Automated Workflows for Reservoir Modeling, Can We Get There?

Abstract

Typical reservoir models are constructed at a scale between core or log profiles and seismic resolution. Outcrop analogues are therefore a crucial source of information to bridge the scale gap that exists and to provide greater detail on reservoir architecture, geometries and inter- and intra-facies connectivity, which in turn are difficult to obtain from subsurface data alone. Typical geostatistical reservoir modeling relies on the definition of object dimensions, variogram ranges or training images following manual outcrop interpretation. Methods in outcrop interpretation and transferring outcrop information to analogue reservoir modelling remains archaic in its approach.

This study presents three approaches for the automatic transfer of outcrop interpretations to geostatistical reservoir modeling workflows to build four sets of three-dimensional (3D) geocellular models of a well-exposed succession of fluvial channels cutting into lower marine shorefaces and associated coastal plain environments at a scale that is analogous to the interwell spacing of a typical North Sea hydrocarbon reservoir.

Digital outcrop data is derived from the Campanian Blackhawk Formation, a well exposed fluvial system that fed wave-dominated Cretaceous shorelines of the Mesaverde clastic wedge. Outcrop interpretations from the UAV-based photogrammetic dataset of the Beckwith Plateau, Utah, was used to derive analogous reservoir information and combined with six producing wells from the Solider Creek Pilot Project (SCPP), Utah.

Results of two methodologies for generating semi-automatic training images for multiple-point statistics (MPS) were compared against two Base Case models using automatic variogram extraction for pixel-based methods and a traditional Boolean model. Models were compared using visual assessment of geological reproduction, static connectivity of pay sandstones within the succession to producing wells.

Waterflooding simulation indicates that the distribution of heterogeneity results in significant parts of moveable oil being trapped in un-swept areas and dead-ends. Training images extracted from virtual outcrops and used to condition MPS modeling approaches can be used to drive scenario modeling for stochastic reservoir modeling and modeling of similar analogue systems in the subsurface, after achieving similar dynamic performance to the base case models.

AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019