Scholarlink

<p>High-resolution characterization of organic facies is fundamental to accurately assessing the hydrocarbon generation potential of source rocks and effectively guiding petroleum exploration strategies. Conventional geochemical techniques rely on discrete core samples and often fail to capture the continuous spatial heterogeneity of organic facies in complex coal-bearing source rocks. To address this limitation, this paper proposes an integrated workflow combining a tailored organic facies classification scheme with machine learning predictions. First, a five-fold classification (types A, B, BC, C, and D) specific to the Jurassic coal measures in the Junggar Basin was established based on total organic carbon (TOC) and hydrogen index values, calibrated by depositional environments. Subsequently, a XGBoost model optimized by particle swarm optimization (PSO) was developed to predict continuous TOC and rock-eval parameter S<sub>2</sub> from well logs. The model demonstrated superior performance over benchmark algorithms (genetic algorithm-XGBoost, random forest, and support vector regression), achieving a <i>R</i><sup>2</sup> of 0.84 for S<sub>2</sub> and 0.76 for TOC on the test set. By applying this model to 34 wells, we mapped the high-resolution spatiotemporal distribution of organic facies. The results revealed an "expansion–contraction–expansion" evolutionary trend across the Jurassic Badaowan, Sangonghe, and Xishanyao Formations, controlled primarily by the interplay of tectonic subsidence and lake-level fluctuations. This workflow provides a robust, data-driven framework for source rock evaluation in lacustrine coal-bearing basins with limited core data.</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 Organic Facies Distribution in Jurassic Coal-Bearing Source Rocks of the Junggar Basin: A PSO-Optimized XGBoost Approach

  • 期刊论文

  • 发表日期：

    2026年2月1日

查看全文
  • Shangbin Wang,
  • Xiangchun Chang,
  • Guanlong Zhang,
  • Ruichao Guo,
  • Bingbing Shi,
  • Junjian Zhang,
  • Tianjiao Liu,
  • Chenyu Wang

摘要

High-resolution characterization of organic facies is fundamental to accurately assessing the hydrocarbon generation potential of source rocks and effectively guiding petroleum exploration strategies. Conventional geochemical techniques rely on discrete core samples and often fail to capture the continuous spatial heterogeneity of organic facies in complex coal-bearing source rocks. To address this limitation, this paper proposes an integrated workflow combining a tailored organic facies classification scheme with machine learning predictions. First, a five-fold classification (types A, B, BC, C, and D) specific to the Jurassic coal measures in the Junggar Basin was established based on total organic carbon (TOC) and hydrogen index values, calibrated by depositional environments. Subsequently, a XGBoost model optimized by particle swarm optimization (PSO) was developed to predict continuous TOC and rock-eval parameter S2 from well logs. The model demonstrated superior performance over benchmark algorithms (genetic algorithm-XGBoost, random forest, and support vector regression), achieving a R2 of 0.84 for S2 and 0.76 for TOC on the test set. By applying this model to 34 wells, we mapped the high-resolution spatiotemporal distribution of organic facies. The results revealed an "expansion–contraction–expansion" evolutionary trend across the Jurassic Badaowan, Sangonghe, and Xishanyao Formations, controlled primarily by the interplay of tectonic subsidence and lake-level fluctuations. This workflow provides a robust, data-driven framework for source rock evaluation in lacustrine coal-bearing basins with limited core data.