CMG: Research on Multiscale Spatial Models for Petroleum Reservoir Mapping Using Static and Dynamic Data

Texas A&M Research Foundation, College Station TX

Investigators

Abstract

This project concerns petroleum reservoir characterization using static and dynamic data with varying scales and precision. Static data refers to time-invariant data such as well logs, cores, and seismic and geologic information. Dynamic data refers to time-varying information such as pressure transient response, tracer data, and multiphase production history. Petroleum reservoirs are complex geological formations encompassing a wide range of physical and chemical heterogeneities. The goal of reservoir characterization is to provide a numerical model of reservoir attributes such as hydraulic conductivities (permeability), storativities (porosity), and fluid saturation. These attributes are then used as input into complex transfer functions represented by various flow simulators to match the dynamic data. In predicting future reservoir performance, it is imperative to have a geological model that can be considered a "plausible" replica of the actual reservoir with acceptable uncertainty. Towards this objective, several flexible spatial modeling approaches are proposed to reproduce complex geological/morphological patterns and the wide variety of architectural heterogeneities observed in reservoirs. To reduce the uncertainty, both static and dynamic data sources are combined to derive an integrated reservoir description. The task is non-trivial because these data sources span different length scales of heterogeneity and can have different degrees of precision. A hierarchical Bayesian model is constructed where the data from different scales will be related to each other by conditional models at different stages of the hierarchy. To preserve the connectivity of the extreme value attributes, and thus preferential fluid flow paths (channels) and barriers, non-stationary and non-Gaussian approaches are explored using partitioning models. The models are extended to categorical facies data with logit or probit structure. Due to the complexity of the problems, the posterior distributions of the unknown parameters are not likely to be explicitly available, and Markov Chain Monte Carlo based computations will be carried out to draw samples from the posterior distributions for quantifying uncertainty. The goal of this work is to provide a systematic approach to petroleum reservoir characterization using diverse data sources. This work will aid in improved oil recovery and will also have direct impact on the design of aquifer remediation methods through better characterization of subsurface heterogeneities. The targeted problem is of practical national interest, as better characterization of petroleum reservoirs will lead to better management and development strategies, leading to increased oil recovery. Currently most of the domestic oil production is from old and partially depleted fields, for which a large amount of data are available. A systematic approach leading to a better reservoir description will have a direct impact on the current industry practice. Despite recent interest in alternative sources of energy, the demand for oil has continued to increase, and the large imbalance between domestic supply and demand is an issue for both national security and the economy. Large amounts of recoverable oil do remain in domestic reservoirs, but those reserves are largely located in highly heterogeneous reservoirs or in remote and expensive locations, such as beneath the deep water of the Gulf of Mexico. The challenges for recovering this oil are to (1) be able to map heterogeneities to improve the locations of wells for reservoir management and to be able to locate unswept reserves, and (2) identify uncertainties in mapping so that risk can be assessed for sound economic decisions.