Scholarlink

<p>Understanding the dynamic behavior of downhole temperature during drilling operations is crucial for optimizing tool configuration and maximizing the acquisition of logging data, thereby eliminating the need for additional tripping or wireline logging runs. This paper presents a comprehensive study on the application of machine learning and deep learning techniques for predicting downhole temperatures in drilling operations. Following an extensive preprocessing stage that included smoothing and normalizing drilling parameters and related well data, the study compares several machine learning algorithms and long short-term memory (LSTM) architectures. Notable models such as random forest, k-nearest neighbors, decision tree regressors, and LSTM (both sequential and encoder-decoder) were found to be effective for temperature prediction. The LSTM Encoder-Decoder model demonstrated the highest accuracy, with a root mean squared error (RMSE) of 0.892, though it requires higher computational resources. Sensitivity analysis of the model identified revolutions per minute (RPM) and borehole deviation as key factors, providing valuable insights for model refinement and improved thermal management.</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

Investigating Downhole Drilling Temperature Prediction: A Data-Driven Trial of Machine Learning and Deep Learning Methods

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Nardthida Kananithikorn,
  • Thitirat Siriborvornratanakul

摘要

Understanding the dynamic behavior of downhole temperature during drilling operations is crucial for optimizing tool configuration and maximizing the acquisition of logging data, thereby eliminating the need for additional tripping or wireline logging runs. This paper presents a comprehensive study on the application of machine learning and deep learning techniques for predicting downhole temperatures in drilling operations. Following an extensive preprocessing stage that included smoothing and normalizing drilling parameters and related well data, the study compares several machine learning algorithms and long short-term memory (LSTM) architectures. Notable models such as random forest, k-nearest neighbors, decision tree regressors, and LSTM (both sequential and encoder-decoder) were found to be effective for temperature prediction. The LSTM Encoder-Decoder model demonstrated the highest accuracy, with a root mean squared error (RMSE) of 0.892, though it requires higher computational resources. Sensitivity analysis of the model identified revolutions per minute (RPM) and borehole deviation as key factors, providing valuable insights for model refinement and improved thermal management.