Вы находитесь на странице: 1из 5

75

th
EAGE Conference & Exhibition incorporating SPE EUROPEC 2013
London, UK, 10-13 June 2013

Th-04-06
Volume Based Modeling - Automated
Construction of Complex Structural Models
L. Souche* (Schlumberger), F. Lepage (Schlumberger) & G. Iskenova
(Schlumberger)
SUMMARY
A new technology for creating, reliably and automatically, structural models from interpretation data is
presented. The main idea behind this technique is to model directly volumes (the geological layers) rather
than surfaces (horizons that are bounding these layers). In order to enforce the geological consistency of
the created models another key element is built into this technology: it guarantees that the variations of dip
and thickness of the created geological layers are minimized, while all seismic and well data are properly
honored. The proposed method enables the construction of very complex structural models, independently
from the geological settings, and even when such models have to be built from sparse or noisy data. The
full automation of the model construction process allows to rapidly update the model, to efficiently
identify the most uncertain parameters, to understand their impact, and to iteratively optimize the model
until it fits all available data. To demonstrate the advantages of this technique the construction of a
complex exploration-scale structural model of a prospect located offshore Australia is detailed.


75
th
EAGE Conference & Exhibition incorporating SPE EUROPEC 2013
London, UK, 10-13 June 2013
Introduction
Structural modeling and reservoir gridding are the centerpieces of the reservoir characterization
workflow. They enable the construction of accurate reservoir models while leveraging all the
information extracted from seismic interpretation. In this area - which is still the most disputed among
reservoir modeling packages - two main challenges remain: (1) the construction of very complex
structural models, especially when such models have to be built from sparse data, and (2) the full
automation of the model construction process, that would enable saving weeks of manual edition and
greatly reduce the risk of errors. The prize of such automation is the ability to rapidly update the
model, to efficiently identify the most uncertain parameters, to understand their impact, and to
iteratively optimize the model until it fits all available data, static and dynamic - an objective the
industry has been striving to reach for decades.

This paper describes a new, inherently robust, methodology for creating faulted structural models that
is being included into an integrated reservoir modeling platform. The main idea behind this technique
is to model directly volumes (the geological layers) rather than surfaces (horizons that are bounding
these layers). In order to enforce the geological consistency of the created models another key element
is built into this technology: it guarantees that the variations of dip and thickness of the created
geological layers are minimized, while all seismic and well data are properly honored.

In this paper, we first describe the inner workings of the Volume Based Modeling technique, we then
illustrate the advantages it brings when building automatically complex reservoir models and finally
we demonstrate how it was used to build a complex exploration-scale structural model of a prospect
located offshore Australia.
Related work
The various approaches that have been described for creating a structural model of the subsurface can
be classified in two categories: surface-based and volume-based modeling. The first stream of
technology, by far the most widespread in the industry, attempts to create a numerical representation
of the surface network formed by faults and horizon surfaces. A popular approach [1] consists in
creating first a smooth, un-faulted surface that is later cut by the fault surfaces before being attracted
by the interpretation points. In practice, this only works (1) when it is possible to create properly the
initial smooth surface and (2) when the topology of the intersections between the faults and the initial
surface does not differ from the one of the final fault polygons. Both assumptions tend to fail when
dealing with complex compressional models or with a large number of X, Y or faults. Another
surface-based approach [2] relies on the partition of the volume of interest into several closed
volumes, delimited by the fault network, so that fault blocks can be modeled independently from each
other. This requires the fault surfaces to be artificially extrapolated so that they fully cross volume of
interest, an inherently non-robust process.

The second family of techniques revolves around the concept of implicit modeling. Implicit
modeling relies on the representation of surfaces as isovalues of a volume attribute the implicit
function. It used to be associated with the construction of multi-z geobody surfaces (i.e. salt bodies)
[3]. The horizon modeling technique underlying the pillar-gridding technology also belongs to this
category, the volume attribute being a thickness proportion, interpolated onto a 2.75D support (the
grid pillars).
Methodology
In the technique described in this paper, the implicit function corresponds to the stratigraphic age of
the formations. It is embedded and interpolated in an unstructured tetrahedral mesh.

The first step of the construction of the structural framework consists in building a tetrahedral mesh
for carrying and interpolating the implicit function (Figure 1a-b). To do so, a 3D Boundary-


75
th
EAGE Conference & Exhibition incorporating SPE EUROPEC 2013
London, UK, 10-13 June 2013

Constrained Delaunay mesh generator is used, with several constraints: firstly, faults affecting the
considered horizons are taken into account as internal boundaries during the mesh generation, in such
a way that some border faces of tetrahedra match the fault geometries in the produced mesh. Using
unstructured grids as interpolation support allows controlling the density of the mesh, for maximizing
the degree of freedom of the interpolation where it matters (i.e. close to the data), and adapting its
anisotropy, in order to better capture the thickness variations in the layering. Note that such a
flexibility and adaptability is extremely difficult or impossible to achieve using a structured mesh
(like Cartesian grids, pillar grids, etc).

The second step consists in interpolating the values of the implicit function on the nodes of the
tetrahedral mesh (Figure 1c). This interpolation is done using a linear least squares formulation, which
will tend to minimize (1) the misfit between the interpretation data and the interpolated surfaces and
(2) the variations of dip and thickness of the layers. The computational cost is typically around 20
seconds to a few minutes depending on the required resolution.

The third step is to generate surfaces representing every implicitly modeled horizon (Figure 1d). Since
the specific value of the implicit function associated to each of them is known, this is simply done
using any iso-surfacing algorithm. Finally, it is possible to take advantage of this Volume Based
Modeling approach to generate a consistent zone model (Figure 1e). Every geological layer of a
model can actually be seen as an interval of values of the implicit function. It is thus very simple,
given its value of the implicit function, to know to which layer an arbitrary point belongs to.
Figure 1 Step by step construction of a volume model: (a) fault model and horizon interpretation, (b)
tetrahedral solid, (c) implicit stratigraphy function (with periodic colorscale), (d) horizon surfaces,
(e) layers
Advantages
Being fully 3D, this method is practically insensitive to the complexity of the fault network (Figure
2). It also benefits from many other desirable features: all conformable horizons belonging to a same
conformable sequence being modeled simultaneously, as several isovalues of the same implicit
attribute, they cannot cross each other [4]. Another advantage is that each and every conformable
horizon constrains the geometry of all other conformable horizons that belong to the same sequence,
and is itself constrained by their geometry. This practically removes the need for isochore or isopach
maps when modeling some horizons from sparse data (i.e. well tops or 2D sections). Finally, this
a b c
d e


75
th
EAGE Conference & Exhibition incorporating SPE EUROPEC 2013
London, UK, 10-13 June 2013

technique does not only output the geometry of the horizon but also the volume attribute, defined
everywhere in the volume of interest, representing the stratigraphic age of the formations (Figure 5).
Figure 2 Model built from a transtensive sandbox experiment. Left: Fault model. Right: Volume
model and interpretation of the top horizon.Sandbox data courtesy pf Pr. K. McClay, Royal Holloway,
Univ. of London.
Case study
The benefits of Volume Based Modeling will be illustrated using an example case study, in which the
3D geological modeling workflow implemented for an exploration project is described. The area
under study is located offshore Western Australia nearby Gorgon Gas Field area. The main objective
of this study is to deliver quickly a comprehensive, geologically consistent 3D structural and
stratigraphic model that can be used for property population, quantification of hydrocarbon volumes
and reservoir simulation.

The survey covers an area of approximately 500km
2
and a thickness of about 3534ms, corresponding
to approximately 4400m depth. The main reservoir formation in this area (Cretaceous formation) is
composed of deltaic to shallow marine sandstone deposits. A total of 161 major faults with significant
displacements are interpreted (Figure 6). The stratigraphy of the study area is complicated by the
presence of a Jurassic unconformity, underlying a Cretaceous deltaic complex (clinoforms). The thick
regional shale forms a seal and unconformably caps the Cretaceous complex. In total 10 major
horizons (including unconformity surfaces) were interpreted on the 3D seismic. To demonstrate the
capabilities of Volume Based Modeling, 4 additional horizons were added, based on sparse well
information.

In order to build an initial representation of the 3D structure, a very coarse (every 2000 m inlines and
crosslines) seismic horizon interpre-tation was performed and fed into the Volume Based Modeler.
This preliminary 3D model was then iteratively refined by comparing the geometry of reconstructed
horizon surfaces with the location of the corresponding reflectors on the seismic image, and adding
interpretation data in places where inconsistencies were detected. This allowed producing an accurate
geological model while minimizing the effort and time spent on interpreting seismic. The next step
consisted in incorporating the well data to control the construction of intermediate reservoir zones.
Thanks to the Volume Based Modeling technology, it was not necessary to compute any isochore
maps.

A fully stair-stepped 3D corner point grid was then directly generated from the structural framework.
The created grid can now be used for property population, volume computation and reservoir
simulation.
Conclusions
A new technology for creating, reliably and automatically, structural models from interpretation data
independently from the structural complexity has been presented. It has also been shown, using an
example case-study, how this new method can help building large-scale, complex models from


75
th
EAGE Conference & Exhibition incorporating SPE EUROPEC 2013
London, UK, 10-13 June 2013

exploration data. This new approach enables a significant speed-up in the model building and gridding
workflow often one of the most time consuming step when creating a reservoir model.

Beyond the structural framework, this new methods yields other valuable outputs: the stratigraphy
attribute, which can be used by consuming applications, and a faulted unstructured mesh, which can
be used as a support for geomechanical applications.








Figure 3 Seismic data courtesy of Geoscience Australia. Top: some of the data used to build the
Volume Based Model. Bottom: built australian model, including horizons based on well tops.
Acknowledgements
The authors would like to thank all their colleagues from the Petrel Structural Framework team that
contributed to the integration of this technology in Petrel: David Desmarest, Sebastien Roret, Dustin
Lister, Thomas Thrams, Michael Palomas and Mohamed Benaichouche. Dr Leigh Truelove is also
warmly acknowledged for providing and interpreting sandbox models shown in this article.
References
1. J-L. Mallet, 2001, Geomodeling, Oxford University Press
2. J.W. Neave, Analysis and characterization of fault networks, US Patent 7512529
3. T. Frank, A-L. Tertois, J-L Mallet, 2007, 3D-reconstruction of complex geological interfaces from
irregularly distributed and noisy point data, Computers and Geosciences, 33, 7
4. F. Lepage, L. Souche, 2012, Method for modeling faulted geological structures containing
unconformities, Provisional Patent Application

Вам также может понравиться