Prediction of capillary pressure for oil carbonate reservoirs by artificial intelligence technique

Several types of Artificial Intelligence (AI) techniques were utilized to predict capillary pressure in carbonate oil reservoirs with complex morphologies. To develop AI models that predict capillary pressure, a training data set is used that comprises of mercury injection drainage capillary pressure data. In this study, the training data included a set of 70% of 202 core samples that included porosity, permeability and grain density measurements from conventional core analysis tests. Models were developed using this data to predict capillary pressure. Using an error minimization routine, a comparison between the predicted results and laboratory measurements was used to show the validity of this analysis. The model was tested against a new dataset, the remaining 30% of the 202 core samples, that was not included in the training phase. This process was performed on mono-modal, bi-modal, and combined modal data sets. The analysis of the AI models used in this study, showed that AI has promising potential; however, there is still room for improvement. Capillary pressure data is one of the most critical parameters used in reservoir characterization and initialization stage of simulation models. Many reservoirs do not have an ample data for a comprehensive evaluation. This method utilizes a small core data set to derive models that can be used to predict capillary pressure in reservoirs that lack these measurements. This proposed approach has three advantages: saves time and money, does not require core samples for new spots in the same area, and uses the available results to their maximum potential from previously destroyed core samples. While previous work did not address the prevalent bimodality of the carbonate rock through the use of Flow Zone Indicator FZI / Rock Quality Index RQI, this paper addresses specifically the AI application to the different rock modals through Thomeer methodology.