Scholarlink

<p>Accurate microseismic monitoring is a key step in unconventional oil and gas fracturing monitoring, including core tasks such as inversions of microseismic source location and source mechanism solution. Accurate P- and S-wave polarities can ensure reliable waveform stacking in subsequent migration-based location and enable fast inversion of the microseismic source mechanism. Currently, existing methods are primarily limited to predicting P-wave first-motion polarities. In this work, we have presented a hybrid deep learning architecture integrating a convolutional neural network (CNN), a long short-term memory (LSTM), and a multi-head attention mechanism, aiming to simultaneously estimate P- and S-wave first-motion polarities. We constructed a comprehensive dataset based on synthetic data and actual fracture monitoring data to carry out experimental tests. Compared with the traditional CNN, CNN-LSTM, or single-head attention mechanism models, the proposed model demonstrated optimal performance in estimating P- and S-wave first-motion polarities. For P- and S-waves, it achieved respective accuracies of 92.45% and 97.43%, with a combined average of 94.94%, a macro precision of 94.96%, a macro recall of 94.89%, and a macro-F1 score of 94.92%. Ten-fold cross-validation demonstrated its robustness and strong generalization ability. Furthermore, the example tests demonstrated the model’s ability to estimate the first-motion polarities of P- and S-waves under different signal-to-noise ratio conditions (encompassing high and low levels).</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

A novel method to predict the P- and S-wave polarities of downhole microseismic events based on hybrid deep learning model

  • 期刊论文

  • 发表日期：

    2026年5月24日

查看全文
  • Junping Zhang,
  • Qinghui Mao,
  • Tahir Azeem,
  • Xiurong Li,
  • Xuliang Zhang,
  • Zhixian Gui,
  • Shijie Zhou

摘要

Accurate microseismic monitoring is a key step in unconventional oil and gas fracturing monitoring, including core tasks such as inversions of microseismic source location and source mechanism solution. Accurate P- and S-wave polarities can ensure reliable waveform stacking in subsequent migration-based location and enable fast inversion of the microseismic source mechanism. Currently, existing methods are primarily limited to predicting P-wave first-motion polarities. In this work, we have presented a hybrid deep learning architecture integrating a convolutional neural network (CNN), a long short-term memory (LSTM), and a multi-head attention mechanism, aiming to simultaneously estimate P- and S-wave first-motion polarities. We constructed a comprehensive dataset based on synthetic data and actual fracture monitoring data to carry out experimental tests. Compared with the traditional CNN, CNN-LSTM, or single-head attention mechanism models, the proposed model demonstrated optimal performance in estimating P- and S-wave first-motion polarities. For P- and S-waves, it achieved respective accuracies of 92.45% and 97.43%, with a combined average of 94.94%, a macro precision of 94.96%, a macro recall of 94.89%, and a macro-F1 score of 94.92%. Ten-fold cross-validation demonstrated its robustness and strong generalization ability. Furthermore, the example tests demonstrated the model’s ability to estimate the first-motion polarities of P- and S-waves under different signal-to-noise ratio conditions (encompassing high and low levels).