Scholarlink

<p>Prediction of in situ minimum horizontal stress (Shmin) is an important task for optimizing many applications related to reservoir geomechanics including wellbore stability analysis, sand production, and well stimulation. Pressure integrity tests are the ideal solutions to accurately determine Shmin, but they provide discrete data points, and they often conducted in a limited number of wells for cost- and time-saving purposes. This study presents an integrated approach to estimating minimum horizontal stress (Shmin) in carbonate reservoirs using a combination of empirical modeling and machine learning techniques. Shmin was calculated using datasets from six wells and calibrated with extended leak-off test (XLOT). The results indicated a strong agreement between the calculated Shmin values and XLOT measurements, achieving high accuracy (<i>R</i><sup>2</sup> = 0.91, RMSE = 0.88). Two predictive models, multiple regression analysis (MRA) and artificial neural networks (ANN), were developed using data from two wells and evaluated using <i>R</i><sup>2</sup> and RMSE metrics. The ANN model outperformed MRA, achieving superior accuracy (<i>R</i> = 0.99, and MSE = 0.435) in all datasets, indicating that true vertical depth (TVD) is the most influential parameter, followed in decreasing order by bulk density (RHOB), gamma ray (GR), and sonic log (<i>∆t</i>). Four additional wells were used to validate the ANN model, confirming its reliability and highlighting its capability to handle geological heterogeneity and stress variations with burial depth. Furthermore, the fivefold cross-validation results confirm that the developed ANN model generalizes well to unseen data, achieving consistently high <i>R</i><sup>2</sup> values across training and test sets and thus indicating strong prediction Shmin with minimal overfitting. However, site-specific calibrations may enhance model accuracy. This study presents the potential of machine learning, as a robust tool for predicting Shmin in complex geological settings, offering valuable insights for future reservoir geomechanics applications in reservoir characterization, wellbore stability analysis, sand production management, and hydraulic fracturing design.</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

An Intensive Study to Determine the In Situ Minimum Horizontal Stress Using Well Logging Data

  • 期刊论文

  • 发表日期：

    2025年4月2日

查看全文
  • Hassan A. Abdul Hussein,
  • Farqad Ali Hadi,
  • Muhsin J. Jweeg,
  • Mustafa Adil Issa,
  • Malik Mustafa Mohammed,
  • Dheyaa J. Jasim

摘要

Prediction of in situ minimum horizontal stress (Shmin) is an important task for optimizing many applications related to reservoir geomechanics including wellbore stability analysis, sand production, and well stimulation. Pressure integrity tests are the ideal solutions to accurately determine Shmin, but they provide discrete data points, and they often conducted in a limited number of wells for cost- and time-saving purposes. This study presents an integrated approach to estimating minimum horizontal stress (Shmin) in carbonate reservoirs using a combination of empirical modeling and machine learning techniques. Shmin was calculated using datasets from six wells and calibrated with extended leak-off test (XLOT). The results indicated a strong agreement between the calculated Shmin values and XLOT measurements, achieving high accuracy (R2 = 0.91, RMSE = 0.88). Two predictive models, multiple regression analysis (MRA) and artificial neural networks (ANN), were developed using data from two wells and evaluated using R2 and RMSE metrics. The ANN model outperformed MRA, achieving superior accuracy (R = 0.99, and MSE = 0.435) in all datasets, indicating that true vertical depth (TVD) is the most influential parameter, followed in decreasing order by bulk density (RHOB), gamma ray (GR), and sonic log (∆t). Four additional wells were used to validate the ANN model, confirming its reliability and highlighting its capability to handle geological heterogeneity and stress variations with burial depth. Furthermore, the fivefold cross-validation results confirm that the developed ANN model generalizes well to unseen data, achieving consistently high R2 values across training and test sets and thus indicating strong prediction Shmin with minimal overfitting. However, site-specific calibrations may enhance model accuracy. This study presents the potential of machine learning, as a robust tool for predicting Shmin in complex geological settings, offering valuable insights for future reservoir geomechanics applications in reservoir characterization, wellbore stability analysis, sand production management, and hydraulic fracturing design.