An enhanced approach to predict permeability in reservoir sandstones using artificial neural networks (ANN)

A key challenge in the oil and gas industry is the ability to predict key petrophysical properties such as porosity and permeability. The predictability of such properties is often complicated by the complex nature of geologic materials. This study is aimed at developing models that can estimate permeability in different reservoir sandstone facies types. This has been achieved by integrating geological characterization, regression models and artificial neural network models with porosity as the input data and permeability as the output. The models have been developed, validated and tested using samples from three wells and their predictive accuracy tested by using them to predict the permeability in a fourth well which was excluded from the model development. The results indicate that developing the models on a facies basis provides a better predictive capability and simpler models compared to developing a single model for all the facies combined. The model for the combined facies predicted permeability with a correlation coefficient of 0.41 which is significantly lower than the correlation coefficient of 0.97, 0.93, 0.99, 0.96, 0.96 and 0.85 for the massive coarse-grained sandstones, massive fine-grained sandstones-moderately sorted, massive fine-grained sandstones-poorly sorted, massive very fine-grained sandstones, parallel-laminated sandstones and bioturbated sandstones, respectively. The models proposed in this paper can predict permeability at up to 99% accuracy. The lower correlation coefficient of the bioturbated sandstone facies compared to other facies is attributed to the complex and variable nature of bioturbation activities which controls the petrophysical properties of highly bioturbated rocks.