The accuracy of the relative permeability curves and rock wettability are of paramount importance in terms of both reducing costs and uncertainty of the assessed reservoirs. They describe the flow of different fluid phases in any reservoir and directly relate the distribution of fluid phases present, including the recovery of hydrocarbons. Therefore, the relative permeability curves along with the wetting preferences of the rock-fluid system define the reservoir flow mechanics, and hence the economics of any field development. Relative permeability data either conventionally obtained from core analysis or approximated from a number of conventionally established correlations depending on the formation characteristics. The aim of this paper is to provide a new method for accurately calculate the relative permeability and wettability in un-cored sandstone reservoirs in Gulf of Suez.
An Artificial Neural Network (ANN) model based on the back-propagation technique trained with the input parameters from relative permeability curves generated from lab experiments. The extracted cores from several reservoirs in Gulf of Suez covers an extensive range of porosity and permeability from different sandstone lithology having diverse wettability. The generated model then tested with input parameters such as the initial connate water saturation (Swc) and the residual oil saturation (Sor).The model predictions define the relative permeability end-points and the intersection point to quantify the wettability and the shape of the relative permeability curves.
A number of correlations like Corey, Pirson's, Wyllie et Al. and Honarpour et Al., which are based on empirical models describing experimentally determined relative permeability curves, which have provided the most successful approximations better than the capillary models and the statistical models that used to determine the relative permeability and the wettability of oil-water system in sandstone formations. Calculations from the ANN model then compared with values calculated from other models currently in widespread use.
The developed ANN model show enhanced predictions in estimating the wettability and end-point relative permeability of sandstone formations. ANN model predictions can be significantly reduce the uncertainties associated in obtaining relative permeability data currently used in reservoir simulators and in the full field development plans and studies better than the currently available empirical correlations. This situation is primarily due to the ability of the neural networks to recognize both linear and non-linear relationships among the different variables, which influence wettability and relative permeability of hydrocarbon reservoirs.
