Scholarlink

<p>With the increasing application of horizontal-well technology in complex geological settings, real-time identification and prediction of drill-bit exit have become critical for maintaining high reservoir-contact ratios and optimizing geosteering decisions. Conventional approaches often suffer from recognition latency, heavy reliance on subjective experience, and limited quantitative capability. Even with near-bit LWD/MWD data, accurately determining bit-exit events remains challenging. To address these limitations, this study proposes an integrated bit-exit identification and prediction framework that incorporates DBSCAN clustering, Transformer networks, and temporal convolutional networks (TCNs). The methodology is applied to near-bit LWD data acquired from horizontal wells in the Daning–Jixian block on the eastern Yishan Slope of the Ordos Basin. First, DBSCAN is employed to classify the logged responses along the drilled intervals, enabling quantitative extraction of amplitude-change features associated with bit-exit events. The resulting identification model achieves high accuracy, with discrepancies between predicted and actual reservoir-contact ratios maintained within 3%, demonstrating strong stability and reliability. Subsequently, a Transformer model is introduced to learn the mapping between near-bit measurements and bit state, enabling dynamic prediction of exit events along newly drilled intervals. A TCN-based forward-prediction module is further developed to forecast bit-exit behavior in undrilled sections. Field applications show that the integrated framework significantly improves reservoir-contact performance in new horizontal wells, achieving reservoir-contact ratios of 97.68%–99.12%, substantially outperforming conventional methods. The results indicate that the proposed data-driven approach enables real-time identification and prediction of drill-bit exit, providing an effective solution for geosteering decision optimization and dynamic formation-model updating. This method reduces reservoir contamination risks and non-productive time, enhances drilling efficiency, and strengthens reservoir exposure, offering robust technical support for intelligent horizontal-well drilling under complex geological conditions.</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

Prediction of Horizontal Well Drill Bit Exit Using a DBSCAN–Transformer–Temporal Convolutional Network Approach

  • 期刊论文

  • 发表日期：

    2026年4月14日

查看全文
  • Jinghua Yang,
  • Bin Gong,
  • Lei Wang,
  • Hu Huang,
  • Hui Li,
  • Jiawei Ren

摘要

With the increasing application of horizontal-well technology in complex geological settings, real-time identification and prediction of drill-bit exit have become critical for maintaining high reservoir-contact ratios and optimizing geosteering decisions. Conventional approaches often suffer from recognition latency, heavy reliance on subjective experience, and limited quantitative capability. Even with near-bit LWD/MWD data, accurately determining bit-exit events remains challenging. To address these limitations, this study proposes an integrated bit-exit identification and prediction framework that incorporates DBSCAN clustering, Transformer networks, and temporal convolutional networks (TCNs). The methodology is applied to near-bit LWD data acquired from horizontal wells in the Daning–Jixian block on the eastern Yishan Slope of the Ordos Basin. First, DBSCAN is employed to classify the logged responses along the drilled intervals, enabling quantitative extraction of amplitude-change features associated with bit-exit events. The resulting identification model achieves high accuracy, with discrepancies between predicted and actual reservoir-contact ratios maintained within 3%, demonstrating strong stability and reliability. Subsequently, a Transformer model is introduced to learn the mapping between near-bit measurements and bit state, enabling dynamic prediction of exit events along newly drilled intervals. A TCN-based forward-prediction module is further developed to forecast bit-exit behavior in undrilled sections. Field applications show that the integrated framework significantly improves reservoir-contact performance in new horizontal wells, achieving reservoir-contact ratios of 97.68%–99.12%, substantially outperforming conventional methods. The results indicate that the proposed data-driven approach enables real-time identification and prediction of drill-bit exit, providing an effective solution for geosteering decision optimization and dynamic formation-model updating. This method reduces reservoir contamination risks and non-productive time, enhances drilling efficiency, and strengthens reservoir exposure, offering robust technical support for intelligent horizontal-well drilling under complex geological conditions.