Scholarlink

<p>The rate of penetration (ROP) is<!--Query ID="Q1" Text="Please check and confirm that the authors and their respective affiliations have been correctly identified and amend if necessary." Resolved="yes"--> a critical parameter for optimizing oil well drilling and the overall cost of drilling operations. In this study, five methodologies, including three artificial intelligence models (artificial neural networks [ANNs], support vector regression [SVR], random forest [RF]), a physical model, and a hybrid model, were evaluated for their ability to estimate the ROP on the basis of drilling data from a complex lithological area. The dataset utilized for model construction was derived from field data from nine wells in southern Iran. Data from wells 1 to 6 were employed for training and testing the models, and unseen data from wells 7 to 9 were used for evaluation. Following preprocessing and outlier removal, six features, namely, the flow rate (Q), weight on bit (WOB), standpipe pressure (SPP), depth, torque (T), and drill string rotation (DSR), were utilized as inputs for estimating the ROP. Sensitivity analysis was used to find the best machine learning (ML) structures. The ROP estimation on evaluation data from three unseen wells demonstrated superior performance of the hybrid model and ANN compared with the SVR, RF, and Physical Model. The hybrid model achieved metrics of RMSE 0.55, MAPE 4.61%, and correlation coefficient 0.94, showcasing its exceptional accuracy. Similarly, the ANN had root mean square errors (RMSEs) of 0.69, mean absolute percentage errors (MAPEs) of 5.01%, and correlation coefficients of 0.93. In contrast, the physical model showed limitations with RMSEs of 4.33, MAPEs of 10.33%, and correlation coefficients of 0.85, highlighting the need for further refinement in deterministic approaches. It is recommended that real-time ROP estimation be repeated in other areas with different hybrid methodologies and ML algorithms. This helps drilling engineers achieve better performance and reduces drilling risk. Future work could focus on addressing the computational complexity of hybrid models, which arises from sequential training steps that combine physical<!--Query ID="Q2" Text="Please check processed list of symbols." Resolved="yes"--> and ML models.</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

Real-time prediction of the rate of penetration via computational intelligence: a comparative study on complex lithology in Southwest Iran

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月10日

查看全文
  • Mohammad Najafi,
  • Yousef Shiri

摘要

The rate of penetration (ROP) is a critical parameter for optimizing oil well drilling and the overall cost of drilling operations. In this study, five methodologies, including three artificial intelligence models (artificial neural networks [ANNs], support vector regression [SVR], random forest [RF]), a physical model, and a hybrid model, were evaluated for their ability to estimate the ROP on the basis of drilling data from a complex lithological area. The dataset utilized for model construction was derived from field data from nine wells in southern Iran. Data from wells 1 to 6 were employed for training and testing the models, and unseen data from wells 7 to 9 were used for evaluation. Following preprocessing and outlier removal, six features, namely, the flow rate (Q), weight on bit (WOB), standpipe pressure (SPP), depth, torque (T), and drill string rotation (DSR), were utilized as inputs for estimating the ROP. Sensitivity analysis was used to find the best machine learning (ML) structures. The ROP estimation on evaluation data from three unseen wells demonstrated superior performance of the hybrid model and ANN compared with the SVR, RF, and Physical Model. The hybrid model achieved metrics of RMSE 0.55, MAPE 4.61%, and correlation coefficient 0.94, showcasing its exceptional accuracy. Similarly, the ANN had root mean square errors (RMSEs) of 0.69, mean absolute percentage errors (MAPEs) of 5.01%, and correlation coefficients of 0.93. In contrast, the physical model showed limitations with RMSEs of 4.33, MAPEs of 10.33%, and correlation coefficients of 0.85, highlighting the need for further refinement in deterministic approaches. It is recommended that real-time ROP estimation be repeated in other areas with different hybrid methodologies and ML algorithms. This helps drilling engineers achieve better performance and reduces drilling risk. Future work could focus on addressing the computational complexity of hybrid models, which arises from sequential training steps that combine physical and ML models.