USE OF NEURAL-NETWORK METHODS TO PREDICT POROSITY AND PERMEABILITY OFA PETROLEUM RESERVOIR

Accurate determination of porosity and permeability values at given we ll locations is a central problem in petroleum reservoir characterizat ion. Identification of individual rock types, or lithofacies, prior to calculating porosity-permeability data can lead to improved estimates . We used the genetic approach in predicting porosity and permeability values from wireline logs and lithofacies information in reservoirs, using a back-propagation neural network method. In order to reproduce the fine-scale variability known to exist in core porosity-permeabilit y data, separate neural nets were used for porosity, followed by perme ability prediction. A simulation technique for adding fine-scale noise was also used. For the reservoir data considered, the fine-scale simu lation approach combined with the use of neural networks provides real istic and accurate porosity-permeability predictions when compared to the core data.