Scholarlink

<p>Geomechanical parameters (GMPs), e.g., Poisson ratio, Young’s modulus, and compressive strength of reservoir rock, are essential for design optimization, reducing operational risks, and wellbore stability assessment in geomechanical studies. Conventionally, these parameters are estimated by time-consuming laboratory experiments, in-situ tests, and empirical equations. However, these methods ignore the influence of scale and exhibit a nonlinear relationship, resulting in uncertain parameter estimations. To evaluate both the potentiality of machine learning (ML) methods and the contribution of various input variables for accurately estimating GMPs in complex geological datasets has not been broadly studied. Therefore, in this research, a two-layer feedforward artificial neural network (ANN) with well-established optimization algorithms, such as Levenberg–Marquardt (LM-BP) and Scaled Conjugate Gradient (SCG), were employed to calculate GMP from complicated data. To develop ML models, 15,705 datapoints of five wells from the complex Mahu field were utilized, and feature selection was used to determine significant input variables that aid in prediction performance, e.g., NPHI, DT, RT, and RHOB, respectively. Moreover, models were validated using 8,592 datapoints from three other wells. Evaluation of results showed that the LM-BP achieved RMSE values of 0.0011–0.00099 and R<sup>2</sup> of 0.995–0.997, indicating superior prediction accuracy compared to SCG. It depicts parallel prediction precision when utilized with unseen datapoints. In addition, efficient and reliable training of ANN with LM-BP and SCG approaches better estimates GMP than empirical methods. Ultimately, this research advances geomechanical modeling and decision-making processes within the oil and gas industry.</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

Predictive modeling of reservoir geomechanical parameters through computational intelligence approach, integrating core and well logging data

  • 期刊论文

  • 发表日期：

    2025年1月7日

查看全文
  • Sayed Muhammad Iqbal,
  • Jianmin Li,
  • Junxiu Ma,
  • Dawei Hu,
  • Shuang Tian,
  • Hui Zhou,
  • Litao Shang

摘要

Geomechanical parameters (GMPs), e.g., Poisson ratio, Young’s modulus, and compressive strength of reservoir rock, are essential for design optimization, reducing operational risks, and wellbore stability assessment in geomechanical studies. Conventionally, these parameters are estimated by time-consuming laboratory experiments, in-situ tests, and empirical equations. However, these methods ignore the influence of scale and exhibit a nonlinear relationship, resulting in uncertain parameter estimations. To evaluate both the potentiality of machine learning (ML) methods and the contribution of various input variables for accurately estimating GMPs in complex geological datasets has not been broadly studied. Therefore, in this research, a two-layer feedforward artificial neural network (ANN) with well-established optimization algorithms, such as Levenberg–Marquardt (LM-BP) and Scaled Conjugate Gradient (SCG), were employed to calculate GMP from complicated data. To develop ML models, 15,705 datapoints of five wells from the complex Mahu field were utilized, and feature selection was used to determine significant input variables that aid in prediction performance, e.g., NPHI, DT, RT, and RHOB, respectively. Moreover, models were validated using 8,592 datapoints from three other wells. Evaluation of results showed that the LM-BP achieved RMSE values of 0.0011–0.00099 and R2 of 0.995–0.997, indicating superior prediction accuracy compared to SCG. It depicts parallel prediction precision when utilized with unseen datapoints. In addition, efficient and reliable training of ANN with LM-BP and SCG approaches better estimates GMP than empirical methods. Ultimately, this research advances geomechanical modeling and decision-making processes within the oil and gas industry.