Using graph-based distances as surrogate information to compare the ow behaviour of structural models

Jérémie Sexe and Gabriel Godefroy and Pierre Anquez and Guillaume Caumon. ( 2018 )
in: 2018 Ring Meeting, ASGA

Abstract

In order to take into account data and interpretation uncertainties, several structural models of subsurface can be generated. In hydrocarbon reservoirs, ow simulations are used to discriminate models that do not fit to exploitation data, then to produce production forecasts and optimize production. As these simulations are time consuming and can be performed only on a limited number of models, ranking techniques can be used to select a subset of representative models. The topology of a reservoir model, i.e. the connectivity of its compartments, impacts the ow behavior of the reservoir. This paper focuses on the comparison of structural models using their topology in order to sort and rank them. We created a set of synthetic layer-cake structural models composed of one fault and twelve horizons using SKUA-Gocad. These models differ by the value of fault drag displacement, i.e. the distance over witch horizons slide on the fault. We use the RINGMesh data structure to obtain the topological information of each model and to extract an associated graph representing the connectivity between its units. This graph makes it possible to define a distance between structural models based on topological information. Flow simulations were run on these models using Eclipse software. A first ranking was made based on the reservoir behavior and a second one by comparing the model topologies. Comparison of these two rankings, suggests that topological distances poorly discriminate models but are not sufficient to explain the range of observed production behavior. We suggest that weighting graph distances based on physical principles could help, but argue that generalizations to arbitrary numbers of region are challenging.

Download / Links

BibTeX Reference

@INPROCEEDINGS{,
    author = { Sexe, Jérémie and Godefroy, Gabriel and Anquez, Pierre and Caumon, Guillaume },
     title = { Using graph-based distances as surrogate information to compare the  ow behaviour of structural models },
 booktitle = { 2018 Ring Meeting },
      year = { 2018 },
 publisher = { ASGA },
  abstract = { In order to take into account data and interpretation uncertainties, several structural models of
subsurface can be generated. In hydrocarbon reservoirs, 
ow simulations are used to discriminate
models that do not fit to exploitation data, then to produce production forecasts and optimize
production. As these simulations are time consuming and can be performed only on a limited
number of models, ranking techniques can be used to select a subset of representative models. The
topology of a reservoir model, i.e. the connectivity of its compartments, impacts the 
ow behavior
of the reservoir. This paper focuses on the comparison of structural models using their topology in
order to sort and rank them. We created a set of synthetic layer-cake structural models composed
of one fault and twelve horizons using SKUA-Gocad. These models differ by the value of fault
drag displacement, i.e. the distance over witch horizons slide on the fault. We use the RINGMesh
data structure to obtain the topological information of each model and to extract an associated
graph representing the connectivity between its units. This graph makes it possible to define a
distance between structural models based on topological information. Flow simulations were run
on these models using Eclipse software. A first ranking was made based on the reservoir behavior
and a second one by comparing the model topologies. Comparison of these two rankings, suggests
that topological distances poorly discriminate models but are not sufficient to explain the range
of observed production behavior. We suggest that weighting graph distances based on physical
principles could help, but argue that generalizations to arbitrary numbers of region are challenging. }
}