Mapping Injection Water Slumping and Reservoir Boundaries Using Real-Time ANN 2D Inversion of Extra Deep Azimuthal LWD Resistivity Measurements

Abstract

Extra deep azimuthal resistivity (EDAR) measurements are widely used for proactive geosteering and reservoir mapping. Real-time decisions are usually supported with data inversion results and uncertainty quantification based on 1D layer-cake models. A richer 2D model is required in complex scenarios, such as the irregular shaped water slumping observed in this Abu Dhabi Lower Cretaceous limestone reservoir. The inversion based on the Levenberg-Marquardt method successively runs forward simulations matching the measured and synthetic data. Inversion with a 1D layered model provides good data match and functional results on most intervals along the horizontal borehole. For intervals where this is not the case, the inversion continues with a 2D model covering a variety of lateral inhomogeneity scenarios. The challenge of fast 2D forward simulation for this model is solved by the application of an Artificial Neural Network (ANN) whose performance is applicable for real-time applications. Injection water override and the base of the reservoir boundary were detected using 1D inversions along the horizontal borehole at distances of 14-43 ft and 6-20 ft TVD respectively. Continuous uncertainty corridors for surfaces of interest were calculated using a linear deterministic approach and visualized on a curtain section. Water-saturated formation is conductive and prevents detection of any layers beyond. Therefore, we calculate only one-sided (downward) uncertainty estimate for the slumping water position which is 2-15 ft along the well. The uncertainty estimate for the base is two-sided and varies from 0.1 to 7 ft along the well. Using statistical analysis of the inversion results, we show joint uncertainty for all model parameters at selected intervals. On exceptional intervals, where the 1D model turned out to be inappropriate, 2D inversion was applied. For example, when the azimuthal measurements indicate the presence of a conductive zone on the side and the top. Our 2D model covers this case and the 2D inversion provides an improved data match and agrees with the a priori information on the reservoir. The results of 2D inversions on consecutive intervals eventually constitute a laterally consistent 3D structure of the reservoir. We demonstrate real-time 2D inversion of EDAR data based on scenario-specific models. Real-time performance is achieved by utilizing a forward solver using a pre-trained ANN. 2D inversion was applied along intervals, where EDAR measurements indicated non-parallel bedding of reservoir base below the well and slumping water above it. Combined 1D and 2D inversion results comprise a laterally consistent 3D model and leads to better understanding of the reservoir structure.