Exploration and Production Technologies
An Intelligent Systems Approach to Reservoir Characterization

DE-FC26-03NT41629

Goal:
The goal of this project is to improve the ability to characterize reservoirs through the use of high-resolution seismic data.

Background:
This project will employ intelligent systems (a combination of a neural network, genetic algorithms and fuzzy set theory) to produce well log data from surface seismic data. A series of 2D vibroseis lines, numerous borehole logs, and a 3-component VSP from one well in the Granny Creek oil field in central West Virginia will be used to test the models developed during this project. The project’s three-step approach involves: (1) processing the seismic data and accompanying well log data, (2) constructing, training, and calibrating an intelligent system to simulate the well-log reflectivity response from seismic traces, and (3) inverting the seismic data to logs using the intelligent system, outside of the training area.

Performer:   West Virginia University – Project management and all research products

Location:
Morgantown, WV

Project Impact:
The project endeavors to build a correlation between high-resolution seismic data and wireline logs using intelligent systems to bridge the resolution gap between the two data sets. The end result will be an ability to simulate well-log response from seismic traces in the vicinity of a proposed well location.

Results:

  • Transferred in-house seismic data from tapes, computed seismic attributes and selected seismic/well log data sets,
  • Performed specialized processing of seismic data and acquired additional well logs data to enhance data sets,
  • Used geological modeling software to create a 2D geologic model of the Granny Creek field and produced synthetic seismic and synthetic logs to be used for the preliminary design of the neural network,
  • Carried out statistical data preparation and pre-processing for use in intelligent system design,
  • Began coding a customized neural network application,
  • Compared trained system with well log data and seismic to just seismic interpretation and results showed better resolution of reservoir character,
  • Detail evaluation complete.
The preliminary design of the neural network was based on a synthetic model of the Granny Smith field. Several possible neural network designs were considered and the choice of a Recurrent Neural Network (RNN) was based on the fact that such networks are most successful in building models that have characteristics of time series. Using a commercial neural network application, simple back propagation and RNNs were applied to the data and successfully used predict VSP attributes from 2D seismic.

The complexity of the problem warranted a high degree of flexibility in the neural network application that was being developed. The lack of such flexibility in off-the-shelf neural network applications prompted the coding of a customized application for the types of networks that will be used in the project.

Current Status and Remaining Tasks: 
This project has been completed

Project Start: September 5, 2003
Project End: September 30, 2005

DOE Contribution: $ 248,514
Performer Contribution: $ 95,060

Contact Information:
NETL – Thomas Mroz (thomas.mroz@netl.doe.gov or 304-285-4071)
WVU – Shahab Mohaghegh (sdmohaghegh@mail.wvu.edu or 304-293-7682 x3405)

Additional Information:
Project results - West Virginia University project website [external site]

Pertinent Publications:
Sanchez, A., Toro, J., T. Wilson, T., and Mohaghegh, S. D., 3D Seismic Interpretation and Modeling of the Atoka-Morrow Sequence in the Buffalo Valley Field, Delaware Basin, NM for Reservoir Characterization Using Neural Networks, AAPG, Eastern Section, Oct. 3-6, 2004, Columbus, OH.

Artun, E.: ‘Reservoir Characterization with Intelligent Seismic Inversion', presented at the SPE Rocky Mountain/Mid-Continent Region Student Paper/Presentation Contest, 21 April 2005, Butte, MT. (Poster is attached)

Sanchez, Alejandro: ‘3D Seismic Interpretation and Synthetic Modeling of the Atoka And Morrow Formations, in the Buffalo Valley Field (Delaware Basin, N.M, Chaves Co.) for Reservoir Characterization Using Neural Networks', M.S. Thesis, West Virginia University.

Artun, Emre: ‘Reservoir Characterization with Intelligent Seismic Inversion', M.S. Thesis, West Virginia University, August 2005. (Abstract is attached)

Artun, E., Mohaghegh, S.D., Toro, J., Wilson, T., Sanchez, A.: ‘Reservoir Characterization with Intelligent Seismic Inversion', paper SPE 98012, to be presented at the SPE Eastern Regional Meeting, 14-16 September 2005, Morgantown, WV. (Abstract is attached)

Artun, E., Mohaghegh, S.D., Toro, J., Wilson, T., Sanchez, A.: ‘Intelligent Seismic Inversion: From Surface Seismic to Well Logs via VSP', to be presented at the AAPG Eastern Section 34th Annual Meeting, 18-20 September 2005, Morgantown, W.V. (Abstract is attached)

 
StayConnected Facebook Twitter LinkedIn RssFeed YouTube