Title: Use of Neural Networks and Hybrid Approaches in Reservoir Property Modeling
Speaker: Fred Aminzadeh, 2500 Tanglewilde, Suite 120, Houston, TX 77063
Date: August 21, 2002
Publication: The Outcrop, August 2002, p. 18-19
I will describe a hybrid method (statistics and physics based) for integrating geoscience and petrophysics data. It uses a geological framework for pattern recognition and to perform segmentation, classification and quantification. Non-linear relations between various data types are established at natural scale levels defined by the integration framework. This geologically driven integration method is used for horizon-based and volume-based stratigraphic and facies analysis, as well as reservoir property prediction.
I will demonstrate how we complement conventional attribute analysis with the unique seismic character analysis. Further, I will show how attributes and/or waveforms extracted from multiple input seismic cubes arc used to improve prediction power and confidence in the estimates. Unique in the method is the pseudo-wells used to relate seismic patterns to the underlying rock and reservoir properties. This is accomplished through training of “supervised neural networks.” The pseudo-well simulator generates stratigraphic columns with the corresponding well logs using a constrained Monte Carlo simulation. For each pseudo-well, synthetic seismograms (both post stack and pre-stack) are generated, yielding a fully integrated data set for quantitative interpretation.
To highlight the approach, case history examples will be given for waveform segmentation, volume trans-formation (porosity/litho class), pseudo well generation, reservoir fluid detection, and 4-D analysis. Some of the new concepts on prediction of % gas saturation using “density volume” and associated confidence levels and gas probabilities will be shown. A new time-lapse analysis method based on the integration of multitude of attributes to highlight dynamic changes of the reservoir will be demonstrated. Finally, dGBs d-Tect package for determining fluid/gas migration path, calculation of fault/fracture cubes, and gas chimney/ hydrate cubes, will be highlighted.
Mr. Fred Aminzadeh, Distinguished Lecturer
Fred Aminzadeh is president and CEO of dGB-USA, specializing in services and software for quantitative seismic modeling, inversion and interpretation, as well as neural networks-based reservoir characterization and anomaly detection. Aminzadeh previously worked for Unocal with both technical and management responsibilities. He has a Ph. D. from the University of Southern California. His thesis dealt with the AVO modeling of layered earth media. He is a member of Russian Academy of Science, Azerbaijan Oil academy, and National Research Council’s Committee on Seismology. He has an extensive list of publications in diverse areas including nine books on topics such as reservoir characterization, seismic data processing, expert systems, pattern recognition and 3-D modeling. He also holds three US patents on AVO modeling, seismic while drilling and pre-stack attribute integration using neural networks. He has served as the chairman of the Society of Exploration Geophysicists Research Committee. Currently, he is Vice President of the SEG.