Neural network prediction of carbonate lithofacies from well logs, Big Bow and Sand Arroyo Creek fields, Southwest Kansas
Section snippets
Introduction and objective
Since its discovery over 50 years ago, hydrocarbon production from St. Louis carbonate shoals in the Hugoton Embayment of southwestern Kansas has increased significantly, especially over the last decade (Morrison et al., 2002). The application of recent technology has had a great impact on exploration success for these stratigraphic traps (e.g., three-dimensional (3D) seismic). However, the complexities of stratigraphy and relatively thin reservoir intervals (generally less than 4.5 m) pose a
Geological background
The reservoir rock at Big Bow and Sand Arroyo Creek fields is composed of the St. Louis Limestone (Meramecian Series). In the Hugoton Embayment, the St. Louis Limestone comprises approximately 200 ft (60 m) of the Mississippian section. The St. Louis thickens southward into the Anadarko Basin. It is erosionally truncated to the north and east across the Las Animas Arch and Central Kansas uplift (Fig. 1). The upper part of the St. Louis Limestone which is the focus of this paper has been named the
Methods and approach
The thin nature and lateral variability of subtidal grainstone of the St. Louis makes it challenging to recognize lithofacies patterns and correlate between wells since it is much more difficult to recognize carbonate lithofacies using wireline logs than siliciclastics because of the difficulty in recognizing micrite (carbonate mud). Neural network analysis provides a method to extract additional information from the digital well logs in order to better recognize lithofacies changes. Artificial
Available data
Digital well logs were assembled from approximately 100 wells in and surrounding Big Bow and Sand Arroyo Creek fields. The minimum suite of digital well logs used for neural network lithofacies prediction consists of gamma-ray (GR), resistivity (deep and medium), density and neutron porosity. Photoelectric (PE) logs were run in a large number of the wells, and were used where available. A total of 750 ft of cored section of the St. Louis Limestone from 10 wells was described (Qi and Carr, 2005a
Lithofacies classification and environments
Six major lithofacies were recognized on half-foot intervals in cores of the St. Louis Limestone. The lithofacies classification, detailed core description, and interpretation of depositional environments are similar to previous work (Handford, 1988; Handford and Francka, 1991; Carr and Lundgren, 1994; Abegg, 1994; Abegg et al., 2001; Abegg and Handford, 2001). The six major lithofacies are: (1) quartz-rich carbonate grainstone deposited by eolian processes adjacent to a coastline; (2)
Statistical data analysis
GR, resistivity logs (deep-ILD and medium-ILM), neutron porosity (Nphi), density porosity (Dphi), and PE factor are commonly available electric logs for wells in Big Bow and Sand Arroyo Creek fields. Basic 1D, 2D, and multidimensional statistical data analyses were performed with the listed curves in the cored wells, and the results were plotted as histogram, dotplot, boxplot, and cross plot to evaluate the predictor variables for neural network modeling.
The histogram plot of GR shows that the
Model building and cross-validation
The basic statistical analysis between digital well logs and lithofacies described from core of 10 wells in Big Bow and Sand Arroyo Creek fields illustrates the limitation of traditional statistical methods. Six well logs (GR, ILM, ILD, Nphi, Dphi, PE curves) were selected as input variables with digitized lithofacies categories to develop a standard single hidden-layer neural network model. The single-layer “back-propagation” neural network was used to predict lithofacies from digital well
Lithofacies prediction
After selecting the neural network parameters through the cross-validation and model testing, the final step is to run neural network models on wells with LAS files and output the predicted lithofacies. LAS log files were batch processed in Kipling, and LAS-formatted prediction files were generated after prediction. Seven curves are contained in the output predicted lithofacies file. For the 10 cored wells six probability curves for lithofacies and one discrete lithofacies curve representing
Cross-sections with interpolated lithofacies
In 10 cored wells, lithofacies curves were composed of description from cored intervals and predicted lithofacies values from neural network for uncored intervals. Lithofacies curves were predicted in approximately 90 wells with available well log suites using the optimal neural network models. Adding the cored wells, a total of 100 wells with lithofacies curves were used to construct 17 stratigraphic cross-sections of St. Louis Limestone in Big Bow and Sand Arroyo Creek fields (Qi and Carr,
Discussion
Compared to conventional linear methods, neural network models provide improved prediction results using training patterns gained from learning and exploring the hidden information within the multi-dimensional data set. As a non-parametric method, neural network analysis is not based on restrictive assumptions about the probability distributions of input variables. Although it is unnecessary to examine the correlation and distribution of input variables when using a neural network, basic
Conclusions
Neural network analysis coupled with detailed lithofacies descriptions is applied to construct improved finer-scale facies models, and provide a better understanding of the relatively thin oolitic grainstone shoal reservoirs in the St. Louis Limestone, of Southwest Kansas. A total of 750 ft of St. Louis Limestone intervals in 10 cored wells were described and six lithofacies were classified and treated as training data with corresponding suites of well logs. Correlations and relationships of
Acknowledgements
We thank the Kansas Geological Survey for financial support as part of Lianshuang Qi's dissertation research. Special thanks to Geoff Bohling for his assistance and comments on the manuscript. Thanks to Marty Dubois for his discussions and comments. Reviewers Rick Abegg and Jeffery Yarus provided significant input that improved the manuscript. BP-American provided the donated core material. We also would like to thank Geoplus Corporation for providing access to Petra.
References (29)
- et al.
Using artificial intelligence to predict permeability from petrographic data
Computers & Geosciences
(2000) - Abegg, F.E., 1991. Sedimentology and lithostratigraphy of the Upper Mississippian Ste. Genevieve and St. Louis...
- Abegg, F.E., 1994. Lithostratigraphy of the Hugoton and Stevens members of the St. Louis Limestone and the Ste....
- Abegg, F.E., Handford, C.R., 2001. Deflation origin of Mississippian carbonate eolianites of Southwestern Kansas. In:...
- Abegg, F.E., Loope, D.B., Harris, P.M., 2001. Deposition and diagenesis of carbonate eolianites. In: Abegg, F.E.,...
- et al.
The New S Language: a Programming Environment for Data Analysis and Graphics
(1988) - Bohling, G.C., Doveton, J.H., 2000. Kipling.xla: an Excel add-in for nonparameter regression and classification....
- Bohling, G.C., Dubois, M.K., 2003. An integrated application of neural network and Markov chain techniques to...
- et al.
Use of gamma ray spectral log to recognize exposure surfaces at the reservoir scale: Big Bow Field (St. Louis Limestone), Kansas
- et al.
Statistical Models in S
(1992)
Pattern Classification
Classification of carbonate rocks according to depositional texture
Mississippian rocks of western Kansas
American Association of Petroleum Geologists Bulletin
Cited by (93)
Classification of shale lithofacies with minimal data: Application to the early Permian shales in the Ordos Basin, China
2024, Journal of Asian Earth SciencesApplication of unsupervised learning and deep learning for rock type prediction and petrophysical characterization using multi-scale data
2023, Geoenergy Science and EngineeringPerformance evaluation of machine learning-based classification with rock-physics analysis of geological lithofacies in Tarakan Basin, Indonesia
2022, Journal of Petroleum Science and EngineeringImputation of missing well log data by random forest and its uncertainty analysis
2021, Computers and GeosciencesDetermination of the elastic parameters of a VTI medium from sonic logging data using deep learning
2021, Computers and Geosciences