Uncertainty Quantification Algorithms for Reservoir Characterization of Interwell Volumes

Abstract

In water injection driven production, evaluating the sweep efficiency and monitoring the flood front is one of the main challenges for oil companies. Most reservoir characterization and saturation logging tools provide data at a specific well location with a limited radius of investigation around the wellbores. When resistivity contrast exists among the saturating fluids, electromagnetic (EM) technologies provide a promising approach to measure the spatial distribution of resistivity away from the wellbore in the interwell region. However, many challenges are associated with the interpretation of the resistivity distribution obtained from EM surveys. In order to generate realistic saturation maps of the interwell regions, assessing the uncertainty in the reservoir properties distribution is essential. In this study, a novel approach to interpret EM surveys is proposed, introducing robust saturation maps that can guide reservoir management policies and ultimately increase hydrocarbon recovery. The proposed workflow heavily utilizes dynamic reservoir simulation and advanced uncertainty quantification methods. In this study, the permeability field is the spatial uncertain parameter being considered. To efficiently quantify uncertainty in the permeability and its effect on the salinity distribution, two methods were applied. The first method is the probabilistic collocation method (PCM) that is based on the use of polynomial chaos expansion to develop a polynomial proxy model. The coefficients of this model are determined by performing reservoir simulation using an optimized set of collocation points, which approximate the sample space. The number of these samples that are required are significantly fewer than those required for a typical Monte Carlo or Latin Hypercube Simulation. The second method we analyzed is the Multi-Level Monte Carlo (MLMC) method, which is based on the use of a combination of multi-level sequentially coarsened (upscaled) grids which shift most of the computational cost to the coarsened grids. Both methods studied Gaussian generated permeability realizations. Based on the results, MLMC was developed further to estimate the fluid distribution in non-Gaussian permeability realizations. The obtained results showed that it is possible to investigate a wide range of uncertainty in a computationally efficient manner providing more robust saturation maps compared to current deterministic practices.

AAPG Datapages/Search and Discovery Article #90332 © 2018 AAPG International Conference and Exhibition, Cape Town, South Africa, November 4-11, 2018