<p>Petroleum reservoir characterization requires reliable methods for fluid identification, yet traditional petrophysical approaches often face limitations. We evaluate whether fractal and multifractal attributes of well log variability, coupled with partial least squares discriminant analysis (PLS-DA), can improve oil–gas discrimination. We analyzed 1085 well logs from 93 wells across seven Algerian basins spanning gamma ray, acoustic, bulk density, neutron porosity, resistivity, photoelectric factor, and gamma-ray spectroscopy. Fractal features included box and regularization dimensions have been computed; multifractal attributes included <i>α</i><sub>min</sub>, <i>α</i><sub>peak</sub>, <i>α</i><sub>max</sub>, width, <i>f</i>(<i>α</i><sub>min</sub>), <i>f</i>(<i>α</i><sub>peak</sub>), <i>f</i>(<i>α</i><sub>max</sub>), left/right heights, and <i>C</i>-value have been extracted. These quantitative attributes formed the predictor matrix for a PLS-DA classifier; model quality was assessed using <i>R</i><sup>2</sup><i>X</i><sub>cum</sub>, <i>R</i><sup>2</sup><i>Y</i><sub>cum</sub>, <i>Q</i><sup>2</sup><sub>cum</sub>, ROC/AUC, and confusion matrices. Key attributes showed the strongest associations with fluid type. The PLS-DA model achieved high predictive and discriminative performance with <i>R</i><sup>2</sup><i>Y</i><sub>cum</sub> &gt; 0.90, <i>R</i><sup>2</sup><i>X</i><sub>cum</sub> &gt; 0.85, <i>Q</i><sup>2</sup><sub>cum</sub> &gt; 0.80, ROC/AUC&#xa0;≈&#xa0;0.98, and overall accuracy &gt; 90%, with higher sensitivity for gas-bearing intervals than for oil-bearing zones. VIP rankings were stable across folds, and the classification function yielded an interpretable per-interval score by aggregating multiscale attributes, indicating that the selected attributes are robust indicators for discriminating oil from gas. Combining fractal and multifractal attributes with PLS-DA provides a robust and interpretable approach for fluid discrimination. This framework enhances well log interpretation, offering a cost-effective complement to traditional reservoir characterization techniques.</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

Fractal and multifractal analysis of well logs coupled with PLS-DA for fluid characterization in petroleum reservoirs

  • Abdelbasset Boulassel,
  • Soraya Makhlouf,
  • Zinelaabidine Boumelit,
  • Amar Boudella,
  • Fethi Ali Cheddad,
  • Saïd Gaci,
  • Abdennour Akli,
  • Mouad Kebir,
  • Salah Boufenchouche,
  • Badis Zegagh,
  • Ayoub Tedjani,
  • Anis Reghis

摘要

Petroleum reservoir characterization requires reliable methods for fluid identification, yet traditional petrophysical approaches often face limitations. We evaluate whether fractal and multifractal attributes of well log variability, coupled with partial least squares discriminant analysis (PLS-DA), can improve oil–gas discrimination. We analyzed 1085 well logs from 93 wells across seven Algerian basins spanning gamma ray, acoustic, bulk density, neutron porosity, resistivity, photoelectric factor, and gamma-ray spectroscopy. Fractal features included box and regularization dimensions have been computed; multifractal attributes included αmin, αpeak, αmax, width, f(αmin), f(αpeak), f(αmax), left/right heights, and C-value have been extracted. These quantitative attributes formed the predictor matrix for a PLS-DA classifier; model quality was assessed using R2Xcum, R2Ycum, Q2cum, ROC/AUC, and confusion matrices. Key attributes showed the strongest associations with fluid type. The PLS-DA model achieved high predictive and discriminative performance with R2Ycum > 0.90, R2Xcum > 0.85, Q2cum > 0.80, ROC/AUC ≈ 0.98, and overall accuracy > 90%, with higher sensitivity for gas-bearing intervals than for oil-bearing zones. VIP rankings were stable across folds, and the classification function yielded an interpretable per-interval score by aggregating multiscale attributes, indicating that the selected attributes are robust indicators for discriminating oil from gas. Combining fractal and multifractal attributes with PLS-DA provides a robust and interpretable approach for fluid discrimination. This framework enhances well log interpretation, offering a cost-effective complement to traditional reservoir characterization techniques.