Data-Driven Characterization of Saturation Distribution for a Heterogeneous Carbonate Reservoir in Middle East

Abstract

Abstract The middle east carbonate reservoirs are of tremendous potential, but due to the complex types of carbonate reservoirs and strong heterogeneity, the distribution characteristics of reservoir saturation are uneven, which makes reservoir development very difficult. The distribution characteristics of remaining oil are of great significance for evaluating development performance and preparing optimization plan. Deep learning offers a novel approach to solving this problem as a method of intelligent forecasting and analysis. In this paper, the dynamic reservoir production data were collected to establish the data foundation for data driven model training and forecasting. Then the Bi-GRU algorithm was utilized to forecast the performance of single well, which achieved high accuracy predictions with R2 of 0.91, RMSE of 198.93, and MAE of 85.22. After that, single-well temporal three-phase saturation inferring method was proposed based on dynamic performance data, relative permeability curves, and reservoir engineering methods. Finally, kriging interpolation algorithm was used to generate reservoir spatial three-phase saturation distribution. Compared with conventional numerical simulation methods, this method exhibits advantages in computational efficiency and prediction accuracy, and also provides a novel direction for saturation prediction research.