--> Abstract: Incorporating Geological Information to Stabilize Neural-Network Pattern Recognition in Fluvial Deposits, by J. Skolnakorn, E. Von Lunen, J. L. Baldwin, C. Kilic, and R. E. Barba; #90937 (1998).

Datapages, Inc.Print this page

Abstract: Incorporating Geological Information to Stabilize Neural-Network Pattern Recognition in Fluvial Deposits

 SKOLNAKORN, JIRAPA, ERIC VON LUNEN, Bureau of Economic Geology, The University of Texas at Austin; JEFFREY L. BALDWIN, Mind & Vision Computer Systems; CEM KILIC and ROBERT E. BARBA, Bureau of Economic Geology, The University of Texas at Austin 

Lithology and porosity determination for reservoir characterization is essential for hydrocarbon reserve estimation. In fluvial-deltaic systems the complex depositional layering often precludes reliable selection of petrophysical parameters such as shale-sandstone points and porosity relations in crossplots. Neural-network pattern recognition techniques can be used to improve petrophysical evaluation in complex and heterogeneous lithology. Neural-network interpretive models thus generated can be improved and made more stable if the interval of investigation is subdivided using key stratigraphic surfaces such as major unconformities and flooding surfaces.

Prior to the inclusion of basic sequence surfaces the neural network prediction confidence was below 80%, with wide variance in comparison to known core calibration examples. In a case study example from the Vienna Basin, which consisted of electric-log and core data sets, a neural-network analysis was contrasted with a limited number of open-hole petrophysical log evaluations for porosity and lithology volumetrics. A second case study from the Maracaibo Basin consisted of open-hole petrophysical suites with volumetric analysis and core data but with a greater number of porosity logs. After sequence stratigraphic boundaries were included, the neural network's lithology and porosity predictions were increased to confidence levels exceeding 95%.

These two examples demonstrated that neural networks could successfully be applied to petrophysical evaluations if properly constrained within major stratigraphic packages. Widespread use of this approach in reservoir characterization studies will improve petrophysical evaluation and gross sandstone-porosity volumetric mapping. Hence, neural network interpretive models ultimately reduce finding costs of petroleum in mature fields.

AAPG Search and Discovery Article #90937©1998 AAPG Annual Convention and Exhibition, Salt Lake City, Utah