Scale Splitting Approach to Reservoir Characterization

Abstract

TX 75083-3836, U.S.A., fax 01-972-952-9435. AbstractProduction forecasts are essential for sound reservoir management. The foundation for such forecasts is a characterization of relevant reservoir properties. These properties are usually determined through history matching, using production data, static well data (hard data), and the upscaled geological model (prior model), simultaneously. This process if often very complex and costly, both in terms of CPU-time and man-hours. To reduce cost and complexity of reservoir characterization, we propose an alternative methodology; scale splitting. This approach utilizes the fact that production data often contain information about variability of reservoir properties on a much coarser length scale than the other data do. Hence, production data alone can be used to determine the large-scale variation in the estimated properties. The selection of parameters that can be estimated from the production data are determined through a data driven, top-down search, starting with a single parameter representing the average property for the reservoir structure in question. One of the primary objectives of this search is to keep the number of parameters as low as possible, without sacrificing the match quality. The prior model and the hard data need to be integrated to allow also for finer-scale property variation, and to match hard data at well locations. With scale splitting this is done at low cost after the history match of production data. We present the scale splitting approach for absolute permeability estimation and give examples where it is compared to other approaches in terms of match quality and work load. y J J J J y y J r

Parametrization of Stochastic Inputs Using Generative Adversarial Networks With Application in Geology

“…The importance of parametrization in subsurface simulations resulted in a variety of methods in the literature including zonation [20,24] and zonation-based methods [5,9,15,16,17,1], PCA-based methods [13,40,43,46,32,55], SVD-based methods [48,47,52,51], discrete wavelet transform [33,31,44], discrete cosine transform [21,22,23], level set methods [35,10,8], and dictionary learning [26,27]. Many methods begin by proposing parametric forms for the random vector to be modeled which are then explicitly fitted to preserve certain chosen statistics.…”

Mapping of Reservoir Properties and Facies Through Integration of Static and Dynamic Data

“…One way to attempt to deal with such problems is via reparameterization by reducing the number of model parameters (reservoir variables). This type of approach has a long history beginning with the pioneering work on zonation of Jacquard and Jain (1965) and Jahns (1966) and continuing today with work on adaptive parameterization; see, for example, Grimstad et al (2001). During the first year of this project report, we investigated reparameterization based on subspace methods (Kennett and Williamson, 1988;Oldenburg et al, 1993;Oldenburg and Li, 1994; to reduce the number of parameters directly estimated in the optimization process.…”