Reservoir Porosity Prediction from Well Logs Using Neural Networks

Abstract

This paper presents neural networks (NN) model for one of the heterogeneous carbonate reservoir in the south of Iraq.

Because of complexity of carbonate rock (Mishrif formation/South of Iraq),porosity can be identified as a complex function of many independent parameters. The ideal results are not expected by using conventional techniques with log analysis, therefore, using the new technique utilizing neural network (NN) model included here to obtain more reliable results.

In order to create this model, there are three layer back propagation network should be followed. Initially Gamma ray (GR), Spontaneous potential (SP), Neutron, Density, Sonic, and Resistivity were applied as input data. Optimization between the estimated and measured porosity from core analysis is reached, and finally, the results has been compared and showed that the accuracy of the model has been improved.

The porosity NN is executed by MATLAB(8.6), with log analysis as input data for one cored well ,Tuba-5. The data for this well was used for training and subsequently prediction and verification was done on the same well. However, porosity was estimated from conventional well logs using intelligent technique (neural network), and this technique is more accurate and reliable for estimating reservoir porosity  compared with classical methods.