Prediction of Formation Permeability While Drilling: Machine Learning Applications

Said Hassaan, Ahmed Farid Ibrahim*, Abdulaziz Mohamed, Salaheldin Elkatatny*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Accurate prediction of formation permeability plays a crucial role in reservoir characterization, well planning, production forecasting, reservoir management, and risk mitigation. Traditional methods including core analysis, well testing, and well logging can be used to estimate the reservoir permeability. However, these methods are time-consuming and expensive, making them impractical for comprehensive reservoir evaluation. This study aims to apply different machine learning tools to predict the formation permeability in a real-time using readily available drilling parameters. The dataset used in this study was gathered from two vertical wells located in the Middle East. It encompasses a variety of drilling parameters such as rate of penetration (ROP), gallons per minute (GPM), revolutions per minute (RPM), strokes per minute (SPP), torque, and weight on bit (WOB). Alongside these parameters, permeability (k) was obtained through core analysis. To predict the formation permeability, three machine learning models were employed: gradient boosting regressors (GBRs), artificial neural networks (ANNs), and function networks (FNs). These models were utilized and assessed to determine their efficacy in estimating the formation permeability from the drilling parameters. Furthermore, an empirical correlation was developed based on the optimized weights and biases derived from the ANN model. Preliminary results showed that the different ML models slightly struggled for direct permeability prediction from the drilling parameters. Adding an intermediate step of predicting the porosity from the drilling parameters and then using the estimated porosity in addition to the drilling parameters improved the permeability prediction. The ANN model showed a superior performance with correlation coefficient (R) value higher than 0.95 and root mean square error (RMSE) less than 0.28 in the different datasets. Similarly, the FN model was able to predict the permeability with R value higher than 0.94 and RMSE less than 0.38 in the different datasets. The GBR model slightly struggled in permeability prediction as the R value decreased to 0.89 in the testing and validation datasets. Accurate prediction of formation permeability enables operators to make informed decisions, enhance hydrocarbon recovery, and optimize overall field development strategies.

Original languageEnglish
Article number015113
JournalArabian Journal for Science and Engineering
DOIs
StateAccepted/In press - 2024

Bibliographical note

Publisher Copyright:
© King Fahd University of Petroleum & Minerals 2024.

Keywords

  • Drilling parameters
  • Formation permeability
  • Machine learning
  • Real-time

ASJC Scopus subject areas

  • General

Fingerprint

Dive into the research topics of 'Prediction of Formation Permeability While Drilling: Machine Learning Applications'. Together they form a unique fingerprint.

Cite this