Geological Realism of Deep Water Channel Reservoir Models with Intelligent Priors

The automation of history matching makes very difficult for modellers to preserve the geological realism of reservoir models. Automation incurs a risk of generating reservoir models with unrealistic geometries based on ad-hoc combination of the model parameter (e.g. channels that are 1 m wide and 200 m thick). Moreover, computational effectiveness of history-matching decreases, as the search for optimum extends to a wider domain. Furthermore, the use of geologically unrealistic reservoir models could mislead the development plan for a specific reservoir.

Use of geological prior information in reservoir models provides a way to control relations between geomodel parameters to ensure their realism. Geological prior information is usually obtained from sources like outcrops, seismic data or modern depositional environments. Geological prior models quantitatively describe the natural relations among the geo-parameters (e.g. channel width, thickness, sinuosity, etc).

Current practice of modelling sanbodies in deepwater channels is based on deterministic or two-dimensional geological priors, which establish relationships between only two parameters at a time.

In this work we propose to tackle the problem of preserving realism in automated history matching by building robust prior models that describe the non-linear multivariate dependencies between geological parameters of the deep water channelized system. We built multi-dimensional realistic priors using intelligent techniques, specifically One-Class Support Vector Machine. OC-SVM allows capturing hidden relations of the deep water channel parameters (Channel Width and Thickness, Meander Amplitude and Wavelength). Furthermore, it is possible to predict realistic parameter combinations, not observed in the available data; but still plausible in nature.

In automated history matching we sample from these realistic priors in order to assure geological realism. A Multiple Point Statistics (MPS) algorithm SNESIM is used to model facies in a deep water channelized reservoir. Variability of the channel geometries are produced by SNESIM algorithm using the affinity parameter, which alters the geometry compared to the training image. We developed a technique to link the MPS affinity parameter with the observed geological characteristics described by the intelligent priors used in history matching. History-matched models produced under geological realistic constraints reduce uncertainty of the production prediction.

AAPG Search and Discovery Article #90142 © 2012 AAPG Annual Convention and Exhibition, April 22-25, 2012, Long Beach, California