SPE-185719-MS

Efficient Modeling and History Matching of Shale Oil Reservoirs Using the
Fast Marching Method: Field Application and Validation

Atsushi Iino, Aditya Vyas, Jixiang Huang, and Akhil Datta-Gupta, Texas A&M University; Yusuke Fujita, JX Nippon
Oil & Gas Exploration Corporation; Neha Bansal and Sathish Sankaran, Anadarko Petroleum Corporation

Copyright 2017, Society of Petroleum Engineers

This paper was prepared for presentation at the 2017 SPE Western Regional Meeting held in Bakersfield, California, USA, 23 April 2017.

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
This paper demonstrates the novelty and practical feasibility of the FMM-based multi-phase simulation for
rapid field-scale modeling of shale reservoirs with multi-continua heterogeneity.
Modeling of unconventional reservoirs requires accurate characterization of complex flow mechanisms in
multi-continua because of the interactions between reservoir rocks, microfractures and hydraulic fractures. It
is also essential to account for the complicated geometry of well completion, the reservoir heterogeneity and
multi-phase flow effects. Currently, such multi-phase numerical simulation for multi-continua reservoirs
needs substantial computational time that hinders efficient history matching and uncertainty analysis. In
this paper, we propose an efficient approach for field scale application and performance assessment of shale
reservoirs using rapid multi-phase simulation with the Fast Marching Method (FMM).
The key idea of the reservoir simulation using the FMM is to recast the 3-D flow equation into 1-D
equation along the ‘diffusive time of flight’ (DTOF) coordinate, which embeds the 3-D spatial heterogeneity.
The DTOF is a representation of the travel time of pressure propagation in the reservoir. The pressure
propagation is governed by the Eikonal equation which can be solved efficiently using the FMM. The 1-
D formulation leads to orders of magnitude faster computation than the 3-D finite difference simulation.
The use of FMM-based simulation also enables systematic history matching and uncertainty analysis using
population-based techniques that require substantial simulation runs.
We first validate the accuracy and computational efficiency of the FMM-based multi-phase simulation
using synthetic reservoir models and comparison with a commercial finite-difference simulator. Next, we
apply our proposed approach to a field example in Texas for a multi-stage hydraulically fractured horizontal
well. The 3-D heterogeneous reservoir model was built and history matched for oil, gas and water production
using the Genetic Algorithm with the FMM-based flow simulation. Multiple history-matched models were
obtained to examine uncertainties in the production forecast associated with respect to the properties related
to hydraulic fractures, microfractures and the matrix.
2 SPE-185719-MS

Introduction
Unconventional reservoirs contribute to a significant share of the energy supply in the US and are predicted
to keep increasing their production in the next 25 years (EIA 2016). The economical exploitation from
such ultra-low permeability formations has been achieved by the technology breakthrough of extended-
reach horizontal drilling and multi-stage hydraulic fracturing (Alexander et al. 2011; King 2010), leading
to the complex fracture network that interacts with pre-existing/ induced fractures and reservoir rock.
Furthermore, the fluid transport modelling used in conventional reservoirs has been found often inadequate
for unconventional reservoirs due to the complicated underlying physics such as gas desorption, Knudsen
diffusion, phase behavior in nano-scale pores and stress-dependent fracture conductivity (Najabaei et al.
2013; Cho et al. 2013; Wang 2015). Thus, robust performance assessment of the unconventional reservoirs
remains a technical challenge as it requires an integrated modelling of a hydraulic fracture network, reservoir
heterogeneity and relevant physics.
In the current industrial practice, various empirical and analytical techniques have been widely used to
predict oil and gas production from unconventional reservoirs because of their simplicity. The commonly
used approach among empirical methods is decline curve analysis. On the basis of the well-known Arps'
decline curve model, many researchers have developed empirical formulations to fit the time-rate history
observed in unconventional reservoirs (Duong 2011; Valko 2009; Ilk et al. 2008). However, the reliability of
the decline curve analysis depends on the quantity and quality of the available production data and may result
in erroneous prediction of future performance due to the absence of underlying physical theory. On the other
hand, analytical methods such as pressure transient analysis (PTA) and rate transient analysis (RTA) are
based on the physical theory described by the diffusivity equation. In these approaches, the parameters in the
well and reservoir models will be calibrated such that the model replicates the observed rate/ pressure history.
However, the conventional PTA and RTA assumes single phase, homogenous reservoir properties, simple
geometries of reservoir boundary and planar hydraulic fractures. Although recent studies have incorporated
the multi-phase effects (Clarkson and Qanbari 2015; Tabatabaie and Pooladi-Darvish 2016), it still requires
the assumption of homogeneous reservoir properties and simple planar geometry for the hydraulic fractures.
In unconventional reservoirs, there is significant impact on the performance prediction due to the reservoir
heterogeneity and complex geometry and network of multi-stage hydraulic fractures (Kam et al. 2015;
Cipolla et al. 2010a; Cipolla 2011).
Numerical simulation is also widely used to evaluate the performance of multi-stage hydraulically
fractured wells in unconventional reservoirs (Wang 2015). The advantage over the empirical and analytical
approaches is the capability of incorporating complex underlying physics. Several recent studies have used
reservoir simulation to model the geometry of hydraulic fractures, stress-dependent reservoir properties,
reservoir heterogeneity and multi-phase effect (Cipolla et al. 2010b; Diaz de Souza et al. 2012; Novlesky et
al. 2011; Du et al. 2009). However, the substantial computational time required is often a bottleneck for the
practical application of the numerical simulation especially when high resolution models are involved to
accurately describe the hydraulic fracture geometry, flow in the vicinity of the hydraulic fractures, reservoir
heterogeneity and multiple continua. Thus, there is an increasing need for a novel approach that offers high
computational efficiency as well as the flexibility to incorporate the relevant physics.
Recently, FMM based simulation has been proposed, which is a bridge between the two approaches
discussed above: analytical approach and full 3-D numerical simulation. Datta-Gupta et al. (2011)
generalized the concept of radius of investigation to heterogeneous reservoirs by introducing a high
frequency asymptotic solution of the diffusivity equation in heterogeneous and fractured reservoirs. The
high frequency solution leads to the Eikonal equation which is solved for the DTOF that governs the
propagation of the ‘pressure front’ in the reservoir and provides the arrival time of ‘peak’ pressure responses
at any location (Vasco and Datta-Gupta 2016). The Eikonal equation can be solved very efficiently using
the FMM, which is a class of front-tracking algorithms and it sequentially looks for the shortest paths from
SPE-185719-MS 3

the source point (Sethian 1996; Sethian 1999; Hassouna and Frag 2007). Once the DTOF is obtained in
the reservoir domain, well-drainage volume at a given time can be readily calculated by summing up the
pore volume within the DTOF contour. The well-drainage volume and associated functions of the DTOF
provides the intuitive understanding on how the drainage volume propagates in the reservoir and assist flow
regime identification (Datta-Gupta et al. 2011; Xie et al. 2015; Yang et al. 2015; Xue et al. 2016).
Analogous to the streamline simulation (Datta-Gupta and King 2007), the DTOF can be utilized as a
1-D spatial coordinate as it embeds reservoir heterogeneity. Zhang et al. (2016) proposed a DTOF-based
numerical simulation by decoupling the 3-D flow equation into 1-D equation along the DTOF coordinate.
This approach consists of two decoupled steps—calculation of the DTOF using the FMM and fully implicit
1-D simulation using the grid blocks defined on the DTOF coordinate. The 1-D simulation demonstrated
orders of magnitude faster computation speed for field-scale application with million-cell models, while
maintaining good accuracy compared to the commercial finite difference (FD) simulator. Fujita et al. (2016)
further extended the DTOF-based 1-D simulation to incorporate the complex flow physics in nano-porous
shale gas such as gas adsorption, Knudsen diffusion and triple-continua system. However, the application
to date has focused on single-phase gas flow.
In this study, the FMM-based 1-D simulation was extended to the three-phase flow problems in
unconventional reservoirs with reservoir heterogeneity and multi-stage hydraulically fractured well. First,
we will briefly describe the background of the FMM and the 1-D simulation along the DTOF. Then we
develop the 1-D mathematical formulation that accounts for the multi-phase flow. Synthetic examples
will be presented to illustrate the validity and the computational efficiency of our proposed approach by
comparison with the commercial FD simulator. Finally, the field application demonstrates the power and
utility of our proposed approach for efficient modeling and history matching of a multi-stage hydraulically
fractured horizontal well in a shale oil reservoir.

Background
Our proposed approach consists of the two main steps: calculation of the DTOF on the original multi-
dimensional grid blocks and 1-D multi-phase simulation on the DTOF coordinate. In this section, we first
describe the concept of the DTOF and Eikonal equation that governs the pressure ‘front' propagation in the
reservoir, followed by the 1-D simulation along the DTOF coordinate.

DTOF and Eikonal Equation Governing the Pressure Front Propagation.

Our proposed simulation approach is based on a high frequency asymptotic solution of the diffusivity
equation in heterogeneous and fractured reservoirs (Datta-Gupta et al. 2011, Vasco and Datta-Gupta 2016).
The high frequency solution leads to the Eikonal equation which is solved for DTOF τ that governs the
propagation of the ‘pressure front’ in the reservoir:

(1)

where α represents the diffusivity defined as α = k/ϕμct that can be spatially heterogeneous. While the
applicability of the traditional concept of radius of investigation is limited to the case with uniform
diffusivity (Lee 1982), eq. (1) is a generalized form of the radius of investigation for heterogeneous
reservoirs. The FMM is an efficient solution technique for the Eikonal eq. (1) and the DTOF τ can be readily
obtained in minutes for high resolution models with millions of grid blocks (Sethian 1996; Sethian 1999).
Figure 1 shows 2-D examples of the DTOF for a homogeneous and a heterogeneous reservoir. For the
homogeneous case with a vertical well (Figure 1a), the DTOF indicates the uniform propagation of the
pressure ‘front’, which is consistent with the calculation by the traditional radius of investigation as depicted
by a solid black line. Our proposed approach has the capability of providing the DTOF for reservoirs with
4 SPE-185719-MS

spatial heterogeneity and arbitrary well completions (e.g. horizontal well and hydraulically fractured well
as illustrated in Figure 1b and 1c). Thus, our approach is well-suited for application to unconventional
reservoirs that require modelling the complicated interaction between reservoir heterogeneity and the well
geometry.

Figure 1—Examples of DTOF for homogeneous and heterogeneous reservoir. (a)

DTOF with vertical well for homogeneous reservoir. (b) Heterogeneous permeability. (c)
DTOF with horizontal well for heterogeneous reservoir (after Datta-Gupta et al., 2011).

Decoupling 3-D Flow Equation into 1-D Equation along the DTOF
Zhang et al. (2016) proposed a novel approach for a DTOF-based numerical simulation associated with
the transformation of the fluid-transport coordinate from the physical 3-D space to the 1-D DTOF space.
Analogous to the streamline simulation (Datta-Gupta and King 2007), the DTOF is utilized as a spatial
coordinate embedding the spatial heterogeneity for computing the pressure (Fujita et al. 2016):

(2)

where subscript ‘ref’ stands for the reference condition at which the DTOF was generated. Using eq. (2),
the pressure diffusivity equation can be written as:

(3)

where the sink/ source terms are expressed in terms of flow rate per unit bulk volume. The diffusivity
equation defined on the 1-D DTOF coordinate has direct analogy to the radial flow equation in homogeneous
medium (Figure 2). The physical distance r in the radial flow equation has been replaced by the DTOF that
is spatially distorted as per the reservoir heterogeneity.
SPE-185719-MS 5

Figure 2—Analogy between the DTOF as a spatial coordinate and

radial coordinate in a homogeneous reservoir (Zhang et al. 2016).

The w(τ) function in eq. (2) is defined as a derivative of the drainage pore volume with respect to DTOF:

(4)

The DTOF is mapped on every grid block after the Eikonal eq. (1) is solved by the FMM. The pore
volumes of each grid block are then sorted and tabulated as per the DTOF. Subsequently, the drainage pore
volume Vp(τ) is calculated by summing up the pore volume of the grid blocks within a specific τ-contour,
followed by the calculation of w(τ) by numerical differentiation of Vp(τ) with respect to the DTOF. As we
deal with the discrete relationship between Vp(τ) and the DTOF, an appropriate smoothing technique must be
applied to the numerical differentiation in order to capture the physical characteristics of Vp(τ) function. In
this study, the smoothing technique that is commonly employed in the well test derivative was used (Horne
1995). Figure 3a and 3b illustrate the examples of Vp(τ) and w(τ) functions that correspond to the cases
shown in Figure 1a and 1c, respectively. The w(τ) function is defined as the derivative of drainage volume
with respect to DTOF and it physically implies the surface area associated with pressure front propagation
and reflects the flow regimes (Yang et al. 2015; Xue et al. 2016). For instance, a linear increase in w(τ)
indicates radial flow, a flat line indicates linear flow and a decrease in w(τ) implies the boundary effects.

Figure 3—Example of the drainage volume and w as functions of the DTOF.

Extension to Multi-phase Flow

Thus far, we have discussed the single-phase case. In this section, we will derive the multi-phase flow
equations along the 1-D DTOF coordinate by a transformation of the 3-D equations using DTOF as the
spatial coordinate.
6 SPE-185719-MS

DTOF for Multi-phase System

For the multi-phase case, the total mobility should be accounted for in the diffusivity computation:

(5)

By substituting the multi-phase diffusivity (5) into eq. (1), we obtain the Eikonal equation that governs
the DTOF for the multi-phase system. The derivation of the multi-phase Eikonal equation is detailed in
Appendix-A. Figure 4 illustrates an example of diffusivity in a 2-D and two-phase reservoir with the
homogeneous porosity and permeability and nonuniform initial water saturation. With the given distribution
of water saturation as depicted in Figure 4a, the two-phase diffusivity was calculated as the summation of
water and oil diffusivities (Figure 4b through 4d). The corresponding DTOFs were also calculated using the
water, oil and two-phase diffusivities (Figure 4e through 4g). Since the propagation of the pressure front in
the multi-phase system is controlled by the total mobility, Figure 4e and 4f are obviously not the appropriate
solutions for multi-phase flow.

Figure 4—Initial water saturation, diffusivity and DTOFs: (a) Initial water saturation, (b) water diffusivity, (c)
oil diffusivity, (d) two-phase diffusivity, (e) water-phase DTOF, (f) oil-phase DTOF and (g) two-phase DTOF

Dual-porosity Formulation along the 1-D DTOF Coordinate

For modelling the shale reservoirs with the complicated flow mechanisms between multi-continua, the dual-
porosity model (Warren and Root 1963) was used in this study. As illustrated in Figure 5, the dual-porosity
system consists of two porous media with distinct reservoir properties: fracture and matrix. The fracture has
a higher conductivity and plays a role of main flow path for fluid flow. The matrix, on the other hand, plays
as a storage or sink/ source term communicating with fracture but no convective flow in-between.
SPE-185719-MS 7

Figure 5—Illustration of dual-porosity system.

Three-phase Dual-porosity Formulation

In the absence of gravity and capillary forces, the mass balance equations in the fracture medium of the dual-
porosity system are written as the following equations (Kazemi et. al. 1976; Gilman and Kazemi 1983):

(6)

(7)

(8)

The fluid transfer term between the fracture and matrix is given by the following (Gilman and Kazemi
1983):

(9)

where the subscript j stands for phases and σ is a shape factor that represents the connectivity between the
matrix block and the surrounding fracture network. The matrix only have fluid transfer with the fracture
and there is no within the matrix:

(10)

(11)

(12)

Coordinate Transformation
We can define the coordinate transformation for the multi-phase system as follows:

(13)
8 SPE-185719-MS

where DTOF τ is used as a spatial coordinate here. The derivation of eq. (13) is detailed in Appendix B.
By applying the coordinate transformation of eq. (13), we can transform the 3-D equations (8) through (8)
for the fracture into 1-D equations defined on the DTOF coordinate:

(14)

(15)

(16)

where the Dirac delta δ(τwb) means that the sink/ source term only appears at the inner boundary of τ = τwb in
the well grid. In the discretized domain along the DTOF coordinate, the sink/ source term can be rewritten
in terms of the bottomhole pressure as:

(17)

where the subscript 1 indicates the grid block in the vicinity of the well on 1-D DTOF grid blocks.
Accounting for the effective radius, the well index WI is defined as follows for the multiple completions:

(18)

where Nc stands for the number of completion and τ1 is the DTOF value at the outer edge of the grid block
adjacent to the well.
In our proposed approach, eqs. (14) through (16) and (10) through (12) are solved for pressures and
saturations on 1-D DTOF grid with the fully implicit method. The important assumption in the coordinate
transformation is that the direction of the saturation change follows the DTOF gradient as well as the
pressure gradient. As verified in the examples presented later, this assumption is reasonable in the problems
in which the viscous force is dominant. Another assumption is that every parameter in the transformed
equations is defined on the 1-D DTOF grid block. Other than the parameters related to the diffusivity in the
fracture that have been already embedded in the DTOF, all the parameters such as rock types, shape factor
and matrix porosity and permeability must be averaged within the DTOF contour.

Simulation Workflow Using the Fast Marching Method

Our proposed simulation approach consists of two decoupled steps: calculating the DTOF on the original
multi-dimensional grid and 1-D simulation on the DTOF coordinate. The illustrative workflow is shown in
Figure 6. First, the multi-phase diffusivity is calculated for every grid block. The Eikonal equation is then
solved by the FMM for the DTOF, followed by the calculation of the drainage pore volume Vp as a function
of the DTOF. Subsequently, the drainage pore volume is discretized on 1-D DTOF coordinate and the w(τ)
SPE-185719-MS 9

function is calculated. The 1-D simulation will be finally performed using the pore volume, transmissibility
and well index defined on the 1-D DTOF system.

Figure 6—Flow chart of multi-phase 1-D simulation based on the FMM.

It should be noted that the DTOF is calculated only at initial condition and not updated during the 1-D
simulation. In theory, the DTOF contour is continuously varying due to the pressure and saturation change.
However, we use the DTOF as a spatial variable for coordinate transformation. As long as the assumption
behind the coordinate transformation is valid, the specific time chosen for DTOF should not matter.

Validation Using Synthetic Examples

We tested our proposed simulation approach using synthetic reservoir models. First, 3-D homogeneous
example with a vertical producer will be presented to show the accuracy and computational efficiency of
our proposed approach. The 3-D heterogeneous example with multi-stage hydraulically fractured well will
follow to demonstrate the applicability to the unconventional reservoir problem.

3-D Homogeneous Example with a Vertical Well

We first present the 3-D homogeneous example to demonstrate the accuracy and computational efficiency
of the FMM-based simulation. Table 1 shows the reservoir properties and initial conditions for the fracture
and matrix. The model size was 4100' ×4100'× 50' dimensioned by 41×41×10 (16,810) grid blocks. The
reservoir is undersaturated with the solution gas-oil-ratio (GOR) of 1,345 scf/stb at the initial pressure of
6,000 psi whereas the bubble point pressure is 2,860 psi. The blackoil table was generated based on the
Bakken (Najabaei et al. 2013). The producer operating with the constant bottomhole pressure of 2,000 psi
was placed at the center of the reservoir and completed vertically in all layers.
10 SPE-185719-MS

Table 1—Reservoir properties and initial condition for 3-D synthetic and homogeneous example

Items Fracture Matrix

Reservoir Property Porosity 0.01 0.10

Permeability (mD) 2 10-4

kv/kh 0.1 -

Shape factor (ft-2) 0.15 ft-2 -

Rock compressibility 1.0×10-6 1.0×10-6

(psi-1)

Initial Condition Pressure (psi) 6000 6000

Water saturation 0.0 0.40

Simulation Results
The DTOF, Vp and w(τ) were computed as depicted in Figure 7. As the producer is completed in entire
thickness, the pressure front expands laterally in the absence of gravity and capillarity. The w(τ) function
shows a straight line until the DTOF of 28 hr-1/2, followed by the decrease towards zero. Again, the w(τ)
function physically implies the surface area of the pressure front and indicates flow regimes. The linear
increase and subsequent decrease of the w(τ) function indicate the radial flow regime and the boundary
effect, respectively.

Figure 7—Calculated DTOF, drainage volume and w function for homogeneous example.

Next, the 1-D simulation was performed on the DTOF coordinate. The oil rate, GOR and water rate
simulated by the FMM-based approach are compared with a commercial FD simulator in Figure 8. For all
the three phases, the FMM-based simulation provides a close matching with the FD simulation.

Figure 8—Comparison between FMM (solid lines) and FD simulation (symbols) for homogeneous example.
SPE-185719-MS 11

CPU Time
Computational efficiency of our proposed approach was also studied. Figure 9 illustrates the CPU time
comparison with the FD simulator. The same dataset as the above example was used except the number
of grid blocks; the reservoir domain was further refined into 101×101×10 (102,010) and 317×317×10
(1,004,890). As the number of grid blocks increases, more significant improvement in computation
efficiency can be seen in our proposed approach that is based on the 1-D formulation. At least two orders
of magnitude faster computation in FMM-based simulation is expected in simulation with millions of grid
cells, as compared to the FD simulation.

Figure 9—Comparison of CPU time (bottom) between commercial

FD simulator and FMM-based simulation and speed-up factor (top).

3-D Heterogeneous Example with Multi-stage Hydraulic Fractured Horizontal Well

We present here a 3-D synthetic example that includes reservoir heterogeneity and a multi-stage
hydraulically fractured well.
A tartan grid was used with 200×394×5 (0.394 million) grid blocks where the grid sizes are uniform in
x- and z-directions (ΔX=10’ and ΔZ=20’) and logarithmically resolved near the hydraulic fractures in y-
direction (Figure 10). A horizontal well was completed with 40 transverse hydraulic fractures where the
cluster and stage spacing were 60’ and 200’, respectively. The individual hydraulic fracture was represented
by the grid blocks with the dimension of 320’× 1’× 100’. Accounting for the fracture width of 1’ that is
much larger than reality, the effective properties rather than intrinsic properties were assigned for hydraulic
fractures as summarized in Table 2. The matrix was assumed to be homogeneous. The heterogeneous
properties for natural fractures are illustrated in Figure 11. In the simulation of shale reservoirs, the initial
water saturation accounting for the completion fluids can be a key variable for the history matching (Diaz
de Souza 2012). In this example, the initial water saturations of 0.9 and 0.8 were assigned for hydraulic
and natural fractures around the wellbore, respectively. The transmissibility reduction in the hydraulic and
natural fractures due to the compaction was modelled as a function of pore pressure (Cipolla 2010), as
depicted in Figure 12. The reservoir was initially undersaturated at the pressure of 4,000 psi and the bubble
point pressure of 2861 psi. The horizontal well was operated at the constant bottomhole pressure of 2,000
psi, neglecting the pressure loss in the wellbore.
12 SPE-185719-MS

Table 2—Properties of hydraulic fractures, natural fractures and matrix for synthetic heterogeneous example

Figure 10—Configuration of hydraulic fractures for 3D synthetic example.

Figure 11—Reservoir properties and initial water saturation for the fracture in synthetic heterogeneous example.

Figure 12—Input data for transmissibility reduction due to compaction

DTOF and w(τ) Function

The DTOF, drainage pore volume and w(τ) function in the fracture system were computed as shown in
Figure 13. Again, we can readily identify the expected flow regimes using w(τ) function and the DTOF map.
In this example, six (6) distinct flow regimes were identified as captioned (a) through (g) in Figure 13 and
SPE-185719-MS 13

the corresponding DTOF maps were depicted in Figure 14. Because of the small contrast of diffusivities
between the hydraulic and natural fractures, no clear indication can be found for linear flow around each
hydraulic fracture. Instead, the sharp increase (a) in the beginning indicates the radial flow in vertical
direction, followed by the boundary effect represented by the steep decrease at (b). After the interference
between stages seen at (c), the formation linear flow in x-direction starts as indicated in the flow regime (d).
When the pressure propagation felt the boundaries in x-direction, subsequent linear flow (e) in y-direction
would start. Finally, the complete pseudo-steady state (f) will be established that leads to w(τ) function
towards zero.

Figure 13—Calculated DTOF (left), drainage pore volume and w(t) function (right) for heterogeneous example.

Figure 14—DTOF map corresponding each flow regime (a) through (f) shown in Figure 13

Simulation Results
The 1-D simulation was conducted using the DTOF as a spatial coordinate and the results were compared
with the commercial FD simulator. Figure 15 shows the simulated well performance. It can be confirmed that
the FMM-based simulation gives consistent results with the FD simulation for all three phase productions.
Initially, high water production rate was seen due to the high water saturation in and around the hydraulic
fractures, followed by the stable rate that comes from the matrix (Figure 15c and 15d). GOR keeps increasing
as the bottomhole pressure (2,000 psi) is less than the bubble point pressure of 2,830 psi. The run time by
the FD simulation was 1,150 seconds while the FMM only needed 134 seconds, which offered an order of
14 SPE-185719-MS

magnitude faster computation. Thus, our proposed approach is capable of efficient simulation with the high
resolution heterogeneous models and multi-phase effects.

Figure 15—Comparisons of simulation results between FMM and commercial FD simulator.

Field Application
Next, we studied the application of our proposed method to the field example of the shale oil reservoir in
Texas where the matrix permeability ranges between micro- and nano-Darcy. We modelled the reservoir
section of the single horizontal well with 10-stage hydraulic fractures. The simulation model was then
history-matched and used for the production forecast.

Production Data
The three-phase production and the wellhead pressure data are available for 380 days as illustrated in Figure
16. The well opened at 1300 bbl/d liquid rate and declined to 100 bbl/d during the observation period. High
water cut during the early production was due to the recovery of completion fluid. The production GOR
stayed constant of 2.1 Mscf/stb for the first three months and then kept on increasing. This indicates that the
reservoir was initially undersaturated and the liberation of solution gas started due to pressure depletion.
SPE-185719-MS 15

Figure 16—Production data of 10-stage hydraulically fractured well in a shale oil reservoir:
wellhead pressure and oil, gas and water rates (Top) and GOR and water cut (bottom).

Base Case Model

The base case model was built after Zhang et al (2016). The modelled reservoir section is dimensioned
7,100’× 2,500’× 180’ and resolved into 71×25×13 (23,075) grid blocks. The reservoir is undersaturated
at the initial pressure of 3,953 psi against the bubble point pressure of 2,930 psi. A dual-porosity model
is assumed. Three distinct regions were defined for the fracture as illustrated in Figure 17: hydraulic
fractures, Stimulated Reservoir Volume (SRV) and non-SRV. Stages of the transverse hydraulic fractures
are represented by the simulation grid blocks highlighted by red color.

Figure 17—Definition of regions for base case model.

Matrix properties were derived from core and log interpretations. For the purpose of the study, average
matrix reservoir properties have been used: porosity, permeability and initial water saturation were 0.08,
2.7×10-5 mD and 0.41, respectively. The connate water saturation was set to be 0.29. The reservoir
heterogeneity was accounted for the fracture as per the regions defined above and shown in Figure 18.
Due to the lack of data such as micro-seismic that indicates the SRV extent and production logging data
to characterize the contribution to the production from individual hydraulic fractures, the base case values
for the fracture and dimensions of hydraulic fractures and SRV were determined by the operator's prior
experiences and to be calibrated in the history matching presented later. The reduction in pore volume and
permeability due to the compaction was modelled as functions of pressure (Figure 19) and to be tuned as
well during history matching.
16 SPE-185719-MS

Figure 18—Reservoir properties of the base case model for fracture.

Figure 19—Rock compaction effect on pore volume (left) and transmissibility (right).

Benchmarking with Commercial FD Simulator

The base case simulation using our proposed approach was conducted for the history period of 380 days
under the wellhead pressure constraint. The DTOF, Vp and w(τ) functions for the base case model are
illustrated in Figure 20. Again, the DTOF is calculated only for the fracture in the dual-porosity model.
From the DTOF map and w(τ) function, we can readily identify the four distinct flow regimes that imply
the hydraulic fracture, SRV and Non-SRV. The Non-SRV region gives no significant contribution to the
fracture drainage volume.

Figure 20—DTOF (left) and Vp and w function (right) for base case model.

Figure 21 shows the well performance simulated by our proposed approach and a commercial FD
simulator. For all the three phase production, the FMM-based simulation provides equivalent results with
the FD simulation. In Figure 21, simulation results are also compared to the observed data where initial
oil and water rate and the oil decline after 250 days from the simulation deviates. In the history matching
presented later, the FMM-based approach was used for the forward simulation and the model parameters
were calibrated to replicate the historical data. The FMM offered a three times faster computation compared
to the FD simulation.
SPE-185719-MS 17

Figure 21—Comparison between FMM and commercial FD simulator for base case
simulation. Blue solid line: FMM, red dashed line: FD simulation, symbols: observed data.

Uncertain Parameters and Sensitivity

Due to the limited data availability, the significant uncertainty lies in the fracture properties and dimensions
of hydraulic fractures and SRV. Table 3 lists the uncertain parameters with the base values and ranges
used for the sensitivity study. The terminologies for the dimensions of hydraulic fractures and SRV were
defined as illustrated in Figure 22. For the sensitivity and history matching purpose, the 10-stage hydraulic
fractures and SRVs were divided into three groups that have uniform properties. The sensitivity of the rock
compaction was investigated by providing different rock tables as illustrated in Figure 23.

Figure 22—Setup of geometric parameters (left) and groups (right) for sensitivity study and history matching.
18 SPE-185719-MS

Table 3—Parameter uncertainties for sensitivity and history matching

Regions Uncertain parameters Base Low High

Porosity (HF_poro1, HF_poro2, HF_poro3) 0.01 0.005 0.02

Permeability (HF_perm1, HF_perm2, HF_perm3), mD 0.20 0.55 3.0

Water saturation (HF_Swi) 0.4 0.2 0.9

Hydraulic Shape factor (HF_sigma1, HF_sigma2, HF_sigma3), ft -2 0.005 0.0025 0.5

Fracture
Compaction table (HF_comp) 2 2 11

Fracture half length (HF_Xf1, HF_Xf2, HF_Xf3), ft 50 50 150

Fracture height (HF_h1, HF_h2, HF_h3), ft 60 40 100

Stage length (HF_len1, HF_len2, HF_len3), ft 300-400 500-600 500-600

Porosity (SRV_poro1, SRV_poro2, SRV_poro3) 0.01 0.005 0.012

Permeability (SRV_perm1, SRV_perm2, SRV_perm3), mD 0.1 0.01 0.2

Water saturation (SRV_ Swi1, SRV_ Swi2, SRV_ Swi3) 0.175 0.35 0.7
SRV Compaction table (SRV_ comp) 2 2 11

Shape factor (SRV_ sigma1, 1.25×10-3 1.25×10-4 0.02

SRV_ sigma2, SRV_ sigma3), ft-2

SRV_Width (SRV_W1, SRV_W2, SRV_W3), ft 500 300 900

Porosity (Mat_poro) 0.08 0.059 0.094

Permeability (Mat_perm), mD 2.7×10 -5

2.3×10 -7
1.3×10-4
Matrix
Water saturation (Mat_Swi) 0.41 0.3 0.77

Connate water saturation (Mat_Swc) 0.7*Swi 0.5*Swi 1.0*Swi

Figure 23—Rock compaction table used for sensitivity study. Grater

table number represents more severe transmissibility reduction.

For the sensitivity study, the objective function was defined as the summation of misfits in the cumulative
production for each of the three phases as:

(19)

where m is a set of reservoir parameters and Δ represents the misfit from the observed data. Figure 24 shows
a tornado diagram of the objective function with respect to reservoir parameters listed in Table 3. For the
hydraulic fractures, water saturation, shape factor and compaction have significant impact on the objective
function. The half-length of hydraulic fractures is also influencing while the height and stage lengths are
less sensitive. For the SRV, the most impacting parameter is the shape factor but all the other parameters
also have significant impact. The matrix porosity is less sensitive as even the low case has enough volume
SPE-185719-MS 19

within the history period, while the matrix permeability is a heavy hitter that directly relates to the pressure
support from the matrix to the fracture.

Figure 24—Sensitivity analysis of individual uncertain parameter.

Low and high cases are colored in blue and orange, respectively.

History Matching
The history matching was carried out following the method outlined by Zhang et al (2016), which utilized
the Genetic Algorithm (GA) facilitated by use of experimental design and response surface modelling (Yin
et al. 2011). For the validation purpose, the observed data for the first 300 days was used for the history
matching and the rest 80 days was used as the prediction where the production data is available.
The well performances simulated with 50 models selected based on the objective function were compared
with the observed data as illustrated in Figure 25 where the best model is highlighted by a purple line. The
selected models showed a good agreement with the three-phase production data with some variations. The
model parameters are summarized in Table 4. As expected, model parameters were not well constrained
due to the non-uniqueness. Figure 26 shows fracture permeability of the best model. The permeabilities of
hydraulic fractures in Groups 1 and 3 were increased while that in group-2 were reduced to the same order
as SRV, which confirms that all the hydraulic fracture stages do not equally contribute to the production.
The fracture height, stage length and SRV width were also altered in the history matching. Again, such
parameters could be better constrained and quantitatively characterized if there existed the additional data
such as micro-seismic and production logging.

Figure 25—Simulated well performances (lines) after history matching compared with the observed data (symbol).
20 SPE-185719-MS

Table 4—Model parameters after history matching

Figure 26—Fracture permeability before and after history matching.

Future Prediction
The production forecast was conducted for another 1620 days using the selected 50 models. The prediction
run was first constrained with the last liquid rate and switched to the wellhead pressure if it reaches the
minimum operating pressure of 200 psi. Figure 27 shows the predicted cumulative production of oil, water
and gas and GOR. For all the three phases, variations can be seen between the selected models, leading to the
range of uncertainty in remaining recovery. The CDF of the expected incremental recoveries are also shown
for oil and gas in Figure 27. Unlike the empirical or analytical methods, our proposed approach involves
a prediction not only for oil but also for gas and water that might influence the effective productivity and
the vertical flow. Thus, the proposed workflow assists the robust and systematic uncertainty analysis that
honors the underlying physics and the available production history.
SPE-185719-MS 21

Figure 27—Future prediction with multiple history matched models (lines) with the observed data (symbol).

Conclusions
In this paper, we have proposed a novel approach that is based on the FMM for rapid multi-phase simulation
in unconventional reservoirs. Our proposed approach is a bridge between simplified analytical methods
and full 3-D numerical simulation and offers rapid computation that incorporates reservoir heterogeneity,
complicated completion geometry, three-phase flow and relevant physics. The conclusions arising from this
study are the followings.

• We have developed the mathematical formulation for FMM-based 1-D simulation that involves
three-phase flow and dual-porosity model. The 3-D flow equations were transformed into 1-D
equation along the DTOF coordinate where the spatial heterogeneity has been reflected.
• The workflow for multi-phase simulation using the FMM was proposed. We first generate the
multi-phase DTOF on the original grid block system. Subsequently, the multi-phase simulation is
carried out using the 1-D DTOF grid blocks. We never revisit the original 3-D grid block system
once the DTOF is generated.
• The proposed approach was validated using synthetic 3-D examples incorporating multi-phase
flow. It was confirmed that the FMM-based simulation shows a good accuracy and orders of
magnitude faster computation at the field scale, as compared to the commercial FD simulator.
Additional advantage of the FMM is flow visualization using the DTOF and w(τ) function, which
assists us to easily identify the flow regimes.
• The field application of history matching was studied for a multi-stage hydraulically fractured well
in a shale oil field in Texas. It demonstrated that the rapid simulation by the FMM-based approach
assists efficient history matching using the population-based technique that requires a large number
of simulation runs. Multiple history matched models were generated and used for the assessment
of uncertainty ranges in the reservoir properties and remaining recovery.
22 SPE-185719-MS

Acknowledgements
The authors would like to acknowledge financial support from members of the Texas A&M University
Joint Industry Project, Model Calibration and Efficient Reservoir Imaging. A special thanks to Anadarko
Petroleum Corporation for providing the data and support for the case study.

Nomenclature
B formation volume factor, rbbl/STB
c compressibility, psi-1
h thickness or height, ft
hf fracture height, ft
k permeability, mD
p pressure, psi
pw bottomhole pressure
q flow rate, ft3/day
r radius, ft
Rs solution gas-oil ratio, scf/STB
S saturation
s skin factor
t time, day
u Darcy velocity, ft/day
Vp pore volume, ft3
w derivative of drainage pore value with respect to t, ft3/hr-1/2
wf fracture width, ft
WI well index, rbbl/day/psi×cp
Xf fracture half length, ft
α diffusivity, mD×psi/cp
δ Dirac delta
porosity
Γ transfer function, 1/day
λ phase mobility, 1/cp
μ viscosity, cp
σ shape factor, ft-2
ρ density, lb/ft3
τ diffusive time-of-flight
ω Frequency, 1/sec

Subscripts
c connate
f fracture
h horizontal
i initial
o oil
g gas
m matrix
mp multi-phase
ref reference condition
SPE-185719-MS 23

t total
w water
wb wellbore
v vertical

References
Alexander, T., Baihly, J., Boyer, C. et al. 2011. "Shale Gas Revolution." Oilfield Review 23 (3): 40–55.
Cho, Y., Apaydin, O. G. and Ozkan, E. 2013. "Pressure-Dependent Natural-Fracture Permeability in Shale and Its Effect
on Shale-Gas Well Production." SPEREE 16 (2): 216–228. http://dx.doi.org/10.2118/159801-PA.
Cipolla, C. L., Fitzpatrick, T., Williams, M. J. et al. 2011. "Seismic-to-Simulation for Unconventional Reservoir
Development." SPE Reservoir Characterisation and Simulation Conference and Exhibition. Abu Dhabi, UAE.
SPE146876. http://dx.doi.org/10.2118/146876-MS.
Cipolla, C. L., M. J. Williams, and X. Weng. 2010a. "Hydraulic Fracture Monitoring to Reservoir Simulation:
Maximizing Value." presented at SPE Annual Technical Conference and Exhibition. Florence, Italy. SPE133877. http://
dx.doi.org/10.2118/133877-MS.
Cipolla, C. L., Lolon, E. P., Erdle, J. C. et al. 2010b. "Reservoir Modeling in Shale-Gas Reservoirs." SPEREE 13 (4):
638–653. http://dx.doi.org/10.2118/125530-PA.
Clarkson, C. R. and Qanbari, F. 2015. "An Approximate Semianalytical Multiphase Forecasting Method for Multifractured
Tight Light-Oil Wells with Complex Fracture Geometry." Journal of Canadian Petroleum Technology 54 (6): 489–508.
http://dx.doi.org/10.2118/178665-PA.
Datta-Gupta, A. and King, M. J. 2007. Streamline Simulation: Theory and Practice. SPE Textbook Series Vol. 11.
Richardson, TX.
Datta-Gupta, A., Xie, J., Gupta, N. et al. 2011. "Radius of Investigation and its Generalization to Unconventional
Reservoirs." JPT 63 (7): 52–55. http://dx.doi.org/10.2118/0711-0052-JPT.
Diaz de Souza, O. C., Sharp, A. J., Martinez, R. C. et al. 2012. "Integrated Unconventional Shale Gas Reservoir
Modeling: A Worked Example From the Haynesville Shale, De Soto Parish, North Lousiana." presented at SPE
Americas Unconventional Resources Conference. Pittsburgh, Pennsylvania. 5-7 June. SPE-154692-MS. http://
dx.doi.org/10.2118/154692-MS.
Du, C., Zhang, X., Melton, B., et al. 2009. "A Workflow for Integrated Barnett Shale Gas Reservoir Modeling
and Simulation." presented at SPE Latin American and Caribbean Petroleum Engineering Conference. Caragena,
Colombia. 31 May-3 June. SPE-122934-MS. http://dx.doi.org/10.2118/122934-MS.
Duong, A. N. 2011. "Rate-Decline Analysis for Fracture-Dominated Shale Reservoirs." SPEREE 14 (3): 377–387. http://
dx.doi.org/10.2118/137748-PA.
EIA. 2016. "Annual Energy Outlook2016 with projections to2040." August. http://www.eia.gov/forecasts/aeo.
Fujita, Y., Datta-Gupta, A. and King, M. J. 2016. "A Comprehensive Reservoir Simulator for Unconventional Reservoirs
That Is Based on the Fast-Marching Method and Diffusive Time of Flight." SPEJ 21 (6): December. 2276–2288. http://
dx.doi.org/10.2118/173269-PA.
Gilman, J. R., and H. Kazemi. 1983. "Improvements in Simulation of Naturally Fractured Reservoirs." SPEJ 23 (4):
695–707. http://dx.doi.org/10.2118/10511-PA.
Hassouna, M. S., and Frag, A. A. 2007. "Multistencils Fast Marching Methods: A Highly Accurate Solution to the Eikonal
Equation on Cartesian Domains." IEEE Transactions of Pattern Analysis and Machine Intelligence 29 (9): 1–12.
https://doi.org/10.1109/TPAMI.2007.1154.
Horne, R. N. 1995. Modern Well Test Analysis: A Computer-Aided Approach. 2nd. Palo Alto, California: Petroway.
Ilk, D., Rushing, J. A., Perego, A. D. et al. 2008. "Exponential vs. Hyperbolic Decline in Tight Gas Sands — Understanding
the Origin and Implications for Reserve Estimates Using Arps' Decline Curves." presented at SPE Annual Technical
Conference and Exhibition. Denver, Colorado. 21-24 September. SPE116731. http://dx.doi.org/10.2118/116731-MS.
Kam, P., Nadeemm, M., Novlesky, A. et al. 2015. "Reservoir Characterization and History Matching of the Horn River
Shale: An Integrated Geoscience and Reservoir-Simulation Approach." Journal of Canadian Petroleum Technology
54 (6): 475–488. http://dx.doi.org/10.2118/171611-PA.
King, G. E. 2010. "Thirty Years of Gas Shale Fracturing: What Have We Learned?" presented at SPE Annual Technical
Conference and EXhibition. Florence, Italy. 19-22 September. SPE-133456. http://dx.doi.org/10.2118/133456-MS.
Lee, J. W. 1982. Well Testing. SPE Textbook Series Vol. 1. Richardson, TX.
Najabaei, B., Johns, R. T. and Chu, L. 2013. "Effect of Capillary Pressure on Phase Behavior in Tight Rocks and Shales."
SPEREE 16 (3): August, 281–289. http://dx.doi.org/10.2118/159258-PA.
24 SPE-185719-MS

Novlesky, A. Kumar, A. and Merkle, S. 2011. "Shale Gas Modeling Workflow: From Microseismic to Simulation -- A
Horn River Case Study." presented at Canadian Unconventional Resources Conference. Calgary, Alberta, Canada.
15-17 November. SPE-148710-MS. http://dx.doi.org/10.2118/148710-MS.
Sethian, J. A. 1996. "A Fast Marching Level Set Method for Monotonically Advancing Front." Proc. Natl. Acad. Sci.
USA 93 (4): 1591–1595.
Sethian, J. A. 1999. "Fast Marching Methods." SIAM Rev 41 (2): 199–235. http://dx.doi.org/10.1137/
S0036144598347059.
Tabatabaie, S. H. and Pooladi-Darvish, M. 2016. "Multiphase Linear Flow in Tight Oil Reservoirs." SPEREE Preprint:
1-13. http://dx.doi.org/10.2118/180932-PA.
Uzun, I., Kurtoglu, B. and Kazemi, H. 2016. "Multiphase Rate-Transient Analysis in Unconventional Reservoirs: Theory
and Application." SPEREE 19 (4): 553–566.
Valko, P. P. 2009. "Assigning value to stimulation in the Barnett Shale: a simultaneous analysis of 7000 plus production
hystories and well completion records." presented at SPE Hydraulic Fracturing Technology Conference. The
Woodlands, TX. 19-21 January. SPE119369. http://dx.doi.org/10.2118/119369-MS.
Vasco, D. W., and A. Datta-Gupta. 2016. Subsurface Fluid Flow and Imaging: Application for Hydrology, Reservoir
Engineering and Geophysics. Cambridge: Cambridge University Press.
Wang, L., Torres, A., Xiang, L. et al. 2015. "A Technical Review on Shale Gas Production and Unconventional Reservoirs
Modeling." Natural Resources 6 (3): 141–151.
Xie, J., Yang, C., Gupta, N., King, M. J. et al. 2015. "Depth of Investigation and Depletion in Unconventional Reservoirs
With Fast-Marching Methods." SPEJ 20 (4): 831–841. http://dx.doi.org/10.2118/154532-PA.
Xue, X., Yang, C., Sharma, V. K. et al. 2016. "Reservoir and Fracture Flow Characterization Using a Novel
w(t) Formulation." Unconventional Resources Technology Conference. San Antonio, Texas. 1-3 August.
URTEC-2440083-MS.
Yang, C., Sahrma, V. K. and Datta-Gupta, A. et al. 2015. "A Novel Approach for Production Transient Analysis of
Shale Gas/Oil Reservoirs." Unconventional Resources Technology Conference. San Antonio, Texas, 20-22 July.
SPE-178714-MS.
Yin, J., Xie, J., Datta-Gupta, A. et al. 2011. "Improved Characterization and Performance Assessment of Shale Gas
Wells by Integrating Stimulated Reservoir Volume and Production Data." presented at SPE Eastern Regional Meeting.
Columbus, Ohio. August 17-19. SPE-148969. http://dx.doi.org/10.2118/148969-MS.
Zhang, Y., Neha., B., Fujita, Y. et al. 2016. "From Streamlines to Fast Marching: Rapid Simulation and Performance
Assessment of Shale-Gas Reservoirs by Use of Diffusivity Time of Flight as a Spatial Coordinate." SPEJ 21 (5): 1–16.
http://dx.doi.org/10.2118/168997-PA.
SPE-185719-MS 25

Appendix A
Derivation of Eikonal Equation for Multi-phase System
In the absence of gravity, capillarity and permeability anisotropy, the general mass balance equation of
phase j can be written as:

(20)

By introducing the pore and fluid compressibilities, eq (20) can be rewritten as:

(21)

Summing up eq (21) for three phases yields the governing equation for the pressure:

(22)

Eq (22) can be approximated as follows by assuming (∇p)2≈0:

(23)

Now we apply a Fourier transform to eq (22) and the pressure can be expressed in the frequency domain:
(24)
where i and ω represent the imaginary unit and frequency, respectively. With the same way for the single
phase case (Vasco and Datta-Gupta, 2016), the asymptotic solution for eq (23) is:

(25)

The infinite series solution (25) was inspired from the Bessel's solution for radial diffusion problem. Al
are real functions relating to the amplitude of pressure wave and τ is a phase function which is the DTOF.
Substituting the asymptotic solution (25) into eq (26) yields:

(26)

If we focus on the pressure wave with high frequency, only the terms of higher order of ω has a significant
physical meaning in eq (26). Collecting the terms of highest order leads to the following expression:

(27)

As A0 is a non-zero function, the summation of the terms inside the bracket needs to be zero:
26 SPE-185719-MS

(28)

Thus, we have come up with the Eikonal equation for the multiphase system:

(29)

The same form of the Eikonal equation can be obtained for the anisotropic case (Zhang et al. 2016).
SPE-185719-MS 27

Appendix B
Coordinate Transformation from 3D to 1D DTOF system
In the absence of gravity, capillarity and permeability anisotropy, the Darcy's velocity u for phase-j is
expressed as:

(30)

If we assume the direction of the pressure gradient can be approximated with the DTOF gradient i.e.:

(31)

The Darcy's velocity can be rewritten on the DTOF coordinate as:

(32)

Mass balance equation over Ωτ that is a domain defined on the DTOF contour:

(33)

where Γτ is a surface of the domain Ωτ and hence is nothing but a DTOF contour. Hence, the unit vector
n can be defined as:

(34)

Substituting eqs. (32) and (34) into (33) yields:

(35)

Rearranging the Eikonal equation, the permeability can be expressed as the following form:

(36)

Eq. (35) becomes:

(37)

Remember the surface area Γτ can be approximated by the derivative of Ωτ with respect to the infinitesimal
distance s:

(38)

By substituting eq. (38) into (35), the surface integration can be expressed in differential form:
28 SPE-185719-MS

(39)

On the other hand, the volume integration on L. H. S. in eq. (35) can be rewritten as follows by assuming
that every properties can be considered constant within the infinitesimal domain of Ωτ:

(40)

Equating eqs. (39) and (40) yields following mass conservation equation defined on the DTOF
coordinate:

(41)

The definition of the coordinate transformation can be obtained by comparing eq. (41) with (33):

(42)

Dividing eq. (42) by the surface density, we finally come up with the definition of eq. (13) that is expressed
in terms of formation volume factor. For anisotropic permeability, we can also get the same expression.