Prediction of Lithofacies Types and Qualities Using a Distributed Neural Network

A distributed neural network, running on a large transputer-based parallel computer, was trained to identify the presence of the main lithographical facies types in a particular oil well, using only the readings obtained by a log probe. The resulting trained network was then used to analyse a variety of other wells, and showed only a small decrease in accuracy of identiication. Geologists classify well structures using rock and fossil samples in addition to the log data that was given to the network. Results are given here for the accuracy with which the learned network agreed with analyses performed by geologists. The study was then extended into two more areas, rstly to investigate the network's success in predicting physical attributes of the rocks, e.g. porosity and permeability, and secondly to investigate the ability of similar networks to isolate particular geological features.