Scholarlink

<p>This study addresses the challenge of improving well-logging data quality under adverse drilling conditions by applying unsupervised machine learning outlier detection techniques (UML-ODTs). Focusing on the Isolation Forest (IF) algorithm, we analyzed 13,566 and 14,880 sample points from two wells in the Songliao Basin, China, to detect and reconstruct anomalies caused by equipment malfunctions, noise, and geological heterogeneity. Four UML-ODTs—Isolation Forest (IF), One-Class Support Vector Machine (OCSVM), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Local Outlier Factor (LOF)—were systematically compared. Detected outliers were reconstructed using Locally Weighted Regression (LOWESS). The results demonstrated the superior performance of the IF model, improving seismic synthetic record correlations from 0.49 to 0.92 (Well-A) and 0.56 to 0.87 (Well-B). This approach not only preserves geological authenticity but also significantly enhances the reliability of seismic data interpretation, offering a robust solution for complex geological settings. These findings highlight the potential of UML-ODTs for future applications in oil and gas exploration.</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

Reconstruction of well-logging data using unsupervised machine learning-based outlier detection techniques (UML-ODTs) under adverse drilling conditions

  • 期刊论文

  • 发表日期：

    2025年3月20日

查看全文
  • Jing-Rui Chen,
  • Rui-Zhao Yang,
  • Ting-Ting Li,
  • Yun-Di Xu,
  • Zhi-Peng Sun

摘要

This study addresses the challenge of improving well-logging data quality under adverse drilling conditions by applying unsupervised machine learning outlier detection techniques (UML-ODTs). Focusing on the Isolation Forest (IF) algorithm, we analyzed 13,566 and 14,880 sample points from two wells in the Songliao Basin, China, to detect and reconstruct anomalies caused by equipment malfunctions, noise, and geological heterogeneity. Four UML-ODTs—Isolation Forest (IF), One-Class Support Vector Machine (OCSVM), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Local Outlier Factor (LOF)—were systematically compared. Detected outliers were reconstructed using Locally Weighted Regression (LOWESS). The results demonstrated the superior performance of the IF model, improving seismic synthetic record correlations from 0.49 to 0.92 (Well-A) and 0.56 to 0.87 (Well-B). This approach not only preserves geological authenticity but also significantly enhances the reliability of seismic data interpretation, offering a robust solution for complex geological settings. These findings highlight the potential of UML-ODTs for future applications in oil and gas exploration.