Augmented Reservoir Characterization Technology for the Stochastic Estimation of Rock Properties

Abstract Understanding spatial reservoir properties presents significant challenges, especially when the required data is scarce. This scarcity renders the problem underdetermined, which impedes informed business decisions. Despite the maturity of classical characterization workflows, they often yield inaccurate estimations. A remedial solution that leverages physics informed artificial intelligence (AI) was employed. In this study, the performance of this new workflow is assessed for Field X in onshore Abu Dhabi, UAE. By incorporating elastic models derived from inversion processes and applying rock physics equations specific to carbonate rocks, several multivariate equations are applied to two- (2D) and three-dimensional (3D) datasets. These serve as constraint equations for the objective function which defines the data augmentation process. Particle Swarm Optimization (PSO) algorithms were utilized for generating AI optimized petrophysical realizations. Specifically, two modified versions of PSO were employed: the Variable Neighborhood PSO (VN-PSO) and the Hierarchical PSO with Time Varying Acceleration Coefficient (HPSO-TVAC). The integration of various equation combinations imposed distinct conditions on the problem, thereby facilitating multiple disparate realizations across 3D design variables. The results demonstrated that the physics-enabled AI workflow significantly improved the correlation between optimized porosity and upscaled porosity logs from wells, with an average increase in correlation of 15%. The optimization process, which considered constraint equations containing permeability, resulted in the classification of rocks into seven types based on porosity and permeability. This novel approach highlights the benefits of multi-dimensional data augmentation and geo-spatial co-generation, leading to improved continuity between well data and multiple 3D petrophysical property realizations compared to traditional methods.