| dc.description.abstract | Multipoint statistics (MPS) provides an approach for pattern-based simulation of complex geologic objects from a training image (TI), which contains the general connectivity structures of complex patterns. While grid-based implementation of the MPS methods facilitates hard-data conditioning, conditioning the simulated facies on flow data poses a challenging problem. The main objective of this dissertation is to develop an inverse modeling framework for conditioning MPS-based facies simulation on dynamic flow data. The developed formulation is then extended to account for uncertainty in the geologic scenario. In the second part of the dissertation, an inverse modeling formulation is presented for estimating large-scale reservoir connectivity from low-resolution pressure data using.
The first contribution of this dissertation is the formulation of a probability conditioning method (PCM). In the PCM approach, the flow data is first inverted to obtain a probabilistic description of facies distribution (a probability map). The resulting probability map is then used to guide the MPS facies simulation from a specified TI. The proposed PCM approach can be used with different inversion algorithms. In this dissertation two alternative implementations are presented: 1) the ensemble Kalman filter (EnKF); 2) a stochastic optimization approach. An important practical limitation of the MPS modeling approach is the uncertainty in the prior TI. To address this problem, a Bayesian mixture modeling formulation is developed. In this approach, the posterior distribution of facies is partitioned into individual conditional densities of the TIs.
In the second part of the dissertation, a novel approach is developed estimating field-scale reservoir connectivity from dynamic data. This is accomplished by reconciling low-resolution dynamic field pressure data with high-resolution static geologic models. Since pressure variation represents a smooth function, a low-resolution (coarse-scale) grid system is adopted for reservoir simulation. To reconcile data and model resolutions, the grid system is generated using the Delaunay triangulation method. The reservoir properties are then scaled up from the fine-scale geological model to this coarse scale unstructured grid system to create an initial static simulation model. In the data integration stage, the EnKF is used to automatically adjust the global parameters of the field to match the static pressure. | en |
