--> ABSTRACT: Application of Neural Networks to Correlate Core Descriptions of Depositional Environments with Log Responses, and to Extrapolate Over Intervals and Wells Where Cores Do Not Exist, by Michael Holmes, Dominic Holmes, and Tony Holmes; #90906(2001)

Datapages, Inc.Print this page

Michael Holmes1, Dominic Holmes1, Tony Holmes1

(1) Digital Formation, Inc, Denver, CO

ABSTRACT: Application of Neural Networks to Correlate Core Descriptions of Depositional Environments with Log Responses, and to Extrapolate Over Intervals and Wells Where Cores Do Not Exist

Neural networks approaches can be used to correlate log responses with core descriptions of lithology and/or depositional environments. A series of training sessions are run over intervals where both core data and a complete log suite exists. Any bad-hole intervals should be excluded in the training process. The training can be over a single or multiple wells, and the separate wells do not necessarily need to have the same suites of rock types. However, they do need to have identical log suites.

Once the training has been established, applications can be run over intervals or wells where no core data exists. If rock types have changed from the intervals of training, the results will be unreliable, and the neural network output will indicate these levels of change.

Examples from both sandstone and carbonate reservoirs are presented. The techniques allow for a number of important reservoir interpretations:

o Field-wide recognition of changing depositional and/or rock type sequences

o Correlation of specific rock types with reservoir quality

o Recognition of transgressive/regressive sequences through the gross intervals, well-by-well

o Detailed field-wide mapping of each different rock type

AAPG Search and Discovery Article #90906©2001 AAPG Annual Convention, Denver, Colorado