Scholarlink

<p>Fracability is the ability to generate fractures with hydraulic fracturing, and it is a crucial index for evaluating a reservoir’s fracturing feasibility at the exploration stage. A high-accuracy prediction for reservoir fracability is essential for the determination of favorable fracturing locations. However, traditional prediction methods are expensive and inefficient due to their strong dependency on fracturing data and high computation cost. This paper proposed a new machine learning-based prediction approach to predict the fracability of deep reservoirs. Firstly, a thermo–hydro–mechanical-damage (THM-damage) coupling model was established to quantitatively characterize reservoir fracability and to generate a dataset by combining logging data for machine learning training. Secondly, the XGBoost model was optimized by the grid search and cross-validation algorithm and used for fracability prediction. Finally, the fracability of deep unconventional gas reservoirs at depths of 5800–6300&#xa0;m in the Tarim Basin was predicted and compared with field-measured fracability. The prediction results indicated that the THM-damage coupling model is valid and reliable in characterizing reservoir fracability. The optimized XGBoost model can predict the fracability of deep reservoirs with high prediction accuracy and generalization ability by using only 279 cases. It was also found that this novel approach can quickly and accurately predict deep reservoir fracability with a small dataset and fast determine the favorable fracturing locations with high accuracy.</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 New Machine Learning-Based Prediction Approach for the Fracability of Deep Reservoirs

  • 期刊论文

  • 发表日期：

    2025年7月13日

查看全文
  • Peibo Li,
  • Jianguo Wang,
  • Wei Liang,
  • Jiajie Yang,
  • Zhizhen Zhang,
  • Ke Xu

摘要

Fracability is the ability to generate fractures with hydraulic fracturing, and it is a crucial index for evaluating a reservoir’s fracturing feasibility at the exploration stage. A high-accuracy prediction for reservoir fracability is essential for the determination of favorable fracturing locations. However, traditional prediction methods are expensive and inefficient due to their strong dependency on fracturing data and high computation cost. This paper proposed a new machine learning-based prediction approach to predict the fracability of deep reservoirs. Firstly, a thermo–hydro–mechanical-damage (THM-damage) coupling model was established to quantitatively characterize reservoir fracability and to generate a dataset by combining logging data for machine learning training. Secondly, the XGBoost model was optimized by the grid search and cross-validation algorithm and used for fracability prediction. Finally, the fracability of deep unconventional gas reservoirs at depths of 5800–6300 m in the Tarim Basin was predicted and compared with field-measured fracability. The prediction results indicated that the THM-damage coupling model is valid and reliable in characterizing reservoir fracability. The optimized XGBoost model can predict the fracability of deep reservoirs with high prediction accuracy and generalization ability by using only 279 cases. It was also found that this novel approach can quickly and accurately predict deep reservoir fracability with a small dataset and fast determine the favorable fracturing locations with high accuracy.