--> ABSTRACT: Use of Artificial Neural Nets to Predict Permeability in Hugoton Field, by Keith A. Thompson, Mark H. Franklin, and Terrilyn M. Olson; #91019 (1996)

Datapages, Inc.Print this page

Use of Artificial Neural Nets to Predict Permeability in Hugoton Field

Keith A. Thompson, Mark H. Franklin, and Terrilyn M. Olson

One of the most difficult tasks in petrophysics is establishing a quantitative relationship between core permeability and wireline logs. This is a tough problem in Hugoton Field, where a complicated mix of carbonates and clastics further obscure the correlation.

One can successfully model complex relationships such as permeability-to-logs using artificial neural networks. Mind and Vision, Inc.'s neural net software was used because of its orientation toward depth-related data (such as logs) and its ability to run on a variety of log analysis platforms. This type of neural net program allows the expert geologist to select a few (10-100) points of control to train the "brainstate" using logs as predicters and core permeability as "truth".

In Hugoton Field, the brainstate provides an estimate of permeability at each depth in 474 logged wells. These neural net-derived permeabilities are being used in reservoir characterization models for fluid saturations.

Other applications of this artificial neural network technique include deterministic relationships of logs to: core lithology, core porosity, pore type, and other wireline logs (e.g., predicting a sonic log from a density log).

AAPG Search and Discover Article #91019©1996 AAPG Convention and Exhibition 19-22 May 1996, San Diego, California