features from MS waveforms to predict event locations. However, in the complex geological environment of southwest China, the CNN based methods exhibit limited robustness, as their localization accuracy is highly sensitive to low signal-to-noise ratio (SNR) and incomplete data. In this study, we proposed a fusion network based on the velocity model constraint as a regression localization model to predict the source locations, namely the parallel transformer-CNN architecture (PTCNets). Initially, the seismic data were synthesized by 3D forward simulation of the study area velocity model, and actual noise was superimposed to construct the training dataset. Subsequently, the velocity model was incorporated into the loss function to achieve dual supervision from both data-driven learning and physical constraints. Finally, the performance of PTCNets was comprehensively evaluated. Ablation experiments demonstrate that PTCNets achieves an average localization error of 3.502&#xa0;m under a complete geophone array. When the SNR drops to <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(-20\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>-</mo> <mn>20</mn> </mrow> </math></EquationSource> </InlineEquation> dB, the localization errors remain within 4&#xa0;m. Even under the extreme condition of a 15% geophone missing rate, the error increases only to 7.3&#xa0;m. Meanwhile, the method was successfully applied to MS monitoring during the 11th stage of hydraulic fracturing in Southwest China, accurately predicting the locations of 57 MS events. These results highlight the method’s potential as a new technological paradigm for MS monitoring in deep geological resource development, as well as its significant value in early warning of fracturing risks and reservoir stimulation evaluation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Microseismic Source Localization via Fusion Networks with Integrated Velocity Model Constraints

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Renjie He,
  • Chunlu Wang,
  • Jiang Wang,
  • Jingchuan Feng,
  • Chunyang Pei,
  • Jiwu Li,
  • Zubin Chen

摘要

Microseismic (MS) source localization plays a crucial role in rockburst risk mitigation and reservoir stimulation evaluation. As a deep learning approach, convolutional neural networks (CNNs) have shown promising capabilities in extracting local spatial features from MS waveforms to predict event locations. However, in the complex geological environment of southwest China, the CNN based methods exhibit limited robustness, as their localization accuracy is highly sensitive to low signal-to-noise ratio (SNR) and incomplete data. In this study, we proposed a fusion network based on the velocity model constraint as a regression localization model to predict the source locations, namely the parallel transformer-CNN architecture (PTCNets). Initially, the seismic data were synthesized by 3D forward simulation of the study area velocity model, and actual noise was superimposed to construct the training dataset. Subsequently, the velocity model was incorporated into the loss function to achieve dual supervision from both data-driven learning and physical constraints. Finally, the performance of PTCNets was comprehensively evaluated. Ablation experiments demonstrate that PTCNets achieves an average localization error of 3.502 m under a complete geophone array. When the SNR drops to \(-20\) - 20 dB, the localization errors remain within 4 m. Even under the extreme condition of a 15% geophone missing rate, the error increases only to 7.3 m. Meanwhile, the method was successfully applied to MS monitoring during the 11th stage of hydraulic fracturing in Southwest China, accurately predicting the locations of 57 MS events. These results highlight the method’s potential as a new technological paradigm for MS monitoring in deep geological resource development, as well as its significant value in early warning of fracturing risks and reservoir stimulation evaluation.