oa Application of Artificial Neural Networks in Water Saturation Prediction in from Iranian Oil Field

In this paper, Artificial Neural Networks (ANNs) has been implemented to calculate water saturation in<br>Gadvan formation in an Iranian oil field. Core data from 3 wells, (AZN01, AZN03 and AZN04), have<br>been used to train and test the Network. The procedure was to use wireline logs such as deep<br>resistivity log (Rt), density log (RHOB), sonic log (DT), gamma ray (GR) and porosity recorded from<br>cores to be used as input to ANNs and water saturation (Sw) measured in laboratory as target. Four<br>networks generated, and then the networks were trained with random 60% of input data and tested<br>with random 20% of data. To make sure of networks well performance, the remained 20% of data<br>were used to validate the network’s ability of determining an acceptable relationship between inputs<br>and target (core Sw). A network with the least prediction error was selected to be used as our network<br>of interest. This network showed the correlation coefficient of 0.979 between the ANNs-predicted water<br>saturation (output) and target (core-derived Sw) data. Using this network, water saturation of Gadvan<br>formation was predicted. ANNs-predicted water saturation revealed lower values of water saturation in<br>the lower part of Gadvan formation in comparison with water saturation computed using Archie’s relationship.