SPE 113461

Heavy-Oil Recovery by Steam Injection in Fractured Reservoirs

A.P.G. van Heel, Shell Technology Oman, Muscat, Sultanate of Oman; J.J. van Dorp, Shell Technology Oman,
Muscat, Sultanate of Oman; and P.M. Boerrigter, Shell Technology Oman, Muscat, Sultanate of Oman; Shell
International Exploration and Production B.V., Rijswijk, The Netherlands.
Copyright 2008, Society of Petroleum Engineers

This paper was prepared for presentation at the 2008 SPE/DOE Improved Oil Recovery Symposium held in Tulsa, Oklahoma, U.S.A., 1923April2008.

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

Using analytical results and thermal reservoir simulations, we study the heating of - and oil recovery from - a vertical stack of
matrix blocks. The stack is surrounded by fractures, where steam is injected at the top and oil recovered from the base of the
fracture system.
We compare fine-grid single-porosity simulations with coarse-grid dual-permeability simulations (where the matrix-fracture
interaction is modelled via shape-factors). We independently validate the simulation results with new analytical results for
the recoveries due to thermal expansion and temperature-induced viscosity reduction.
Our dual-permeability results show that the early-time heating of the matrix cannot be captured using a constant shape-factor.
We analytically derive the time-dependent (transient) shape-factor that captures the heating of the blocks for all time-scales.
When this transient shape-factor is used in combination with an analytically derived viscosity correction (to capture the effect
of the temperature profile inside a matrix-block), the coarse-grid dual-permeability simulations accurately reproduce the fine-
grid single-porosity simulations and analytical results.

Introduction

With fine-grid single-porosity models (SP) for fractured reservoirs being computationally prohibitive in terms of computer
time and memory, the simulation of oil recovery from fractured reservoirs typically requires a dual-permeability formulation
(DP). In such DP models, the matrix-fracture interaction is modelled via shape-factors [1, 2]. In a thermal setting, the matrix-
fracture interaction comprises both hydraulic interaction (modelling the fluid exchange between matrix and fractures), as well
as thermal interaction (modelling the energy and heat exchange between matrix and fractures). This implies that in a thermal
DP simulation two kinds of shape-factors are required: hydraulic shape-factors and thermal shape-factors. Since a shape-
factor is dependent on the underlying physical process that it models (and not only on the shape and dimensions of a matrix-
block), hydraulic- and thermal shape-factors should a priori be regarded as independent quantities that are not necessarily
equal.
There has been much discussion about (hydraulic) shape-factors in the literature [3-15]. Although they have typically been
treated as constant, time-independent quantities, it follows from its definition that a shape-factor is essentially a time-
dependent quantity. This was already emphasised by several authors, who derived explicitly time-dependent expressions for
the shape-factor for pressure diffusion processes [7-9].
An explicitly time-dependent formulation, however, is quite cumbersome for implementation in a reservoir simulator because
each grid-block would have its own 0 = t , corresponding to that moment in time where the pressure (or temperature) in
the fractures surrounding that particular grid-block changes.

As we have shown earlier [16], an exact analytical transient shape-factor can be derived from first-principles for the
temperature diffusion in 1D matrix-blocks (bounded by two parallel fractures), which is valid for all time-scales. In that
approach we used the analytical relation between time and the (volume-averaged) matrix-block temperature to eliminate time
in favour of the volume-averaged matrix-block temperature, thus obtaining an exact expression for the transient shape-factor
in terms of the (volume-averaged) matrix-block temperature.
2 SPE 113461
A similar (approximate) approach had been introduced by Zimmerman et al. [7,17] to describe pressure diffusion (assuming
spherical matrix-blocks). In their approach, Zimmerman et al. used an approximate simple algebraic expression for the exact
solution for single-phase pressure diffusion in a spherical block (which had been proposed by Vermeulen [18]), to eliminate
the explicit appearance of time in the matrix-fracture interaction and to formulate it in terms of the (volume-averaged)
matrix-block pressure. This approach was recently applied in a multi-phase setting [19].

A limitation of Vermeulens approximation is that it is essentially limited to spherical matrix-blocks. In an appendix to their
paper, Zimmerman et al. show that for rectangular blocks, no Vermeulen-approximation exists which provides a good
approximation to the exact response in both the short- and the long-time regimes [7].
However, as we have shown in [16], an exact analytical transient shape-factor can be derived from first-principles for the
temperature diffusion in 1D rectangular matrix-blocks (bounded by two parallel fractures), which is valid for all time-scales.
In this paper, we extend our previous work and derive the exact analytical transient shape-factor for diffusion in 2D
rectangular matrix-blocks (which are surrounded by four fractures).

Recovery mechanisms induced by steam-injection in heavy-oil, fractured reservoirs

Recovery mechanisms induced by steam-injection in fractured reservoirs are: thermal expansion, thermal viscosity reduction,
solution-gas drive and steam-stripping. For heavy oils, the dominant recovery methods are thermal expansion and thermal
viscosity reduction. The driving force behind these phenomena is the temperature-increase in the reservoir, and that is why
the accuracy of thermal simulation results critically depends on the accuracy with which the temperature-evolution in the
matrix-blocks is modelled.
In this paper we consider the characteristic building block of a heavy-oil fractured reservoir: a single (homogeneous) vertical
matrix-stack, surrounded by fractures. Steam is injected at the top, and oil recovered from the base, of the fracture system.
We consider two idealised geological settings: the 1D case where there is only one set of vertical fractures (with a fracture
spacing
x
L ), so that the vertical stack is bounded by two fractures; and the 2D case where there are two perpendicular sets
of vertical fractures (with a fracture spacings
x
L and
y
L ), such that the stack is bounded by four fractures.
As a mathematical idealisation of the steam-injection into the fracture system surrounding the matrix-stack, we assume that
the matrix-stack initially has a uniform temperature
i
T , and that from the start of steam-injection ( 0 = t ) onwards, the
temperature in the fractures is changed to, and maintained at a temperature
f
T . Under these idealisations we can derive
analytical expressions for the evolution of the temperature profiles inside the matrix-blocks ( ) , ( t x T and ) , , ( t y x T for the
1D and 2D stacks, respectively). As we will show below, from these temperature profiles we can obtain a closed-form
analytical expression for the oil recovery by thermal expansion, and a semi-analytical expression result for the oil-recovery
by viscosity reduction.

Evolution of the temperature profile inside a matrix-block (1D)

As shown in [16] and repeated in Appendix A
1
, the temperature profile inside a 1D matrix-block that is heated from its two
sides at 0 = x and
x
L x = is given by (when expressed in dimensionless quantities, see Appendix A):
( ) ( ), exp sin
1
4 ) , (
2
0
1 , d k
k
d k
k
d d D d
t x t x T

=
(1)
where ( ) 1 2 + = k
k
, and
2
x d
L t t = denotes the dimensionless time, based on the fracture spacing
x
L .
Eq. 1 has the following short- and long-time approximations:

1
Note that in the current paper we use a different definition of the dimensionless temperature than that in [16]. In the present
paper we define ( ) ( )
i f f d
T T T T T = . The advantage of this definition is that the applied temperature-boundary
condition in the fractures becomes homogeneous. In [16] we used ( ) ( )
i f i d
T T T T T = ; one can easily switch from
one definition to the other by making the following substitution: ( )
d d
T T 1 .
SPE 113461 3
( ) ( )

>

=
, 05 . 0 exp sin
4
, 05 . 0
2
1
2
1
) , (
2
1 ,
d d d
d
d
d
d
d
d d D d
t t x
t
t
x
erfc
t
x
erfc
t x T

(2)
where erfc denotes the complementary error-function.
The (volume-averaged) matrix block temperature is obtained by integration of Eq. 2 over the matrix-block, and is given by
(see Appendix A):
( )
( )

>

=
. 05 . 0 exp
8
, 05 . 0 4 1
2
2
1 ,
d d
d
d
d D d
t t
t
t
t T

(3)
Eq. 3 shows that the short-time (volume-averaged) temperature has a square-root time response, and that the long-time
response is exponential.

The derivation of the temperature evolution inside 2D matrix-blocks follows similar lines, and is given in Appendix B. In the
2D case, we distinguish three time regimes: a short-time regime (where the volume-averaged matrix-block temperature has a
square-root time response), a transitional medium-time response, and a long-time regime where the volume-averaged matrix-
block temperature has an exponential response.

Oil recovery by thermal expansion

In Appendix C we derive analytical expressions for the oil rate due to thermal expansion from 1D and 2D matrix-blocks
which are heated from their sides. Assuming that the specific volumes of oil and water are linear functions of the temperature
over the range of interest (i.e. assuming constant thermal expansion coefficients), the oil rate due to thermal expansion is
found to be proportional to the time-derivative of the (volume-averaged) matrix-block temperature, and for 1D matrix-blocks
we find:

>

=
, 05 . 0 ) exp( 8
, 05 . 0
2
) (
2
1
1
d d D
d
d
D
d o
t t K
t
t
K
t q

(4)
where

( )
o
i f wp cw op o x
D
B
T T S S WHL
K
] [
0
1
+
=

(5)
The
d
t 1 short-time response of Eq. 4 illustrates that oil recovery by thermal expansion is (under the mathematical
idealisations made here) both a very fast (in fact instantaneous), and very significant (indeed resulting in an infinite rate at
start of steam-injection). The generalisation to oil recovery by thermal expansion in 2D matrix blocks is similar and is given
in Appendix C.

Thermal viscosity reduction

When a matrix-block is heated from its sides, a temperature profile will develop inside the matrix as given by Eq. 1. This
temperature profile will be reflected by other properties of the oil inside the matrix-block, and notably by the oil viscosity.
Indeed, for heavy-oils the oil viscosity typically is very sensitive to temperature changes, and can decrease by orders of
magnitude upon heating. Thus, a temperature profile ) , ( t x T inside a matrix-block will induce a viscosity profile inside that
matrix-block: )), , ( ( ) , ( t x T t x = where ) (T denotes the viscosity-temperature relationship. This phenomenon is
illustrated in Figure 1 and Figure 2. In turn, the induced viscosity-profile will increase the gas-oil gravity drainage rate from the
matrix-block. This gas-oil gravity drainage from the matrix-block (assuming a purely vertical flow) is given by [20]
.
)) , ( (
1
) (
0

=
x
L
og
r
x
gd
dx g
t x T
kk
L
t q

(6)
4 SPE 113461
Eq. 6 is general, but semi-analytical since with the exception of a few special viscosity-temperature correlations- the
integration will generally need to be evaluated numerically.






Dual-permeability simulation of heavy-oil recovery by steam-injection in fractured reservoirs

In a thermal DP simulation, the matrix-fracture interaction comprises both hydraulic interaction (matrix-fracture fluid-flow)
and thermal interaction (matrix-fracture heat-flow), implying the need for two types of shape-factors: hydraulic shape-factors
and thermal shape-factors. In our previous work we have shown that DP simulations of (both isothermal and heated) draining
stacks that use a hydraulic shape-factor
2
2
L
h

= (7)
provide the best approximation to the corresponding fine-grid SP reference results [16]. In the same paper we have shown
that the commonly-used hydraulic shape-factor
2
4 L
h
= leads to significant discrepancies between the DP results and the
corresponding fine-grid SP results. Based on these observations, in this paper we use have used Eq. 7 (for 1D matrix-blocks)
and its 2D generalisation for 2D matrix-blocks, viz.
.
1 1
2 2
2

+ =
y x
h
L L
(8)

The focus in this paper is on the thermal shape-factor, and in particular we will derive the transient thermal shape-factor. The
importance of using a transient thermal shape-factor rather than a constant thermal shape-factor can be motivated as follows
(we consider the 1D case, the reasoning for the 2D case is analogous).
If we use a constant (time-independent) thermal shape-factor in DP simulations, then by integrating the shape-factor
definition, Eq. D-4 (see Appendix D), the (volume-averaged) dimensionless temperature of a matrix-block is found to have a
purely exponential character for all time-scales:
). exp( ) (
2
d x d d
t L t T = (9)
This result is to be contrasted with the exact analytical
d
t early-time response of the dimensionless average matrix-block
temperature (see Eq. 3). Since the oil-rate by thermal expansion is proportional to the time-derivative of the volume-averaged
temperature of the matrix-block (see Appendix C), a time-independent shape-factor will predict an oil-rate by thermal
expansion having a purely exponential character, in significant contrast to the exact short-time behaviour which exhibits a
t 1 -response. From this analysis (and as will further be illustrated below) we can conclude that an accurate simulation of
temperature-induced recovery mechanisms requires the transient thermal shape-factor.


















SPE 113461 5








The transient shape-factor for thermal diffusion

The idea behind the derivation of the analytical transient shape-factor is most easily explained by considering the heating of a
1D matrix-block, and therefore we here first briefly summarise our earlier 1D results (see [16]).

As shown in Appendix D, the (explicitly time-dependent) shape-factor describing the thermal matrix-fracture interaction is
defined (when expressed in dimensionless quantities) by
.
) (
) (
1
) (
,
,
2
d m d
d
d m d
x
d
t T
dt
t T d
L
t

= (10)
Substituting the analytical short- and long-time expressions for the average 1D matrix-block temperature (Eq. 3) into Eq. 10,
we obtain:

( )

>

=
. 05 . 0
05 . 0
4 1
2 1
2
2
2
d
x
d
d
d
x
d
t
L
t
t
t
L
t

(11)
Eq. 11 is the transient shape-factor which should be used in DP simulations in order to reproduce the exact analytical
volume-averaged matrix-block temperature, Eq. 3.

Eq. 11 is impractical for simulator implementation because the shape-factor is explicitly formulated in terms of
(dimensionless) time. Moreover, each grid-block has its own time for which 0 =
d
t , because 0 =
d
t corresponds to that
moment in time at which, for that particular grid-block, the temperature in the surrounding fractures is raised.
This difficulty, however, can be circumvented by inverting Eq. 3 (i.e. we express the dimensionless time
d
t analytically as a
function of the dimensionless temperature
d
T ) and by using this expression to subsequently eliminate
d
t in Eq. 11 in favour
of the dimensionless temperature
d
T :
( )
( )

<

= .
5 . 0
5 . 0
1
1 8 1
2
2
2
d
x
d
d d x
d
T
L
T
T T L
T

(12)
In going from Eq. 11 to Eq. 12 we made use of the fact that in the 1D case 5 . 0 ) 05 . 0 ( =
d d
t T (which is readily obtained
from Eq. 3).
Note that the formulation in terms of the dimensionless temperature
d
T removes the practical difficulties associated with the
formulation in terms of dimensionless time
d
t . In the simulator, the shape-factor of a given grid-block is updated every
timestep according to Eq. 12, based on its own (momentaneous) dimensionless temperature
2
.

2
Due to the mathematical idealisations made in the derivation, a division by zero would occur if 1 =
d
T (corresponding to
an infinite heat-flux immediately following raising of the temperature in the fracture). In the simulations we have prevented
6 SPE 113461

Eq. 12 is the analytical formulation for the transient shape-factor for 1D diffusion. If this transient shape-factor is used in DP
simulations, the DP simulations will reproduce the analytical results for the volume-averaged matrix-block temperature
exactly. We want to point out that from a mathematical point of view, the above transient shape-factor is strictly applicable
for an instantaneous step-change in the temperature boundary condition (which also implies an infinite matrix-fracture heat-
flux immediately following the temperature step-change). In actual field applications and simulations, however, the matrix-
fracture heat-flux is always finite and the temperature in the fracture doesnt change instantaneously. However, as was
shown by Zimmerman et al., using detailed theoretical analysis and numerical simulation [7] (for the Vermeulen
approximation and spherical matrix-blocks), the approach remains applicable even when the applied boundary conditions
vary with time (e.g. for linearly increasing and even for oscillating boundary conditions), provided that the characteristic time
of the variations in the applied fracture boundary conditions,
f
, is large compared to the characteristic time for temperature
diffusion inside the matrix-block (
2
x
L = ), i.e. as long as
f
x
L

<<
2
. (13)
In that case the matrix-block can keep up with changes in the applied boundary conditions. In our simulations, the
dimensionless matrix-block temperature is therefore calculated based on the (time-varying) momentaneous fracture- and
matrix-block temperatures, ) (t T
f
and ) (t T
m
as follows:
.
) (
) ( ) (
) (
i f
m f
d
T t T
t T t T
t T

= (14)
The derivation of the 2D transient shape-factor for diffusion processes is more complicated, but proceeds along lines similar
to those given above, and is detailed in Appendix D. In the 2D case, we distinguish three time-regimes: a short-, medium-
and long-time regime (and therefore also three temperature-regimes). We have implemented the 1D and 2D transient shape-
factors in Shells in-house reservoir simulator MoReS (Modular Reservoir Simulator).

Simulation of thermal expansion in DP mode

As shown in Appendix C, the oil rate due to thermal expansion is determined by the time-derivative of the volume-averaged
matrix-block temperature. Since the analytical volume-averaged matrix-block temperature
d
T is exactly reproduced in DP
simulations if the above-derived transient shape-factor (Eq. 12) is used, it then follows that the analytical oil rate due to
thermal expansion is also exactly reproduced. In other words, if the transient thermal shape-factor is used in the DP
simulations, the simulated oil rate due to thermal expansion is automatically correct.


Simulation of thermal viscosity reduction in DP mode

In this paper we model the viscosity of the heavy-oil by the ASTM correlation [21], and we fix this viscosity-temperature
relation by specifying the oil viscosity for initial- and steam-temperature: 100 ) 323 ( = K cP and 2 ) 506 ( = K cP. The
evolution of the dimensionless temperature profiles (inside a 1D matrix-block which is heated from its two sides) and the
resulting viscosity profiles for the heavy-oil considered here are shown in Figure 1 and Figure 2.
In a DP simulation the actual temperature profile ) , ( t x T inside a matrix block is not resolved; instead the grid-block
temperature represents the volume-averaged temperature T (this equally applies to the pressure). Therefore the actual
viscosity profile inside a matrix-block, )) , ( ( ) , ( t x T t x , is not captured either. In DP simulations, the viscosity in a
grid-block is evaluated based on the prevailing grid-block temperature: ) (T = . This viscosity is used to calculate the
gas-oil gravity-drainage rate through the bottom of the grid-block. However, as noted by Wassing et al. [20], the gas-oil
gravity drainage rate from the grid-block calculated using a viscosity ) (T = , is not equal to the gas-oil gravity drainage
rate if the exact viscosity (temperature) profile were used.





this by using a large but finite value for for 1
d
T .
SPE 113461 7





































Figure 1: Evolution of the dimensionless temperature inside a 1D matrix-block which is heated from its sides.





















Evolution of temperature profiles inside matrix-block
0.0
0.2
0.4
0.6
0.8
1.0
0 0.2 0.4 0.6 0.8 1
dimensionless distance (xd)
(
1
-
T
d
)
td=0.001
td=0.01
td=0.05
td=0.1
td=0.25
td=0.5
Evolution of viscosity profiles inside matrix-block
0.0
20.0
40.0
60.0
80.0
100.0
0 0.2 0.4 0.6 0.8 1
dimensionless distance (xd)
V
i
s
c
o
s
i
t
y

(
c
P
)
td=0.001
td=0.01
td=0.05
td=0.1
td=0.25
td=0.5
8 SPE 113461





Figure 2: Evolution of the induced viscosity profiles inside a 1D matrix-block which is heated from its sides. An ASTM
viscosity relation is assumed in this paper where 100 ) 323 ( = K cP and 2 ) 506 ( = K cP.

To address the above-mentioned issue, Wassing et al. [20] have recently shown that a viscosity-correction factor can be
introduced in DP simulations which corrects the calculated viscosity in the simulator, ) (T , in such a way that the gravity
drainage rate from the DP simulation reproduces the gravity drainage rate from the same matrix-block if the exact 1D
temperature ) , ( t x T profile were used. The 1D viscosity correction factor introduced by Wassing et al. was derived
assuming that the volume-averaged simulator grid-block temperature is equal to the exact (analytical) volume-averaged
temperature. This is generally not the case (it is only so if the transient shape-factor is used in the DP simulations), and the
viscosity correction factor should actually be formulated in terms of the simulator grid-block temperature, as this is the only
temperature that the simulator knows. In this section we generalise Wassings approach to arbitrary 2D rectangular matrix-
blocks and we calculate the viscosity correction factor based on the (volume-averaged) simulator grid-block temperature.

The viscosity correction factor for use in DP simulation is motivated as follows. The exact (volumetric) oil drainage rate
through the bottom of a 2D matrix block is given by (assuming purely vertical flow):
dxdy g
t y x T
kk
L L
t q
x
y L
L
og
r
y x
exact

=
0 0
)) , , ( (
1
) (

, (15)
where we have suppressed the arguments of the quantities under the integral, except those for the viscosity, in order to
highlight its dependence on the temperature profile.
In a dual-permeability simulation, however, the oil drainage rate would be calculated based on the prevailing simulator grid-
block temperature:
og
sim
r
L
L
og
sim
r
y x
sim
g
T
kk
dxdy g
T
kk
L L
t q
x
y

= =

) ( ) (
1
) (
0 0
. (16)
Comparing Eqs. 15 and 16 for a homogeneous matrix block, and further assuming that the product
og r
kk can be taken in
front of the integral and is equal to the value of the corresponding product in the simulator, the factor R which with the
simulator viscosity, ) (
sim
T , must be multiplied in order to give the same drainage rate as the analytical drainage rate (Eq.
15) is given by :
dxdy
t y x T L L
t T
t R
x
y L
L
exact y x
sim

=
0 0
)) , , ( (
1 1
)) ( (
1
) (

. (17)

The viscosity reduction factor, Eq. 17, is written explicitly as a function of time t . This, however, is not practical for
simulator implementation: the viscosity-correction clock for a given grid-block starts ticking only when the fractures
surrounding that grid-block are heated, so this time will be different for different grid-blocks.
The driving force behind the viscosity reduction is of course the temperature rise, and since there exists a one-to-one
relation between the heating time t and the simulator grid-block temperature
sim
T , we can eliminate the parameterisation in
terms of t in favour of the grid-block simulator temperature
sim
T .
In other words, in our simulations, the viscosity reduction factor will be not be formulated in terms of time, but in terms of
the simulator grid-block temperature
sim
T :
SPE 113461 9
dxdy
t y x T L L
T
T R
x
y L
L
exact y x
sim
sim

=
0 0
)) , , ( (
1 1
) (
1
) (

(18)
In this expression,

t corresponds to that particular time for which the corresponding simulator grid-block temperature is
equal to
sim
T . For simulator implementation it is convenient to express the viscosity reduction factor as a function of the
dimensionless temperature,
d
T ( 1 0
d
T ).
The above derivation shows that the viscosity correction factor is actually an exact upscaling method (for the stipulated
conditions, assuming a purely vertical flow through the matrix): if the viscosity correction factor is applied to the viscosity in
(coarse-grid) dual-permeability simulations, then these (coarse-grid) dual-permeability simulations will reproduce the exact
oil-rates. Moreover, since we have formulated the viscosity correction factor in terms of the simulator grid-block
temperature, our approach will also give the correct gravity drainage rate in DP simulations in which a constant thermal
shape-factor is used (and not necessarily the transient diffusion shape-factor).

We want to mention that the viscosity correction table needs to be calculated only once, before the start of the simulation, and
it does not need to be re-calculated during the simulation. At the start of the simulation the reduction factor is tabulated as a
function of the dimensionless simulator grid-block temperature: ) (
,sim d
T R . During the dual-permeability simulation, a
reduced-viscosity table is then calculated (again before the start of the simulation) by multiplying the actual oil viscosity (at
a given dimensionless temperature) with the viscosity correction factor (at the same dimensionless temperature):
) ( ) ( ) (
d d d red
T R T T = . (19)

Although we have formulated and implemented the DP viscosity correction in terms of the dimensionless temperature, it is
instructive to investigate its range of influence in the terms of time (following steam-injection).
Only during a small time-period, immediately following contact with steam, the implemented viscosity-reduction is effective.
This time-period can be estimated from the results shown in Figure 1 and Figure 2: for times 25 . 0
d
t the temperature-
profile can be approximated by a uniform (flat) temperature profile, equal to the (volume-averaged) matrix-block
temperature. Similarly, the viscosity inside the matrix-block can then also be approximated by a uniform profile. As soon as
these approximately uniform profiles develop, the viscosity inside the matrix-block is roughly constant (independent of
position) so:
.
) , ( (
1 1
) (
1
0 0

x
y L
L
y x
dxdy
y x T L L T
(20)

From Eq. 17 it follows that under these condition the viscosity reduction factor reduces to 1. In other words, the viscosity-
reduction effect is most pronounced for times 25 . 0
d
t ; for larger times, the viscosity reduction factor essentially
approaches unity.
















10 SPE 113461











Simulation results: heavy-oil recovery by steam-injection in fractured reservoirs

To illustrate the above concepts we consider the heating of - and oil recovery from - a vertical stack of matrix blocks. The
stack is surrounded by fractures, where steam is injected at the top and oil recovered from the base of the fracture system; this
is schematically shown in Figure 3.



Figure 3: schematic representation of steam-injection into a fractured reservoir.
We will compare coarse-grid DP simulations with the corresponding fine-grid SP simulations. We independently validate the
simulation results with the analytical results for the heating of the matrix-blocks.
Both the matrix and fractures were assigned homogenous and isotropic hydraulic and thermal properties, as summarised in
Table 1.

Parameter Value
m
k
10 mD
m

0.3
f
k
500 D
f

0.01
m

2.3 W/K/m
grain
M
1.8 E6 J/K/m3
Table 1: Some key matrix- and fracture properties used in the simulations.
In all simulations, a zero capillary pressure was assigned both in the matrix and the fractures. Straight-line relative
permeabilities were used in the fractures and the matrix relative permeabilities were modelled using a Corey correlation (see
Appendix E). The same relative permeabilities were applied both in the DP and the SP simulations (i.e. no pseudos were
used). The viscosity of the oil was modelled by the ASTM correlation (and was fixed by specifying that at 323 K the oil
viscosity is 100 cP and at 506 K the oil viscosity is 2 cP; see Appendix E).

Steam
injection
production
heat
Fracture spacing
Height
Steam
injection
Steam
injection
production
heat
Fracture spacing
Height
SPE 113461 11
Before the start of the simulation, the fractures are fully oil-saturated and the matrix contains connate water and oil. The
initial reservoir temperature is 323 K. From 0 = t onwards, oil is produced from the fracture at the bottom of the stack, and
the fractures are quickly emptied. Gas is simultaneously injected in the fracture at the top of the stack to ensure that,
following the initial pressure drop during the emptying of the fractures, the pressure is subsequently kept at a lower, fixed
level. During the first five years only gas is injected and during this period the stack drains under isothermal conditions. The
oil rate during this period is very low because of the high oil viscosity (100 cP). After five years, the gas-injection is stopped
and is replaced by steam-injection (also into the top of the fracture system). Steam is injected at a rate of 40m
3
/day (cold-
water equivalent). At the prevailing conditions, the steam temperature is 506 K, so the matrix-blocks will gradually be heated
to 506 K.

The matrix-stack is 100 m high, and is uniformly divided into 20 grid layers with 5 = z m (both in the SP and DP
simulations). In the SP simulations, we exploited the symmetry of the problem to simulate half of the stack only. The DP and
SP simulation grids are shown in Figure 4. In the SP simulations, each matrix layer is (horizontally, i.e. in the x -direction)
subdivided into small grid blocks to capture the variation of pressure, temperature and saturation (and therefore their
gradients) inside the matrix and close to the matrix-fracture interface. Small grid-blocks were used next to the fracture, which
gradually increase in size when moving towards the centre of the matrix. In the DP simulations, each matrix layer is
discretised into a single grid block (the matrix-block is represented by a single grid-block with
x
L x = ), and each grid
block contains a matrix and a fracture node. As was mentioned above, in the DP simulations the horizontal pressure-,
temperature- and saturation-profile inside the matrix are not resolved, and in DP simulation these quantities are to be
interpreted as volume-averaged matrix properties.
The fine-grid SP simulation results and the analytical results are considered as the reference solutions against which the DP
results are compared.



Figure 4: The dual-permeability grid (top) and the corresponding fine-grid single-porosity grid (bottom) used in the
simulations.







12 SPE 113461









Case1: Simulation of steam-injection into a 1D heavy-oil fractured stack

As an introduction to the 2D simulations to be discussed in the next section, we first briefly review some of the 1D results
discussed in our earlier paper [16]. In that work we considered various fracture spacings (5, 20 and 50 m) for a fixed stack-
height of 100 m; here we will briefly review the 20 m case.

For the 1D case, we compare and contrast the following three choices for the thermal shape-factor (in all simulations we used
a hydraulic shape-factor
2 2
x h
L = ):
,
4
2
x
th
L
= commonly used in commercial reservoir simulators;
,
2
2
x
th
L

= the long-time asymptotic shape-factor for diffusion [8,9,16];

( )

<

=
5 . 0
5 . 0
1
1 8 1
) (
2
2
2
d
x
d
d d x
d th
T
L
T
T T L
T

, the transient shape-factor introduced here.

In Figure 5, we show the simulated DP mid-stack matrix-block temperature for the above three different thermal shape-
factors, along with the corresponding fine-grid simulation result and the 1D analytical result (see Appendix A) for a fracture-
spacing 20 =
x
L . In the simulations the steam does not instantaneously fill the 100 m high fracture system because the
steam is injected at a finite rate, so it needs time to travel from the top to the bottom of the stack. This means that the mid-
stack grid-block is not instantly surrounded by steam from 0 = t onwards, and in our simulations the steam-front in the
fractures needs approximately 2 months to reach the mid-stack. To take this delay into account in Figure 5, the analytical
results have been time-shifted by this delay in steam-arrival time. The DP simulation using the transient thermal shape-factor
and the fine-grid SP simulation are in good agreement with the analytical result. A thermal shape-factor
2 2
x th
L = gives
a good overall temperature response, although it somewhat underestimates the initial heating rate. It is clear from the results
that a thermal shape-factor
2
4
x th
L = is inadequate for simulating the heating of a matrix-block by diffusion.

The corresponding oil-rate responses are shown in Figure 6 (focussing on the short-time behaviour, following start of steam-
injection) and Figure 7 (showing the longer time behaviour). These results show a very good agreement between the DP
simulation using the transient thermal shape-factor and the corresponding fine-grid SP (reference) simulation. The initial
under-prediction of the temperature using a thermal shape-factor
2 2
x th
L = (shown in Figure 5) is reflected in a
significantly lower peak in the oil rate (approximately a factor 2 smaller). This initial peak in oil rate corresponds to thermal
expansion of the oil and the simulated slower-than-actual heating of the matrix is translated in a slower oil production. If a
thermal shape-factor
2
4
x th
L = is used, the resulting DP oil rate response does not even qualitatively agree with the fine-
grid SP results.







SPE 113461 13


































Figure 5: Mid-stack temperature for 20 =
x
L m. Shown are DP results using shape-factors
2
4 L
th
= (dark blue),
2 2
L
th
= (green), the transient shape-factor (red), as well as the fine-grid single-porosity result (black) and the
corresponding analytical solution (light blue).



















Mid-stack temperature, Lx=20 m
323
373
423
473
523
5 6 7 8 9 10
time, year
t
e
m
p
e
r
a
t
u
r
e
,

K
4/L^2
(pi/L)^2
transient
fine-grid
analytical
Oil rate, Lx=20 m
0
5
10
15
5 6 7 8 9 10
time, year
O
i
l

r
a
t
e
,

m
3
/
d
4/L^2
(pi/L)^2
transient
fine-grid
14 SPE 113461





Figure 6: Oil rates for 20 =
x
L m. Shown are DP results using shape-factors
2
4 L
th
= (dark blue),
2 2
L
th
= (green), the transient shape-factor (red), as well as the fine-grid single-porosity result (black).
























Figure 7: Oil rates for 20 =
x
L m. Shown are DP results using shape-factors
2
4 L
th
= (dark blue),
2 2
L
th
= (green), the transient shape-factor (red), as well as the fine-grid single-porosity result (black).





















Oil rate, Lx=20 m
0.01
0.1
1
10
100
0 10 20 30
time, year
O
i
l

r
a
t
e
,

m
3
/
d
4/L^2
(pi/L)^2
transient
fine-grid
SPE 113461 15









Case2: Simulation of steam-injection into a 2D heavy-oil fractured stack

The 2D matrix-stack is bounded by two perpendicular fracture sets with fracture spacings
x
L and ,
y
L and without loss of
generality we assume .
y x
L L To simplify the notation in the derivations in the Appendices, we have introduced the ratio
of the two fracture spacings:
y x
L L = : , so we have . 1 0 The results for the 1D stack presented in the previous
section clearly illustrate that the choice
2
4
x th
L = for the thermal shape-factor is inappropriate for modelling the thermal
matrix-fracture interaction in DP models, so we in the 2D simulations, we will only focus attention on the 2D version of the
asymptotic diffusion shape-factor,

+ =
2 2
2
1 1
y x
th
L L
, and the 2D transient shape-factor (derived in Appendix D).

We consider two fracture spacings: 20 =
x
L m, and 50 =
x
L m, and we investigate the impact of the ratio of the fracture
spacings (i.e. the anisotropy of the fracture spacings in the x- and y-directions) by varying
10
1
2
1
, , 1 = . We independently
check the DP simulation results by comparing the evolution of the volume-averaged matrix-block temperature with the
analytical results. The impact of the viscosity reduction factor is illustrated by comparing simulations with and without
viscosity reduction.

In Figure 8 and Figure 9, we compare the 20 =
x
L m DP simulation results for the grid-block temperature (of the mid-stack
grid-block) with the corresponding analytical results for 2D matrix-blocks . The DP simulation results shown in Figure 8
were obtained using a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = . For independent validation of these simulation
results, we have also plotted the corresponding analytical results which, when expressed in terms of dimensionless variables,
are given by ) exp( ) (
2
d th d d
t t T = ; this latter result is obtained by integration of Eq. D-4 (see Appendix D). The
simulation- and analytical results are in good agreement for all values of . In the plotting of the analytical results in Figure
8, the time it takes the steam to travel from the top of the stack to the mid-stack (which is of the order of two months) has
been taken into account by time-shifting the analytical results by this travel time (like in the presentation of the 1D results in
the previous section).
Note that since we inject at the same steam-rate of 40 m
3
/d (cold-water equivalent) in all simulations, the case
1 = (implying matrix-blocks of 20 m by 20 m) heats up faster than the cases
2
1
= (matrix-blocks of 20 m by 40 m) and
10
1
= (matrix-blocks of 20 m by 200 m).

Figure 9 shows the comparison between the DP simulation using the transient 2D shape-factor (derived in Appendix D) and
the analytical results for the mid-stack grid-block (see Appendix B). As mentioned above, the analytical results were time-
shifted by approximately 2 months to take into account the time the steam-front needs to travel from the top of the stack to
the mid-stack grid-block.
From the results in Figure 8 and Figure 9, we can conclude that for a fracture spacing =
x
L 20 m and for all values of , the
simulation results reproduce the analytical results very well. We can therefore already conclude upfront that the simulated
oil-rate due to thermal expansion (which is proportional to the time-derivative of the average matrix-block temperature) must
be accurate.
The corresponding oil rates (for 20 =
x
L m) for the constant and the transient thermal shape-factor are shown in Figure 10
and Figure 11, respectively. The figures show the comparison between simulations without the DP viscosity-correction
(annotated with MuC=0) and with DP viscosity-correction (labelled MuC=1). As we have shown in the discussion of the
viscosity-correction factor, the viscosity-reduction effect is essentially limited to times 25 . 0
d
t . In the present case,
16 SPE 113461
25 . 0 =
d
t approximately corresponds to 2 years, so any DP viscosity-correction effects are limited to some 2 years
following first contact with steam. It turns out, however, that for 20 =
x
L m, the viscosity-correction has no visible impact
on the results (at least for the initially 100 cP oil considered here, the viscosity of which is reduced by a factor of 50 when it
reaches steam temperature). This means that for 20 =
x
L m (and equally for smaller
x
L , not shown here), the viscosity
calculated based on the volume-averaged grid-block temperature (for this particular oil) gives the same results as volume-
averaged viscosity if the exact 2D temperature-profile in the grid-block were used.
Although the viscosity-reduction effect is not visible from the results for 20 =
x
L m, the impact of using the transient
thermal shape-factor as compared to using a constant thermal shape-factor is significant, as can be seen by comparing
corresponding curves between Figure 10 (constant thermal shape-factor) and Figure 11 (transient thermal shape-factor). The
use of the constant thermal shape-factor
2 2 2
) 1 (
x th
L + = leads to an under-prediction of the heating-rate of the
matrix-blocks and hence under-predicts the initial oil peak-rates, which are dominated by thermal expansion. As discussed
above, the transient shape-factor reproduces the analytical heating rate and hence captures the initial peak in oil rate.
These results show that to accurately predict the initial oil-rate response following steam-injection, the use of the transient
thermal shape-factor is essential.











































SPE 113461 17
































Figure 8: Mid-stack temperature for 20 =
x
L m and
10
1
2
1
, , 1 = . The dotted lines correspond to DP simulation
results using a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = , the solid lines show the corresponding
analytical results.























Mid-stack temperature; Lx=20 m
323
373
423
473
523
5 6 7 8 9 10
time, year
T
e
m
p
e
r
a
t
u
r
e
,

K
Lambda=1.0
Lambda=0.5
Lambda=0.1
Lambda=1.0, analytical
Lambda=0.5, analytical
Lambda=0.1, analytical
Mid-stack temperature (transient shape-factor); Lx=20 m
323
373
423
473
523
5 6 7 8 9 10
time, year
T
e
m
p
e
r
a
t
u
r
e
,

K
Lambda=1.0, simulation
Lambda=1.0, analytical
Lambda=0.5, simulation
Lambda=0.5, analytical
Lambda=0.1, simulation
Lambda=0.1, analytical
18 SPE 113461

Figure 9: Mid-stack temperature for 20 =
x
L m and
10
1
2
1
, , 1 = . The dotted lines correspond to DP simulation
results using the transient thermal shape-factor, the solid lines show the corresponding analytical results.



























Figure 10: Simulated DP oil rate using a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = , for 20 =
x
L m and
10
1
2
1
, , 1 = . Dark colours correspond to simulations where no viscosity correction was applied (MuC=0); light
colours correspond to simulations where viscosity correction was applied (MuC=1).























Oil rate; Lx=20 m
0
2
4
6
8
10
12
0 5 10 15 20 25 30
time, year
O
i
l

r
a
t
e
,

m
3
/
d
Lambda=1.0, MuC0
Lambda=0.5, MuC0
Lambda=0.1, MuC0
Lambda=1.0, MuC1
Lambda=0.5, MuC1
Lambda=0.1, MUC1
Oil rate (transient shape-factor); Lx=20 m
0
2
4
6
8
10
12
0 5 10 15 20 25 30
time, year
O
i
l

r
a
t
e
,

m
3
p
d
Lambda=1.0; MuC=0
Lambda=0.5: MuC=0
Lambda=0.1, MuC=0
Lambda=1.0, MuC=1
Lambda=0.5, MuC=1
Lambda=0.1, MuC=1
SPE 113461 19

Figure 11: Simulated DP oil rate using the transient thermal shape-factor, for 20 =
x
L m and
10
1
2
1
, , 1 = . Dark
colours correspond to simulations where no viscosity correction was applied (MuC=0); light colours correspond to
simulations where viscosity correction was applied (MuC=1).

The results for 50 =
x
L m are shown in Figure 12 - Figure 15. Like in the 20 =
x
L m case, the simulated and analytical
matrix-block temperatures initially agree very well. Note, however, that in the present case, at a certain moment, all simulated
temperature curves start deviating from their corresponding analytical curves. This deviation is remarkable, because the
simulation results over-predict the corresponding analytical temperatures, although one would expect the analytical results to
provide upper-limits for the simulation results (for they allow for an infinite heating rate immediately following steam-
injection). It turns out that this sudden rise above the analytical temperature curves is correct, and corresponds to a real
physical effect that is not taken into account in the derivation of the analytical results: the gas-oil contact moves down the
stack, so at a certain moment in time, at any given location, the local oil saturation will start to decrease, resulting in a
corresponding decrease of the effective heat-capacity
3
.

Again, because the simulated matrix-block temperatures initially agree well with the analytical results, we can already
upfront be confident that the simulated oil rates are correct (because the derivative of the temperature determines the oil-rate
by expansion, which dominates the early-time response).

The oil rates themselves are shown in Figure 14 (for a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = ) and Figure 15
(for the transient thermal shape-factor). For the 50 =
x
L m cases considered here, the impact of the DP viscosity-correction
factor is more pronounced than in the 20 =
x
L m case: the oil rates of the simulations where DP viscosity correction is
applied (labelled MuC=1), is higher (because of the reduced viscosity) than those in which viscosity correction is not
applied (MuC=0). As mentioned above, the DP viscosity-correction effects are expected to be limited to times 25 . 0
d
t ,
which for the 50 =
x
L m case, approximately corresponds to 12 years following start of steam-injection, in agreement with
the results in Figure 14 and Figure 15. Although the impact of the viscosity correction does not appear very significant in our
simulations, at larger fracture spacings and/or higher initial oil viscosity, the effect is further amplified and should be taken
into account.

The impact of using the transient thermal shape-factor is significant, as can be seen by comparing the curves in Figure 14
(where a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = was applied) with the corresponding curves in Figure 15
(using the transient thermal shape-factor). The constant thermal shape-factor predicts initial peak oil-rates (due to thermal
expansion) which are approximately lower by a factor of 2 than those using the transient shape-factor.













3
Indeed, the simulator temperature starts deviating from the analytical curve when the gas-oil gravity drainage saturation
front reaches the mid-stack grid-block. At that moment, the oil-saturation in the mid-stack grid-block starts to decrease (and
the gas saturation increases). To illustrate this phenomenon, we have also plotted the oil-saturations in the mid-stack grid-
block in Figure 12 and Figure 13. When the oil-saturation decreases, the effective heat-capacity of the matrix-block starts
decreasing (the heat-capacity of the oil is of the same order of magnitude as the heat-capacity of the matrix-grains, while the
gas heat-capacity is much smaller). This in turn implies that the rate of temperature-increase of the partially de-saturated
matrix-block will increase, and this is why the simulator grid-block temperatures cross (as they should) the analytical results
(derived under the assumption of a constant oil saturation and heat-capacity).
20 SPE 113461
































Figure 12: Mid-stack temperature and oil saturation for 50 =
x
L m and
10
1
2
1
, , 1 = . The dark colours correspond
to DP simulation results using a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = , the light colours show the
corresponding analytical results.




















Mid-stack temperature, saturation; Lx=50 m
323
373
423
473
523
5 10 15 20 25 30
time, year
T
e
m
p
e
r
a
t
u
r
e
,

K
0.5
0.6
0.7
0.8
0.9
1
O
i
l

s
a
t
u
r
a
t
i
o
n

Lambda=1.0
Lambda=0.5
Lambda=0.1
Lambda=1.0, analytical
Lambda=0.5, analytical
Lambda=0.1, analytical
Mid-stack temperature, saturation
(transient shape-factor); Lx=50m
323
373
423
473
523
5 10 15 20 25 30
time, year
T
e
m
p
e
r
a
t
u
r
e
,

K
0.5
0.6
0.7
0.8
0.9
1
O
i
l

s
a
t
u
r
a
t
i
o
n

Lambda=1.0, simulation
Lambda=1.0, analytical
Lambda=0.5, simulation
Lambda=0.5, analytical
Lambda=0.1, simulation
Lambda=0.1, analytical
SPE 113461 21




Figure 13: Mid-stack temperature and oil saturation for 50 =
x
L m and
10
1
2
1
, , 1 = . The dark colours correspond
to DP simulation results using the transient thermal shape-factor, the light colours show the corresponding analytical
results.

























Figure 14: Simulated DP oil rate using a constant thermal shape-factor
2 2 2
) 1 (
x th
L + = , for 50 =
x
L m and
10
1
2
1
, , 1 = . Dark colours correspond to simulations where no viscosity correction was applied (MuC=0); light
colours correspond to simulations where viscosity correction was applied (MuC=1).




















Oil rate; Lx=50 m
0
2
4
6
8
10
12
0 5 10 15 20 25 30
time, year
O
i
l

r
a
t
e
,

m
3
/
d
Lambda=1.0, MuC0
Lambda=0.5, MuC0
Lambda=0.1, MuC0
Lambda=1.0, MuC1
Lambda=0.5, MuC1
Lambda=0.1, MuC1
Oil rate (transient shape-factor); Lx=50 m
0
2
4
6
8
10
12
0 10 20 30
time year
O
i
l

r
a
t
e
,

m
3
p
d
Lambda=1.0; MuC=0
Lambda=0.5: MuC=0
Lambda=0.1, MuC=0
Lambda=1.0, MuC=1
Lambda=0.5, MuC=1
Lambda=0.1, MuC=1
22 SPE 113461




Figure 15: Simulated DP oil rate using the transient thermal shape-factor, for 20 =
x
L m and
10
1
2
1
, , 1 = . Dark
colours correspond to simulations where no viscosity correction was applied (MuC=0); light colours correspond to
simulations where viscosity correction was applied (MuC=1).

Conclusions

We have introduced an upscaling method for dual-permeability simulation of heavy-oil recovery by steam-injection in
fractured reservoirs. In particular:

1. We have analytically derived the transient thermal shape-factor that reproduces, for all time-scales, the analytical
heating of 1D and 2D matrix-blocks.
2. We have generalised the analytically-based dual-permeability viscosity correction introduced in [20] to 1D and 2D
matrix-blocks.

We have shown that, when this transient shape-factor and the dual-permeability viscosity correction are implemented in dual-
permeability simulations, these coarse-grid dual-permeability simulations accurately reproduce the fine-grid single-porosity
simulations and analytical results.


Acknowledgements
We would like to thank Shell International Exploration and Production for permission to publish this paper.

Nomenclature

a : ASTM model parameter [-]
b : ASTM model parameter [-]
o
B : oil formation volume factor [m
3/
m
3
]
th
c : heat capacity [J/kg/K]
D D
C C
2 1
, : lumping constants [m
3
]
g : gravity acceleration [m/s
2
]
H : stack height [m]
k : permeability [m
2
]
r
k : relative permeability [-]
D D
K K
2 1
, : lumping constants [m
3
/s]
r
K : end-point relative permeability [-]
L : fracture spacing [m]
M : volumetric heat capacity [J/K/m
3
]
n : Corey exponent [-]
p : pressure [Pa]
p : volume-averaged pressure [Pa]
q : volumetric flow rate [m
3
/s]
R : viscosity reduction factor [-]
S : saturation [-]
t : time [s]
d
t : dimensionless time [-]
T : temperature [K]
T : volume-averaged temperature [K]
SPE 113461 23
d
T : dimensionless temperature [-]
V : volume [m
3
]
W : stack width [m]
x : position [m]
d
x : dimensionless length [-]





Greek letters
: thermal diffusivity [m
2
/s]
: volumetric expansion coefficient [K
-1
]
p
: ) 1 2 ( + p [-]
t
: partial time derivative [s
-1
]
x
: partial spatial derivative [m
-1
]
: difference operator
: hydraulic diffusivity [m
2
/s]
: thermal conductivity [J/s/K/m]
: ratio of fracture spacings [-]
: viscosity [Pa.s]
: mass density [kg/m
3
]
: shape-factor [m
-2
]
: porosity [-]

Subscripts

m : matrix
f : fracture
d : dimensionless
i : initial
sim : simulator
o : oil
p : pore
p : phase
w : water
g : gas
cw : connate water
1D : one dimensional
2D : two dimensional
h : hydraulic
th : thermal
red : reduced (viscosity)



Superscripts

s : short
m : medium
l : long

24 SPE 113461










References


1. Barenblatt, G.I., Zheltov, Iu.P. and Kochina, I.N., Basic concepts in the theory of seepage of homogeneous liquids in fissured
rocks, Journal of applied mathematics and mechanics, vol. 24, pp. 1286-1303, 1960.
2. Warren, J.E. and Root, P.J., The behavior of naturally fractured reservoirs, SPE Journal, vol. 3 (3), pp. 245-255, September
1963.
3. Kazemi, H., Merrill, J.R, Porterfield, K.L. and Zeman, P.R., Numerical simulation of water-oil flow in naturally fractured
reservoirs, SPE Journal, 16 (6), pp. 317-326, December 1976.
4. Thomas, L.K., Dixon, T.N. and Pierson, R.G., Fractured reservoir simulation, SPE Journal, pp. 42-54, February 1983.
5. Coats, K.H., Implicit compositional simulation of single-porosity and dual-porosity reservoirs, SPE 18427, presented at the 10
th

SPE symp. Reservoir Simulation, Houston, Texas, Feb. 1989.
6. Ueda, Y., Murata, S., Watanabe, Y. and Funatsu, K., Investigation of the shape factor in the dual-porosity reservoir simulator,
SPE 19469, paper presented at the SPE Asia-Pacific Conference, Sydney, Australia, September 1989.
7. Zimmerman, R.W., Chen, G., Hadgu, T., and Bodvarsson, G.S., A numerical dual-porosity model with semianalytical treatment
of fracture/matrix flow, Water Resources. Res., vol. 29, pp.2127-2137, 1993.
8. Chang, M., Deriving the shape factor of a fractured rock matrix, Technical Report NIPER-696 (DE93000170), NIPER,
Bartlesville, Oklahoma.
9. Lim, K.T, and Aziz, K., Matrix-fracture transfer functions for dual porosity simulators, J. Petroleum Sci. Eng., vol. 13, pp. 169-
178, 1995.
10. Gilman, J.R. and Kazemi, H., Improvements in simulation of naturally fractured reservoirs, SPE Journal, pp. 695-707, August
1983.
11. Beckner, B.L., Ishimoto, K., Yamaguchi, S., Firoozabadi, A. and Aziz, K., Imbibition-dominated matrix-fracture fluid transfer
in dual porosity simulators, SPE 16981, presented at the 62
nd
SPE Annual Technical Conference, Dallas, September 1987.
12. Beckner, B.L., Firoozabadi, A. and Aziz, K., Modeling transverse imbibition in double-porosity simulators, SPE 17414,
presented at the SPE California Regional Meeting, Long Beach, California, March 1988.
13. Rossen, R.H., and Shen, E.I.C., Simulation of gas/oil drainage and water/oil imbibition in naturally fractured reservoirs, SPE
Reservoir Engineering, pp. 464-470, November 1989.
14. Bech, N., Jensen, O.K., Nielsen, B., Modeling of Gravity-Imbibition and Gravity-Drainage Processes: Analytic and Numerical
Solutions, SPE 18428, presented at the Society of Petroleum Engineers Symposium on Reservoir Simulation, Houston, TX, Feb.
6-8, 1989.
15. Bourbiaux, B., Granet, S., Landereau, P., Noetinger, B., Sarda, S. and Sabathier, J.C., Scaling up matrix-fracture transfers in
dual-porosity models: theory and application, SPE 56557, presented at the 1999 SPE Annual Technical Conference, Texas,
October 1999.
16. van Heel, A.P.G., Boerrigter, P.M., and van J.J. van Dorp, Thermal and hydraulic matrix-fracture interaction in dual-
permeability simulation, accepted for publication in SPEREE journal.
17. Zimmerman, R.W., Hadgu, T., and Bodvarsson, G.S., Development of a dual-porosity model for vapor-dominated fractured
geothermal reservoirs using a semi-analytical fracture/matrix interaction term, Proceedings Eighteenth Workshop on
Geothermal Reservoir Engineering Stanford University, Stanford, California, January 26-28, 1993.
18. Vermeulen, T., Theory for irreversible and constant-pattern solid diffusion, Ind. And Eng. Chem., vol. 45, pp. 1664-1670.
19. Lu, H., Di Donato, G., and Blunt, M.J., General transfer functions for multi-phase flow, SPE 102542, SPE ATCE, San
Antonio, September, 2006.
20. Wassing, B., Van Wunnik, J., Warrlich, G., Lamki, A., Johnson, T., Gittins, J., and Riyami, M. Preparing a fractured carbonate
field for steam injection, 14
th
EAGE European Symposium on Improved Oil Recovery, Cairo, April 2007.
21. ASTM Viscosity Temperature Charts for Liquid Petroleum Products, ASTM Standard D341-87 in 1989 Annual Report of
Standards, vol. 05.01, Philadephia, ASTM, 1989.
22. Carslaw, H.S. and Jaeger, J.C., Conduction of heat in solids, Oxford University Press, Second edition, 1959.







SPE 113461 25











Appendix A Temperature diffusion in 1D matrix blocks

We consider a one-dimensional matrix block, bounded by two fractures at 0 = x and L x = , which initially has a uniform
temperature
i
T . From 0 = t onwards, the temperature in the fractures is changed to, and maintained at temperature
f
T .
The temperature evolution in the matrix block is then given by
( ) . 0 , , ) , 0 (
) 0 , (
2
> = =
=
=
t T t L T t T
T x T
T T
f
i
x t

(A-1)
To simplify the analysis we introduce a characteristic unit of length, L , and a characteristic unit of time:
2
L = , and
define a dimensionless time and length as follows:
. t t
L x x
d
d
=
=
(A-2)
We also introduce a dimensionless temperature [ ] 1 , 0
d
T as follows:
.
i f
f
d
T T
T T
T

= (A-3)
This choice of the dimensionless temperature is convenient because it makes the temperature boundary conditions
homogeneous.
Expressed in these dimensionless variables, the above equations become
( ) . 0 , 0 , 1 ) , 0 (
1 ) 0 , (
2
> = =
=
=
d d d d
d d
d x d t
t t T t T
x T
T T
d d
(A-4)

The solution of this system is given by (see e.g. [22])
( ) ( ), exp sin
1
4 ) , (
2
0
1 , d k
k
d k
k
d d D d
t x t x T

=
(A-5)

where ( ) 1 2 + = k
k
.

For sufficiently small times, ( 05 . 0
d
t ), Eq. A-5 can be approximated by
,
2
1
2
1 ) , (
1 ,

d
d
d
d
d d
s
D d
t
x
erfc
t
x
erfc t x T (A-6)
where () erfc denotes the complementary error-function.

For times 05 . 0
d
t , the lowest-order term of Eq. A-5 is an excellent approximation:
26 SPE 113461
( ) ( ) ( )
d d d d
l
D d
t x t x T
2
1 ,
exp sin
4
,

(A-7)
In our discussion, the (volume) averaged temperature in the matrix block, ( ) ( )
d d d d d D d
dx t x T t T

=
1
0
1 ,
, , is an important
quantity. Integrating Eq. A-5, we find
( ). exp
1
8 ) (
2
0
2
1 , d k
k
k
d D d
t t T

=
(A-8)
The corresponding short- and long-time approximations for ( )
d d
t T are obtained by integrating Eqs. A-6 and A-7:

( )
( ) ( ) 05 . 0 , exp
8
05 . 0 , 4 1
2
2
1 ,
1 ,


d d d
l
D d
d
d
d
s
D d
t t t T
t
t
t T

(A-9)

Appendix B Temperature diffusion in 2D matrix blocks

We consider a two-dimensional matrix block, bounded by four fractures at 0 = x and
x
L x = , 0 = y and
y
L y = , which
initially has a uniform temperature
i
T . From 0 = t onwards, the temperature in the fractures is changed to, and maintained at
temperature
f
T . Assuming homogeneous and isotropic matrix properties, the evolution of the dimensionless temperature is
given by
( ) ( ) . 0 , , , ) , 0 , ( , , ) , , 0 (
) 0 , , (
} {
2 2
> = = = =
=
+ =
t T t L x T t x T t y L T t y T
T y x T
T T
f y x
i
y x t

(B-1)
Using separation of variables, the solution to the corresponding set of equations for the dimensionless temperature is found to
be

=
t
L L L
y
L
x
t y x T
y
l
x
k
y
l
x
k
l k l k
D d
2
2
0 0
2
2 ,
exp sin sin
1 1
4 ) , , (

(B-2)
where ) 1 2 ( + = p
p
.
Without loss of generality we take
y x
L L , and define a characteristic time based on the smallest grid-block dimension:

2
x
L
= . Similar to the 1D case treated in the previous section, we define dimensionless distances in the x and y directions,
and dimensionless time as follows:

t
t
L
y
y
L
x
x
d
y
d
x
d
= = = ; ; .

Using these variables, the 2D dimensionless temperature profile is given by
( ) ( )

+ =

=
d l
y
x
k d l d k
l k l k
d d d D d
t
L
L
y x t y x T
2
2
2
0 0
2
2 ,
exp sin sin
1 1
4 ) , , (

. (B-3)
Eq. B-3 can be written as the product of two 1D temperature profiles:

) , ( ) , ( ) , , (
2
2
1 , 1 , 2 , d
y
x
d D d d d D d d d d D d
t
L
L
y T t x T t y x T

= , (B-4)
where the 1D dimensionless temperature profile is given by Eq. A-5
SPE 113461 27


In the 1D case, the evolution of the temperature profile was split in a small-time region ( 05 . 0
d
t ) and a large-time region
( 05 . 0 >
d
t ), where the exact temperature profile is approximated by short- and long-time approximations.
( ) ) 05 . 0 ( 1
2
1
2
,
1 ,
<

d
d
d
d
d
d d
s
D d
t
t
x
erf
t
x
erf t x T K K K K K (B-5)
( ) ) 05 . 0 ( ) exp( ) sin(
4
,
2
1 ,
>
d d d d d
l
D d
t t x t x T K K K K K K

(B-6)
Note that erf in Eq. B-5 denotes the error-function (and not the complementary error-function).

In the 2D case, we have to distinguish three time-intervals:

I : 05 . 0
d
t , in which case
( ) ( )

d
y
x
d
s
D d d d
s
D d d d d
s
D d
t
L
L
y T t x T t y x T
2
1 , 1 , 2 ,
, , , , (B-7)

II :
y d d
t t
,
05 . 0 05 . 0 <

( ) ( )

d
y
x
d
s
D d d d
l
D d d d d
m
D d
t
L
L
y T t x T t y x T
2
1 , 1 , 2 ,
, , , , (B-8)

III:
y d
t t
,
05 . 0 > ,

( ) ( )

d
y
x
d
l
D d d d
l
D d d d d
l
D d
t
L
L
y T t x T t y x T
2
1 , 1 , 2 ,
, , , , (B-9)
where we have introduced a dimensionless time which is defined based on the length in the y -direction:
y
y d
t
t

= :
,

with
2
:
y y
L = (and since we have chosen
y x
L L , also
y x
).

The average 2D matrix-block temperature is easily obtained from the average 1D matrix block temperature.
By integration of the above 1D small- and long-time approximations, we find the average 2D matrix-block temperature for
the three time-intervals:

I :
x
t 05 . 0 , in which case
( ) ( )

d
y
x s
D d d
s
D d d
s
D d
t
L
L
T t T t T
2
1 , 1 , 2 ,
(B-10)

II :
y x
t 05 . 0 05 . 0 <

28 SPE 113461
( ) ( )

d
y
x s
D d d
l
D d d
m
D d
t
L
L
T t T t T
2
1 , 1 , 2 ,
(B-11)

III:
y
t 05 . 0 > ,

( ) ( )

d
y
x l
D d d
l
D d d
l
D d
t
L
L
T t T t T
2
1 , 1 , 2 ,
(B-12)
where

2
:
x
x
L
= and

2
:
y
y
L
= (and since we have chosen
y x
L L , also
y x
).
Appendix C Oil recovery due to thermal expansion

We assume that the specific oil- and water volumes are linearly dependent on the temperature:
( )
( ), ) ( ) (
, ) ( ) (
i w i w w
i o i o o
T T T V T V
T T T V T V
+ =
+ =

(C-1)
where
w o
, are the thermal expansion coefficients of oil and water, respectively. We assume that the water that is present is
connate water which remains immobile. The thermal expansion of the pore volume is similarly assumed to be linearly
dependent on the temperature:
( ). ) ( ) (
i p i o
T T T T + = (C-2)
Then the cumulative net oil volume expulsed (in reservoir volumes) from a matrix block at time is given by
( ) ( ) , ) (
0 ,

+ = d T T S S t Q
i wp cw op o r o
(C-3)
where denotes the volume of the matrix block, and where we have introduced the notation
. :
, :
p w wp
p o op


=
=
(C-4)
The net cumulative oil volume expulsed, when expressed in terms of surface volumes, is given by
( )
( ) , ) (
0

+
= d T T
B
S S
t Q
i
o
wp cw op o
o

(C-5)

C1 Oil rate due to thermal expansion for 1D matrix-blocks

In the case where we have 1D matrix blocks, Eq. C-3 becomes:
( ) ( ) , ) , ( ) (
0
0

+ =
x
L
i wp cw op o o
dx T t x T S S WH t Q (C-6)
where H denotes the height of the stack, and W denotes the width of the stack in the neutral (translation-invariant) direction
parallel to the matrix-fracture interfaces at 0 = x and
x
L x = .
This equation can be written in terms of dimensionless time, distance and temperature:
{ } , ) , ( 1 ) (
1
0
1

=
d d d d D d o
dx t x T C t Q (C-7)
where

( )
.
] [
0
1
o
i f wp cw op o x
D
B
T T S S WHL
C
+
=

(C-8)

SPE 113461 29
Eq. C-7 shows that the cumulative oil expelled by thermal expansion is a function of the average temperature of the matrix-
block.
The momentaneous oil rate is given by the time-derivative of Eq. C-7:
{ } . ) , ( 1 ) (
1
0
1

=
d d d d
d
D d o
dx t x T
dt
d
C t q (C-9)
Using the short- and long-time approximations for the dimensionless temperature (Eq. A-9), we obtain:
05 . 0
2
) (
1
<
d
d
D
d o
t
t
C
t q K K

(C-10)
05 . 0 ) exp( 8 ) (
2
1
>
d d D d o
t t C t q K K K K K K (C-11)

In the above equations,
d
d d d
d o
t
t Q t t Q
t q

) ( ) (
) (
+
= , so to obtain ) (t q
o
, we have to replace
d
t by
2
x
L
t
:
) ( ) (
2 d o
x
o
t q
L
t q

= , thus: (C-12)
05 . 0
2
) (
1
< =
d
d
D s
o
t
t
K
t q K K

(C-13)

(C-14)

where

( )
o x
i f wp cw op o
x
D
D
B L
T T S S WH
L
C
K
] [
0
2
1
1
+
= =

(C-15)

C2 Oil rate due to thermal expansion for 2D matrix-blocks

The derivation of the oil rate by thermal expansion for 2D matrix-blocks is analogous to that of the 1D case discussed in the
previous section: the net cumulative oil volume expelled from the matrix-block (expressed in terms of surface volumes) is
given by:
{ }

=
x
y L
L
d d d d d D d D d o
dy dx t y x T C t Q
0 0
2 , 2
, ) , , ( 1 ) ( (C-16)
where
( )
o
i s wp cw op o y x
D
B
T T S S L HL
C
] [
0
2
+
=

(C-17)
where
x
L and
y
L denote the fracture spacings.
The volumetric oil rate is obtained by taking the time derivative of Eq. C-16:
{ }

=
x
y L
L
d d d d d D d
d
D d o
dy dx t y x T
dt
d
C t q
0 0
2 , 2
. ) , , ( 1 ) ( (C-18)
In this case we obtain for the short-, medium- and long-time regimes:
I : Short-time regime (
x
t 05 . 0 ):
( )

d
d
d
s
D d
d
t
t
t T
dt
d
8 1
2
2 ,
(C-19)

II : Medium-time regime (
y x
t 05 . 0 05 . 0 < ):

05 . 0 ) exp( 8 ) (
2
1
> =
d d D
l
o
t t K t q K K K K K K
30 SPE 113461
( )

+
d
d
d d
m
D d
d
t
t
t t T
dt
d

2
2
2 ,
2
4 1 ] exp[ 8 (C-20)

III: Long-time regime (
y
t 05 . 0 > ):
( ) ( ) ( ) ] 1 exp[ 1
64
2 2 2
2
2 , d d
l
D d
d
t t T
dt
d

+ +

(C-21)

where

2
:
x
x
L
= and

2
:
y
y
L
= (and since we have chosen
y x
L L , also
y x
).



Thus, the expressions for the oil-rate by thermal expansion in the 2D case are given by

In the short-time regime, i.e. for 05 . 0
d
t :
( )

d
d
D s
o
t
t
K
t q 8 1
2
2
, (C-22)
In the medium-time regime, i.e. for
2
05 . 0 05 . 0
d
t :

+
d
d
d D
m
o
t
t
t K t q

2
2
2
2
4 1 ] exp[ 8 ) ( , (C-23)
For the long-time regime, i.e. for
2
05 . 0
d
t
( ) ( ) ( ) ] 1 exp[ 1
64
2 2 2
2
2
d
D l
o
t
K
t q

+ + , (C-24)
where

( )
.
] [
0
2
2
2
o x
i s wp cw op o y
x
D
D
B L
T T S S HL
L
C
K
+
= =

(C-25)


Appendix D The transient shape-factor

The (time-dependent, thermal) shape-factor is defined as follows [8, 9,16]:
( )
.
) (
) (
) (
f m
m
T t T
dt
t T d
t

(D-1)
When expressed in terms of the dimensionless time (using
d x
dt
d
L dt
d
2

= ), Eq. D-1 becomes:

( )
.
) (
) (
1
) (
2
f d m
d
d m
x
d
T t T
dt
t T d
L
t

= (D-2)
Using the definition of the dimensionless temperature,
,
i f
f
d
T T
T T
T

= (D-3)
Eq. D-2 for the shape-factor can be written as
SPE 113461 31
.
) (
) (
1
) (
,
,
2
d m d
d
d m d
x
d
t T
dt
t T d
L
t

= (D-4)

D.1 Transient shape-factor for 2D diffusion
The 2D case is more elaborate, but it follows the same reasoning as used in the 1D case: we first derive the shape-factor
expression in terms of the dimensionless time
d
t (for the short-, medium- and long-time regime), and subsequently invert the
relation ) (
2 , d D d
t T to obtain ) (
2 , D d d
T t . Using this latter expression, we next eliminate
d
t from the shape-factor expression
in favour of
d
T . Without loss of generality we take
y x
L L , and we define the characteristic time

2
x
x
L
= and the
dimensionless time
x
d
t
t

= . We also define

2
:
y
y
L
= (and since we have chosen
y x
L L , also
y x
). When
expressed in the dimensionless time,
d
x
y
y
t
L
L
2

= .

I :short-time regime:
x
t 05 . 0 , i.e. 05 . 0
d
t :
Substituting Eq. B-10 into Eq. D-1, we find
.
) (
8 1
2
1
) (
2 ,
2 2
d D d
d
y
x
y
x
d
x
d
s
D
t T
t
L
L
L
L
t
L
t

+
=

(D-5)
For simulator implementation, we want to formulate
s
D 2
in terms of the dimensionless temperature
s
D d
T
2 ,
, rather than in
terms of the dimensionless time
d
t . This is achieved by using Eq. B-10 to express
d
t as a function of
s
D d
T
2 ,
: Eq. B-10
states that
s
D d
T
2 ,
is a quadratic equation in
d
t ; inverting this relation gives
d
t as a function of
s
D d
T
2 ,
:

), ; ( :
16
) 1 (
8
1
8
1
2 ,
2 ,
2

s
D d
s
D d
d
T
T
t
=

+
= (D-6)

8
) 1 ( 4 ) 1 ( 1
) ; (
2 ,
2
2 ,
+ + +
=
s
D d
s
D d
T
T (D-7)
where we have used the short-hand notation
y
x
L
L
= : .
Thus,
[ ]
.
8 1 2
) ; (
2 ,
2
2 , 2
s
D d x
s
D d D
T L
T

+
= (D-8)

[ ]
.
) 1 ( 4 ) 1 ( 1
) 1 ( 4 ) 1 (
16
) ; (
2 ,
2
2 ,
2 ,
2
2 2 , 2
+ + +
+ +
=
s
D d
s
D d
s
D d
x
s
D d D
T T
T
L
T

(D-9)

32 SPE 113461
II : Medium-time regime
y x
t 05 . 0 05 . 0 < , i.e.
2
05 . 0 05 . 0

<
x
y
d
L
L
t :

d
d d
m
D d
t
t t T 4 1 )) exp(
8
( ) (
2
2
2 ,
(D-10)
and substituting this in Eq. D-1 we find:
.
4 1
1 2
4 1
) (
2
2
2
2

+
=

d
d
d
x
d
m
D
t
t
t
L
t (D-11)
In this medium-time regime
d
t cannot be written in closed-form as a function of
m
D d
T
2 ,
, and therefore we cannot explicitly
express
m
D 2
in terms of
D d
T
2 ,
.
III: Long-time regime:
y
t 05 . 0 > , i.e.
2
05 . 0

x
y
d
L
L
t :

= ) exp(
8
)) exp(
8
( 1 ) (
2 2
2
2
2
2 , d d d
l
D d
t t t T

, (D-12)
and substituting this expression in Eq. D-1, we obtain:
( )
.
1
2
2 2
2
x
l
D
L

+
= (D-13)

Appendix E: Saturation functions, PVT and viscosity model

E-1: Saturation functions

In all simulations the capillary pressures were set to zero, and we use the same relative permeability relations in the SP and
DP simulations. In the fractures, we use straight-line relative permeabilities:
p p r
S k =
,
, with [ ] 1 , 0
p
S .
The relative permeabilities in the matrix are modelled using Coreys model:

( ) ,
,
p
n
p p r
S k

= (E-1)
where

p
S denotes the reduced (normalised) saturation of phase p .

The saturation parameters for the matrix relative permeabilities summarised in Table 2.


parameter value
cw
S
0.1
org
S
0.1
cg
S
0
rwo
K
0.5
row
K
0.7
rgo
K
1.0
rog
K
row
K
SPE 113461 33
wo
n
3
ow
n
3
go
n
1.5
og
n
4
Table 2: Summary of matrix relative permeability parameters (using a Corey model parameterisation).

E-2: PVT

The oil is modelled as a 2 component (heavy, light) dead oil: the light component is assumed not to dissolve in the oil-phase.
Water properties are calculated from the steam-tables.






E-3: Viscosity model

The water and steam viscosity are calculated from the steam-tables.

The oil viscosity is modelled by the ASTM correlation [21]:
( ) ( ) ( ) , log 7 . 0 log log
10 10 10
b T a + = + (E-2)
where is expressed in centi-Poise andT in Kelvin.
The ASTM viscosity is uniquely determined by specifying the viscosity at two different temperatures (which fixed a and
b ). In this paper we defined the temperature-viscosity relation by the values as shown in Table 3.

Temperature (K) Viscosity (cP)
323 100
506 2

Table 3: Temperature-viscosity pairs used to fix the ASTM viscosity correlation used in the simulations.