A Deterministic Method for Calculating Facies from Wireline Logs
Robert Y Elphick: Scientific Software-Intercomp, Denver, Colorado, David
L. Carr and Ronald A. Johns: Bureau of Economic Geology, Austin, Texas,
David E. Lancaster: S.A. Holditch & Associates, Inc., College Station,
Texas
Abstract: When facies vary widely across a field, defining well-to-well stratigraphic correlations can be difficult. During the secondary gas recovery (SGR) study of Boonsville field, workers encountered a complex stratigraphic situation in which finding correct well-to-well correlations was essential to identifying reservoir compartments within the Bend conglomerate, Fort Worth basin, Texas. Sequence stratigraphy methods require identification of key time surfaces (that is, areally consistent flooding surfaces and erosional surfaces) before full depositional cycles can be identified and interpreted. Correlations are traditionally done by hanging all, or some of, the wireline curves on an interpreted stratigraphic datum in a cross section and identifying curve patterns that belong to given facies. This identification is easier when the logs are of similar types and are presented on similar scales. When logs are from different vintages (from the 1940s to the present in the case of Boonsville field), are of different types, or require different scales and presentations, then the task of correlation becomes difficult _ sometimes impossible. Identifying various facies by curve pattern alone, even when they have consistent well log curves, is often difficult.
Pattern-recognition software designed to identify facies has existed for a number of years. Most of these packages, which involve sophisticated stochastic techniques, typically provide impressive results when applied by skilled users. When the SGR analysis team used a principal component analysis technique, the results were good when applied to well logs of similar curve suites but were inconsistent when applied to wells that were nearby but had radically different log suites. A deterministic method for identifying facies from logs was therefore developed for use in the Boonsville Bend conglomerate, and out of several different approaches one system was chosen. The final system consisted of five different computational models, all using the same numerical technique but different wireline curve suites.
Transformation of Geochemical Log Datato Mineralogy Using Genetic
Algorithms
J. H. Fang: Department of Geology, University of Alabama , C. L. Karr:
Department of Engineering Science and Mechanics, University of Alabama
, D. A. Stanley: US Bureau of Mines, Tuscaloosa, Alabama
Abstract: A technique based on a genetic algorithm (GA) is proposed to transform geochemical data into a rock's mineral composition. Genetic algorithms are a family of combinatorial optimization methods that emulate the principle of survival of the fittest found in nature. These algorithms solve various data fitting, parameter estimation, and optimization/search problems from a wide variety of scientific fields. Genetic algorithms first generate a random population of trial solutions_in the context of this paper solutions are the relative amounts of minerals in the rock_and evolve toward better solutions by mimicking the genetic process of natural selection. A set of hypothetical rocks is used to demonstrate the power of the method.
Virtual Measurement of Heterogeneous Formation Permeability Using
Geophysical Well Log Responses
Shahab Mohaghegh and Sam Ameri: Petroleum & Natural Gas Engineering,
West Virginia University Reza ArefiLCC, Inc.
Abstract: Virtual measurement is a term that describes indirect and inferential determination of parameters that are usually measured in the laboratory. Unlike conventional measurement, which requires actual rock samples and a laboratory with specialized equipment, virtual measurement uses indirect information, such as geophysical well log responses, to obtain laboratory measurable formation parameters. Permeability is the most important reservoir parameter that can be virtually measured using information provided by well logs. Conventional predictions of permeability from logs are tied to explicit functions from engineering theory or statistical models, whose forms are assumed to be both known and universal. In cases where the interrelations between the dependent (permeability) and the independent variables (well logs) are not well understood, this modeling approach often performs poorly. By contrast, a virtual measurement technique requires no specific mathematical function and can be implemented by an artificial neural network. Neural networks can be trained using the available data to learn their interrelations automatically. In cases where there is a high information content on permeability embedded within the input logging data, virtual measurement can achieve results with an accuracy comparable to actual, physical laboratory measurement of permeability but without the high cost. In a case study on a highly heterogeneous formation in West Virginia, virtual measurement showed promising results.