Ensemble machine learning: An untapped modeling paradigm for petroleum reservoir characterization

For help in preparation of this article, thanks to Amy Andrews, Karen Glaser, Ralf Heidler, Martin Isaacs and Jack LaVigne, Sugar Land, Texas, USA; Bill Batzer and Claude Signer, Ridgefield, Connecticut, USA; Tim Diggs, Shell International Exploration and Production Inc., Houston, Texas; Nick Heaton, The Hague, The Netherlands; David C. Kopaska-Merkel, Geological Survey of Alabama, Tuscaloosa, Alabama, USA; Bill Murphy and Bruce Ward, Earthworks LLC, Sandy Hook, Connecticut; Richard Netherwood, Cairo, Egypt; and Rachel Wood, Cambridge, England. CMR (Combinable Magnetic Resonance), ELANPlus (Elemental Log Analysis), FMI (Fullbore Formation MicroImager), GeoFrame, MDT (Modular Formation Dynamics Tester), MRF (Magnetic Resonance Fluid), MRX (Magnetic Resonance eXpert), Platform Express, proVISION and Well-ID (Well-Informed Decisions) are marks of Schlumberger. Windows is a registered trademark of Microsoft Corporation. Heterogeneity of carbonate reservoirs at all scales makes it difficult for experts

Confronting the carbonate conundrum

Ensemble model of non-linear feature selection-based Extreme Learning Machine for improved natural gas reservoir characterization

Development of a Papua New Guinean onshore carbonate reservoir: A comparative borehole image (FMI) and petrographic evaluation

Investigating NMR-based absolute and relative permeability models of sandstone using digital rock techniques

Using high-resolution electrical borehole images and other open-hole log data in 13 wells, a case study was conducted to analyze the porosity and permeability heterogeneity in the Means field in the Permian Basin, west Texas. Because of negligible clay and other conductive minerals in the predominantly dolomitic rocks, the electrical images are converted into microporosity maps of the boreholes using a modified Archie equation. The vug porosity is then partitioned from the total porosity by analyzing the porosity spectra around the boreholes in the sliding windows of 1-in. (2.5-cm)-thick interval. A practical workflow was also developed to compute a sonic vug-porosity index using the sonic, neutron, and density logs. The two vug porosities based on the independent measurements support each other in the Means field except that the sonic vug-porosity index has much lower resolution than the image-derived porosity. Through iterative experiments, a synthetic permeability model is established using the total porosity and vug-porosity logs. The core permeability in the vuggy dolomites varies exponentially with the product of the vug-porosity and vug-connectivity factor. Thin superpermeable vuggy zones below the resolution of the conventional logs are revealed by the electrical images and the permeability quantification. These superpermeable zones sandwiching thick oil-bearing reservoir rocks are the primary cause for the early high water cut and the bypassed oil. Integration of the continuous high-resolution permeability logs, the petrophysical facies from supervised neural network (SNN) processing, and the production data from both producers and injectors provide valuable insights to reservoir modeling.

Chunming Xu

Papers

Porosity Partitioning and Permeability Quantification in Vuggy Carbonates, Permian Basin, West Texas, U.S.A.