Scholarlink

<p>Petrophysical parameters play an important role in characterizing geological formations. Water and oil saturation, permeability, effective porosity, and total porosity are among the parameters that are frequently evaluated during the field development planning using wireline measurements. It is generally acceptable that wireline measurement of these parameters may be subjected to various errors including human errors, hence compromising the accuracy of hydrocarbon reserves. In this study, machine learning has been employed due to its ability to identify and train on the controlling features that are strong for the petrophysical parameters. Well logs data with the data obtained from special core analysis in the little knife field are employed. 5 well log datasets were used as independent variables such as depth, resistivity, compressional slowness, neutron-porosity, gamma ray, and formation density. Grain density, permeability, water saturation, and porosity are obtained from core measurements. Water saturation, porosity, and permeability are the target variables. Preprocessing involves cleaning and ensuring the quality of the data, followed by subdividing the dataset into training and testing sets. Six machine learning techniques including decision tree regression, random forest regression, extreme gradient boost, extra-trees, light gradient boost machine and K-nearest neighbor were trained on the datasets. The models varied in their performance to predict water saturation, porosity, and permeability. Model effectiveness was evaluated, extra trees algorithm demonstrates superior performance in water saturation porosity and permeability predictions with R<sup>2</sup> of 98%, 89% and 86% respectively with the lowest error metrics. This study is novel in its multi-model, multi-parameter evaluation approach within the Bakken interval, combining well log and core data for improved validation, applying hyperparameter tuning to enhance model accuracy, and benchmarking predictions against empirical methods. The results highlight the strength of ML models in capturing non-linear subsurface relationships critical to reservoir characterization.</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

Advanced machine learning-based predictive tools for estimating petrophysical parameters: a case study from the Little Knife Field, North Dakota

  • 期刊论文

  • 发表日期：

    2025年6月4日

查看全文
  • Luc Yvan Nkok,
  • Adesina S. Fadairo,
  • Tomiwa Isaac Oguntade

摘要

Petrophysical parameters play an important role in characterizing geological formations. Water and oil saturation, permeability, effective porosity, and total porosity are among the parameters that are frequently evaluated during the field development planning using wireline measurements. It is generally acceptable that wireline measurement of these parameters may be subjected to various errors including human errors, hence compromising the accuracy of hydrocarbon reserves. In this study, machine learning has been employed due to its ability to identify and train on the controlling features that are strong for the petrophysical parameters. Well logs data with the data obtained from special core analysis in the little knife field are employed. 5 well log datasets were used as independent variables such as depth, resistivity, compressional slowness, neutron-porosity, gamma ray, and formation density. Grain density, permeability, water saturation, and porosity are obtained from core measurements. Water saturation, porosity, and permeability are the target variables. Preprocessing involves cleaning and ensuring the quality of the data, followed by subdividing the dataset into training and testing sets. Six machine learning techniques including decision tree regression, random forest regression, extreme gradient boost, extra-trees, light gradient boost machine and K-nearest neighbor were trained on the datasets. The models varied in their performance to predict water saturation, porosity, and permeability. Model effectiveness was evaluated, extra trees algorithm demonstrates superior performance in water saturation porosity and permeability predictions with R2 of 98%, 89% and 86% respectively with the lowest error metrics. This study is novel in its multi-model, multi-parameter evaluation approach within the Bakken interval, combining well log and core data for improved validation, applying hyperparameter tuning to enhance model accuracy, and benchmarking predictions against empirical methods. The results highlight the strength of ML models in capturing non-linear subsurface relationships critical to reservoir characterization.