Neural Networks for Hydrocarbon Net Pay Prediction
Back to Applications
Principal Contact Person and Organization (including e-mail address):
Jorge A. Pita Mobil Technology Company japita@dal.mobil.com
Brief Description of Application:
Suite of HPF routines that correlates 3D attributes generated from depth seismic data with
borehole rock and fluid properties via neural networks, which compete to generate the best
predictor of hydrocarbon distribution for the reservoir after passing a bootstrap validation (or
cross-validation) process. The predictions are used to select drilling targets, quantify oil reserves
and provide detailed stratigraphic models for reservoir simulation.
Number of Lines of Code: 2350
Target Platforms and HPF Compilers Used:
Cray T3E, Portland Group HPF Compiler.
Coding Styles (data decompositions, computational methods):
Data Parallel model wherever possible. !HPF INDEPENDENT a useful directive in many loops.
Extrinsic Interfaces Used (and reasons):
Please comment on any aspects of the application that might be interesting, including any problems using HPF effectively:
Efficient/fast parallel I/O is very important for this and other applications involving 3D seismic data and its attributes.