Neural computing has made a major step forward by the introduction of multi-net systems in practical applications. In this study we developed and tested a modular artificial neural network system for predicting underground fluids, water, oil and gas, and their partial saturations, directly from well logs, without explicit knowledge of the fluid and rock properties as required by conventional methods. Based on laboratory data on relative permeability for alternative fluid systems –oil–water or gas–oil – respectively, relative permeability logs may also be provided for input to reservoir simulation while drilling.
Simple three-layer back-propagation artificial neural networks (ANN) constitute the building blocks of a modular system, where the input logs are sonic, density, neutron porosity and resistivity. By numerical experiments using synthetic logs we have determined the optimal architecture of the ANN. We find that the overtraining strategy is a suitable technique for bias reduction and an unconstrained optimal linear combination is the best method of combining outputs in the committee neural net. The accuracy of the net is restricted only by accuracy of data. Comparison between ANN predictions of fluid saturation with those of conventional petrophysical analysis, in wells unknown to the network, indicates a standard error of less than 0.03.