Using a Genetic Algorithm in Constructing Ensemble Combinations of Neural Networks to Estimate Reservoir Rock Porosity

Artificial neural networks (ANNs) can be used for reservoir characterization. Ensemble combination of ANNs, which is a type of committee machine having parallel structure, are used for reservoir characterization and may lead to better results compared to single ANNs. In this research, ensemble combinations of single ANNs were used to estimate Kangan gas reservoir rock porosity in Southern Pars hydrocarbon field, in Iran. Well logging data, related to Kangan formation, were given as the input of the networks whiles the porosity data were considered as the output. Back propagation method was used for training single networks having different structures. Then, 7 networks, which had the best results, i.e. contained minimum mean square error (MSE) in the test step, were selected. A genetic algorithm was used to obtain the coefficients of the ensemble combinations of the selected single networks. Thus, we obtained optimal linear combinations of the networks, in which maximum reductions in MSE of the test step compared to the best single networks were achieved. The best ensemble combination, compared to the best single neural network, reduced MSE in the training and test steps 14.4% and 12.5%, respectively.