You are on page 1of 15

SPE-177480-MS

Sensitivity Analysis of Unstructured Meshing Parameters on Production


Forecast of Hydraulically Fractured Horizontal Wells
Jianlei Sun, and David Schechter, Texas A&M University; Chung-Kan Huang, ConocoPhillips

Copyright 2015, Society of Petroleum Engineers

This paper was prepared for presentation at the Abu Dhabi International Petroleum Exhibition and Conference held in Abu Dhabi, UAE, 9 –12 November 2015.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
In the context of multi-stage hydraulically fractured horizontal wells, unstructured grids such as PEBI
(perpendicular bisector) grids or Voronoi grids have been widely applied to generate simulation meshes
for fracture networks. In previous work we developed a robust optimization-based 2D meshing algorithms
to handle non-orthogonal, low-angle intersections of extensively-clustered fractures with non-uniform
aperture distributions. However, vertically extruded 2.5D PEBI grids become necessary with more wells
being drilled into dipping layers. Besides, it is necessary to provide practical application suggestions for
choosing an unstructured mesh to accurately resolve flow regimes of production.
In this work we extended 2D PEBI workflow to 2.5D, and then validated it against two models: (a) a
synthetic model with one horizontal well and 120 orthogonal intersected hydraulic fractures built by
Tar-tan grid and (b) a field-scale model with three horizontal wells and 120 non-orthogonal intersected
hydraulic fractures in a slightly dipping reservoir created by a commercial software plug-in. For the
synthetic model, we simulated production performance at a constant bottom-hole flowing pressure up to
20 years. For the field-scale model, we first performed history matching and then predicted production at
a constant rate up to 20 years. We compared pressure diffusion front, bottom-hole flowing pressure as well
as CPU performance.
Reasonably good matches between PEBI grids and other grids are observed in both pressure diffusion
front and production behavior. Sensitivity analysis suggests that refinement around the fractures has
modest impact to early time production while background density has dominant impact to the late time
production. Background grid type and grid orientation have less influence as long as they have the same
grid density. Less number of 2.5D PEBI cells can be achieved by removing unnecessary refinement
around fractures, increasing reservoir background size and reservoir background size ratio, replacing
unstructured background grids with structured grids, and reducing the complexity of the fracture networks
without loss of the accuracy and therefore result in much favorable CPU performance. The existence of
opened natural fractures was evaluated to simulate the out-of-zone production. The results suggest that
effective drainage area can be significantly greater with the synergy of the natural fractures.
This study is the first to apply unstructured grids to simulate multiple horizontal wells with irregular
hydraulic fractures. Besides, this paper provides detailed discussions of implementation algorithms as well
as comparisons between 2.5D PEBI and LGR based grids in the context of fracture modeling. And most
2 SPE-177480-MS

importantly, this study answers the question regarding how to choose an appropriate 2.5D PEBI mesh to
yield both accurate results and good CPU performance.

Introduction
Unconventional resource plays require extensive hydraulic fracturing treatment in order to produce at
commercially viable rates. During hydraulic fracturing, hydraulic fractures interact with in-situ natural
fractures, resulting in complex fracture networks(Olson and Wu 2012). One way to simulate production
performance of the fractured reservoirs is to upscale and then use the dual porosity/permeability approach
(Yan et al. 2013). Besides, in order to more accurately conform to fracture geometries, unstructured grids
such as PEBI (perpendicular bisector) grids or Voronoi grids have been widely used for generating
simulation meshes by both academic researchers (Branets et al. 2008, Olorode et al. 2012, Romain et al.
2011) as well as commercial software packages (Cipolla et al. 2011, KAPPA 2015, Moog 2013). Such a
grid system is flexible, locally conforming to the fracture networks, but also capable to reduce grid
orientation effect(Heinemann et al. 1991).
The PEBI grid or Voronoi grid was introduced to the petroleum industry by Heinemann et al. (1991).
As seen in Figure 1a, the blue points represent the cell centers or the locations where we populate reservoir
properties and simulate reservoir production performance. The red triangles comprise the triangular mesh,
which is constructed by connecting all the cell centers. The black polygons form the polygon mesh, which
is constructed by making perpendicular bisectors of the triangular mesh. Therefore, we usually call the
polygon mesh as 2D PEBI mesh. The 2D PEBI mesh is often considered as the dual mesh of the 2D
triangular mesh. Such duality characteristics is the most important reason for us to choose PEBI mesh as
the most common unstructured grid. Because we can run reservoir simulations on the PEBI mesh, and at
the same time run geomechanics simulations on the triangular mesh.

Figure 1a—Unstructured PEBI grid and Delaunay Triangulation

Besides, as seen in Figure 1b. The shape of the 2D PEBI is very flexible, and thus we can use PEBI
cells to model complex fracture networks of hydraulically fractured horizontal wells. In order to model
geological layers, a 2.5D PEBI grid was introduced to the petroleum industry. The 2D PEBI grid was
projected and extruded to each layer of the reservoir to construct the 2.5D PEBI grid.
SPE-177480-MS 3

Figure 1b—flexible shapes of the PEBI grid

Appendix–A tabulates the main unstructured gridding algorithms for modeling fractures in the
literature. When it comes to fracture modeling, Muralidharan et al. (2004) mentioned that it is important
to accurately model both fracture aperture and fracture geometry. Karimi-Fard et al. (2003) introduced a
lower-dimensional approach to model fracture aperture, which have been used by many researchers (Jiang
and Younis 2015, KAPPA 2015). This approach models each fracture gridblock as a 2D line segment, and
thus facilities mesh generation process. On the other hand, fracture aperture can be explicitly modeled and
gridded (Cipolla et al. 2011, Olorode et al. 2012, Sun and Schechter 2014, Sun et al. 2014). One advantage
of the second approach is more straightforward to model and visualize distributed apertures as seen in
Figure 2. The green ellipse highlights a 2.5D PEBI mesh along a non-uniform fracture aperture.

Figure 2—Unstructured PEBI mesh for a non-uniform fracture aperture(Sun and Schechter 2015).

In the previous study (Sun and Schechter 2014), we demonstrated an optimization-based 2D PEBI
meshing algorithms to grid up non-orthogonal, low-angle intersections with extensively clustered fracture
networks. To start with, fixed points are first computed to honor both fracture intersections and fracture
aperture distribution. And then flexible points are generated and optimized based on pre-defined mesh size
functions to achieve both local grid refinement features around fractures, and most importantly well-
orientated cells around fracture tips and fracture intersections. We also tested the 2D algorithms with
synthetic models, and presented a detailed sensitivity study of natural fracture related properties such as
natural fracture density or spacing, length, and strike on production performance.
However, besides of the properties of natural fractures, unstructured mesh itself might introduce
uncertainties or inaccuracies on simulation results. For example, unstructured mesh density, background
grid type, refinement method around fractures might have either significant or slight effect on the
4 SPE-177480-MS

production performance, which is required to be investigated. Moreover, reservoir engineers might be


more interested in how much is the difference between the conventional grid solutions such as tar-tan grid
or LGR girds and unstructured PEBI grid for both synthetic and real reservoir models, and in whether or
not the unstructured PEBI grid can accurately reproduce the results of the conventional grid solutions.
Most importantly, in terms of the advantages of using unstructured PEBI grids, not only do we need to
present its flexibility, but also carry out a detailed CPU performance analysis to investigate if flexibility
will yield better CPU performance, or we have to compromise CPU performance by using more flexible
unstructured grids.
To these ends, we will first extend the previously developed 2D algorithms to 2.5D, and demonstrate
the unstructured workflow to simulate unconventional reservoirs. Then we will see whether or not the
unstructured grids are accurate enough to reproduce conventional grid solutions by comparing to the tartar
grid, whether or not robust enough to model a real field problem by comparing to the LGR grid, whether
or not flexible enough to change mesh related parameters on the real field problem, and finally whether
or not efficient enough to improve the CPU performance.
The Workflow of 2.5D PEBI Grid
In this study we first extended our previous developed 2D PEBI mesh generation algorithms to 2.5D.
Figure 3 is an overall workflow chart, which includes components for modeling and simulating hydraulic
fractured horizontal wells. The most important components are the preprocessor and the postprocessor.
The preprocessor takes reservoir properties, fracture properties, and user-defined meshing parameters and
generates input files for finite volume based simulators. The postprocessor generates files with the
extension of VTK and visualizes simulation results in open-source software packages. Sensitivity analysis
yields numerous realizations of discrete fracture network (DFN) to investigate the effect of fracture related
parameters on reservoir production performance. Among those realizations, the best realization can be
selected by history matching against production data, and then used for future prediction forecast.

Figure 3—Mesh generation workflow with 2.5D PEBI grids for hydraulically fractured horizontal wells

Figure 4 is a detailed workflow within the preprocessor – 2.5D PEBI mesh generator. Starting from the
input data, we first reduce the 3D meshing problem to a 2D meshing problem by projecting the 3D fracture
geometry to the horizontal plane. Then we deal with the fractures and reservoir background, respectively.
For the fractures, we first compute fixed Voronoi cell centers and deal with fracture refinement, fracture
SPE-177480-MS 5

intersection and clustering, as well as variable fracture aperture. Then we compute flexible Voronoi cell
centers for reservoir background from forced-based optimization algorithms, where different reservoir
background mesh density, orientation, and type can be achieved by iteratively updating flexible Voronoi
cell centers. After we combine all the Voronoi cell centers, we can construct the 2D PEBI grid, and
extrude it to each geological layer to yield the 2.5 PEBI grid. Finally, we can compute simulator inputs
such as pore volume of each PEBI cell, transmissibility between two PEBI cells, cell center depth, as well
as well-related properties.

Figure 4 —Workflow for the 2.5D PEBI mesh generator

If there is a dipping angle for 3D fractures, or in other words, the fractures are not vertical or
perpendicular to bedding layers any more, 3D PEBI grids are usually required to conform to the fracture
boundaries. Merland et al. (2011) and Romain et al. (2011) have investigated the application of 3D PEBI
grids to resolve geological features. However, there are still lack of documented approaches to model fully
3D fractures using 3D PEBI grids. Another approach is to use 3D tetrahedral grids to model fractures,
which have already been investigated by many publications(Jiang et al., Moog 2013).

2D Synthetic Model
In the following sections, we will apply the developed 2.5D PEBI workflow to solve both synthetic and
real reservoir problems. To start with, we will take a look at a 2D synthetic model with a horizontal well
and 55 orthogonal hydraulic fractures. We will validate the developed PEBI workflow by comparing
pressure graphs with known grid (i.e., Tartan grid) solutions. Besides, the main advantage of unstructured
grids is to conform to fracture geometry and provide more accurate well testing results for complex
fracture networks. Therefore, we will also compare flow regimes between the two grid systems.
6 SPE-177480-MS

Table 1—Reservoir and Fracture Properties for 2D Synthetic Model


Reservoir Properties Fracture Properties

KX ⴝ KY ⴝ KZ, md porosity Length, ft Width,ft Num. kf,md xf,ft porosity aperture

0.0009 0.0009 0.055 5280 1320 55 2754 302.6 0.3 0.083

As seen Figure 5, we discretized the 2D synthetic model with both the Tartan grid and PEBI grid. We
can reduce the cell number from 29,000 tartan grid cells to 19,000 PEBI grid cells. After 20 years’
production at a constant BHP, we plotted oil rate vs. time on a log-log scale in Figure 6. We also prepared
pressure graphs at 1 day, 10 days, 100 days, and at the end of 20 years’ production. The top pressure graph
is for Tartan grid, and the bottom one is for PEBI grid.

Figure 5—Comparison between the Tartan mesh and 2D PEBI mesh for the 2D synthetic model

Figure 6 —Comparison of flow regimes between the Tartan mesh and 2D PEBI mesh

From the oil rate plot from 0.1 day up to 10 days, we see a half-slope, or in other words, the linear flow
regime. From 10 days up to 100 days, we see significantly pressure depletion around the hydraulic
SPE-177480-MS 7

fractures. At the end of 20 years’ production, we see a complete depletion for the whole reservoir. For all
the flowing regimes, we see a good match between the Tartan grid and the PEBI grid. For this model, the
simulations took only several minutes, thus it makes no sense to compare CPU performance. Instead, we
will look at a detailed analysis of CPU performance for the following field case.
2.5D Field Model
The previous study case is based on a synthetic model. From this section, we will look at a 2.5D field
model, and compare the PEBI grid with Petrel LGR Grid. There are two main goals of this comparison.
First, we want to see if the 2.5D PEBI workflow is robust enough to handle complicated field problems.
Second, we want to investigate if the 2.5D PEBI grid system is more flexible than the Petrel LGR grid.
At the same time, we will perform three sensitivity studies to discuss the effect of unstructured meshing
related parameters on both accuracy and CPU performance of simulation results. This model is made up
with three horizontal wells, 40 hydraulic fractures per well, and 8 layers with a small dipping angle.
Figure 7 shows several snapshots of the same locations of the reservoir. We will compare the
differences between PEBI mesh and Petrel LGR mesh. From two graphs at the top, we have two snapshots
around non-orthogonal intersected fractures. As you can see, Petrel plug-in software simplifies intersected
fractures to a zig-zag shape. Instead, in the PEBI grid, we consider them as straight line segments. From
two graphs at the bottom, we have two enlarged views around a fracture intersection. For Petrel, the
connections around the intersection are very complicated. However, in the PEBI grid, we explicitly model
the intersection as a single grid block. Besides, for the Petrel grid, the background mesh depends on the
mesh size between fractures. However, in the PEBI grid, we have the option to use fine mesh around
fractures and coarse mesh far away from fractures. On the other hand, both PEBI grid and Petrel LGR grid
use rectangular type of cells for fracture refinement. Refinement is flexible enough to model the lineal
flow regime at the early period of production.

Figure 7—Comparison between 2.5D PEBI mesh and Petrel LGR mesh

Sensitivity of Background Grid Density


As mentioned before, one of the advantages of the unstructured PEBI grid is its flexibility in cell shape
and size. The motivation of this sensitivity study is to reduce the total number of simulation cells without
compromising the accuracy of simulation results, which is very hard to achieve using the conventional
LGR mesh.
As seen in Figure 8 from left to right, we have Petrel LGR mesh with 401,384 cells, PEBI fine mesh
with 409,920 cells, PEBI meshes with 281,336 and 255,088 cells, and PEBI coarse mesh with 190,440
cells. By using large initial gridblock size and mesh size progression ratio(two input parameters for the
8 SPE-177480-MS

2.5D PEBI mesh generator), we can reduce the total number of cells by half from around 410K to 190K.
For the four PEBI meshes as well as the Petrel LGR mesh, we will first run history production for around
a year, and then we will make a 20 years’ prediction at a constant rate using an in-house black-oil
simulator. Figure 9 shows the comparisons between the pressure diffusion graphs at the end of the
simulation.

Figure 8 —Petrel LGR mesh vs. 2.5D PEBI meshes for different background grid density

Figure 9 —The effect of background grid density on pressure graphs

We see a good match in pressure diffusion front between the Petrel model and the PEBI fine model.
The background grid density plays an important effect in long-term production. For example, for the PEBI
coarse mesh, we see that the background mesh is too coarse to accurately capture the pressure diffusion
front. However, for the rest PEBI cases, we see a reasonable match. Besides, we also see good matches
of both history production and long-term prediction between the Petrel model and PEBI models, except
that the PEBI coarse mesh is off the trend at the later simulation period due to extreme low resolution of
the reservoir background.
Sensitivity of Background Grid Type
Even though PEBI grid is flexible and has demonstrated its capability to reduce the total number of
simulation gridlocks, the Jacobian matrix formulated from the unstructured PEBI grids usually ends up
with unequal bandwidth, and thus compromise computational efficiency. One straightforward approach is
to replace the background unstructured grids with Cartesian or hexagonal grids. This way, the resulting
Jacobian matrix might be as narrow in bandwidth as possible without compromising accuracy requirement
around fractures.
As seen from Figure 10a, we choose to define a region labeled by the green dotted box. On the region
boundary, we have a constant distance to the fracture. Within this region, we keep using unstructured
PEBI grids. Outside this region, we have the option to use either structured rectangles or hexagons. Figure
10b is the reference case with 255,088 cells from Figure 9. Following the hybrid grid approach, we reduce
SPE-177480-MS 9

the cell number to 231,048 in Figure 10c, and 232,080 in Figure 10d. Again, we will run similar
simulations and compare the pressure graphs in Figure 11. From the pressure diffusion front, we see
reasonable matches between the Petrel model and the three PEBI models. The hybrid approach not only
reduces the cell number requirement, but also yields better resolution of simulation results for the reservoir
background. We don’t see too much difference between hexagon and rectangle background. Both cases
yield almost the same pressure graphs as seen in Figure 11 as well as BHP responses with production time,
because they are of almost the same background density.

Figure 10a—hybrid grids

Figure 10b—PEBI reference case

Figure 10c—PEBI and rectangles

Figure 10d—PEBI and hexagons


10 SPE-177480-MS

Figure 11—The effect of fracture background type on pressure graphs

Sensitivity of Fracture Refinement


In terms of well testing analysis, fracture refinement is usually required to accurately capture early flow
regimes. In this study, we investigated the effect of fracture refinement on both short-term and long-term
production. Figure 12a and Figure 12b correspond to the cases without refinement and with three
refinement gridblocks on both sides of fractures, respectively. If fracture refinement is required around
fractures, we basically combined the Tar-tan grid with PEBI grids to model fractures, and for the reservoir
background we use the structured Cartesian grids.

Figure 12a—Hydraulic fractures without refinement

Figure 12b—Hydraulic fractures with refinement

From the pressure graphs after 20 years’ simulation in Figure 13, we didn’t observe any significant
difference in either reservoir background or around fractures. The case with only 155,080 cells can
produce similar simulation results as the Petrel case and PEBI refinement case with 388,008 cells.
Obviously, the extra refinement around fractures don’t contribute too much to the later time production
behavior. Even for early time, we only see modest difference between the two cases with and without
refinement, because bottom-hole flowing pressure of the three wells is always above the bubble point and
we have single-phase fluid flow within the fractures.
SPE-177480-MS 11

Figure 13—The effect of fracture refinement on pressure graphs

In Figure 14, we summarized CPU performance for all the cases in the previous sensitivity studies. We
also scaled the CPU time according to the Petrel model. From the left to the right, we have the Petrel
model, the four cases for background density, the two cases for background type, and the last two cases
for fracture refinement. The less number of cells, the better the CPU performance. We don’t see too much
difference in CPU performance between rectangle background and hexagonal background. Besides, for
the best four cases in CPU performance, we also compared their pressure graphs. Except for the case with
190,440 cells with low background density, we see a reasonably close result. Among all the cases, the
PEBI case with 155,080 cells yields the best CPU performance without compromising too much accuracy
of the simulation results. Therefore, for this field case, PEBI mesh with the 155,080 cells should be the
best mesh for history matching and production prediction.

Figure 14 —Comparison of CPU performance for all the PEBI cases and the Petrel case

Sensitivity of Natural Fracture Density


One of the advantages of unstructured PEBI grid is to incooperate the effect of natural fractures. As seen
in Figure 15, if we have information about how the natural fractures are distributed, for example, from
seismic attribute maps or outcrop maps, we can generate different realizations of natural fractures to
history matching production data. In the center of each graph in Figure 15, we have the previous field
12 SPE-177480-MS

model with three horizontal wells. Then we created natural fractures from the seismic covariance maps.
We assume the same properties for natural fractures as hydraulic fractures. From left to right of Figure
15, we have four cases corresponding to 0, 20, 40, and 80 natural fractures. For each case, we generate
unstructured grid with the previous 2.5D PEBI meshing algorithms. The cell number ranges from 228,752
for the case without natural fractures to 826,288 for the case with 80 natural fractures.

Figure 15—2.5D PEBI grid for the field cases with extensively distributed natural fractures

Figure 16 shows pressure graphs after 20 years’ simulation. Characterization of natural fractures is very
important to accurately predict the Stimulated Reservoir Volume (SRV). With the increase of the number
of natural fractures, we see a significant increase in SRV. Three horizontal wells can produce reservoir
fluids from the regions far away from hydraulic fractures with the help of the opened natural fractures.
The CPU performance is around 12.9 hours for the case with 80 natural fractures, compared to 2.7 hours
for the case without natural fractures.

Figure 16 —The effect of natural fractures on production performance

Conclusions
A 2.5D PEBI grid workflow has been developed for flow simulation of unconventional reservoirs. First
of all, we discussed the main components of this workflow, especially the 2.5D PEBI mesh generator,
which is an extension of the previous 2D mesh generation algorithms. Throughout this paper, we
demonstrated the proposed gridding approach is accurate, robust, and efficient to discretize complex
SPE-177480-MS 13

fracture networks. Besides, we also compared the conventional grid solutions such as Tartan grid and
Petrel LGR grids with the PEBI grids. During the comparison, we applied the workflow to simulate two
cases – 1) a synthetic model to compare the flow regimes against known Tartan grid solutions, and 2)
another field case to investigate the effect of unstructured mesh density, type, refinement, natural fractures
on production performance.
The uncertainties or inaccuracies from the meshing related parameters are far less important than
fracture properties themselves. In other words, more effort should be focused on characterizing the
complexity of the fracture networks, especially when natural fractures are extensively distributed.
Speaking of the meshing related parameters, we found that the less number of cells, the better the CPU
performance. Fracture refinement shows modest influence to the early time production while background
density has dominant influence to the late time production. Background grid type and orientation have less
effect as long as they have the same mesh density. After comparing pressure graphs, bottom-hole flowing
pressure and CPU performance, we chose the hybrid mesh (PEBI grid ⫹ Rectangle grid) as the best mesh
yielding both better CPU performance and accurate simulation results. In general, we might want to
choose a PEBI mesh with the lowest number of cells, which can be achieved by removing unnecessary
refinement around fractures, increasing reservoir background size and reservoir background size ratio,
replacing unstructured background type with structured grids, and reducing the complexity of the fracture
networks.

References
Branets, L. V., Ghai, S. S., Lyon, S. L. et al. 2008. Challenges and Technologies in Reservoir
Modeling. Communications in Computational Physics 6:23.
Cipolla, C. L., Fitzpatrick, T., Williams, M. J. et al. 2011. Seismic-to-Simulation for Unconventional
Reservoir Development. Presented at the SPE Reservoir Characterisation and Simulation Confer-
ence and Exhibition, Abu Dhabi, UAE, 9-11 October. http://dx.doi.org/10.2118/146876-MS.
Fung, L. S. K., Ding, X. Y., Dogru, A. H. 2014. Unconstrained Voronoi Grids for Densely Spaced
Complex Wells in Full-Field Reservoir Simulation. 19 (05): 803–815. SPE-163648-PA. http://
dx.doi.org/10.2118/163648-PA.
Heinemann, Z. E., Brand, C. W., Munka, M. et al. 1991. Modeling Reservoir Geometry with Irregular
Grids. SPE Reservoir Engineering 9 (02):225–232. http://dx.doi.org/10.2118/18412-PA.
Jiang, J., Shao, Y., Younis, R. M. Development of a Multi-Continuum Multi-Component Model for
Enhanced Gas Recovery and Co2 Storage in Fractured Shale Gas Reservoirs, 2014/4/12/. 10.2118/
169114-MS.
Jiang, J., Younis, R. M. 2015. Numerical Study of Complex Fracture Geometries for Unconventional
Gas Reservoirs Using a Discrete Fracture-Matrix Model. Journal of Natural Gas Science and
Engineering 26:1174 –1186. http://dx.doi.org/10.1016/j.jngse.2015.08.013.
Kappa Reading Room Brochure. 2015. Kappa Unconventional Resources, October 2013, http://
www.kappaeng.com/documents/flip/ur-brochure/ (accessed Apri 23 2015).
Karimi-Fard, M., Durlofsky, L. J., Aziz, K. 2003. An Efficient Discrete Fracture Model Applicable for
General Purpose Reservoir Simulators. Presented at the SPE Reservoir Simulation Symposium,
Houston, Texas, 3-5 February. 79699. http://dx.doi.org/10.2118/79699-ms.
Merland, R., Lévy, B., Caumon, G. 2011. Building Pebi Grids Conforming to 3d Geological Features
Using Centroidal Voronoi Tessellations. IAMG Proceedings.
Moog, G. J. E. A. 2013. Advanced Discretization Methods for Flow Simulation Using Unstructured
Grids.
14 SPE-177480-MS

Muralidharan, V., Schechter, D. S., Chakravarthy, D. et al. 2004. Investigating Fracture Aperture
Distributions under Various Stress Conditions Using X-Ray Ct Scanner. Presented at the Canadian
International Petroleum Conference, Calgary, Alberta, 8 –10 June. http://dx.doi.org/10.2118/2004-
230.
Olorode, O. M., Freeman, C. M., Moridis, G. J. et al. 2012. High-Resolution Numerical Modeling of
Complex and Irregular Fracture Patterns in Shale Gas and Tight Gas Reservoirs. Presented at the
SPE Latin America and Caribbean Petroleum Engineering Conference, Mexico City, Mexico,
16 –18 April. http://dx.doi.org/10.2118/152482-ms.
Olson, J. E., Wu, K. 2012. Sequential Vs. Simultaneous Multizone Fracturing in Horizontal Wells:
Insights from a Non-Planar, Multifrac Numerical Model. Presented at the SPE Hydraulic Frac-
turing Technology Conference, The Woodlands, Texas, 6 – 8 February. http://dx.doi.org/10.2118/
152602-MS.
Palagi, C. L., Aziz, K. 1994. Use of Voronoi Grid in Reservoir Simulation. 2 (02):69 –77. SPE-22889-
PA. http://dx.doi.org/10.2118/22889-PA.
Romain, M., Caumon, G., Levy, B. et al. 2011. Building Centroidal Voronoi Tessellations for Flow
Simulation in Reservoirs Using Flow Information. Presented at the SPE Reservoir Simulation
Symposium, The Woodlands, Texas, USA, 21–23 February. http://dx.doi.org/10.2118/141018-
MS.
Sun, J., Schechter, D. S. 2014. Optimization-Based Unstructured Meshing Algorithms for Simulation
of Hydraulically and Naturally Fractured Reservoirs with Variable Distribution of Fracture
Aperture, Spacing, Length and Strike. Presented at the SPE Annual Technical Conference and
Exhibition, Amsterdam, The Netherlands, 27–29 October. http://dx.doi.org/10.2118/170703-MS.
Sun, J., Schechter, D. S. 2015. Optimization-Based Unstructured Meshing Algorithms for Simulation
of Hydraulically and Naturally Fractured Reservoirs with Variable Distribution of Fracture
Aperture, Spacing, Length and Strike. SPE Reservoir Evaluation & Engineering. SPE-170703-PA.
Sun, J., Schechter, D. S., Hu, K. et al. 2014. Investigating the Effect of Improved Fracture Conduc-
tivity on Production Performance of Hydraulic Fractured Wells through Field Case Studies and
Numerical Simulations. Presented at the SPE Hydrocarbon Economics and Evaluation Sympo-
sium, Houston, Texas, 19 –20 May. http://dx.doi.org/10.2118/169866-MS.
Verma, S., Aziz, K. 1997. A Control Volume Scheme for Flexible Grids in Reservoir Simulation.
Presented at the SPE Reservoir Simulation Symposium, Dallas, Texas, 8 –11 June. http://dx.do-
i.org/10.2118/37999-MS.
Vestergaard, H., Olsen, H., Sikandar, A. S. et al. 2008. The Application of Unstructured-Gridding
Techniques for Full-Field Simulation of a Giant Carbonate Reservoir Developed with Long
Horizontal Wells. 11 (06):958 –967. SPE-120887-PA. http://dx.doi.org/10.2118/120887-PA.
Yan, B., Killough, J. E., Wang, Y. et al. 2013. Novel Approaches for the Simulation of Unconven-
tional Reservoirs, 12-14 August. http://dx.doi.org/10.1190/URTEC2013-131.
SPE-177480-MS 15

Appendix A
Unstructured Grid Generation Algorithms for Modeling Fractures

Type Authors (Year) Significance Limitations

PEBI Heinemann et al. ● Reduce grid orientation effect. ● Grid refinement around internal and external features
and (1991) ● Better conform to geological features is required to avoid circumcenter lies outside the PEBI
GPEBI ● Better computation of inter-block transmissibility with cells.
TPFA. ● GPEBI run into problems for models with high per-
● Proposed k-orthogonal or GPEBI grids to handle full meability anisotropy.
anisotropic and symmetric permeability tensors.
CVFE Palagi and Aziz ● Solve problems regarding anisotropic and asymmetric -
(1994), and permeability, and permeability heterogeneity.
Verma and Aziz ● Proposed pre-defined geometrical grid modules.
(1997)
CVFE Karimi-Fard et al. ● Don’t explicitly represent the thickness of fractures, ● Require volume correction for fracture grid blocks.
(2003) and avoid computation of fracture intersections. ● Introduce errors in fracture pore volume if there are
● Applied a special technique based on the ⬙star-delta⬙ large number of fractures.
transformation to compute transmissibilities. ● Have to use MPFA to yield optimal accuracy.
PEBI Branets et al. (2008) ● Proposed to describe 2D features as Planar Straight ● For non-vertical fractures and partial well completion,
Line Graphs (PSLG). 3D PEBI grids are required.
● Proposed a new approach to model feature ● For 3D PEBI grids, gridding algorithms become very
intersections. complicated.
PEBI Vestergaard et al. ● Applied 2.5D PEBI to model large field reservoirs
(2008) with numerous horizontal wells to improve near-well
flow accuracy.
PEBI Cipolla et al. (2011) ● Explicit represent fractures with a constant aperture.
● Apply multipliers for both pore volume and
transmissibilities.
PEBI Olorode et al. (2012) ● Explicitly represent fractures.
● Applied very high grid resolution to model non-planar
fractures.
● Investigated characteristic flow regimes.
PEBI Fung et al. (2014) ● Proposed multi-level quad-tree LGR approach to
honor complex well geometry.
● Coupled with a paralleled unstructured reservoir sim-
ulation workflow.
PEBI Sun and Schechter ● Explicit represent fractures.
(2014) ● Proposed algorithms to discretize extensively clus-
tered fracture networks with non-uniform fracture ap-
erture and lower-angle fracture intersections.
PEBI KAPPA (2015) ● Same as Karimi-Fard’s approach except that addi-
tional refinement around fractures is used to capture
the transient effect.

You might also like