Investigation of Recovery Mechanisms in Fractured Reservoirs

Investigation of Recovery
Mechanisms in Fractured Reservoirs
by
Huiyun Lu
A dissertation submitted in fulfillment of the requirements for

the degree of Doctor of Philosophy of the University of London
and the Diploma of Imperial College
November 2007
Abstract
My work focuses on developing a physical-motivated approach to modeling displacement

processes in fractured reservoirs. I use analytical expressions for the average recovery
as a function of time to find matrix/fracture transfer functions for different recovery
mechanisms such as gas gravity drainage and counter-current imbibition.
To account for heterogeneity in wettability, matrix permeability and fracture geometry
within a single grid block a multi-rate model is proposed, where the matrix is composed of
a series of separate domains in communication with different fracture sets with different
rate constants in the transfer function.
I show example results using a streamline-based dual porosity model for single and
multi-rate models and demonstrate that a multi-rate model predicts rapid initial recovery
followed by very low production rates at late time. The use of streamlines elegantly allows
the transfer between fracture and matrix to be accommodated as source terms in the one-
dimensional transport equations along streamlines that capture the flow in the fracture.
I studied the Clair oil field operated by BP using a fine-grid reservoir model. The
method is very efficient allowing million-cell models to be run on a standard PC. I also
applied this methodology to simulate recovery in a Chinese oil field to assess the efficiency
of different injection processes.
On the basis of the transfer functions study, the formulation for the matrix-fracture
transfer function in dual permeability and dual porosity reservoir simulation was rewrit-
ten. Based on one-dimensional analytical analysis in the literature, expressions for the
transfer rate accounting for both displacement and fluid expansion were found. The
1
resultant transfer function is a sum of two terms: a saturation-dependent term repre-
senting displacement and a pressure-dependent term representing fluid expansion. The
expression reduces to the Barenblatt form for single-phase flow at late times, but more
accurately captures the pressure-dependence at early times. For displacement, both imbi-
bition and gravity drainage processes were considered. The transfer function is validated
through comparison with one and two-dimensional fine-grid simulations and compared
with predictions using the traditional Kazemi et al. formulation. This method captures
the dynamics of expansion and displacement accurately, giving better predictions than
current models, while being numerically more stable.
Publication as a results of this research is as follows:
Di Donato, G., Lu, H., Tavassoli, Z., Blunt, M. J. Multi-rate transfer dual porosity
modeling of gravity drainage and imbibition. SPEJ, Vol: 12, Pages: 77-88.
Lu H., Di Donato,G., Blunt Martin J. General Transfer Functions for Multiphase Flow.
SPE102542, proceedings of the SPE Annual Technical Conference and Exhibition held in
San Antonio, Texas, U.S.A., 24-27 September 2006.
Lu H., Blunt Martin J. General Fracture/Matrix Transfer Functions for Mixed-Wet
Systems. SPE107007, proceedings of the SPE Europec/EAGE Annual Conference and
Exhibition held in London, United Kingdom, 11-14 June 2007.
2
Acknowledgements
I would first like to thank my supervisor Professor Martin J. Blunt for his constant
support, guidance and encouragement over the last three years. This project would not
have gone as smoothly as it did without his brilliant leadership and management skills.
I would also like to thank Ahmed Sami Abushaikha, Qatar Petroleum, and Olivier
Gosselin, Total, for making us aware of the interesting behaviour of mixed-wet systems
and inspiring us to do this work. I am grateful to the EPSRC, DTI and the sponsors of
the ITF project on ’Improved Simulation of Flow in Fractured and Faulted Reservoirs’,
including - BP, especially Peter Cliford who helped to set up the Clair field model. I am
very grateful to Petro-China for providing me with data on the Liu7 oil field.
My great gratitude extends to Olivier Gosselin from TOTAL and Dr. Stephan Matthäi
from Imperial College London for serving on my examination committee.
I would like to thank all the friends and colleagues for their great help and support. It
is impossible to put in words my gratitude for their help. It is hard to think how I would
have survived without their friendship.
I would also like to thank my family for supporting me from the beginning of my study.
Bingjun and Lufu showed their patience and love to support and encourage me. I would
like to dedicate my thesis to them for their endless love!!!
Finally I would like to thank my father Huanxiang who came here twice to help me
during my transfer and the final stage of my PhD study, my sister in law Fengying who
came here to help me in spite of her busy work schedule and my brother, sister and all
the relatives who give me their endless encouragement.
3
Contents
1 Introduction 1
1.1 Recovery mechanisms in fractured reservoirs . . . . . . . . . . . . . . . . 1
1.2 Dual porosity modelling and shape factors . . . . . . . . . . . . . . . . . 5
1.3 Transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Streamline-based simulation . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Streamline-based dual porosity simulation 16

2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Streamline-based dual porosity formulation . . . . . . . . . . . . . . . . . 17
2.3 Single and multi-rate transfer functions . . . . . . . . . . . . . . . . . . . 19
2.3.1 Capillary-controlled imbibition . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Gravity drainage . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 Extensions to the approach . . . . . . . . . . . . . . . . . . . . . 26
2.3.4 Multi-rate transfer function . . . . . . . . . . . . . . . . . . . . . 27
2.3.5 Multi-rate model . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.6 Coupling with the fracture fractional flow . . . . . . . . . . . . . 29
3 Oil field case studies 32

3.1 Synthetic fractured reservoir simulation . . . . . . . . . . . . . . . . . . . 32
3.1.1 Reservoir description and computing comparison . . . . . . . . . . 32
3.1.2 Transfer rates and different transfer functions comparison . . . . . 34
i
3.1.3 Permeability-dependent transfer rates . . . . . . . . . . . . . . . . 35
3.1.4 Results for a single rate model . . . . . . . . . . . . . . . . . . . . 37
3.1.5 Results for multi-rate models . . . . . . . . . . . . . . . . . . . . 40
3.1.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Liu7 oil field simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.1 Reservoir description . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.2 Geological characteristics . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.3 Fluid characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.4 Parameters for imbibition modeling . . . . . . . . . . . . . . . . . 48
3.2.5 Parameters for gravity drainage . . . . . . . . . . . . . . . . . . . 51
3.2.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 Clair field recovery investigation . . . . . . . . . . . . . . . . . . . . . . . 60
3.3.1 Reservoir description . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.3.2 Reservoir simulation . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3.3 Recovery comparison with different simulators . . . . . . . . . . . 67
3.3.4 Initial conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4 General Transfer Functions for Multiphase Flow 72

4.1 Formulation for multiphase flow . . . . . . . . . . . . . . . . . . . . . . . 73
4.1.1 Traditional formulation . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1.2 Our formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.1.3 Numerical implementation . . . . . . . . . . . . . . . . . . . . . . 81
4.1.4 Numerical tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.1.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5 General Fracture/Matrix Transfer Functions for Mixed-Wet Systems 95

5.1 Formulation for mixed-wet media in two dimensional models . . . . . . . 96
5.1.1 Horizontal and vertical displacement . . . . . . . . . . . . . . . . 96
ii
5.1.2 Transfer rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.2 Test cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.1 Base case study for the general transfer function . . . . . . . . . . 99
5.2.2 One-dimensional and two-dimensional model tests with capillary
pressure and gravity . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.2.4 Comparison with other models . . . . . . . . . . . . . . . . . . . . 112
5.2.5 Sensitivity studies for the capillary pressure . . . . . . . . . . . . 119
5.2.6 Sensitivity studies for gas gravity drainage . . . . . . . . . . . . . 128
5.2.7 Correct factor of the gas gravity drainage . . . . . . . . . . . . . . 135
6 Conclusions and future work 139

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.1.1 Development of a general transfer function . . . . . . . . . . . . . 139
6.1.2 Developed a multi-rate model for displacement processes in frac-
tured reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.1.3 Sensitivity studies of the transfer function . . . . . . . . . . . . . 140
6.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.2.1 Upscaling strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 142
A Barenblatt et al.’s analytical solution for counter-current imbibition 1
B Numerical implementation 5
C Downscaling permeability data 10
D Calculation of the average saturation in gravity/capillary equilibrium 11
E Input deck of the Base Case (mixed-wet) for Eclipse model 13
iii
List of Figures
1.1 Schematic of different recovery processes in fractured reservoirs . . . . . . 2

1.2 Capillary imbibition from the fracture to matrix . . . . . . . . . . . . . . 3
1.3 Displacement of oil by gas gravity drainage forces . . . . . . . . . . . . . 4
1.4 Idealized fractured reservoir model represented by Warren and Root(1963) 6
1.5 Streamline-based dual porosity model . . . . . . . . . . . . . . . . . . . . 12
1.6 The principle of streamline-based simulation . . . . . . . . . . . . . . . . 13
2.1 Oil production rate comparisons of grid-based and streamline-based model 22

2.2 Run times of different grid numbers . . . . . . . . . . . . . . . . . . . . . 23
2.3 A single grid block in a field-scale simulation . . . . . . . . . . . . . . . . 27
2.4 Intersection fracture sets in a limestone and shale outcrop . . . . . . . . 28
2.5 Schematic of the conceptual model used to develop our dual porosity model 30
3.1 The porosity distribution of North Sea reservoir model . . . . . . . . . . 33

3.2 Comparison of oil rate using a fixed βav . . . . . . . . . . . . . . . . . . 37
3.3 Fracture and matrix saturation . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Recovery curve for different transfer rate . . . . . . . . . . . . . . . . . . 39
3.5 Oil production rate as a function of time for different model . . . . . . . 41
3.6 Comparisons of oil production for different model . . . . . . . . . . . . . 42
3.7 Fracture permeability distribution field . . . . . . . . . . . . . . . . . . . 44
3.8 Capillary pressure curve of Liu7 . . . . . . . . . . . . . . . . . . . . . . . 45
3.9 Oil and water relative permeability curves in the matrix . . . . . . . . . 46
iv
3.10 Sections of Liu7’s x direction permeability . . . . . . . . . . . . . . . . . 48
3.11 Water saturation of horizontal slices in the reservoir . . . . . . . . . . . . 53
3.12 Oil rate prediction of the field by different model . . . . . . . . . . . . . 54
3.13 Oil rate prediction for imbibition model . . . . . . . . . . . . . . . . . . . 55
3.14 Oil rate of the field by different wettability . . . . . . . . . . . . . . . . . 56
3.15 Field recovery of imbibition vs. injected pore volumes . . . . . . . . . . . 57
3.16 Vertical slices of gas saturation . . . . . . . . . . . . . . . . . . . . . . . 58
3.17 Field recovery for gravity drainage . . . . . . . . . . . . . . . . . . . . . 59
3.18 Clair field development location . . . . . . . . . . . . . . . . . . . . . . . 61
3.19 Schematic of N-S cross-section of Clair field . . . . . . . . . . . . . . . . 62
3.20 Clair field relative permeability curves in the matrix. . . . . . . . . . . . 63
3.21 Clair field capillary pressure curves in the matrix. . . . . . . . . . . . . . 63
3.22 Clair field permeability distribution . . . . . . . . . . . . . . . . . . . . . 65
3.23 Well placement for water flooding . . . . . . . . . . . . . . . . . . . . . . 66
3.24 Oil production comparison for different model . . . . . . . . . . . . . . . 68
3.25 Oil production prediction with different grid refinement . . . . . . . . . . 69
3.26 Oil production for dual porosity model with different grid number . . . . 70
4.1 Predicted and simulated oil transfer rate for Case 1 . . . . . . . . . . . . 87

4.2 Predicated and simulated oil transfer rate for Case 2 . . . . . . . . . . . 88
4.3 Predicted and simulated oil transfer rate for Case 3 . . . . . . . . . . . . 89
4.4 Predicted and simulated matrix water saturation for case 3 . . . . . . . . 90
4.5 Predicted and simulated average oil pressure for Case 3 . . . . . . . . . . 91
4.6 Predicted and simulated matrix water saturation for Case 3 . . . . . . . 92
4.7 Predicted and simulated matrix gas saturation for Case 4 . . . . . . . . . 93
5.1 Capillary pressure curve for the mixed-wet case . . . . . . . . . . . . . . 100

5.2 Base case for a block length of 1 m . . . . . . . . . . . . . . . . . . . . . 102
5.3 Base case for a block lenght 10 m . . . . . . . . . . . . . . . . . . . . . . 102
v
5.4 The three cases considered in this section . . . . . . . . . . . . . . . . . . 103
5.5 Grid blocks for one and two-dimensional models . . . . . . . . . . . . . . 105
5.6 The oil recovery for Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.7 The oil recovery for Case 1 of 10 m . . . . . . . . . . . . . . . . . . . . . 108
5.8 The oil recovery for Case 2 of 1 m . . . . . . . . . . . . . . . . . . . . . . 109
5.12 Case1: Oil recovery comparison by Sonier function . . . . . . . . . . . . . 113
5.13 Case1:Oil recovery comparison using Quandalle and Sabathier function . 114
5.14 Case2:Oil recovery comparison by Sonier function . . . . . . . . . . . . . 115
5.15 Case2:Oil recovery comparison of Case2:Oil recovery comparison using
Quandalle and Sabathier function . . . . . . . . . . . . . . . . . . . . . . 116
5.16 Case3:Oil recovery comparison by Sonier function . . . . . . . . . . . . . 117
5.17 Case3:Oil recovery comparison by Quandalle and Sabathier function . . . 118
5.18 Case3:Oil recovery comparison by Quandalle and Sabathier function . . . 119
5.19 The oil recovery for different capillary pressure curves . . . . . . . . . . . 120
5.20 The oil recovery for Case 1 of different capillary curves . . . . . . . . . . 120
5.24 Case 3 for a block height of 1 m with different capillary pressures. . . . . 122
5.25 Different capillary pressure curves for the base case . . . . . . . . . . . . 123
5.26 The different capillary pressure for Case 3 of 10 m . . . . . . . . . . . . . 124
5.27 The different capillary pressure for Case 3 of 10 m . . . . . . . . . . . . . 125
5.28 Matrix relative permeabilities with different exponents . . . . . . . . . . 126
5.29 The curves of different relative permeabilities exponents for 10 m block . 126
5.30 The perm curves used for the mixed-wet system . . . . . . . . . . . . . . 127
vi
5.31 Oil recovery curve with permeabilities curves in Fig. (5.28) . . . . . . . . 128
5.32 The oil viscosity is 10−3 P a.s . . . . . . . . . . . . . . . . . . . . . . . . 129
5.33 The oil viscosity is 10−2 P a.s. . . . . . . . . . . . . . . . . . . . . . . . 130
5.34 The oil viscosity is 0.1 P a.s . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.35 The oil viscosity is 1 P a.s . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.36 The matrix block is 10 m high . . . . . . . . . . . . . . . . . . . . . . . . 132
5.41 Gravity drainage with an oil viscosity of 10−3 P a.s . . . . . . . . . . . . 136
5.44 Gravity drainage with an oil viscosity of 0.1 P a.s . . . . . . . . . . . . . 137
5.45 Gravity drainage with an oil viscosity of 0.1 P a.s . . . . . . . . . . . . . 138
6.1 Simulations using a discrete fracture model at water breakthrough after

136 days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.2 Relative permeability curves determined from waterflood simulation with
the model above . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
D.1 The capillary curves used for the mixed-wet system . . . . . . . . . . . . 11
vii
List of Tables
3.1 The parameters used in dual porosity simulation . . . . . . . . . . . . . . 34

3.2 Parameters used in the simulations . . . . . . . . . . . . . . . . . . . . . 51
3.3 Parameters used in the simulations . . . . . . . . . . . . . . . . . . . . . 62
3.4 Parameters used to calculate the transfer rate . . . . . . . . . . . . . . . 64
3.5 Run time of different simulators for 6000 days (minutes) . . . . . . . . . 71
4.1 Parameters used for case 1 and 2 (compressible fluids/no capillary pressure) 85
4.2 Parameters used for Cases 3 and 4 (capillary or gravity-driven flow with
compressible fluids) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.1 Parameters used for the mixed-wet case . . . . . . . . . . . . . . . . . . . 100

5.2 End-point saturations computed for the mixed-wet cases . . . . . . . . . 104
viii
Nomenclature
a inverse time constant

A area, L2 , m2
b oil relative permeability exponent, dimensionless
B Vermeulen correction factor, dimensionless
c component concentration (fractional density), M L−3 , kgm−3
D diffusion coefficient, L2 T −1 , m2 s−1
f fractional flow, dimensionless
F weighting function, dimensionless
g acceleration due to gravity, Lt−2 , ms−2
G gravity fractional flow, dimensionless
h height, L, m
H capillary rise, L, m
J∗ entry dimensionless capillary pressure
K permeability, L2 , m2 or D
kr relative permeability, dimensionless
L, l length, m
Lc effective length, L, m
M mass per unit volume, M L−3 , kgm−3
P pressure, M L−1 T −2 , P a
Pc capillary pressure, M L−1 T −2 , P a
qt total velocity, LT −1 , ms−1
Q injection rate, L3 T −1 , m3 s−1
r gravity/capillary ratio, dimensionless
R recovery, dimensionless
S saturation, dimensionless
S∗ maximum saturation of invading fluid, dimensionless
t time, T, s
ix
∆t time step, T, s
T transfer function, T −1 , s−1
V volume, L3 , m3
Vt total velocity, LT −1 , ms−1
α rate in gravity transfer function, T −1 , s−1
β rate in imbibition transfer function, T −1 , s−1
γ compressibility, M L−1 LT 2 , P a−1
ξ small convergence parameter, dimensionless
η thermal diffusivity, m2 /s
θ temperature, centigrade
φ porosity, dimensionless
λ mobility, M −1 LT, P a−1 .s−1
µ viscosity, M L−1 T −1 , P a.s
ρ density, M L−3 , kg.m−3
δ coefficient,dimensionless
σ interfacial tension, L−1 t−2 , N m−1
Γ transfer rate, M L−3 T −1 , kgm−3 s−1
χ shape factor, L−2 , m−2
τ time of flight, t, s
ψ potential, M L−1 T −2 , P a
Subscripts
d diffusion
D dimensionless
e expansion
f flowing or fracture
g gas
i initial
x
j grid block label
k fracture set label
m matrix or stagnant
N number of fracture sets
o oil
p, q phase label
r residual
s displacement
t total
th heat
w water
∞ ultimate (at infinite time)
superscripts
n time level
H horizontal
V vertical
∗ end point
xi
Chapter 1
Introduction
1.1 Recovery mechanisms in fractured reservoirs
In fractured reservoirs there are four principal recovery processes, illustrated schemati-
cally in Fig. (1.1): fluid expansion, capillary imbibition, diffusion and gravity-controlled
displacement. We will describe each of these processes in turn.
Initially the reservoir is at high pressure with oil in both fracture and matrix. During
primary recovery, the pressure will drop. Since the fractures are well connected, the
pressure will drop rapidly in them, while the lower permeability matrix will remain at
high pressure. This leads to a pressure difference between the matrix rock and the
fractures: slowly there will be flow of oil from matrix to fracture as the fluids expand.
When we drop below the bubble point, gas will evolve from solution and the expanding
gas will lead to further recovery from the matrix. This process is effective, but once the
gas is connected in the system, principally only gas will be produced, leaving significant
quantities of oil in the matrix.
1
Oil (a) Oil (b)
Matrix
Fracture
Gas
Oil (c) Oil (d)
Figure 1.1: Schematic of different recovery processes in fractured reservoirs: (a) fluid
expansion; (b) imbibition; (c) gravity-controlled displacement; and (d) diffusion. We
consider matrix/fracture transfer as a sum of the contributions from these different effects.
In order to improve recovery, it is necessary to maintain pressure - ideally above the

bubble point - by injecting another phase to displace the oil. One possible injectant is
water. Since the fractures have a high permeability, the water will rapidly invade the
fractures, surrounding the matrix block. If the block has water-wet characteristics (that
is a capillary pressure that is positive for some range of saturation), water will enter the
block by capillary imbibition. Oil will be displaced from the block and be recovered from
the fractures. This process is illustrated in Fig. 1.2: it is relatively straightforward to
study this process experimentally by immersing a core sample in brine and using the
weight of the core to monitor recovery. There have been many studies of this process
on different rock types, cores of different shape and wettability. Fig. 1.2 illustrates
a typical recovery profile that has a characteristic shape regardless of the block size,
shape, composition or wettability (assuming that there is some imbibition): there is a
rapid initial recovery followed by an approximately exponential relaxation to the ultimate
recovery. We will describe this process physically and mathematically in more detail later
in the thesis. The effectiveness of waterflooding will depend on the amount of water that
will imbibe into the matrix and the rate at which this occurs. We will quantify this
effect and introduce a transfer rate which allows an engineer to assess the effectiveness of
2
waterflooding in a fractured field.
water
Oil recovery - % recoverable oil

oil
numerical solution
experimental data
fracture matrix
Dimensionless time (tD)
Normalized oil recovery vs. dimensionless time for very strong imbibition
Figure 1.2: Connected fracture network that carries the fluid flow in communication
with a stagnant matrix, showing transfer of fluids from fracture to matrix (left). On the
right, experimental data from Zhou et al. are plotted as a function of dimensionless time
(R/R∞ = 1 − e−0.05tD ). The recovery can be matched empirically by a simple exponential
function (solid line) (Zhou,1997).
It is also possible to inject gas into a fractured reservoir. Like water, the gas will
rapidly invade the fractures, particularly near the top of the formation. Gas injection
is only effective if buoyancy forces allow the gas to enter the lower permeability matrix,
displacing oil from the bottom of the block. Fig. 1.3 shows this gravity drainage behaviour
at different times for an oil/water system. There are two remarkable features to the
displacement pattern: one is that gas rapidly reaches the equilibrium position H beyond
which the column is saturated with the wetting phase - the oil is retained in the matrix
due to capillary forces; another is that the saturation profile is almost uniform with
distance above the distance H, but this saturation slowly decreases over time. The gas
overcomes the capillary pressure and keeps pushing the oil to the bottom of the core
until oil and gas are in capillary/gravity equilibrium. Again, later in the thesis we will
show how to describe this process mathematically and use the information to analyze
gravity-controlled displacement in fractured reservoirs.
3
ｇａｓ
ｏｉｌ
１，１７　ｈｏｕｒｓ２．５０　ｈｏｕｒｓ１０．５　ｈｏｕｒｓ１１７　ｈｏｕｒｓ
Ｆｒｏｍ　ｔｈｅ　ＰｈＤ　ｔｈｅｓｉｓ　ｂｙ　Ａ．Ｓａｈｎｉ，　１９９８
Figure 1.3: Displacement of oil by gas under the influence of gravity . The gas moves in
from the top of the core, since the buoyancy force exceeds the capillary pressure. This
process continues until gas and oil are in gravity/capilary equilibrium (From the PhD
thesis by Sahni, 1998).
The final recovery process is diffusion. This will not be discussed in such detail in the
thesis, but is important in miscible gas injection processes. Here the injected gas and
oil can mix to form a single hydrocarbon phase and this hydrocarbon may be swept into
the fracture and produced. The gas and oil reach thermodynamic equilibrium through
the diffusion of individual components of both phases through the relatively stagnant
matrix. This is potentially a slow process but effective is miscibility is reached. Even in
immiscible systems, components of the injected gas may diffuse into the oil and cause it to
swell - such as in carbon dioxide injection - forcing oil out of the matrix and reducing its
viscosity, increasing flow rates. Implicit in this discussion has been the concept of a dual
porosity system which is described in more detail in the next section: we assume that a
fractured medium is composed of connected high permeability fractures that carry the
flow with a relatively low permeability matrix that contains most of the fluids. Recovery
is controlled by the transfer of oil out of the matrix by the physical mechanisms described
above. In this thesis we will develop mathematical models - called transfer functions - to
describe these processes quantitatively. The main concept will be to treat each physical
4
effect separately and make the overall transfer the sum of these.
1.2 Dual porosity modelling and shape factors
Fractured reservoir development is one of the most difficult technologies in oil field
exploration. Numerical simulation is more difficult than for unfractured reservoirs since
there is typically greater uncertainty in both the reservoir description and the proper
modeling of the reservoir dynamics.
Numerous papers on single- and two-phase flow in naturally fractured porous media,
have appeared in the literature especially when a medium approach is used. In single-
phase flow, Barenblatt et al.(1960) proposed the dual porosity system to model fractured
media. They represented the reservoir by two overlapping continua which are matrix and
fracture. They then formulated the flow equations for each continuum using conservation
of mass principles; and flow between the matrix blocks and the fractures is accounted for
by source functions. These source or transfer functions are derived using Darcy’s law
expressed over some mean path between the matrix-blocks, and the density difference
between liquid phases.
Warren and Root(1963) developed an idealized model to study the characteristic behav-
iour of fractured reservoirs. They used Fig. (1.4) to illustrate their model of a fractured
reservoir which is still used in numerical simulation today. This is a dual porosity model
which assumes that the reservoir is composed of two regions by primary porosity and sec-
ondary porosity. The primary porosity represents the matrix with low permeability, and
secondary porosity indicates fracture and high permeability. In their model the matrix
contributes significantly to the pore volume of the system but contributes negligibly to
the flow capacity, while the secondary porosity carries all the fluid flow. They used dual
porosity transfer functions for single-phase flow based on an assumption of quasi-steady
transfer flow:
5
Kχ
τ = φm (Pm − Pf ) (1.1)
µ
where τ is transfer function with unit s−1 , φm is the ratio of matrix porosity, χ is the
shape factor which is defined by:
4n(n + 2)
χ= (1.2)
l2
n is the number of flow directions and l is a characteristic length. The shape factor
was derived based on an integral material balance on this length. Then they related this
characteristic length to the sides of a cubic matrix block. It assumes a continuous uniform
fracture network oriented parallel to the principal axes of permeability. The matrix blocks
in this system occupy the same physical spaces as the fracture network and are assumed
to be identical rectangular parallelpipeds with no direct communication between matrix
blocks. The matrix blocks are also assumed to be isotropic and homogeneous.
x
Vugs
∆z Matrix
∆z
∆y Fracture
∆x
∆x ∆y
x
Actual reservoir
Reservoir Model on grid block
Matrix
Fracture
Lz ∆z
Ly ∆x ∆y
Lx
One matrix block Idealized model reservoir

surrounded by fracture one gridblock
Figure 1.4: Idealization a fractured heterogeneous porous medium. The figure on the
right presents the geological model which is represented by a cubic block on the left
(Warren, 1963)).
6
Kazemi et al.(1976) extended Warren and Root’s (1963) single-phase flow model to
simulate multi-phase flow. They solved the dual porosity system in three dimensions
numerically. As with the Warren and Root model, two sets of different equations are
required to define the complete system, which define a flow in the fracture and in the
matrix. The transfer function for a black-oil system was given by:
Km Krp χ
τp = φm (Ppm − Ppf ) (1.3)
µ
In addition, Kazemi et al. also gave a new definition for the shape factor for paral-
lelepiped and isotropic matrix blocks as:
1 1 1
χ=4 2 + 2 + 2 (1.4)
Lmx Lmy Lmz
derived based on a direct material balance on a rectangular parallelpiped matrix block

by assuming a pseudo-steady state.
Since then, many researchers have attempted to improve the dual porosity model pro-
posed by Kazemi et al. from different aspects: (1) shape factor calculation; (2) physical
modelling of multi-phase flow in naturally fractured reservoirs.
The shape factor is always hard to define since there is no clear definition of the value, it
is frequently used as a tuning factor for history matching. Several authors have proposed
different expressions for the shape factors.
Thomas et al.(1983) studied fine-grid single porosity upscale to a single-block dual-
porosity model. They gave 25/L2 for a three-dimensional oil and water model with
near unit mobility ratio displacement where L is block side length. Udea et al.(1989)
studied the shape factor for one and two-dimensional models . They also estimated that
the Kazemi shape factor needs to be multiplied by 3. Coats (1989) on the other hand
obtained a shape factor that is exactly twice the Kazemi model.
Lim and Aziz (1995) derived the shape factor by applying analytical solutions for the
single phase pressure diffusion equation for different parallelepiped geometries for the
7
matrix blocks:
" #
π2 Kx Ky Kz
χ= + 2 + 2 (1.5)
(Kx Ky Kz )1/3 L2x Ly Lz
This shape factor is only used for single phase flow of parallelepiped geometries. It is
not suggested to use it for other geometries.
Chang et al.(1993) avoided the assumption of pseudo-state by combining the geomet-
rical aspects of the system with analytical solutions of the pressure diffusion equation for
flow between matrix and fracture. They proposed a time-dependent shape factor. Using
the complete solution to the diffusion equation, they derived for one-dimensional flow:

exp − (2m + 1)2 πtD
P
π2
χ= 2P 1 m = 1, ∞ (1.6)
Lx (2m+1)2
exp − (2m + 1)2 πtD
where tD is the dimensionless time.

Although this equation counts for the transient flow and gives an analytical form of the
transient shape factor, the expression is too complicated to be used in simulators. It can
not be validated for non-orthogonal systems.
Van Heel et al.(2006) studied the shape factor for a both convection and diffusion
processes for a steam-enhanced gravity drainage model:
dθ̄m
χth (t) = − /ηth (θ̄m − θf ) (1.7)
dt
where ηth is the thermal diffusivity of the rock, and θ is temperature.

They found that shape factor is not just determined by the geometric form of a matrix
block, for the same matrix different processes require different shape factors. The shape
factor is not always a constant as assumed in current simulators.
From these studies we can conclude that the shape factor can not be used by an engineer
reliably since the definition is not clearly related to any physical processes. There are
lots of arguments about how to use the shape factor in certain circumstances, but it is
hard to say which one is the standard.
8
1.3 Transfer functions
Litvak (1985) presented a formulation for the simulation of natural fractured reservoirs
for a matrix block immersed in water:
!
krp
τp = φm Km χ(Pm − Pf ) + CGamf (1.8)
µp Bp
m
The term CGamf includes the effects of different forces. In his transfer functions, Litvak
(1985) shows a gravity term that involves a product of the difference in density between
phases and the difference in saturation heights between matrix and fracture.
However, Litvak (1985) did not show how to calculate the saturation height in matrix
and fracture. He gave a procedure to implement capillary and gravity forces (CGamf ) in
the equation by single matrix block simulations, considering the number of matrix blocks
contained in a grid cell and the water (gas) level in the grid block.
There are several assumptions in Litvak’s work (1985): (1) water can move rapidly
through the highly permeable fractures and water imbibes over the entire height in non-
fractured reservoirs; (2) the dual porosity treatment of capillary and gravity forces as-
sumes that imbibition of water (oil drainage in the gas case) in the matrix can occur only
in a portion of the oil zone invaded by water (displaced by gas); (3) water saturation in
the matrix blocks is not related to the oil-water contact due to the separation of matrix
blocks by fractures (matrix discontinuity). This is defined only by the properties of the
matrix rock. Tighter matrix may have higher water saturation because of higher capillary
pressure. As a result, higher initial water saturation can be observed in zones above a
low water saturation zone.
Sonier et al. (1986,1988) also proposed the following transfer function for oil, gas, and
water respectively:
" #
k k
m ro
τo = φm χ Pom − Pof + ρo g(zwf + zgf − zwm − zgm ) (1.9)
µo Bo m
9
" #
kk
rg
τg = φm χ Pom − Pof − (Pcgof − Pcgom ) − ρg g zgf − zgm (1.10)
µg Bg m
" #
kk
rw
τw = φm χ Pom − Pof − (Pcowf − Pcowm ) − ρw g zwf − zwm (1.11)
µw Bw m
where z is the height of fluid and is calculated by the corresponding fluid saturation:
swf − swf i
zwf = h (1.12)
1 − sorwf − swf i
Quandalle and Sabathier (1989) defined a transfer function which separates viscous,
capillary and gravity within a grid block. The model defines flow toward all six faces
of a three-dimensional parallelepiped-shaped block. They used coefficients for each force
acting in each flow direction:
Vb krp cp ρp
τp = K m
χ φpm − φpf (1.13)
V µp
The second term in parenthesis is defined for different faces of the parallelepiped, in
the x+ direction, it can be written as:
φpm − φpf = Pom − Pof − δv (p+

f z − Pf ) − δc (Pcpof − Pcpom ) (1.14)
and the in the z + direction,
∗ ∆z ∆z
φpm −φpf = Pom −Pof −δv p+
f z −Pf +ρ g −δg (ρpm −ρ∗ g) −δc (Pcpof −Pcpom ) (1.15)
2 2
The coefficients (δv , δg , δc ) were used to match fine grid simulations; they do not
necessarily have a physical meaning.
Gilman and Kazemi (1983) updated the earlier dual porosity model of Kazemi et al.
(1992) by refining the treatment of mobility. They alternated the transfer functions to
10
include fracture relative permeability when fluid is flowing from the fracture to the matrix.
Birks (1955), Mattax et al. (1955) proposed practical ways of calculating oil recovery
from the matrix blocks. Birks formulated the mechanics of oil displacement from the
matrix blocks first by an idealized capillary model, and second by a simple relative-
permeability model. Mattax et al. (1955) were concerned with imbibition oil recovery
from matrix blocks in water-drive reservoirs. They developed an oil-recovery prediction
technique based on the semi-empirical relation that the time required to recover a given
fraction of oil from a matrix block is proportional to the square of the distance between
fractures.
The dual permeability model, proposed by Hill et al. (1985) differs from the dual
porosity approach by considering matrix block-to-block flow. This model is useful where
there is significant permeability in the matrix.
1.4 Streamline-based simulation
The coupling between fracture and matrix generally leads to dual porosity models
being more time consuming and memory intensive than single porosity (unfractured)
simulations. This frequently means that dual porosity simulations are restricted to very
coarsely gridded approximations that fail to capture the geological complexity of the
fracture network (Sonier, 1986).
To overcome these run-time limitations, the use of streamline-based simulation has
been considered (Di Donato et al., 2003,2004; Huang et al. 2004; Lake et al, 1981; Thiele
et al., 2004). The streamline simulator solves a three-dimensional problem by decoupling
it into a series of 1-D problems, each one solved along a streamline (Pollock, 1988). The
transfer between fracture and matrix is represented by source or sink terms in the one-
dimensional transport equations, making the method both fast and elegant see Fig. (1.5)
(Di Donato et al., 2003, 2004, 2007).
Streamlines were first applied in the study of well patterns and total recovery by Muskat
11
Fracture fluid transfer between the flowing and
stagnant regians along streamlines
(a) Matrix (b) Streamline
Figure 1.5: Streamline dual-porosity model for fractured reservoirs. (a) The real field
contains both a fracture network and a relatively low permeability matrix. Effective
transport properties are defined at the grid block scale (dashed lines). (b) Streamlines
follow the flow field computed at the grid block scale. This captures the movement
through the fracture network and high permeability matrix: the flowing fraction of the
system.
et al. (1935). Higgins et al. (1962) used stream tubes, which carry a fixed fluid flux to
treat two-phase flow in a homogeneous medium. The method is composed of dividing
the stream tubes into elements of equal volume. Average mobility and geometric shape
factors were calculated for each element and the total resistance along each stream tube
was used to calculate the total flow rate for each stream tube.
Gelhar et al. (1971) introduced the concept of the time-of-flight along the streamline
by modeling a reservoir with anisotropic permeability. Time-of-flight was defined as the
time which a particle travels from an injection point to a sink or production point, and
it was different for each streamline. It is related to the reservoir heterogeneity directly,
and accounts for the reservoir driving forces when the velocity field is calculated -see Fig.
(1.6).
Bommer et al. (1962) solved the transport equations along streamlines to account for
chemical reactions and physical diffusion for modeling in-situ uranium leaching. Lake et
al. introduced the concept of decoupling the vertical response from the areal one in a
12
Permeability field Pressure solve Saturation along SL
Saturation for
Initial saturation SL tracing
the next time step
Figure 1.6: The principle of streamline-based simulation is: (1) at the beginning of each
time step the saturation, permeability and porosity are defined on an underlying grid.
Just as in conventional grid-based methods, total mass balance in each grid block is used
to construct an equation for pressure and this is solved on the grid with known boundary
conditions at wells; (2) then, from Darcy’s law, the total velocity is found at each cell
face and these velocities are used to trace streamlines from injectors to producers; (3)
Saturations are mapped from the grid to streamlines and the conservation equation is
solved along each streamline ignoring gravity; (4) The saturation is then mapped down
onto the grid and including only the gravity terms is solved on the grid; (5) The simulation
returns to step (1).
polymer/surfactant displacement (Lake, 1981).

Pollock (1988) used a linear interpolation of the velocity field vector within each grid
block to trace three-dimensional streamlines. His algorithm has been widely used in
most streamline simulation models since it is simple and accurate. Batycky et al. (1997)
introduced a three-dimensional streamline-based fluid flow simulator. It accounted for
the effects of changing well conditions as well as gravity for incompressible multi-phase
flow. An operator splitting technique was used to separate the calculation of the viscous
and gravity flow components (Bratvedt,1996). Ponting (2004) extended the streamline
technique to handle compressibility and depletion by a hybrid method.
Di Donato et al.(2003,2004) presented a dual porosity streamline-based model for frac-
tured reservoirs. They applied this methodology to study capillary-controlled transfer
13
between fracture and matrix and demonstrated that using streamlines allowed multi-
million cell models to be run using standard computing resources. They showed that the
run time could be orders of magnitude smaller than equivalent conventional grid-based
simulation (Belayneh, 2004). This streamline approach has been applied by other au-
thors that have extended the method to include gravitational effects, gas displacement
and dual permeability simulation, where there is also flow in the matrix (Di Donato,
2003; Lake,1981).
Thiele et al. (2004) have described a commercial implementation of a streamline dual
porosity model based on the work of Di Donato that efficiently solves the one-dimensional
transport equations along streamlines (Pollock, 1988).
1.5 Overview
In this work I will present a physically-motivated approach to modeling displacement

processes in fractured reservoirs. This main contribution of this thesis is the development
of a general transfer function that accurately captures the average recovery from the ma-
trix due to fluid expansion and the combined effects of gravitational and capillary forces.
A analytical expression for the average recovery for gas gravity drainage and counter-
current imbibition are used to derive the transfer functions. For capillary-controlled
displacement the recovery tends to its ultimate value with an approximately exponential
decay (Barenblatt, 1960). When gravity dominates the approach to ultimate recovery is
slower and varies as a power-law with time. I will apply transfer functions based on these
expressions for core-scale recovery in field-scale simulation using streamlines (Di Donato,
2004).
On the basis of this work, I extended the formulation for the matrix-fracture trans-
fer function in dual permeability and dual porosity reservoir simulation. The current
Barenblatt-Kazemi approach uses a Darcy-like flux from matrix to fracture, assuming a
quasi steady-state between the two domains (Barenblatt,1960; Warren, 1963, Kazemi,
14
1992). However, this does not correctly represent the average transfer rate in a dynamic
displacement. Based on one-dimensional analytical analysis in the literature, expressions
for the transfer rate accounting for both displacement and fluid expansion are found
(Vermeulen, 1953; Zimmerman, 1993).
The resultant transfer function is a sum of two terms: a saturation-dependent term rep-
resenting displacement and a pressure-dependent term representing fluid expansion. The
expression reduces to the Barenblatt (1960) form for single-phase flow at late times, but
more accurately captures the pressure-dependence at early times. The transfer function is
validated through comparison with one-dimensional fine-grid simulations and compared
with predictions using the traditional Kazemi et al. (1976) formulation. I will show
that our method captures the dynamics of expansion and displacement accurately, giving
better predictions than current models, while being numerically more stable.
I also extend a model of fracture/matrix transfer in dual porosity and dual permeability
systems to mixed-wet media, where there can be displacement due to imbibition when
the capillary pressure is positive combined with gravity-controlled displacement. This can
lead to a characteristic recovery curve from the matrix, with a period of rapid imbibition
followed by slower recovery where gravitational effects dominate. The general transfer
function model is refined to accommodate such cases by including transfer due to hori-
zontal and vertical displacement separately (Lu et al. 2006). The model is tested against
fine-grid simulation in one and two dimensions and accurate predictions are made in all
cases; in contrast the conventional Kazemi et al. (1976) model gives poor predictions of
rate and ultimate recovery.
15
Chapter 2
Streamline-based dual porosity

simulation
2.1 Background
Streamline-based simulation is now established as an attractive alternative to conven-

tional grid-based techniques for simulating displacements in highly heterogeneous reser-
voirs (Batycky, 1997; King, 1998). For incompressible or nearly incompressible flow
simulated through more than around 100, 000 grid blocks, streamlines have been shown
to be generally faster than grid-based methods and have been successfully used to study
several field cases in recent years (Baker, 2002; Grinestaff et al., 2000; Samier et al.,
2002).
Streamline-based models have recently been generalized to model fluid flow in fractured
reservoirs including matrix-fracture interactions. This methodology assumes that the
fluid flow occurs primarily through the high permeability fractured system and the matrix
acts as fluid storage (Barenblatt, 1960; Dean, 1988; Kazemi et al., 1992). A matrix-
fracture transfer function is used to calculate the fluid exchange between the matrix and
fracture.
As discussed in the previous chapter, many different transfer functions have been pro-
16
posed to address different problems such as water counter current imbibition, gravita-
tional segregation and viscous displacement (Bratvedt, 1996; Di Donato, 2003, 2004;
Sahni, 1998). The main characteristic of these transfer functions is that they are an ex-
tension of Warren and Root’s model that assume quasi-static Darcy-like flow and neglect
saturation and pressure gradients in the matrix blocks.
Di Donato et al. (2003, 2004) developed a physically-motivated approach to modeling
displacement process in fractured reservoirs. They used expressions of matrix/fracture
transfer functions that match capillary imbibition experiments. These transfer functions
were used in field-scale simulation. They proposed a multi-rate model to account for
heterogeneity in wettability, matrix permeability and fracture geometry within a single
grid block. This allows the matrix to be composed of a series of separate domains in
communication with different fracture sets with different rate constants in the transfer
function. This is the main methodology I am going to use to investigate fractured reser-
voir recovery mechanisms in my project. I will summarize the streamline-based model
methodology in the subsequent sections.
2.2 Streamline-based dual porosity formulation
Conceptually, in a dual porosity simulator the flowing (fracture and high permeability
matrix) and stagnant (low permeability matrix) domains are each defined by different
blocks with specific porosity, permeability, depth, etc. It assumes that there is no signif-
icant fluid flow between matrix grid blocks. Transport occurs through exchange of fluid
from the matrix to the fractures and in the fracture network itself. In the streamline-
based dual porosity model, high permeability matrix is combined into the flowing fraction
conceptually by considering its contribution to the effective permeability of the flowing
fraction (Di Donato, 2003, 2004).
The streamline code that I use assumes incompressible two-phase flow of oil and wa-
ter (Batycky, 1997; Di Donato, 2003, 2004; Sahni, 1998). While gravitational forces
17
are included in the transport of fluid in the flowing regions, the only mechanism for
fracture/matrix transfer that we allow is capillary-controlled imbibition (extensions to
gravity-controlled flow are discussed later). The volume conservation equations are:
∂Swf
φf + vt · ∇fwf + ∇ · gG = −T (2.1)
∂t
∂Swm
φm =T (2.2)
∂t
where T is the transfer function. Define a time-of-flight flight as the time taken for neutral
tracer to move a distance s along a streamline (Gelhar, 1971):
Z s
φf
τ (s) = ds (2.3)
0 Vt
Then we can transform Eq. (2.1) into an equation along a streamline as follows:
∂Swf ∂fwf 1 T
+ + ∇ · gG = − (2.4)
∂t ∂τ φf φf
Single porosity streamline simulation ( equivalent to setting T = 0; m = 0 in Eq. (2.1)

and (2.2)) can be described in the following five steps: (1) At the beginning of each time
step the saturation, permeability and porosity are defined on an underlying grid. Just
as in conventional grid-based methods, total mass balance in each grid block is used to
construct an equation for pressure and this is solved on the grid with known boundary
conditions at wells. (2) Then, from Darcy’s law, the total velocity is found at each cell
face and these velocities are used to trace streamlines from injectors to producers. (3)
Saturations are mapped from the grid to streamlines and the conservation Eq. (2.4) is
solved along each streamline ignoring gravity. (4) The saturation is then mapped down
onto the grid and Eq. (2.4) including only the gravity terms is solved on the grid. (5)
The simulation returns to step (1). In our dual porosity simulator, we follow exactly
the same approach. In computing the pressure field and tracing streamlines we only
18
use information on the flowing fraction saturation, porosity and permeability. The only
change is to modify step (3). Both the flowing and stagnant fraction saturations are
mapped onto streamlines and we solve the following two conservation equations along
each streamline:
∂Swf ∂fwf T
+ =− (2.5)
∂t ∂τ φf
∂Swm T
= (2.6)
∂t φm
Swf and Swm are then mapped back to the grid and the simulation follows the single
porosity methodology. Note that the effects of gravity in the fracture flow are accounted
for in step (4) (as in single porosity streamline methods), while gravity affecting frac-
ture/matrix transfer is included in the form of the transfer function T.
2.3 Single and multi-rate transfer functions
2.3.1 Capillary-controlled imbibition
Ma et al. presented an analysis of capillary-controlled imbibition, where water is im-

bibed into the matrix by capillary pressure and oil comes out into the fractures and high
permeability matrix (Ma, 1997). This mechanism is in line with the streamline conceptual
model which captures movement through the flowing fraction while the transfer of fluid
from flowing to stagnant region is modeled as a source/sink term in the one-dimensional
transport equation along each streamline.
Fig. (1.2) illustrates an idealized representation of a fractured reservoir. During water-
flooding of a fractured reservoir, most of the water flow is in the high permeability chan-
nels called fractures. The water from the fractures imbibes into the matrix by capillary
action and the oil comes out provided the matrix is water-wet. In this chapter we will
make this assumptions later (Chapter 4) this will be related.
19
A number of authors have performed counter-current imbibition experiments where
water-wet cores have been surrounded by water (Aronofsky, 1958; Morrow, 2001, Ding,
2005). The recovery of oil can be matched by a simple exponential function of time (Ding,
2005). Zhang et al. (1996) proposed an expression that matched a range of imbibition
experiments on samples with different geometry and fluid properties (Di Donato, 2003).
We will use a transfer function based on a semi-analytical solution for counter-current
imbibition proposed by Barenblatt et al. (1963). Since this solution has been ignored in
the petroleum literature, it is reviewed in Appendix A. The recovery, R, is:
R = R∞ (1 − e−βt ) (2.7)
where R∞ is the ultimate recovery. The rate constant β is defined by:
s
Km σow J 0 λo λw
β=3 · · (2.8)
φm L2c λt

∗
Swm =Sw
where σow is the oil/water interfacial tension. The mobilities λ = kr /µ are defined at
Sw∗ , the maximum saturation reached in the matrix during imbibition; we assume that
the system is not strongly water-wet and so this saturation is lower than 1 − Sorm , where
Sorm is the residual oil saturation in the matrix after waterflooding, including forced
displacement due to viscous r gravitational forces. The case where Sw∗ = 1 − Sorm , giving
β = 0 in Eq. (2.8), has been considered by Tavassoli et al. (2005). J 0 is the dimensionless
gradient of the capillary pressure at Sw∗ , defined by:
r r
dPcm φm dJ(Swm ) φm
= · σow · =− · σow J 0 (2.9)
dSwm Km dSwm Km

∗
Swm =Sw ∗
Swm =Sw
where we assume Leverett J-function scaling of the capillary pressure:
r
φm
Pcm = σow J(Swim ) (2.10)
Km
The definition of the rate constant is similar to that derived by Zhou et al. (2000),
20
although we evaluate the mobility at the end of imbibition.
Only a linear system of length L was considered in the analytical analysis of Barenblatt
et al. (1963). To account for more complex fracture geometries we define Lc as an
effective length given by (Ma, 1997):
n
X
Ai
L2c =V (2.11)
l
i=1 i
where V is the matrix block volume, Ai is the area open to flow in the it h direction and
li is the distance from the open surface to a no-flow boundary. If we assume Swm as the
average saturation in the matrix, then we can write:
R Swm − Swmi
= ∗ (2.12)
R∞ Sw − Swmi
where Swmi is the initial water saturation in the matrix. Thus:
Swm = Swmi + (Sw∗ − Swmi ) · (1 − e−βt ) (2.13)
then from Eq. (2.12):
∂Swm
φm = T = βφm (Sw∗ − Swm ) (2.14)
∂t
This assumes that the transfer function is independent of the flowing saturation, as
long as Swf > 0. This is consistent with the assumption that the capillary pressure in the
low permeability matrix is much higher than in the fractures. The imbibition continues
until the capillary pressure reaches its equilibrium between matrix and fracture, when
Swf = 0 and Swm = Sw∗ . Thus we take:

βφm (Sw∗ − Swm ) Swf > 0,


T = (2.15)

0
 Swf = 0
The transfer function varies linearly with matrix saturation. We will call this the
21
linear transfer function. This linear transfer function represents the similar physics to
the conventional functions: transfer controlled solely by the matrix capillary pressure
with negligible fracture capillary pressure, so the results should be similar.
To check the validity of the approach, Di Donato et al. (2003) first single porosity
simulations on the same reservoir model with all properties the same as the SP E 10th
Comparative Solution project (Christie et al., 2001). As well as the fine grid model
60 × 220 × 85, they generated a series of coarser grids: 20 × 55 × 17, 20 × 55 × 85,
60 × 55 × 85. Permeability was uspcaled using geometric averaging. The results and run
time were compared with conventional grid-based simulation using Eclipse. The results
are identical to those obtained from Christie and Blunt’s work (2001). Both streamline
and grid-based methods gave very similar predictions of oil recovery for the same sized
grid. They then ran the streamline-based dual porosity model (see Chapter 3.1) for the
different grids using the linear model with β = 5 × 10−9 s−1 they obtained very similar
predictions of oil recovery for the same sized grid compared to conventional grid-based
simulation. Fig. (2.1) shows the oil production rate. For the same transfer function,
grid-based and streamline dual porosity models gave virtually identical results.
Figure 2.1: Comparisons of oil production rate for grid-based and streamline-based dual
porosity models for different grid sizes. In both cases a conventional transfer function
with the same shape factor was used. Both simulation methods give very similar results
for the same grid size (Di Donato et al., 2003).
22
Fig. (2.2) shows comparisons of the run times for the various simulations. All the
times are for a 3GB RAM, 2.4 GHz Pentium Xeon workstation. For both single and
dual porosity models the run time of the streamline code scales approximately linearly
with the number of grid blocks. The dual porosity model is slower because of the more
complex one-dimensional transport solver, in particular the iterative procedure for the
conventional model makes it significantly slower than the other models. The grid-based
code gives run times that are approximately proportional to the number of grid blocks
squared. This leads to run times that are one to two orders of magnitude slower for the
finest grids.
1.E+07
Linear model,streamline
1.E+06 Conventional model, streamline
CPU time (min)
Dual porosity model

1.E+05 Single porosity,streamline
Single porosity,Eclipse
1.E+04
1.E+03
1.E+02
1.E+01
1.E+00
1.E-01
1.E-01
1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
Number of grid blocks
Figure 2.2: Comparisons of run times as a function of grid block numbers for a simulation
time of 2,000 days (Di Donato et al., 2003).
23
The transfer rate constant β is a function of the fracture spacing, interfacial tension
and matrix permeability. It is also a strong function of wettability and viscosity contrast
through the introduction of mobilities in Eq. (4.8) (Birks, 1955). In mixed-wet systems,
the transfer rate can be 100 − 10, 000 times lower than that in similar strongly water-wet
media since the water mobility is low (Behbahani, 2005; Morrow, 2001).
It is important to input a proper rate constant β in fracture modeling, since β depends
on the matrix, fracture and fluid properties. It is different from the traditional approach
of using a shape factor, which is physically opaque, to represent fracture geometry since
it does not relate easily to a transfer rate (Gilman, 1983).
The matrix relative permeabilities and capillary pressure can be adjusted to account for
wettability and mobility in a conventional model. However, the impact of such changes on
recovery is obscured by the unnecessarily complex non-linearity of the resultant transfer
function that simply adds computational difficulty while failing to represent the correct
physics (Di Donato, 2007).
2.3.2 Gravity drainage
The displacement of oil by gas under gravity in the presence of water is an important
recovery process in oil reservoirs (Ding, 2005). Compared with water flooding, the gas
gravity drainage process can be very slow, and the ultimate recovery is very high. Sahni
et al. (1998) used CT scanning to image saturations directly in a long vertical column.
Fig. (1.3) shows the behaviour at different times for air/water gravity drainage. There
are two remarkable features to the displacement pattern: one is that gas rapidly reaches
the equilibrium position H beyond which the column is saturated with the wetting phase;
another is that the saturation profile is almost uniform with distance above the distance
H, but this saturation slowly decreases over time. The gas overcomes the capillary
pressure and keeps pushing the oil to the bottom of the core until oil and gas are in
capillary/gravity equilibrium.
In our streamline model gas injection is also considered. The gas is assumed as in-
24
compressible and water is everywhere at its irreducible saturation. Each matrix block is
treated separately and there is no capillary continuity between blocks allowed (Por, 1989;
Horie, 1990). In the governing transport Eq.(2.5) and (2.6) we simply substitute gas for
the water phase.
As for water flooding, we base our transfer function on analytical solutions to the flow
equations in an idealized geometry - in this case vertical gravity drainage of oil in the
presence of gas. The ratio of gravitational to capillary forces in each matrix block, r, is
defined as:
s
L ∆ρgL Km
r= = (2.16)
H σgo J ∗ φm
where L is the block height, H is the amount of capillary rise of oil in the presence of gas,
and J ∗ is the dimensionless entry pressure for gas invasion into oil. The density difference
∆ρ = ρo − ρg .
For r ≤ 1 gas cannot enter the block and there is no gravity drainage. For r 1,
gravitational forces dominate and we can write a solution for the average recovery at late
times due to Hagoort (1980):
1
R (αt) a−1
=1− (2.17)
R∞ Sg∗
where the rate constant α is:
aa Km Krom
max
∆ρg
α= (a−1)
(2.18)
(a − 1) φµo L
The exponent a (it is assumed that a > 1) is defined such that at low oil saturation
the oil relative permeability is:
max
kro = krom (Som − Sorgm )a (2.19)
where Sorgm is the residual oil saturation in the presence of gas. Note that we define
25
the ultimate recovery by finding the maximum gas saturation Sg∗ at infinite time from
capillary/gravity equilibrium, not simply by assuming that the oil saturation eventually
reaches Sorgm everywhere:
Z L
Sg∗ = 1 − Swmi − −1
Pcgo (∆ρgh)dh (2.20)
0
Then following the same approach as for capillary-controlled imbibition:
R Sgm
= ∗ (2.21)
R∞ Sg
1
Sgm = Sg∗ − (αt)− a−1 (2.22)
and hence writing the transfer as a function of gas saturation:
∂Sgm αφm ∗
φm =T = (Sg − Sgm )a (2.23)
∂t a−1
Then we define:

 αφm (S ∗ − S a

gm ) Sgf > 0,

a−1 g
T = (2.24)

0
 Sgf = 0
Note that while the original expression for gas saturation, Eq. (2.21), diverges at t = 0,
the transfer function itself is always well behaved. This is one reason why the transfer
function is written as a function of saturation and not of time explicitly.
2.3.3 Extensions to the approach
It is possible to extend this approach to study other physical processes, such as transfer
due to fluid expansion (implicitly included in the traditional formulations), displacement
due to a combination of capillary and gravitational forces and interaction between blocks
due to capillary continuity (Barenblatt, 1863; Warren, 1963; Gilman, 1983; Por, 1989).
26
We treat each matrix block as independent source term. Some of these extensions will
be discussed in Chapter 4.
2.3.4 Multi-rate transfer function
Realistically, the transfer function is not a constant, even within a single grid block,
because of different fracture sets with different spacing, or sub-grid block variability in
wettability and matrix permeability (Barenblatt, 1963).
Fig. (2.3) illustrates the complex interaction between fracture and matrix of a single
grid block in field-scale simulation; a single grid block tens to hundreds of meters in extent
would include many fractures. Capturing the transfer from these different fracture sets
is the key to solving this kind of problem. Di Donato et al. (2007) derived a multi-rate
transfer function to account for heterogeneity in a single grid block.
ｆｒａｃｔｕｒｅ
ｃｏｒｒｉｄｏｒ
Figure 2.3: A single grid block in a field-scale simulation, there are different spacing
fractures in one grid block, so it is not accurate to set the fracture property by one value.
The multi-rate transfer model will describe the complexity physically by different transfer
rates within a single block (Matthai, 2005).
2.3.5 Multi-rate model
The single-rate model assumes that there is just a single set of fracture and matrix
properties within each grid block and the matrix permeability may have small-scale vari-
ations (Daly, 2004). Lumping all the transfer between fracture and matrix in a single
27
rate may therefore be inaccurate, and, as we show later, will tend to over-predict oil re-
covery at late time, since a single rate model will not adequately account for the regions
of the matrix that are in poor communication with the fractures. As an example of this
complexity, Fig. (2.4) illustrates intersecting fracture sets in a limestone outcrop near
the Bristol Channel in England (Belayneh, 2004).
ＢＥＤ１ＢＥＤ２
ａｂ
ＢＥＤ３ＢＥＤ４
ｃｄ
Figure 2.4: Intersection fracture sets in a limestone and shale outcrop in the Bristol
channel (Belayneh, 2004). The fracture/crack spacing is between 0.3 and 1m. We use a
multi-rate model to accommodate transfer from different fracture sets.
In dual porosity simulation, a single grid block tens to hundreds of meters in extent
would include many fractures. A single rate is insufficient to capture the transfer from
these different fracture sets. Although it is more realistic to capture the geometry explic-
itly, there is no proper computing resource to cope with this complicated problem.
To account for sub-grid-block heterogeneity we follow the approach of Ponting (2004)
and propose a multi-rate model where the matrix is assumed to be composed of N
domains each with different transfer rates. The transfer into each of the domains is
handled separately:
N
X N
X N
X
T = Tk ; φm = φmk ; φm Swm = φmk Swmk ; (2.25)
k=1 k=1 k=1
28

∗

βk φmk (Swk
 − Swmk ) Swf > 0,
Tk = (2.26)

0
 Swf = 0
∂Swmk Tk
= (2.27)
∂t φmk
Mathematically, Eq. (2.24)-(2.26), with two or three domains are similar to models
proposed by other authors to describe multiple transfer rates in a single block, although
the physical interpretation is very different (Terez, 1999; Civan, 2001).
The same equations can be used for gas injection with water substituted by gas and
the transfer function is:

 αk φmk (Sgk
∗


ak −1
− Sgmk ) Swf > 0,
Tk = (2.28)

0
 Sgf = 0
2.3.6 Coupling with the fracture fractional flow
The multi-rate model leads to the regions of the matrix in contact with fractures having
the largest surface area and consequently the highest transfer rates being drained first,
while the regions with the lowest transfer rates are drained last. The lowest transfer rates
may come from the largest aperture fractures. Hence, during waterflooding the largest
fractures become saturated by water first and the smaller aperture fractures, with the
largest surface area, are only contacted by water later. Thus the multi-rate model does
not capture the proper sequence of displacement within a grid block.
We propose another model where in the upscaled fracture fractional flow curve different
average saturation regions represent the flooding of different fracture sets. Corresponding
to each fracture set would be transfer into the matrix - in this scenario the smaller matrix
blocks, with the highest transfer rates, may be recovered last, since imbibition will not
start until their average fracture saturation was sufficiently high. Fig. (2.5) illustrates
this model schematically. This formulation allows for a more realistic multi-rate model,
29
where which portions of the matrix are recovered is controlled by the average fracture
saturation that in turn represents which fracture sets are flooded.
ｆｗｆ
Ｓｗｆ
Ｓｗｆ０Ｓｗｆ１Ｓｗｆ２Ｓｗｆ３
ｌａｒｇｅ　ｍｅｄｉｕｍ　ｓｍａｌｌ　

ｆｒａｃｔｕｒｅｓｆｒａｃｔｕｒｅｓｆｒａｃｔｕｒｅｓ
Figure 2.5: Schematic of the conceptual model used to develop our dual porosity model.
The left hand figure would represent a single grid block in a field-scale simulation. The
average fracture fractional flow (right-hand figure) has contributions from all fracture
sets - the larger aperture fractures are filled first at low average saturation with smaller
aperture fractures filled at higher saturations. Associated with each fracture set is a
different matrix/fracture transfer rate. We represent the complex interaction between
fracture and matrix by a series of linear functions representing transfer from different
fracture sets at different rates (Di Donato et al., 2007)
Mathematically, the fractional flow curve is divided into N sections Swf k (by defin-
ition Swf N = 1 and Swf 0 = 0, assuming no residual saturation in the fractures) with
corresponding matrix porosities and transfer rates. The expressions are only a simple ex-
tension of Eq. (2.24)-(2.26) above, but now have the structure to represent very complex
interactions between fracture and matrix. We also allow for reduced transfer if a frac-
ture set is not completely saturated, assuming now that locally fractures are either fully
saturated or dry and that fractures at the small scale are rarely partially saturated-the
single-rate version of this model then reduces to the empirical model of Kazemi et al.
(1992).
30

∗




βk φmk (Swk − Swmk ) Swf ≥ Swf k

 " #

Swf −Swf k−1 ∗
Tk = βk φmk Swf k −Swf k−1
(Swk − Swmi ) + Swmi − Swmk Swf k > Swf > Swf k−1 (2.29)






0
 Swf ≤ Swf k−1
∂Swmk Tk
= (2.30)
∂t φmk
For gravity drainage, Eq. (2.29) becomes:


 αk φmk (Sgk∗
− Sgmk )ak



 ak −1
Sgf ≥ Sgf k

 !ak

Sgf −Sgf k−1 Sgf k −Sgf k−1
Tk = αakkφ−1
mk
· Sgf k −Sgf k−1
∗
Sgk − S
Sgf −Sgf k−1 gmk
Sgf k > Sgf > Sgf k−1 (2.31)






0
 Sgf ≤ Sgf k−1
The numerical implementation of these transfer functions in the one-dimensional trans-

port equations along streamlines is described in Appendix B.
31
Chapter 3
Oil field case studies
Three different cases are studied in this chapter: 1. A non-fractured sandstone oil field
with synthetic fractures; 2. A Chinese fractured dolomite oil field; and 3. The Clair field
with high matrix permeability.
3.1 Synthetic fractured reservoir simulation
3.1.1 Reservoir description and computing comparison
The reservoir model we use is derived from the 10th SPE Comparative Solution Project
(Christie, 2001). The model is a two-phase (oil and water) model with no dipping or
faults. The dimensions of the reservoir are given by 366m × 670m × 52m; the number
of grid cells used for the simulations was 1, 122, 000, given by 60 × 220 × 85 in the x, y
and z directions respectively. It is a non-fractured sandstone fluvial reservoir based on
a North Sea oilfield with high permeability meandering and channels surrounded by low
permeability shale.
The original data set includes a set of heterogeneous permeability data, a set of het-
erogeneous porosity data, compressible two-phase PVT data and relative permeability
data. We turn the original model into quasi-incompressible data set by changing the
compressibility for both rock and fluid to nearly zero to satisfy the needs of testing the
32
new streamline-based code. This adjustment has no big impact on the oil recovery pre-
dictions, since the field is operated above the bubble point.
Fig. (3.1) (a) shows the porosity for the whole model, and Fig. (3.1) (b) shows part of
the Upper Ness sequence, with the channels clearly visible.
Figure 3.1: The porosity distribution of North Sea reservoir model: (a) the porosity for
the whole model; (b) part of the Upper Ness sequence well distribution (Christie, 2001).
We made an artificial dual porosity model to test the streamline-based research code
by doing the following: (1) use a constant porosity for both matrix and fracture; (2) the
original heterogeneous permeability data is treated as defining the fracture permeability
distribution, with the exception that we set a minimum fracture permeability of 1 mD,
while a single uniform permeability of 1 mD is assigned to the matrix; (3) take the original
relative permeability curve as the one in the matrix. For the fracture, we assume a linear
fractional flow for water, even if the oil and water viscosities are not equal. From this
33
assumption, we derived the following expressions to calculate the relative permeabilities
in the fracture:
" #
1 − Swf µo
krwf = 1+ · (3.1)
Swf µw
" #
Swf µw
krof = 1+ · (3.2)
1 − Swf µo
The capillary pressure in the fracture impact is ignored by setting it to zero. The
reference pressure is 41.3 × 106 P a at a depth of 3657.6 m. For all models, one injector
is located at the center of the model; four producers are at the corners of the model - see
Fig. (3.1). All wells are vertical and perforated throughout the formation. In addition,
we use a constant well bore rate for the injector and a constant well bore pressure for
the producers as the inner boundary conditions as well as no flow on the outer boundary.
The parameters used in the simulations are listed in Table 3.1.
Table 3.1: The parameters used in dual porosity simulation

φm = 0.2 φf =0.02
Km = 1.0 mD Kf = 1 − 20, 000 mD
Swim = Sorm = 0.2 Swif = Sorf = 0.0
krom (Swi ) = krwm (Sor ) = 1.0 krof (Swi ) = 1.0
−3
µw = 0.3 × 10 cp
µo = 3 × 10−3 cp
ρw = 1026 kg.m−3
ρo = 1026 kg.m−3
Bw = 1.01
Q = 300 m3 .day −1
Bottom Hole Pressure =27.6 × 106 P a
3.1.2 Transfer rates and different transfer functions comparison
As described earlier, this model was tested by Di Donato et al.(2003) to compare

the difference between streamline-based and grid-based dual porosity simulators. They
designed several different cases to compare the difference between a streamline-based
34
model and a conventional grid-based model and found that the streamline-based model
can easily cope with incompressible displacement in a highly heterogeneous reservoir,
and streamline-based simulation gave almost identical results to commercial grid-based
simulation for single and dual porosity models.
Based on the work of Di Donato et al. (2003, 2004), we design two cases to study
which are transfer rate sensitivity studies. This research give us a more comprehensive
understanding of fractured reservoir recovery mechanisms, especially where the multi-rate
model represents the complexity in one grid block in a field scale model.
3.1.3 Permeability-dependent transfer rates
Mathematical model used to assign the transfer rate
The transfer rate is a parameter which is related to the geometry and fluid properties -
see Eq. (2.8). We have tested the one-, two- and three-rate models for the field. We also
assigned different rates to each grid block dependent on the fracture spacing and matrix
permeability. The aim of this section is to explore the impact of using a multi-rate model.
We use Eq. (2.8) to find the average transfer rate for the whole reservoir βav . β is
√
proportional to Km and 1/L2c . Kf is proportional to the fracture spacing Lc , so we
used:
r 2
Km K f
β = βav (3.3)
Kma Kf a
to assign the transfer rate for each grid block where Ka is the average permeability. The
range of the transfer rate is from 1.7 × 10−13 s−1 ∼ 7.0 × 10−5 s−1 . βav is the average
transfer rate for the whole reservoir calculated by Eq. (2.8) to be 5×10−9 s−1 . Kma = Km
while Kf varies from 1 mD to 2000 mD and Kf a is 170 mD.
35
Complement the equation into the streamline-base code
The work to code the permeability dependent transfer rate equation into the streamline-
based model is composed of three parts:
• Go through the code and understand the main structures of the program, draw a
whole picture of the structure.
• Select the module related to the transfer rate calculation. Make a plan to modify
the module.
• Code the relevant equation into the module and set up different options for the
single, multiple, permeability dependant transfer rate. Test the code and make
sure the code work normally. After completion the code I run a few cases of single
transfer rate, multi-rate to test the validation of the model and they work all normal.
Results for the permeability dependant transfer rate
We compared the results using a single average transfer rate βav in each block to that
obtained using Eq. (3.3) on a block-by-block basis. The results, shown in Fig. (3.2), show
that assuming a fixed βav for all blocks gives similar recoveries. as assuming a variable β.
In all subsequent work we will assume β is independent of fracture permeability. However,
we will allow multiple rates in each block.
36
350
300
250 Permeability dependent β

Oil production (m /day)
Average βav
200
3
150
100
50
0
0 2000 4000 6000 8000
Time (days)
Figure 3.2: Comparison of oil rate using a fixed βav for all blocks and a β that varies from
block to block using Eq. (3.3), dependant on fracture permeability.The results are similar.
In all the subsequent simulations, we assume β is independent of fracture permeability.
3.1.4 Results for a single rate model
Fig. (3.3) shows horizontal slices of the matrix and fracture saturations from the single-
rate model with different transfer rates (5 × 10−8 s−1 , 5 × 10−9 s−1 , 5 × 10−10 s−1 ). The
water injected into the fractures moves along the high permeability channels. Where
there is water in the fractures, there is transfer to the matrix. The transfer rate decreases
as the water saturation in matrix contacted with fractures. As the transfer rate decreases
there is less and less transfer to the matrix and the recovery is lower.
37
Figure 3.3: Fracture (left column) and matrix (right column) saturation after 4,500 days
(3.89×108 s) at the 60th layer of the reservoir for the transfer rate:(a) β=5×10−8 s−1 ;
(b) β=5×10−9 s−1 ; and (c) β=5×10−10 s−1 . Notice that the water rapidly saturates the
high permeability fracture regions, while the saturation in the matrix is controlled by the
transfer rate.
The time taken to allow significant recovery in the matrix due to imbibition is given by
Eq. (2.13) as approximately 1/β. If the injection rate of water is Q and the gross rock
volume of the reservoir is V , then the time taken to inject one fracture pore volume of
V φf
water is Q
. Then, we define a dimensionless ratio of the time for significant imbibition
to the time of injecting a fracture pore volume:
Q
r= (3.4)
βφf V
The recovery is favorable for r of order 1 or less, since there is sufficient time for
38
the water to be imbibed into the matrix before water breakthrough. The recovery is
unfavorable when r is greater than 1, since water breaks through long before there is
appreciable water imbibed into the matrix. Fig. (3.4) shows a dramatic reduction in
recovery for low values of r. Essentially only oil in the fractures is recovered for r 1 .
0.4
-8 -1
linear model β = 5x10 s , r = 0.27
-9 -1
linear model β = 5x10 s , r = 2.7
0.3 -10 -1
linear model β = 5x10 s , r = 27
Oil recovery
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
PVI
Figure 3.4: Recovery curves for different values of the transfer rate β. We define r as
the dimensionless ratio of the time to inject one fracture pore volume of water. For r
of order 1 or lower, recoveries are favorable. For higher values of r recoveries are poor,
since water breaks through well before significant quantities of oil are recovered from the
matrix (Di Donato, 2003).
39
3.1.5 Results for multi-rate models
We also ran a two-rate model by choosing a value of φm1 = φm2 = φm /2 = 0.1 and
β1 = 9 × 10−9 s−1 , β2 = 10−9 s−1 to investigate the impact on oil production rate.
We used a three-rate model with the value of φm1 = φm2 = φm3 = φm /3 = 0.067
and β1 = 9 × 10−9 s−1 , β2 = 5 × 10−9 s−1 , β3 = 10−9 s−1 to account for the geological
complexity in one grid block. Also single-rate models are run as a comparison with rates
chosen to span the maximum and minimum rates encountered in the multi-rate cases;
β = 5 × 10−9 s−1 is the arithmetically averaged rate for both the two- and three-rate
models.
Fig. ( 3.5) and (3.6) show predicted oil rates using the multi-rate models. It is initially
assumed that the initial and residual saturations are the same for each matrix domain
and hence from Eq. (2.14) and (2.15) in regions contacted by water the total transfer is:
N
X N
X
T = Tk = (1 − Somr − Swmi ) βk φmk = (1 − Somr − Swmi )φm βav (3.5)
k=1 k=1
T is equivalent to a single-rate model with a constant transfer rate βav . This is obvious
in Fig. ( 3.5), where the oil production rate at early time is the same for the multi-rate
models and a single-rate model with βav = 5 × 10−9 s−1 . However, the multi-rate models
predict lower recovery at later times, since the single rate model allows more imbibition
in matrix and the multi-rate models have some areas will never be imbibed over the
timescales of production.
In a time scale 1/βk , the k th matrix domain is filled with water. It means that further
recovery only comes from regions with a slower transfer rate. At late times, assuming
that all the βk values are very different from each other and βN is the lowest rate, then
only the N th domain contributes to recovery and the behaviour is similar to a single
rate model with β = (φmN /φm )βN . Note that this is a lower rate than given by a
corresponding single-rate model with the lowest value of β in the multi-rate formulation.
This is evident in Fig. ( 3.5), where at the latest times the multi-rate model predicts a
40
recovery rate that is slightly lower than using a single-rate model with β=βmin = βN =2 .
300
single_beta_av
single_beta_max
single_beta_min
200 multi_two rates
Oil rate (m3 /day)
100
0
0 2000 4000 6000 8000
Time (days)
Figure 3.5: Oil production rate as a function of time for single and two-rate models,
the single rate model has the arithmetically-averaged transfer rate that gives the same
early-time behaviour as the two-rate model. Also shown are single-rate models with the
larger and smaller rates in the two-rate model.
Fig. (3.6) shows the two- and three-rate cases oil rate results. The recovery plots are
very similar, which indicates that perhaps only a two-rate formulation is sufficient to
capture the main characteristics of the oil rate in this case.
The implication of this analysis is that a dual porosity model using a transfer rate
based on, for instance, an average fracture spacing, or tuned to initial production history,
will tend to over-predict recovery at intermediate times, since it does not adequately
account for the slow transfer encountered from portions of the matrix that are in poor
communication with the fracture network.
41
300
single_beta_av
200
Oil rate (m 3/day)
multi_two rates
multi_three rates
100
0
0 2000 4000 6000 8000
Time (days)
Figure 3.6: Comparisons of oil production rates for single-, two- and three-rate models,
single rate model gives high oil production rate.
3.1.6 Conclusions
We have extended the streamline-based dual porosity model of Di Donato et al. (2003,
2004) to account for multi-rate transfer between matrix/fracture following the approach
of Ponting (2004). We put multi-rate transfer in a single grid block to account for sub-
block scale heterogeneities in the matrix properties and fracture spacing, since it is not
realistic to use a single fracture property in a single grid block which contains many
different fracture sets. A single-rate model will significantly over-estimate recovery at
late time by using an equivalent average rate that matches the early-time behaviour,
since it will not capture the long time scales for recovery from regions of the matrix in
poor communication with the fractures.
All the simulations presented were performed on a standard Windows PC; it is not
possible to do this using a conventional dual porosity simulator; it implies that streamline-
based approach can cope with million-cell dual porosity models using standard computer
resources.
42
3.2 Liu7 oil field simulation
We now study a Chinese oil field to assess the efficiency of different injection processes.
We are going to use the field data to test the predictions of the multi-rate transfer
functions.
3.2.1 Reservoir description
Liu7 is a small oil field located in a remote, mountainous region of Northwest China
which was discovered in 1980. The estimated initial oil in place is about 22, 000, 000 m3 .
The oil accumulation is located between 4000 m and 5000 m below the ground surface;
there was no connecting aquifer or gas cap found.
There are 31 major partially seal faults crossing the reservoir. The reservoir consists
of layers of extensively fractured dolomite which includes three main fracture sets: (1)
vertical fractures within layers with a density of 0.2 − 45 m−1 (average 4.1 m−1 ); (2)
oblique fractures with density 0.2 − 5.7 m−1 (average 1.3 m−1) ); (3) vertical fractures
between layers of density 0.1 − 3 m−1 (average also 1.3 m−1 ).
The fractures have typical apertures in the range 15 − 200 µm and provide good con-
nectivity in all directions. The overall fracture porosity is approximately 0.4% . The
matrix porosity lies in the range 10 − 18% with permeabilities between 1 and 10 mD.
The data set was taken from a CMG format. The original number of grid blocks was
58 × 37 × 20. It was run using a dual porosity model. The data sets include vertical
thickness; matrix and fracture permeability and porosity distribution. The simulations
were run on a downscaled Cartesian grid with 116 × 74 × 126 blocks (102, 000 active cells)
of size 40 m by 39 m with a variable height of average 4 m. The downscaling method is
described in Appendix C.
This reservoir is a tilted reservoir where an artificial gas cap can be maintained. As
a consequence, gas gravity drainage may be a promising recovery method. Fig. ( 3.7)
shows the Liu7 oil field vertical permeability distribution.
43
Permeability (mD)
0 10 3 5x10 3
604 m
4640 m
Figure 3.7: XZ sections of the fracture permeability field, the permeability increases with
reservoir depth.
3.2.2 Geological characteristics
After investigating three-dimensional seismic data, core, well test, and productivity test
data, we interpret the main characters of the oil reservoir, in terms of rock type, porosity,
compressibility, permeability, and wettability as described below.
Liu7 is a super-deep fractured reservoir located in the Cretaceous. It is composed of
conglomerate, sandstone, shale and dolomite. Dolomite is the main rock which contains
most of the fractures. The thickness of a single layer is less than 3 m.
The porosity in matrix varies between 10−18%; the average fracture porosity is around
0.4%.
The rock effective compressibility is 8.29 × 10−10 P a−1 , water compressibility is 5.76 ×
10−10 P a−1 and oil compressibility is 6.55 × 10−9 P a−1 . We revise the compressibility to
nearly incompressible. It will not impact the calculation results significantly because the
reservoir will be produced above the bubble point pressure.
The permeability in matrix is from 1 to 10 mD; the fracture permeability is between 2
and 4000 mD. The fracture permeability increases with reservoir depth. Because of the
large contrast between fracture and matrix permeabilities in many parts of the field, it is
44
assumed that a dual porosity model is appropriate. Fig. (3.7) shows two cross-sections
of the fracture permeability distribution after downscaling. We can see that the reservoir
is highly heterogeneous.
The initial water saturation in the matrix is 0.544. It is a mixed-wet oil reservoir,
which means that only a fraction of the mobile oil can be recovered by imbibition alone.
Average initial pressure is 54 × 106 P a; initial formation volume factor is 1.3 rm3 /sm3 ;
the initial temperature is 120 ◦ C. It is a high pressure, high temperature oil field.
Fig. (3.8) gives the imbibition capillary pressure curve which shows that the capillary
pressure is very high in such a low permeability matrix. The maximum water-flooding
recovery is about 0.22. The rock wettability is mixed-wet. We use this capillary pressure
curve for the matrix to compute β and set the capillary pressure in the fracture as zero
since fractures have a relatively high permeability.
12
9
(Bar)
6
Capillary pressure
End of imbibition
3
-3
0.4 0.5 0.6 0.7 0.8
Sw
Figure 3.8: Capillary pressure vs. water saturation curve in the matrix, which presents
the imbibition behaviour in the matrix. Capillary pressure is one of the key factors to
determine the fractured reservoir recovery.
We ignore the efforts of gravity on the matrix recovery, we discuss in Chapter 4 how to
incorporate this effect.
45
3.2.3 Fluid characteristics
No water/oil contact has been found. The gas/oil ratio is 211 m3 /m3 ; bubble point
pressure is 25 × 106 P a. The difference between initial reservoir pressure and the bubble
point pressure is huge. So what we can do is to use the natural energy of the reservoir to
develop the oil reservoir then just maintain the oil field pressure above the bubble point
pressure.
The oil density at the surface is 817.4 − 898.2 kg/m3 ; the average value is 852.8 kg/m3 .
The stratum oil density is 680.7 − 687.5 kg/m3 , and the water density in the reservoir is
1032 kg/m3 , water compressibility is 5.76 × 10−10 P a−1 .
There are four oil and water relative permeability curves that were measured for the
matrix. The irreducible water saturation averages about 0.544 and the average residual
oil saturation is 0.24. So the mobile oil is only about 0.22. Fig. (3.9) gives the average
relative permeability curve in the matrix.
1.2
1
Krw
Kro
0.8
Kr
0.6
Relative permeability at the end of imbibition
0.4
0.2
0
0.4 0.5 0.6 0.7 0.8
Sw
Figure 3.9: Oil and water relative permeability curves in the matrix. We assume linear
relative permeabilities in the fractures.
Fig. (3.7) illustrates a layered system with wide variations in fracture permeability
both within and between layers with a tendency that the bottom of the field has higher
permeability. The fracture permeability varies from over 4000 mD to a minimum of 2 mD
46
(the average matrix permeability). The model uses a Cartesian grid with 116 × 74 × 126
blocks of size 40 m by 39 m with a variable height of average 4 m. The average x direction
permeability (in the direction of the main fracture set) is 440 mD, the average y direction
permeability is 150 mD, while the average z direction (vertical) permeability is 27 mD -
this is a value averaged over several layers.
We design two different cases to study possible recovery processes. For water flood
simulations, we put 5 vertical injectors with a constant rate of 2, 000 m3 /day each and 13
vertical producers with a constant bottom-hole pressure of 17 × 106 P a, as are currently
used in the field. For gas injection we put 2 vertical injectors near the top of the field
with a constant rate of 5, 000 reservoir m3 /day each and 4 horizontal producers near the
bottom of the field with a pressure of 17 × 106 P a. The well placements for these two
cases are shown in Fig. (3.10).
47
P11 P3 P13
I2 I3 I1 P12 P7
P8 P2 P9 P5
P1 I4 P6 I5 P4
P10
(a)
I2 I1
P3 P1 P4
P2
(b)
(b)
2 100 1000 4000

Kx (mD)
Figure 3.10: Sections of the Liu7 oil field model showing the fracture permeability in the
x direction (along the direction of the main fracture set). (a) The well placement for
water-flooding. Water injectors are shown in blue and producers in red. (b) The well
placement for gas injection. The vertical gas injectors are shown in green. The horizontal
producers (red) are placed at the bottom of the field.
3.2.4 Parameters for imbibition modeling
We now discuss how to assign parameters in the transfer functions. ∆ρgl is small com-
pared to a typical capillary pressure. Since the gravity is neglected, because the average
block height is 4 m. The transfer rate can be calculated directly from the experiment
data, so the following equation was used to calculate the value (Di Donato, 2007):
s
3λ∗om λ∗wm σow J ∗ Km
βwo = (3.6)
λtm H2 φm
where J ∗ is calculated by:
r r
dPcm φm dJ(Swm ) φm
= σ =− σow J ∗ (3.7)
dSwm Swm =Sw∗
Km dSwm Swm =Sw∗
Km
48
The value of Sw∗ = 0.6217, Pc (Sw∗ ) = 0, J ∗ = 7.4495, krw
∗ ∗
= 0.2, kro = 0.4 - see Fig.
(3.8)-(3.9), other values come from Table 3.2. The transfer rate is 3.73 × 10−4 s−1 . This
corresponds to a time scale of only around 0.744 hour. It is a very favourable imbibition
rate giving essentially instantaneous displacement. However, the experimental data shows
a very steep capillary pressure curve and well-connected flow with almost linear relative
permeabilities. These relative permeabilities are likely water-flooding a cleaned water-wet
core and are representative of counter-current imbibition in a mix-wet system.
In this case a more representative estimate of transfer rate can be obtained using a
correlation that does not require a measurement of relative permeability or capillary
pressure. In such cases we can define a transfer rate based on the correlation proposed
by Zhang et al. (1996) that works well for cases where the oil and water viscosities are
not very different from each other:
s
Km σ
β=b √ (3.8)
φm L2c µo µw
where the parameter b depends on the reservoir wettability and varies from 0.05 for
strongly water-wet systems (Zhang, 1996) to 10−5 and lower for mixed-wet systems
(Morrow, 2001). Measurements of waterflood capillary pressure for this field indicate
an Amott water wettability index of approximately 0.5 - the amount of water displaced
by spontaneous imbibition is similar to that displaced by subsequent forced water injec-
tion. Experimentally (albeit in sandstones) imbibition for such systems is approximately
100 times slower than for water-wet media (Morrow, 2001, Behbahani, 2005). Hence we
take b = 5 × 10−4 . Then in Eq. (3.8) and using the parameters listed in Table 3.1, we
find that β = 2.6 × 10−9 /L2c s−1 with Lc measured in m. Using Eq.( (2.11) where l is
half the fracture spacing and assuming that we have three independent fracture sets with
the average density, we find 1/L2c = 40.4 m−2 and β ≈ 1.1 × 10−7 s−1 = 1/108 days−1 .
This means that imbibition will take around 108 days to give significant recovery. This
timescale is short relative to the typical periods for water injection and implies that im-
bibition will be an effective displacement process. However, the analysis relies on using a
49
single effective rate, which is high because the average fracture density is large. A more
conventional analysis using a shape factor would come to the same conclusion. In reality
though, there is a huge variation in the fracture spacing from 2.5 cm to 5 m within each
simulation grid block that should be reflected in the model.
To capture the observed range in fracture spacing we consider two multi-rate models.
In both cases we constrain the parameters such that φm βav = N
P
k=1 φmk βk where βav is
the rate constant for the single-rate case, βav = 1.1×10−7 s−1 . Then the initial imbibition
rate:
N
X N
X
T = Tk = (1 − Somr − Swmi ) βk φmk = (1 − Somr − Swmi )φm βav (3.9)
k=1 k=1
is the same as for the single-rate model. This allows us to compare the impact of the
multi-rate model on recovery - by definition recovery will be similar to begin with, but, as
we demonstrate later, will show considerable variations at later times. The first, two-rate
model has one third of each grid block containing matrix with fractures whose spacing
is around 30 times larger than average - representing a fracture spacing close to the
maximum observed value of 5 m. We then have φm1 = 0.043, β1 = 1.1×10−10 s−1 (1,000
times less than βav ), φm2 = 0.087 and β2 = 1.6 × 10−7 s−1 (chosen to obey Eq. (3.9).
The second, three-rate model is similar with one third of the matrix contacted by fractures
with a spacing of 5 m and another third having a fracture spacing of approximately 1
m: φm1 = 0.043, β1 = 1.1 × 10−10 s−1 , φm2 = 0.043, β1 = 3.6 × 10−9 (30 times less
than βav ), φm3 = 0.043 and β3 = 3.1 × 10−7 s−1 . For a model coupled to the fracture
fractional flow, we use the same three-rate model with Swf 1 = 1/3 and Swf 2 = 2/3.
50
Table 3.2: Parameters used in the simulations
φf = 0.004 φm =0.13
−15 2 max
Km = 2 × 10 m a = 2, 4; Krom =1
∗
Swmi = 0.544 Sw = Swmi + 0.123 = 0.657
µw = 0.3 × 10−3 P a.s µo = 3 × 10−3 P a.s
µg = 2 × 10−5 P a.s ρw = 1026 Kg.m−3
ρo = 850 Kg.m−3 ρg = 250 Kg.m−3
" #−1 " #−1
1−Swf µo Swf µw
krwf = 1 + Swf
· µw
Krof = 1 + 1−Swf
· µo
σow = 0.03 N.m−1 σgo = 0.001 N.m−1

Somr = 0.23 Sorgm = 0
L = 2, 4 m J ∗ = 0.3
To reduce our uncertainty in the assignment of transfer rates we need to characterize the
variability in fracture spacing accurately. This could be done through a careful analysis
of fracture patterns (see, for instance, Daly, 2004). A more rigorous approach would be to
simulate displacement through realistic fracture geometry at the grid block scale, where
all the fractures are explicitly modeled. From this the effective fracture fractional flow
could be computed and rate constants for transfer determined to match the observed
average recovery in the discrete fracture model. While the tools for such an upscaling
approach have been developed (Di Donato, 2007) we have not performed this analysis
for this study. Mathematically we would be reproducing average recovery by a series of
linear functions, which should be a reliable and robust approach. In this work we use
plausible values of parameters based on experimental data and the results illustrate the
uncertainty in our performance predictions from using different models.
3.2.5 Parameters for gravity drainage
For gravity drainage we only consider a single-rate model. The transfer rate is less
sensitive to variations in grid block size than in imbibition - α is independent of areal
fracture spacing and only scales as 1/L2 as against 1/L for β. Hence consideration of
multi-rate models has less impact than in imbibition. The base case has the height of the
matrix block, L = 4 m, corresponding to the height of each grid block, and the oil relative
51
permeability exponent a = 2 - this corresponds to approximately to experiment - see Fig.
(3.9). Using Eq. (2.15) and the parameters listed in Table 3.1, we find r = 9.73 and so
gravitational forces dominate over capillary effects. The rate constant α in Eq. (2.17) is
3 × 10−8 = 1 year−1 . We ran two other cases for comparison. Since there is layering
within each grid block we consider a case with L = 2m (r = 4.86 and α = 6 × 10−8 s−1 ).
Also we assess the impact of oil relative permeability by running a case with a = 4 and
L = 4m (r = 9.73 and α = 2.8 × 10−7 s−1 ).
3.2.6 Results
Fig. (3.11) shows horizontal slices through the field showing the fracture and matrix
saturations after 1, 800 days of water-flooding for the single-rate model, the three-rate
model and for a water-wet case. The water-wet system allows more imbibition having
Sw∗ = 1 − Somr = 0.77 and b = 0.05 with a single rate of β = 1.1 × 10−5 s−1 . In all cases
water breaks through very rapidly as it channels through the fractures and the water
cut is over 90% beyond 2, 000 days. The volume swept by the water in the fractures is
controlled by the reservoir heterogeneity. Where there is water in the fracture, imbibition
can occur and oil is recovered from the matrix.
52
(a)
(b)
(c)
0 0.2 0.4 0.6 0.8 1.0

　Sw
Figure 3.11: Horizontal slices through the reservoir showing the fracture (left) and matrix
(right) saturations after 1,800 days of water-flooding. The top figure (a) shows the single-
rate model, the middle figure (b) the three-rate model and the bottom figure, (c) a
water-wet case.
Fig. (3.12) shows the field oil rate from all wells for the one-, two- and three-rate
models. The single-rate model gives higher recoveries since imbibition happens instantly.
It takes of order 108 days, and regions of the reservoir contacted by water in the fractures
see recovery from the associated matrix.
For the two and three-rate models one third of the matrix has an imbibition time 1, 000
times larger, of around 140, 000 days (400 years) and so effectively this portion of the
matrix is never recovered. This result in lower matrix water saturations - see Fig. (3.10).
In the three-rate model the imbibition time for the second region is approximately 4, 200
days, which is similar to the overall time for the simulation, allowing some of this section
of the matrix to be recovered. This still gives a slightly lower recovery than the two-rate
model, where recovery is rapid in two-thirds of the matrix.
53
10000
single rate model

8000
two rate model
three rate model
6000
Oil rate (m3 /day)
4000
2000
0
0 1000 2000 3000 4000 5000
Time (days)
Figure 3.12: Oil rate from all wells for simulations of imbibition for the one-, two- and
three-rate models. In all cases water breakthrough is very rapid, but the multi-rate
models give lower recovery since portions of the matrix are effectively inaccessible to
water over the time-scale of the simulation.
Fig. (3.13) shows the oil rate for the model coupled to the fractional flow (the fff
model, Appendix B). Also shown for comparison is the linear-Kazemi model - this is the
fractional flow model with a single rate βav that corresponds to the model proposed by
linear-Kazemi model (Kazemi, 1992). Coupling with the fractional flow leads to even
lower recoveries since rapid transfer only occurs once the fracture saturation is above
2/3. Interestingly though the Kazemi model gives a higher recovery than the equivalent
single-rate model ( see Fig. (3.11)) even though, by definition, the transfer rate in this
model is always the same or lower - see Eq. (2.29). This has been discussed previously
(Di Donato, 2003, 2004) - the Kazemi model predicts a more smeared out saturation
profile in the fractures which lessens the apparent mobility contrast between injected
and displaced fluids, leading to a higher overall sweep efficiency. Since in both cases
imbibition is rapid, the improved sweep efficiency more than compensates for the slightly
slower transfer rates in the Kazemi et al. model.
54
10000
8000
three rate model
fff model
Kazemi linear model
Oil rate (m /day)
6000
3
4000
2000
0
0 1000 2000 3000 4000 5000
Time (days)
Figure 3.13: Oil rate from oil field for simulations of imbibition for different models. The
fff model refers to a case where the transfer rates are coupled to the fracture fractional
flow. The Kazemi model is equivalent to the fff model with a single average transfer rate.
Fig. (3.14) illustrates the effect of wettability on the results. The water-wet case gives
higher recovery since more oil can be recovered from the matrix by spontaneous imbibition
(Sw∗ is higher - see Fig. (3.11). Note that in Fig. (3.11) fewer of the fractures are swept
by water in the water-wet case. This is because the large amount of imbibition slows
down the advance of water in the fractures. Water is continuing to invade more and more
of the fractures after 2,000 days, resulting in the higher oil rate evident in Fig. (3.14).
Once water occupies the fracture, imbibition is exceptionally rapid, with a time constant
of only 1.4 days.
55
10000
8000
mixed -wet single rate

Oil rate (m3/day)
6000 water-wet single rate
4000
2000
0 1000 2000 3000 4000 5000

Time (days)
Figure 3.14: Oil rate from all wells for simulations of imbibition showing the effect of
wettability. A water-wet model gives a higher recovery than the corresponding mixed-
wet model since the final saturation reached in the matrix is higher.
Fig. (3.15) shows the recovery curves for the single-rate, three-rate, fff and water-
wet cases as a function of pore volumes injected. One pore volume injected (including
fracture and matrix volumes) corresponds to nearly 10,000 days. The water-wet case
gives the best recovery since more water can enter the matrix, it is also the only case
where recovery continues beyond one pore volume injected, since more of the fractures
continue to be swept. Including a plausible amount of heterogeneity in the sub-grid scale
fracture spacing leads to much less favorable predictions of recovery.
56
0.5
single rate model

0.4 fff model
three rates model
water-wet model
Oil recovery
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
PVI
Figure 3.15: Field recovery of imbibition vs. pore volumes injected. The water-wet model
gives the best recovery, since more water can enter the matrix. The fractional flow model
shows the worst recovery since imbibition is only rapid once the fracture water saturation
is quite high.
For gravity drainage we use two vertical wells to inject gas at the top of the reservoir
and four horizontal wells to produce at the bottom of the reservoir. Fig. (3.15) shows a
vertical slice through the field at two different times for simulations of gravity drainage
using the base case model. Gas is is injected at the top of the field and slowly displaces oil
downwards. Where gas has entered the fractures, slow drainage allows oil to be recovered
from the matrix.
Fig. (3.16) shows the recovery curves for the three cases considered. Notice that in
contrast to imbibition, recovery continues at late times. This is a result of the power-law
functional form for recovery, Eq. (2.16). The oil relative permeability has a dramatic
impact - changing the exponent a in Eq. (2.18) from 2 to 4 results in approximately
half the recovery, even though the rate constant in the later case, Eq. (2.17), is higher.
For a = 4 the oil relative permeability is much lower at low oil saturation and gravity
drainage is very slow. Decreasing the block height L also leads to lower recovery, by
reducing the fraction of oil that can be recovered from each block. However, overall the
recoveries are more favorable than for water-flooding.
57
0.2 PVI (a)
1.0 PVI (b)
0.0 0.2 0.4 0.6 0.8 1.0
Sg
Figure 3.16: Vertical slices through the reservoir showing the fracture (left) and matrix
(right) gas saturations for simulations of gravity drainage at two different times, one PVI
corresponds to almost 10, 000 days.
For this field gravity drainage is more efficient than water-flooding - Fig. (3.17). This
is because we have a tilted reservoir where an artificial gas cap can be maintained. The
ultimate recovery for gravity drainage is high, giving final recoveries of up to 0.4. The
power-law recovery in each matrix block allows oil to continue to be recovered, albeit
slowly, for long times. The dolomite rock is mixed-wet and only half the oil initially
in place can be recovered by spontaneous imbibition. This coupled with a poor sweep
efficiency in the very heterogeneous fracture network makes waterflood recoveries rather
low.
The recovery is controlled by both the distribution of fracture permeability and the
matrix properties - in particular the matrix wettability (for water-flooding) and oil relative
permeability (for gravity drainage) control the recovery rate and have as significant an
effect on performance as the fracture geometry.
58
0.6
L=4, a=4, r=9.7

0.5
L=4, a=2, r=9.7
L=2, a=2, r=4.9
0.4
Oil recovery
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
PVI
Figure 3.17: Field recovery for simulations of gravity drainage. The recovery is affected
by block height and relative permeability exponent a.
3.2.7 Conclusions
The Liu7 oil field study represents a streamline-based simulation of a fractured reservoir
using multi-rate transfer functions. We use transfer functions for imbibition and gravity
drainage based on analytical solutions to the transport equations, and develop a multi-
rate model that is coupled to the fracture fractional flow. This approach suggests a
rigorous method of upscaling fracture flow using a combination of theoretical analysis,
experimental results and discrete fracture network simulation.
We applied the methodology to an extensively fractured mixed-wet dolomite oil field in
China. We demonstrated that using a single average rate based on the average fracture
density would over-estimate recovery compared to the use of a model that captured the
heterogeneity of the fracture network within each grid block. Coupling to the fracture
fractional flow predicted even lower recovery. For this field, gas gravity drainage was
more favorable than water-flooding, since it had a higher sweep efficiency and ultimate
recovery.
Possible future work includes assigning different transfer rates to different grid blocks,
upscaling and the development of a more general methodology to account for fluid ex-
59
pansion, possibly using conventional grid-based simulation.
3.3 Clair field recovery investigation
The Clair field is located approximately 75 km west of the Shetland Islands in about
140 m of water. It covers an area of 220 km2 . It was discovered in 1977. Between 1977
and 1990, 10 appraisal wells drilled, followed by a further 5 appraisal wells between 1990
and 1995. A 3D seismic survey was also carried out during this period. A further 2
appraisal wells drilled in 1997 and an extended well test conducted. In May 1997, it
was agreed to develop the field. The five Clair field partners have the following fixed
equity interests: BP: 28.6% (Operator); ConocoPhillips: 24.0%; ChevronTexaco: 19.4%;
Enterprise: 18.7%; Amerada Hess: 9.3% (Barr, 2005).
Clair is a large oil field (more than 0.63 × 109 m3 STOIIP); it is the largest naturally
fractured reservoir developed in the UK, and is recovered by waterflood. Its 28 years
appraisal period reflects the high reservoir complexity, relatively poor quality conventional
seismic image, and uncertain impact of the conductive fractures - see Fig. (3.18) (Clifford,
2005).
It started producing in February 2005 under Phase 1, and will focus on the Core,
Graben and Horst reservoir areas which have estimated oil in place of approximately
0.278 × 109 m3 , of which 39.8 × 106 m3 is predicted to be recovered. It is planned to drill
15 producing wells, 8 water injectors and 1 drill cuttings re-injection well.
3.3.1 Reservoir description
The reservoir is made up of fractured sandstones of Devonian to Carboniferous age,

and is mixed-wet. Table 3.3 shows the reservoir properties we are going to use in the
simulation. However, there is significant uncertainty both in terms of reserves and the
ability to commercially produce this highly fractured reservoir.
The Clair reservoir lies on and alongside the Rona Ridge, a fault-bounded basement
60
Figure 3.18: The Clair field located west of the Shetland Islands. The ar-
eas with green colour shows the parts which are going to develop subsequently.
(www.woodmacresearch.com)
high that separates the West Shetland Basin from the Faeroes-Shetland Basin. Fig. (3.19)
shows a schematic N-S cross section of the field showing the base (units I and II), lower
(units III to VI) and upper Clair groups. The field is divided into nine fault-bounded
segments, which have a common free water level and a maximum oil column of 600 m.
A gas cap is present in the structurally elevated ridge segments.
BP provided the data set, which used the VIP (Landmark) simulator using a dual
porosity/ dual permeability approach. The number of grid blocks is 20 × 10 × 17 (total
3,400, 3,212 active) with a block size of 120×120×15 m. The matrix/fracture permeability
and porosity were set grid by grid: the value of matrix permeability is in the range
20 − 120 mD and the fracture permeability varies from 30 to 130 mD. The average
porosity in the matrix is about 13%, and the average fracture porosity is about 0.25%.
The relative permeability curves for the matrix as shown in Fig. (3.20) were obtained
from a BP internal report. In this study, the data from a 1993 32 mD core sample
from unit V of Clair was used. An assumption that both phases are equally mobile
for the entire range of saturations for the fractures was made for the fracture relative
permeability curves. Fig. (3.21) shows the matrix capillary pressure curve; the fracture
61
Figure 3.19: Schematic N-S cross-section of the Clair Field (BP unpublished report)
Table 3.3: Parameters used in the simulations

P arameter V alue P arameter V alue
W OC , m 2336 Km , 10−15 m2 (mD) 20 − 120
GOC , m 1533 φf , % 0.1 − 0.3
Ref erence pressure , P a 191 × 105 Kf , 10−15 m2 (mD) 30 − 130
T , ◦C 101 φm , % 12 − 14
Swi , % 18 − 38 Sor , % 18 − 30
−2
χ , m 0.018 Cr , P a−1 5.8 × 10−10
µo , P a.s 3.04 × 10−3 µw , P a.s 0.5 × 10−3
−1 −10
Crw , P a 4.64×10 Bo , rm3 /sm3 1.058
capillary pressure was assumed to be zero.
62
1.0
Measured kro
0.9 Measured krw
Corey kro fitted to data
0.8
Corey krw fitted to data
0.7 Corey kro adjusted to Swi
Corey krw adjusted to Swi
kr 0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Sw
Figure 3.20: Clair field relative permeability curves in the matrix.

200
150
Drainage Pc
Input imbibition Pc
Measured imbibition Pc
Pc Psia
100
50
0
0.0 0.2 0.4 0.6 0.8
Sw
-50
Figure 3.21: Clair field capillary pressure curves in the matrix.
The results from Eclipse and streamline-based model were compared by assuming sim-
ilar physics: transfer controled solely by the matrix capillary pressure with negligible
fracture capillary pressure. Hence the two models should have similar results. However,
to do this we need to relate the transfer rate β used in the streamline-based model to the
properties used in the Eclipse, principally the shape factor χ. We set up the two models
to be equivalent by using the flowing equation (Di Donato, 2003):
χKm Pcm (S wm )krom (S wm )

β= (3.10)
µo φm (1 − Somr − S wm )
63
These two models are equivalent at some typical water saturation around S wm = 0.5.
Table 3.4 lists the parameters used to calculate the transfer rate. So the value of β is
5 × 10−8 s−1 .
Table 3.4: Parameters used to calculate the transfer rate

φf = 0.004 φm = 0.13 Km = 8 × 10−14 m2 krom = 0.032
Somr = 0.21 S wm = 0.5 µw = 0.5 × 10 P a.s µo = 3×10−3 P a.s
−3
σow = 0.03 N.m−1 5

Pcm = 1.28 × 10 P a L2c = 0.159 m2
Fig. (3.22) shows the reservoir permeability distribution in the matrix and fracture.
This reservoir is quite different from Liu7, since the matrix permeability is quite high
with an average value of 80 mD. The fracture permeabilities are similar to matrix peme-
abilities because the model is relatively coarse. The figure gives the matrix permeability
distribution in the matrix in three directions, and shows the permeability distribution in
the fractures, which is the average value in a 120 × 120 × 15 m grid block, so it can not
represent the fracture characteristics accurately.
Reservoir model preparation
The data set was provided by BP is dual porosity model.The model was converted to
single porosity model to study as well. We transferred this coarse model by: 1. the
porosity of fracture and matrix was combined to a single grid porosity to keep the same
pore volume; 2. the total permeability in each direction was calculated by equation:
φm Km + φf Kf
Kt = (3.11)
φm + φf
3.3.2 Reservoir simulation
We design different injection strategies to investigate the reservoir behaviour on the base
of the grid-block system which BP provided. Five wells were placed in the reservoir: one
injector in the middle and four producers at the four corners (Fig. (3.23)), the injection
rate is set as 1600 m3 /day and producers are controlled by the bore hole pressure of
64
(a)
(b)
Figure 3.22: Clair field permeability distribution in matrix and fracture, which shows
that the permeability in both fracture and matrix is significant: (a) is the permeability
distribution in x, y, and z directions for the matrix; and (b) is the permeability distribution
in the same three directions. It is not realistic to represent one 120 × 120 × 15 m block
with one value, since each grid block might contain several different fracture sets.
13.7 × 106 P a.
It is obvious that the model is too coarse to imitate the heterogeneity of the field. So
we downscaled the model from 20 × 10 × 17 to 40 × 20 × 17 and 40 × 20 × 68 fine grid
models.
Permeability distribution downscaling
We downscale both matrix and fracture permeability to maintain the similarity between
them. We take the original permeability distribution and downscale it. This is done step
by step:
65
Figure 3.23: Well placement for water-flooding. The water injector is shown in green and
producers in red. The injector is given a rate of 500 m3 /day and the producers have a
fixed BHP value of 1.37 × 107 P a. The grid block size is 120 × 120 × 15 m in each layer.
1. Divide each grid into two layers, we put

= Kxold ea(x−x̄)

K

x1
(3.12)
Kx2 = Kxold e−a(x−x̄)


where x is a random number between 0 and 1, a is a constant (we choose a = 2.5) and
x̄ = 0.5. This method preserves the geometric average permeability only if x̄ = 0.5.
2. Compute the y direction permeability based on the old values:

= Kx1 · Kyold /Kxold

K

y1
(3.13)
Ky2 = Kx2 · Kyold /Kxold


3. The vertical permeability is treated by maintaining the harmonic average instead of

the geometric average which is
2
Kzold = (3.14)
1/Kz1 + 1/Kz2
Let
66

Kzold = rold Kxold







Kz1 = rnew Kx1 (3.15)




Kz2 = rnew Kx2


Then 
 rold ( 1 + f ) rnew ≥ rold


2 f
,
new
r = (3.16)
new


 1 , r >1
where f = ea(x−0.5) , and x is a random number between 0 and 1.
new rold · f
r = , if rnew ≥ rold (3.17)
2(1 + f 2 )

 f 2 ·rold


2f −r old
, f >1
rnew = (3.18)
r old



2f −f 2 ·r old )
, f <1
4. Each grid block is further divided into four using the downscaling method described
in Appendix C and the same ratio of Ky /Kz is kept as the original block.
The maximum permeability is 400 mD in the 40 × 20 × 17 model, and 800 mD is the
maximum value of the 40 × 20 × 68 model. Other properties - such as porosity - are kept
the same as the original values. The size of each grid block is split proportional to the
number of grids and layers.
3.3.3 Recovery comparison with different simulators
We investigate the Clair field using single porosity, dual porosity and dual permeability
models with different numbers of grid blocks. We use a conventional grid-based simulator,
Eclipse, to compare single porosity and dual permeability models and the streamline
simulator to study single porosity and dual porosity models.
67
Grid-based simulation
The predicted production for all models is unfavorable, because the reservoir is poorly
connected. It was not possible to run a dual porosity model with the right producing rate
because the simulator experiences stability problems, since the fracture porosity is much
lower than the matrix porosity. This is a problem that I have encountered commonly
when running grid-based dual porosity simulations.
Fig. (3.24) shows the oil production from the single porosity model with different levels
of grid refinement. The fine grid model can produce more oil because the high perme-
ability regions are better connected. The figure also shows that the dual permeability
models give similar results to the single porosity models, since there is significant flow
in both fracture and matrix. The flow can be accurately described by a single effective
permeability that accommodates the fracture and matrix contributions.
3000
2500
Oil production m /D
2000
3
1500 20x10x17_single porosity

40x20x68_single porosity
20x10x17_dual permeability
1000 40x20x68_dual permeability
500
0 1000 2000 3000 4000 5000 6000
Time ( days )
Figure 3.24: Oil production comparison for single porosity model and dual permeability
for different levels of grid refinement using grid-based simulation. Both methods give
smiliar results.
68
Streamline-based simulation
Fig. (3.25) shows three single porosity models obtained by the streamline simulator.
The coarse grid block model gives more favourable recovery because the model is more
homogenous, so the water can sweep the reservoir efficiently. The fine grid model is more
heterogeneous, and hence the water preferentially moves along high permeability channels
giving poorer recovery. Note that this is the opposite trend as seen using grid-based
simulation. However, overall the results of grid-based and streamline-based simulation
are similar.
3000
2500
Field oil production m /D
3
2000
20x10x17
1500
40x20x17
40x20x68
1000
500
0 1000 2000 3000 4000 5000 6000
Time (days)
Figure 3.25: Oil production for single porosity models with different levels of grid refine-
ment using streamline-based simulation.
69
Fig. (3.26) shows the predicted recovery using a dual porosity model showing typical
characteristics: very early breakthrough and a rapid decline in production. The oil
production decreases very fast, because it does not produce much oil from matrix. This
model does not account for the matrix contribution to the fluid flow. It might be quite
different from the real reservoir behaviour because the matrix permeability is high in this
case.
2000
1600
1200
Oil production m /D
3
800 20x10x17
40x20x17
40x20x68
400
0
0 1000 2000 3000
Time (days)
Figure 3.26: Oil production for dual porosity models with different levels of grid refine-
ment using streamline-based simulation. The recoveries are very low for this type of
model, since viscous flow in the matrix is ignored.
70
3.3.4 Initial conclusions
The reservoir model is relatively coarse so it can not describe the fracture geometry
properly. It could overestimate the oil recovery since the whole model is relatively homo-
geneous. It is necessary to have a fine grid model to describe the reservoir heterogeneity.
Eclipse can treat the data set with single porosity and dual permeability models and the
results are similar. But the dual permeability model takes more computing time. Table
3.5 shows the computing time for different models for different number of grid blocks.
Eclipse cannot handle a dual porosity model since it is numerically difficult to account
for a situation where all the flow is in the fractures that occupy only a tiny fraction of the
pore space. The streamline-based simulator is faster, although the speed-up is modest in
comparison to the results in Chapter 2.
Overall, the coarse model used is poorly connected. We suggest the development of a
geologically informed fine-grid model that could possibly be run in single porosity model
with effective permeabilities that account for contributions from both fracture and matrix.
The streamline-based simulator can handle the dual porosity model. The results, how-
ever, could under-estimate the oil recovery, because the Clair field has high matrix per-
meability, the fluid transport between matrix blocks is not negligible.
Table 3.5: Run time of different simulators for 6000 days (minutes)
N umber of Single porosity Dual permeability Single porosity Dual porosity
grid blocks (Eclipse) (Eclipse) (Streamline) (Streamline)
20 × 10 × 17 2.4 5.0 1.1 2.81
40 × 20 × 17 8.2 78.9 3.2 53.5
40 × 20 × 68 53.0 399.0 9.1 218.9
71
Chapter 4
General Transfer Functions for

Multiphase Flow
The work described so far in this thesis concerns incompressible displacement imple-
mented into a dual porosity streamline-based simulator. I would, however, like to extend
the scope of this research to encompass all types of displacement processes, including
compressible flow, for both dual porosity and dual permeability models.
In this Chapter a general transfer function for all types of recovery process is formulated
and tested, extending the work described previously.
The current Barenblatt-Kazemi approach assumes a Darcy-like flux from matrix to frac-
ture, assuming a quasi steady-state between the two domains (Barenblatt, 1960, Kazemi,
1992). However, this does not correctly represent the average transfer rate in a dynamic
displacement.
Based on one-dimensional analytical analysis in the literature, expressions for the trans-
fer rate accounting for both displacement and fluid expansion will be found. A new,
physically-motivated formulation for the matrix-fracture transfer function in dual per-
meability and dual porosity reservoir simulation is proposed. The resultant transfer
function is a sum of two terms: a saturation-dependent term representing displacement
and a pressure-dependent term representing fluid expansion.
72
The expression reduces to the Barenblatt (1960) form for single-phase flow at late times,
but more accurately captures the pressure-dependence at early times. For displacement
we consider both imbibition and gravity drainage processes. The novel transfer function is
validated through comparison with one-dimensional fine-grid simulations and compared
with predictions using the traditional Kazemi formulation.
It is shown that the new method is more numerically stable, being almost linear in the
solution variables (saturation and pressure) while capturing the dynamics of expansion
and displacement accurately.
4.1 Formulation for multiphase flow
4.1.1 Traditional formulation
The conventional macroscopic treatment of flow in fractures reservoirs assumes that

there are two communicating domains: a flowing region containing connected fractures
and high permeability matrix; and a stagnant region of low permeability matrix. Conven-
tionally these are referred to as fracture and matrix. There is transfer between fracture
and matrix mediated by gravitational and capillary forces. In a dual porosity model it
assumes that there is no viscous flow in the matrix; a dual permeability model allows
flow in both fracture and matrix. In a general compositional model (where black-oil and
incompressible flow can easily be treated as special cases) we can write:
∂Mcf X
+ ∇ · cpcf fp qt = −Γc (4.1)
∂t p=o,w,g
where Γc is a transfer term with units of mass per unit volume per unit time - it is a
rate times a density. c is a component density (concentration) with units of mass of
component per unit volume. The subscript p labels the phase and c the component. Γc
represents the transfer of component c from fractures to the matrix. The subscript f
refers to the flowing or fractured domain, as before. The first term is accumulation and
73
the second term represents flow - this is the same as in standard (non-fractured) reservoir
simulation. We can write a corresponding equation for the matrix (m):
∂Mcm
= Γc (4.2)
∂t
where we have assumed a dual porosity model (no flow in the matrix); for a dual
permeability model a flow term is added to Eq. (4.2).
The mass per unit volume of component c in the fracture is:
X X
Mcf = Mpcf = φf Spf cpcf (4.3)
p=o,w,g p=o,w,g
while for the matrix:
X X
Mcm = Mpcm = φm Spm cpcm (4.4)
p=o,w,g p=o,w,g
All that is required now is an expression for the transfer rate. The original Kazemi
formulation (1990) was for a black-oil model, but is extended here to a compositional
treatment:
X
Γc = χKm λp cpc (ψpf − ψpm ) (4.5)
p=o,w,g
where χ is the shape factor with units of length−2 (m−2 in SI units). The shape factor
is a cross-sectional area for transfer per unit volume divided by a typical distance for
flow between matrix and fracture. λ is a mobility - a relative permeability divided by a
viscosity. ψ is the potential: ψ = P − ρgh where P is the pressure and h is a height. The
gravitational term is necessary to accommodate transfer due to gas gravity drainage. The
values of mobility and concentration are normally upstream-weighted (Kazemi, 1990): if
the flow of a given phase is from matrix to fracture (ψpm > ψpf ), then λp = λpm and
ccpc = ccpcm , whereas if the flow is from fracture to matrix (ψpf > ψpm ), then λp = λpf
and ccpc = ccpcf .
74
Eq. (4.5) assumes a pseudo-steady-state with the flow between fracture and matrix
governed by an average potential difference. This formulation is correct in the limit
that each matrix block is discretized into many simulation grid blocks, but it does not
necessarily accurately represent the average behaviour when the matrix block is treated
as a point. In single-phase flow it has been shown that the transfer function is only valid
at late times with a suitably chosen value of the shape factor that depends on the fracture
spacing and geometry and even then there is still disagreement over its value, as discussed
in Chapter 2 (Zimmerman et al., 1989, 1993). In multiphase flow there is a displacement
front that moves through the system and this is not adequately captured by an average
pressure difference multiplied by a mobility at an average saturation.
4.1.2 Our formulation
In fractured reservoirs there are four principal recovery processes, illustrated in Fig.
(1.1): fluid expansion, capillary imbibition, diffusion and gravity drainage. The overall
transfer from fracture to matrix will be given by the combination of these different effects.
Mathematically the evolution of pressure, saturation and concentration due to fluid
expansion, imbibition and diffusion respectively are all governed by a linear or non-linear
diffusion equation. In this case the generic behavior of the recovery for constant fracture
conditions is well-known (Vermeulen, 1953, Zimmerman et al., 1993): initially it scales
√
at t before the advancing front reaches any boundaries and, approximately, as 1 − e−βt
at late time with some geometry and fluid-dependent rate β (Di Donato, 2006, Hagoort,
1980).
Our formulation will treat the transfer function as a sum of contributions due to dif-
ferent physical effects with a functional form that gives the correct early and late-time
behaviours:
X X
Γc = Γpcd + cpcm Tpe + φm Tpqs (4.6)
ρ=o,w,g q6=p
75
Γpcd is transfer due to molecular diffusion. T is a transfer function with the units of a
rate (inverse time). Tpe is the transfer rate of phase p due to fluid expansion. Tpqs is the
transfer of phase p due to displacement by phase q governed by saturation changes. Sarma
and Aziz (2006) also proposed a transfer function where the expansion and displacement
terms were treated separately. They derived time-dependent shape factors to match
analytical solutions. In this work we will follow a similar approach, but write the transfer
functions as functions of pressure and saturation.
Fluid expansion.
To account for fluid expansion for both early and late times we propose a multiphase
extension of an expression first introduced by Vermeulen (1953)and implemented in an
air/water dual porosity model by Zimmerman et al. (1993):
Tpe = χBe Km λp m(ψof − ψom ) (4.7)
Eq. (4.7) is the same as the conventional model, Eq. (4.5) except for two differences.
First Eq. (4.7) is written in terms of the oil potential ψo and not the particular phase
pressure. This is important. If we wrote the transfer in terms of phase pressures, then
capillary pressure would be included implicitly (as in the conventional formulation). In-
stead, we account for capillary-driven transfer separately in the displacement terms. It
is assumed that the simulator solves for the oil pressure; if not, the equations above are
identical, but ψo is replaced by the potential that is solved for. Although Eq. (4.7) is
written for the potential, normally gravitational effects are accounted for in the displace-
ment terms only and we ignore any height difference between matrix and fracture for
expansion: ψ = P .
The second difference with the conventional model is the inclusion of a boost or cor-
rection factor Be given by:
76

ψof − ψoim + ψom − ψoim
Be = (4.8)
2max[ξψom ; ψom − ψoim ]
where the subscript i refers to initial conditions.

The correction factor is close to 1 at late time when the fracture and matrix pres-
sures are similar, so the transfer function then reverts to the traditional Barenblatt form
(1960). At early time, however, the matrix pressure is close to its initial value and the
√
correction factor is large, giving a pressure evolution ψom = ψoim + A t with a constant
A (Vermeulen, 1953, Zimmerman et al. 1993). ξ is a small parameter, of order 0.01 − 0.1,
that ensures stability of numerical codes at very early times; otherwise the correction
factor diverges at initial conditions causing convergence problems.
We use the same shape factor χ. This can be derived analytically for flow in simple
geometries. In this Chapter we will consider linear flow in a system of length L with one
end in connection with the fracture: in this case χ = π 2 /4L2 .
Diffusion
By analogy to our treatment of pressure diffusion, we can use a similar expression for
molecular diffusion since we are solving similar equations:
Γpcd = χBd Dpc (cpcf − cpcm ) (4.9)
|cpcf − cpcmi | + |cpcm − cpcmi |

Bd = (4.10)
2M ax[εcpcm ; |cpcm − cpcmi |]
where again i refers to the initial equilibrium conditions. D is the molecular diffusion
coefficient in the porous medium.
Fluid displacement
We write the displacement terms as follows. For water displacing either gas or oil:
77
∗

Twqs = Bs βwq Swim + F (Swf )(Swm − Swim ) − Swm (4.11)
for q = g, o. Note once again the introduction of a Vermeulen-type correction factor

(Vermeulen, 1953):
∗
Swm − Swim + Swm − Swim
Bs = (4.12)
2M ax ξSwm ; Swm − Swim
For gas/oil displacement:
∗
b
Tgos = βgo Sgim + F (Sgf )(Sgm − Sgim ) − Sgm (4.13)
We define, since the displacement terms alone only account for incompressible flow:
Tpqs = −Tqps (4.14)
F (S) is a smooth function that allows fluid to transfer both from fracture to matrix and
from matrix to fracture. We assume that displacement continues until the invading phase
saturation in the fracture is very low - this corresponds physically to having a much lower
capillary pressure in the fractures than the matrix, or allowing gravity drainage even at
low fracture gas saturation. We propose:
√
1 − e− Kf /Km S
F (S) = √ (4.15)
1 − e− Kf /Km
b is the exponent of the oil relative permeability, defined with that at low oil saturation
krom ∝ (Som − Sorm )b .
β are transfer rates that account for both capillary and viscous forces:
s !
3λ∗pm λ∗qm σpq J ∗
Km Km |∆ρpq |g
βpq = + (4.16)
λ∗tm H2 φm φm H
S ∗ corresponds to the final saturation in the matrix at the end of the displacement. This
78
is not simply the residual saturation, but the saturation in capillary/gravity equilibrium
and needs to be pre-computed for each matrix block. λ∗ is the corresponding mobility,
it is a typical mobility for the displacement. For gas gravity drainage, where the gas is
much more mobile than the oil, λ∗gm λ∗om /λ∗tm ≈ kro
max max
/µo , where kro is the maximum
oil relative permeability. For capillary imbibition, the mobilities can be taken at the end
of imbibition for mixed-wet systems. Where the system is strongly wetting and the end-
point mobility of the displaced phase is zero, mobilities for a saturation mid-way between
the initial and final saturations can be used. J ∗ is the gradient of the dimensionless
capillary pressure at the saturation for which the mobilities are computed for imbibition
( it is assumed to be positive ), or the entry pressure for gas gravity drainage. We assume
Leverett J-function scaling:
r
φm
Pcpqm (Spm ) = J(Spm ) (4.17)
Km
H is the distance through which the displacing phase actually travels. While for
capillary-controlled displacement this is simply L, for gravity drainage this may be smaller
than the total height of the matrix block, since the wetting phase is retained in capil-
lary/gravity equilibrium at the base of the block:
r
φm
σgo J ∗ = ∆ρgo gh (4.18)
Km
The displacement rate for gravity drainage is zero if the gas cannot enter the matrix:
for a block of height L we set βgo = 0 if:
r
φm
σgo J ∗ ≥ ∆ρgo gh (4.19)
Km
For a one-dimensional system with a fracture at one end, L is simply the length (or
height) of the matrix block. For more complex geometries, we use an expression due to
Zhang et al. (1996), described in Chapter 2 (Eq. (2.10)).
79
V
L2 = Pn Ai
(4.20)
i=1 li
where V is the matrix block volume, Ai is the area open to flow in the iith direction and
li is the distance from the open surface to a no-flow boundary.
In many cases, the matrix relative permeabilities are not known. In this case, for
moderate mobility ratios, a simpler form for the rate constant can be used (Zhang et al.,
1996):
s !
α σpq Km Km |∆ρpq |g
βpq = √ + (4.21)
µp µq H2 φm φm H
where α is a dimensionless coefficient whose value is around 0.05 for water-wet media,
decreasing to 10−4 for mixed-wet systems.
Justification
The expressions for capillary-controlled displacement come from a combination of theo-

retical, numerical and experimental analysis (Zhang et al., 1996; Behbahani et al., 2005):
the linear dependence of transfer rate on matrix saturation has been shown to reproduce
experimentally-measured recoveries while being based on an approximate analytical solu-
tion to the governing flow equations in the matrix, the numerical factor of 3 in Eq. (4.16)
is equivalent to a dimensionless shape factor and comes from the analytic derivation
(Aronofsky, 1958; Barenblatt, 1960; Tavassoli, 2005).
The introduction of the fracture saturation dependence through F (S) is an empirical
device to ensure that the transfer rate depends smoothly on the system parameters:
previous treatments of this problem had a finite transfer rate until the fracture wetting
phase saturation was zero (Di Dinato, 2003,2006; Huang,2004; Behbahani, 2005), which
leads to a non-monotonic variation of transfer rates that causes convergence problems in
general-purpose simulators. F (S) is close to 1 until the fracture saturation is very small.
This formulation allows transfer to occur in both directions as the fracture saturation
80
changes.
We have added a Vermeulen-type correction to the linear transfer rate; this will allow
us to reproduce more accurately the early-time behavior. For gravity-controlled displace-
ment the form of the transfer function is also derived from an analytical analysis of the
flow equations, confirmed by numerical and experimental tests ( Hagoort, 1980; Di Do-
nato, 2006, 2007). There is no need for a Vermeulen correction, since the power-law
behavior reproduces the behavior quite accurately.
Our formulation, at first sight, seems more complex than the current Kazemi et al.
model (1992). However, since our transfer rate varies almost linearly with saturation
and pressure, it is easier to implement numerically. The model forces the engineer to
consider carefully the physical processes that are occurring in the system. While all the
parameters in the equations are generally known, it is advisable to treat displacement in
terms of an effective rate β, rather than the shape factor. The transfer rate accounts for
both the geometry of the system and, equally important, the fluid and rock properties.
The time-scale for significant recovery is then estimated as 1/β .
As written, the model assumes that water will imbibe to displace gas or oil controlled
principally by capillary forces, while gas displaces oil by gravity drainage. However, the
transfer rates and the final saturations do account for both gravity and capillary forces.
For a system with known fracture geometry and fluid properties this model has no
adjustable parameters.
4.1.3 Numerical implementation
We will test the transfer function against one-dimensional simulations where the frac-
ture saturation and pressure are held constant. The averaged matrix properties will be
compared to predictions using the transfer function in a zero-dimensional model (the
matrix block is treated as a point). The transfer function for displacement, without the
Vermeulen correction terms, has also been incorporated in a three-dimensional streamline-
based simulator (Di Donato, 2007), as described previously. We assume a linearly com-
81
pressible system:
ρp = ρpi eγp (Po −Poi ) (4.22)
where γ is a compressibility assuming no changes in composition or capillary pressure.

Similarly for the rock:
φ = φi eγφ (Po −Pof ) (4.23)
Since the displacement terms are for incompressible flow, when volume conservation
over all phases is summed, only the expansion term is left. We find the following equation
for pressure from Eqs. (4.2), (4.4),(4.6), (4.7), (4.22), (4.23):
∂Pom X 2Poim − Pof − Pom )

γt = Tpe = Km λt (Pof − Pom ) (4.24)
∂t p=o,w,g
2(Poim − Pom )
where it assumes that the fracture and matrix are at the same height, Poi ≥ Pof , Poi ≥ Pom
and:
X
γt = φm (γφ + Spm γp ) (4.25)
p=o,w,g
The pressure equation is solved implicitly assuming constant saturation and porosity.
Since we have a quadratic expression this can be done analytically. We find:
n
p
n+1 a ∗ (Poim − Pom )− a2 (Poim − Pom
n )2 + 4(a + 1)(P
oim − Pof )
2
Pom = Poim − (4.26)
2(a + 1)
where the subscript n indicates the time level and:
2γt
a= (4.27)
χKm λt ∆t
The saturations and component concentrations are then updated using the diffusion
82
and displacement terms. To accommodate diffusion an expression similar to Eq. (4.26)
with c substituted for P can be used to update concentration.
Since the fracture saturation is held at a constant high saturation, we take F = 1
in Eq. (2.15). For two-phase oil/water displacement, the saturation equation from Eqs.
(4.2), (4.6) and (4.11) is:
∂Swm ∗ (S ∗ + Swm − 2Swim )

= Twos = βwo (Swm − Swm ) wm (4.28)
∂t 2(Swm − Swim )
from which we find:
n
p
− Swim + n −S 2 ∗ 2
n+1 Swm (Swm wim ) + βwo ∆t(2 + βwo ∆t)(Swm − Swim )
Swm = Swim +
2 + βwo ∆t
(4.29)
It is also possible to solve the pressure equation analytically if the saturation is held
constant, or the saturation equation if the pressure is held constant (Vermeulen, 1953;
Zimmerman et al., 1993). The Vermeulen correction gives the correct behaviour at early
time with an exponential relaxation at later time, as observed experimentally and derived
analytically from solutions of the transport equations in the matrix (Barenblatt, 1960;
Vermeulen, 1953; Zimmerman et al., 1993; Morrow, 2001).
For gravity drainage the numerical implementation has been described in detail else-
where, see Appendix B (Di Donato, 2007). For the Kazemi et al. model the oil pressure
equation is (Kazemi, 1993):
∂Pom X
γt = χKm λt (Pof − Pom ) − λq Pcoqm (4.30)
∂t q6=o
Note the presence of capillary pressure terms in Eq. (4.30) in contrast to Eq. (4.24).
For two-phase oil/water flow, Eq. (4.30) in discretized form becomes:
n
n+1 aPom + 2Pof + 2λg Pcowm /γt
Pom = (4.31)
a+2
83
The capillary pressure term means that the pressure evolution is different from our
model, even at late time. Furthermore, it leads to numerical convergence problems: near
end-points where Pc diverges, it is necessary to use an iterative solution technique. We
will show later that the Kazemi et al. (1993) model gives erroneous predictions of the
average pressure response.
For gas/oil gravity drainage the corresponding equation for potential is:
n
n+1 aψom + 2ψof + 2(∆ρgo gh − λg Pcgom /γt )
ψom = (4.32)
a+2
where h is the matrix block height.
4.1.4 Numerical tests
Predictions of average matrix saturation, pressure and production using these trans-
fer functions were compared to numerical simulation. In all cases homogeneous one-
dimensional two-phase systems were considered with a fracture whose saturation and
pressure were held constant. The porosity was assumed to be constant and independent
of pressure, but the oil and water were compressible. The matrix was given uniform initial
conditions and was allowed to relax to pressure/capillary/gravity equilibrium. Pressure
was a volume-weighted average. Saturation was computed as the total mass of each phase
divided by the pore volume of the matrix times the phase density at the average matrix
pressure.
Four cases were considered whose parameters are given in Tables 4.1 and 4.2. The
matrix porosity and permeability were 0.2 and 10−13 m2 (100 mD) respectively. The
relative permeabilities were krw = Sw2 , kro = So2 and krg = Sg2 . The matrix had cross-
sectional area 1 m2 and height 1 m, except for gravity drainage where the matrix block
height was 5 m. There was a fracture in communication with one face of the matrix.
The model was discretized into 40 blocks parallel to the fracture: the block spacing was
smaller nearer the fracture and decreased away from it to capture early-time behaviour
84
(Zhou et al., 2002; Behbahani,2006).
Grid refinement studies demonstrated that 40 grid blocks was sufficient to give the
average behaviour accurately. Cases 1 and 2 represented compressible multiphase flow
but with no displacement (there was no capillary pressure and the system was horizon-
tal). Case 1 had slightly compressible oil and water with similar viscosities. Case 2 was
more challenging: the oil phase was highly compressible but had a large viscosity (this
represented a foam/water system where the oil phase was the foam). Case 2 was chosen
to test how well the transfer functions predicted very compressible flow where there was
a significant change in saturation due to oil expansion, but since it was immobile, it dis-
placed water that was produced. Case 3 modeled capillary imbibition with compressible
fluids and Case 4 was gas gravity drainage.
Table 4.1: Parameters used for case 1 and 2 (compressible fluids/no capillary pressure)
Case1 Case2
Initial matrix pressure (P a) 1, 100, 000 1, 100, 000
Initial f racture pressure (P a) 100, 000 100, 000
Initial matrix water saturation 0.5 0.5
Initial f racture water saturation 0.5 0.5
W ater density at initial matrix pressure (kg/m3 ) 1, 000 1, 000
3
Oil density at initial matrix pressure (kg/m ) 800 800
−9
W ater compressibility (1/P a) 10 10−9
Oil compressibility (1/P a) 10−8 10−6
W ater viscosity (P a.s) 0.001 0.001
Oil viscosity (P a.s) 0.001 1
85
Table 4.2: Parameters used for Cases 3 and 4 (capillary or gravity-driven flow with
compressible fluids)
Case3 Case4
Initial matrix pressure (P a) 1, 100, 000 1, 100, 000
Initial f racture pressure (P a) 100, 000 100, 000
Initial matrix water saturation 0 0
Initial f racture water saturation 1 0
Initial matrix oil saturation 1 1
F racture gas saturation 0 1
3
W ater density at initial matrix pressure (kg/m ) 1, 000 n/a
Oil density at initial matrix pressure (kg/m3 ) 800 800
3
Gas density at initial matrix pressure (kg/m ) n/a 100
−9
W ater compressibility (1/P a) 10 n/a
−8
Oil compressibility (1/P a) 10 10−8
Gas compressibility (1/P a) n/a 10−7
W ater viscosity (P a.s) 0.001 n/a
Oil viscosity (P a.s) 0.001 0.001
Gas viscosity (P a.s) n/a 0.00005
∗
End − point saturation, S 0.7(water) 0.919(gas)
∗
Dimensionless capillary pressure, J 0.148 ∼ 1 1
Interf acial tension (N/m) 0.05 0.001
T ransf er rate, s−1 8.14×10−6 ∼ 1.27×10−6 2.24 × 10−6
4.1.5 Results
Compressible flow
Fig. (4.1) shows the simulated and predicted transfer rates for Case 1. There is
production due to a rapid pressure transient through the system. The production rate is
initially high and then drops sharply once pressure equilibrium is reached.
86
1
Simulation
Transfer rate (Kgm -3 s-1 )
0.1 General transfer function
0.001 Kazemi et al.
0.0001
0.01 0.1 1 10 100
Time (s)
Figure 4.1: Predicted and simulated oil transfer rates for Case 1. Our model, Eq. (4.7),
accurately predicts the rapid initial expansion of the system, whereas the Kazemi et
al.(1992) formulation under-estimates the transfer at early times.
The correction in our transfer function accurately predicts this behaviour - see Eq.
(4.8), captures the expansion at early times. In contrast, the conventional Kazemi et al.
model (Kazemi et al., 1992) is only accurate at late time. In Case 2, the high viscosity
compressible oil expands, displacing water. When the oil saturation is close to 1, the total
mobility in the system is very low and it takes a long time to reach pressure equilibrium.
Fig. (4.2) shows that our expression gives a much better prediction of the transfer rates
than the Kazemi et al. model.
87
10
General transfer function

Transfer rate (Kgm s )
0.1
-3 -1
Simulation
0.001
Kazemi et al.
0.00001
0.01 10 10000
Time (s)
Figure 4.2: Predicted and simulated oil (red) and water (blue) transfer rates for Case 2.
In this example pressure equilibrium takes much longer to be established, since expanding
oil displaces water, which has the higher transfer rate, leaving the system with a high
but almost immobile oil saturation. Our model, Eq. (4.7), gives a better prediction of
production rates than the Kazemi et al. formulation which again under-estimates the
transfer at early times.
Capillary-driven flow
For Case 3 the capillary pressure in the matrix is:
r
φm n ∗n n ∗n
Pcowm = · σow (Swm − Swm ); J = Swm − Swm (4.33)
Km
where σow = 0.05 N/m and Sw∗ = 0.7 for different exponents, n < 0. We also considered
a linear capillary pressure (n = 1, but the pressure has the opposite sign to Eq. (4.33)).
The dimensionless capillary pressure gradient at Sw∗ is:
J ∗ = −dJ/dSwm |Swm =Swm

∗ = |n|Sw∗n−1 (4.34)
Initially the matrix is entirely saturated with oil and the fracture saturation is held at
1. The system is held horizontal and so there are no gravity effects.
Fig. (4.3) shows the transfer rates for n = −0.1. There is an initial pressure transient
giving a high transfer rate. As discussed before, this is well captured by our model but
88
is underestimated by the conventional formulation. At later times, there continues to be
appreciable production for around one year. Our model excellently predicts the behav-
iour, with no adjustable parameters, over seven orders of magnitude in time. The Kazemi
et al. (1992) model shows a steadily decreasing transfer rate, as does the simulation, but
the quantitative match is poor.
Simulation
Transfer rat (Kgm s )
-3 -1
0.01
Kazemi et al.
0.0001
0.1 100 100000
Time (s)
Figure 4.3: Predicted and simulated oil transfer rates for Case 3 with a capillary pressure
exponent n = −0.1 in Eq. (4.33). The new model gives excellent predictions over seven
orders of magnitude in time. The Kazemi et al. model gives less accurate predictions.
Fig. (4.4) shows the corresponding average matrix saturation, which is also excellently
predicted by our model. The conventional model could be improved by adjusting the
shape factor to match certain parts of the recovery, but it can not give the reasonable
results as the shape factor do not change the recovery curve. The general transfer function
does not need any tuning factors.
89
0.8

Water saturation in the matrix
0.6
Simulation
0.4
0.2 Kazemi et al.
0.0
10 1000 100000 10000000
Time (s)
Figure 4.4: Predicted and simulated matrix water saturation for Case 3 with a capillary
pressure exponent n = −0.1 in Eq. (4.33). The new model gives excellent predictions.
Fig. (4.5) shows the oil pressure prediction by different models, note that the Kazemi
et al. (1992) model prediction takes a long time to reach pressure equilibrium with the
matrix, which is not seen in the simulation. This is a direct result of the unphysical
inclusion of capillary pressure terms in the governing pressure Eq. (4.30).
90
1000000
800000
Pressure in the matrix (Pa)
Simulation
600000 General transfer function
400000
Kazemi et al.
200000
0
1 10 100 1000 10000 100000
Time (s)
Figure 4.5: Predicted and simulated average oil pressure for Case 3 with a capillary
pressure exponent n = −0.1 in Eq. (4.33). The Kazemi et al. (1993) model over-predicts
the pressure at late times after the relaxation of the initial transient.
Fig. (4.6) shows the simulated and predicted matrix water saturation for n = 1, linear
capillary pressure. Again our model gives excellent predictions. We studied other cases
with different values of n. In all cases we made excellent predictions of saturation and
transfer rate. We could adjust the shape factor in the Kazemi et al.(1992) model to give
a better prediction of average saturation at some average time. However, this would not
affect the shape of the curve and so we would be unable to achieve as good a prediction
as the new formulation. Furthermore, this would require different ad hoc adjustments for
each case, whereas we are able to make predictions without any tuning parameters.
91
0.8
0.6
Water saturation in the matrix
0.4
Simulation
Kazemi et al.
0.2
0.0
10 1000 100000
Time (s)
Figure 4.6: Predicted and simulated matrix water saturation for Case 3 with a capillary
pressure exponent n = 1 in Eq. (4.33). The new model gives excellent predictions.
Gravity-driven flow
The final Case 4 is for gas gravity drainage. We use Eq. (4.33) for the gas/oil capillary
pressure with n = −0.5 and σgo = 0.001 N/m. The fluid properties are shown in Table
4.2; for this example the final matrix gas saturation in gravity/capillary equilibrium is
0.919. Gas enters a fracture at the top of a vertically-oriented block while oil is produced
to a fracture at the bottom. In the Kazemi et al. model we assume that the matrix is at
a height h above the fracture, where h is the matrix block height.
Fig. (4.7) shows the average matrix gas saturation. The predicted saturations are
in excellent agreement with simulation. The Kazemi et al. model (1992) gives poor
predictions and seriously over-estimates the transfer rate. Changing the potential to
model the average height difference, h/2, gives only a small improvement. Furthermore,
the Kazemi et al. model allows transfer until the matrix capillary pressure is ∆ρgh -
virtually no oil is left in the matrix at this capillary pressure and hence the final recovery
is over-estimated.
Further tests for different relative permeability exponents, b, and where the relative
strength of capillary and gravitational forces is varied, are going to be presented in Chap-
ter 5 and in Di Donato’s paper (2006).
92
1.0
0.8
Gas saturation in the matrix
0.6
Kazemi et al.
0.4
Simulation
0.2
0.0
100 10000 1000000

Time (s)
Figure 4.7: Predicted and simulated matrix gas saturation for Case 4 - gravity drainage.
The new model gives good predictions.
Discussion and conclusions
We have proposed a general transfer function for fracture/matrix flow that accounts for
fluid expansion, diffusion and displacement. The principal features of the work are the
decoupling of the transfer rate into contributions due to different physical effects and the
use of a Vermeulen-like functional dependence of the transfer on pressure, concentration
and saturation to capture accurately both the early and late time behaviour (Vermeulen,
1953).
Although our formulation contains algebraically more terms than the conventional
Kazemi et al. model, it is easy to implement numerically: displacement is decoupled
from the solution to the pressure equation in the matrix and the transfer rate depends
quasi-linearly on the solution variables, avoiding the use of iterative techniques (Lu, 2006).
We tested the model for compressible flow, capillary imbibition and gravity drainage.
We demonstrated that for the cases studied we were able to predict transfer rate, pres-
sure and average saturations accurately. The performance of the model was superior to
conventional methods that tend to under-predict transfer at early times. Furthermore,
the method, although it contains physically-relevant parameters, has no adjustable fac-
tors. Further work is required to test the method for cases where there is a competition
93
between capillary and gravitational forces. For more complex geometries the predictions
of this work need to be tested against discrete fracture simulations (Matthai, 2007).
The success of this study suggests that there is no need to use several grid blocks
in communication with each fracture block to capture correctly transient displacement
behavior. Such methods assume a Darcy or Kazemi-like transfer across several matrix
grid blocks (Pruess, 1985; Wu, 1988; Chen, 1994); we propose that the matrix can be
represented by a single cell with a suitably chosen transfer function.
94
Chapter 5
General Fracture/Matrix Transfer

Functions for Mixed-Wet Systems
In this chapter, the model of fracture/matrix transfer in dual porosity and dual perme-
ability systems is extended to mixed-wet media, where there can be displacement due to
imbibition when the capillary pressure is positive combined with gravity drainage. This
can lead to a characteristic recovery curve from the matrix, with a period of rapid imbi-
bition followed by slower recovery as buoyancy overcomes capillary forces. We refine the
current general transfer function model described in the previous chapter to accommodate
such cases by including transfer due to horizontal and vertical displacement separately
(Lu et al., 2006).
The model is tested against fine-grid simulation in one and two dimensions and accu-
rate predictions are made in all cases; in contrast the conventional Kazemi et al.(1976)
model gives poor predictions of rate and ultimate recovery. Two-dimensional predictions
are more difficult, since capillary and gravity forces may act in different directions, but
refinements to the model enable good predictions to be made in such cases.
95
5.1 Formulation for mixed-wet media in two dimen-
sional models
5.1.1 Horizontal and vertical displacement
There are two distinct types of displacement: capillary imbibition that acts in all
directions and gravity drainage that acts vertically. In mixed-wet oil/water systems we
can have rapid capillary imbibition, until the capillary pressure is zero, followed by a
slower approach to capillary/gravity equilibrium where water is pushed into the matrix
at a negative capillary pressure due to buoyancy forces. To account for this we extend
the formulation presented in Chapter 4 and write the displacement terms as a sum of
horizontal and vertical contributions:
H V
Tpqs = Tpqs + Tpqs (5.1)
where H refers to horizontal and V to vertical. Quandalle and Sabathier (1989)suggested

a similar splitting of vertical and horizontal displacement, although they did this in the
context of a Kazemi-like transfer function.
We assume that the horizontal displacement is controlled entirely by capillary forces.
For water displacing gas or oil:
H ∗H
Twqs = BsH βwq
H
(Swim + F (Swf )(Swm − Swim ) − Swm ) (5.2)
for q = g, o where:
∗H
|Swm − Swim | + |Swm − Swim |
BsH = (5.3)
2 max[εSwm ; |Swm − Swim |]
We use the same functional form for vertical displacement, where gravitational forces
are also important:
96
V ∗V
Twqs = BsV βwq
V
(Swim + F (Swf )(Swm − Swim ) − Swm ) (5.4)
∗V
|Swm − Swim | + |Swm − Swim |
BsV = (5.5)
2 max[εSwm ; |Swm − Swim |]
For gas/oil displacement, assuming that it acts only vertically:
V V ∗V
Tgos = βgo (Sgim + F (Sgf )(Sgm − Sgim ) − Sgm )b , H
Tgos =0 (5.6)
Again b is the exponent of the oil relative permeability, defined such that at oil low
saturation kom ∝ (Som − Sorm )b .
We define, since the displacement terms alone only account for incompressible flow:
Tpqs = −Tqps (5.7)
F (S) is a smooth function that allows fluid to transfer both from fracture to matrix
and from matrix to fracture; it is given in Eq. (4.15).
5.1.2 Transfer rates
For vertical displacement, the transfer rate β accounts for both capillary and viscous
forces (similar to Eq. (4.16)):
s !
V
3λVpm λVqm ∗V
σpq J Km Km |∆ρpq |g
βwq = + (5.8)
λVtm H2 φm φm H
For horizontal displacement we only account for capillary forces. Assuming that only
water imbibes into gas or oil (although other cases can be accommodated):
s
H
3λH H
pm λqm σpq J
∗H
Km
βwq = · (5.9)
λHtm L2c φm
S ∗V corresponds to the final saturation in the matrix at the end of the displacement.
97
This is the saturation in capillary/gravity equilibrium and needs to be pre-computed for
each matrix block , details on its calculation for specific cases is provided later. S ∗H
corresponds to the saturation in the matrix if only imbibition acts, it is the saturation for
which the capillary pressure is zero: Pcpq (S ∗H ) = 0. However, if gravity acts to give a
lower recovery in equilibrium we set S ∗H = S ∗V ; in other words S ∗H = min(S ∗V ; Pcpq
−1
(0)).
λH is the typical mobility for the horizontal displacement . For mixed-wet systems it is
the mobility computed at S ∗H . Where the system is strongly wetting and the end-point
mobility of the displaced phase is zero, mobilities for a saturation mid-way between the
initial and final saturations can be used.
λV is a typical mobility for the whole displacement. For gas gravity drainage, where
max
the gas is much more mobile than the oil, λVgm λVom /λVtm ≈ kro max
/µo , where kro is the
maximum oil relative permeability. If there is capillary imbibition, in this thesis, we
compute mobilities at a saturation S V = 1/2(S ∗H + S ∗V ).
J ∗ is the gradient of the dimensionless capillary pressure at the saturation for which
the mobilities are computed (it is assumed to be positive), or the entry pressure for gas
gravity drainage. As before Leverett J-function scaling is assumed:
r
φm
Pcpqm (Spm ) = J(Spm ) (5.10)
Km
As in Chapter 4, H is the distance through which the displacing phase actually trav-
els vertically. While this is simply the matrix block height, L, for water displacement
controlled by gravity and capillary forces, for gas gravity drainage this may be smaller
than the total height of the matrix block, since the wetting phase (oil and/or water) is
retained in capillary/gravity equilibrium at the base of the block:
r
φm
σgo J ∗V = ∆ρgo g(L − H) (5.11)
Km
The displacement rate for gravity drainage is zero if the gas cannot enter the matrix:
for a block of height L we set βgo = 0 if:
98
r
φm
σgo J ∗ ≥ ∆ρgo gL (5.12)
Km
For horizontal displacement, Lc is related to the shape factor. As in Chapter 4, we use

an expression due to Zhang et al.(1996) - see Eq. (4.20).
The model does appear to have quite a bewildering array of terms at first sight, but
this is necessary to capture, correctly, the complexities of all types of fracture/matrix dis-
placement. It is easiest to explain how to determine all the terms, there are no adjustable
parameters through some example test cases.
5.2 Test cases
5.2.1 Base case study for the general transfer function
The general behaviours of both water-wet and mixed-wet porous media should not be
so different except the final water saturation in the matrix. They will all end at the zero
capillary pressure point. So I chose a mixed-wet case horizontally as the base case to
make sure the general transfer functions for multiphase flow works.
On the basis of the former tests of the Chapter 4, I tested fluid expansion, capillary
imbibition and gas/oil gravity drainage in one dimension. Here I will extend the test
cases to consider capillary imbibition and gravity drainage acting together and both one
and two-dimensional systems.
I designed a horizontal model as the base case for further studies in this chapter. The
matrix porosity and permeability were 0.2 and 10−13 m2 (100 mD) respectively. The
relative permeabilities were krw = Sw2 , kro = So2 , the capillary pressure was used as
following equation (see Fig. (5.1))
r
φm 1
Pcowm = 0.3σ ( − 2) (5.13)
Km S w
The matrix block had a cross-sectional area of 1 m2 and lengths of 1 m and 10 m.
99
180000
Capillary pressure (Pa)
110000
40000
0.0 0.2 0.4 0.6 0.8 1.0

Water saturation
-30000
Figure 5.1: Capillary pressure curve for the mixed-wet model from Eq. (5.13)
Table 5.1: Parameters used for the mixed-wet case

Case1
Initial matrix pressure (P a) 1, 100, 000
Initial f racture pressure (P a) 1, 100, 000
Initial matrix water saturation 0.5
Initial f racture water saturation 1.0
3
W ater density at initial matrix pressure (kg/m ) 1, 000
Oil density at initial matrix pressure (kg/m3 ) 600
W ater compressibility (1/P a) 10−9
Oil compressibility (1/P a) 10−9
W ater viscosity (P a.s) 0.001
Oil viscosity (P a.s) 0.001
First we consider a base case, which is a test of the original general transfer function
presented in Chapter 4. The system is horizontal and so there are no gravitational
forces. Capillary forces allow a final saturation of 0.5 ( see Eq. (5.15)) to be reached: in
equilibrium Pcowm = 0.
Fig. (5.2) and (5.3) show the base case results: the system is horizontal with no
gravitational forces. The average water saturation in the matrix as a function of time
is shown. In this and all other figures, the simulation results are shown as a solid line,
the predictions of the general transfer function as a dashed line, while predictions using
the conventional Kazemi et al (1976) model are dotted. Note that, with no adjustable
parameters, the general transfer function predicts the early and late-time behaviour; in
100
contrast the conventional model under-predicts recovery at early time.
101
0.5
Kazemi et al.
0.4

Oil recovery in the matrix
0.3
0.2
Simulation
0.1
0.0
0.1 10 1000 100000
Time (s)
Figure 5.2: Base case for a block length of 1 m. The general transfer function gives
accurate predictions with no adjustable parameters.
0.5
Simulation
0.4
0.3
Kazemi et al.
0.1
0.0
10 1000 100000 10000000
Time (s)
Figure 5.3: Base case for a block length of 10 m. Note the much slower recovery than
the 1 m model. The conventional, Kazemi, model seriously under-estimates the fluid
recovery at early time since it computes transfer at an average potential difference.
102
5.2.2 One-dimensional and two-dimensional model tests with
capillary pressure and gravity
We consider three different boundary conditions, illustrated in Fig. (5.4): case 1 is

one-dimensional where water imbibes against gravity and the capillary pressure is zero
at the base of the matrix block - there is a water-filled fracture at the base and an oil-
filled fracture at the top; Case 2 which is also one-dimensional but with a zero capillary
pressure at the top of the block, which means that water displaces oil helped by both
capillary and buoyancy forces; and Case 3 which is two-dimensional where water can also
enter the matrix from fractures along the sides.
Pc=0 Pc=0
Pc=0
(a) (b) (c)
Figure 5.4: The three cases considered in this section. The initial conditions are shown:
red represents oil in the matrix, while blue represents water in the fracture. The frac-
ture saturation and pressure remains constant and there is no capillary pressure in the
fractures. (a) Case 1: the system is one-dimensional. The oil and water pressures are
the same at the base of the matrix block of height L. The capillary pressure at the top
of the block is initially (ρw − ρo )gL and water imbibes against gravity. (b) Case 2: the
system is one-dimensional and the water and oil pressures are the same at the top of the
matrix block. Water has a higher pressure at the base of the block, the capillary pressure
is initially −(ρw − ρo )gL, and water displaces oil helped by both capillary and buoyancy
forces. (c) Case 3 is similar to Case 2, but the system is two-dimensional with imbibition
from the sides of the matrix block.
Table 5.2 lists the end-point saturation for the different models used. The matrix block
always has a width of 1 m and two different heights are considered: 1 m and 10 m.
103
The matrix block is homogenous with a permeability of 10−13 m2 and a porosity of 0.2.
The initial matrix water saturation is 0 and the fracture water saturation is held at 1
constantly. The water and oil viscosities are 1 mP a.s. The reference, datum, pressure
is set to 1.1 mP a where the capillary pressure is zero. At the datum pressure, the
water density is 1, 000 kg · m−3 and the oil density 600 kg · m−3 . The oil and water
compressibilities are 10−9 P a−1 . The values are listed in the Table 5.1. Table 5.2 gives
the end-point and typical water saturations for each case.
Table 5.2: End-point saturations computed for the mixed-wet cases

Case1 Case2 Case3
Lc (m) N/A N/A N/A
∗H
End − point saturation Sw ( H ) N/A N/A 0.50
∗V
End − point saturation Sw f or L = 1 m ( H/G ) 0.478 0.525 0.525
∗V
End−point saturation Sw f or L = 10 m ( H/G ) 0.354 0.834 0.834
T ypical saturation SwV f or L = 1 m ( H/G ) 0.484 0.512 0.512
V
T ypicalsaturation Sw f or L = 10 m ( H/G ) 0.427 0.667 0.667
H – horizontal capillary forces only; H/G – gravity and capillary forces together.
Predictions using the general transfer function were compared with explicit simulation
of the fracture/matrix transfer.
For the simulations, in one dimension, 42 blocks were used; the blocks were refined near
the fracture. For the 1 m grid block, the fracture and the first matrix grid block sizes
were set to 0.19622117 cm, then we use equation zn = zn−1 · e0.1 (z is the grid block height
and n is the number of the grid blocks size) to increase the grid size gradually from the
fracture. For the 10 m block, the first grid blocks size for fracture and matrix were set to
3.527724336 cm and zn = zn−1 · e0.08 was used to assign the grid sizes. A number of grid
refinement studies confirmed that sufficient blocks were used to obtain converged results.
For the two dimensions case, a grid of 44 × 44 was used. As in one dimension, the
grid was refined near the fractures as shown in Fig. (5.5). Again, grid refinement studies
showed that this number of blocks is sufficient to obtain converged results.
104
Distance (cm) Distance (cm)
0 20 40 60 80 100 0 20 40 60 80 100
- 20 - 20
- 40 - 40
Depth (cm)
- 60
Depth (cm)
- 60
- 80
- 80
-100
- 100
Figure 5.5: Grid blocks for one and two-dimensional models of height 1 m, the grid was
refined near the fractures to better capture the early-time behaviour.
We use quadratic relative permeabilities in the matrix:
2
krwm = Swm ; krom = (1 − Swm )2 (5.14)
A mixed-wet system is considered with the following capillary pressure:
r
∗ φm 1
Pcowm = J σow ( − 2) (5.15)
Km S w
where the interfacial tension σow is 50 mN/m, J ∗ = 0.3, the capillary pressure is zero
when the water saturation in the matrix is 0.5.
Capillary and gravity forces are chosen to be of similar magnitude. The gravitational
potential difference is approximately 4×103 P a for L = 1 m and 40×103 P a for L = 10 m.
This compares to an end-point capillary pressure, at Swm = 1, of around −2 × 104 P a.
In all Cases S ∗H = 0.5 and H = L. The final water saturation in capillary/gravity
equilibrium, S ∗V , was computed as follows. For Case 1:
Z L
1 a L
Sw∗V = −1
Pcowm (ρw − ρo )gzdz = ln(1 + ) (5.16)
L 0 L 2a
where a is a representative height of the transition zone:
105
r
J ∗ σow φm
a= (5.17)
(ρw − ρo )g Km
which has a numerical value of 5.4 m in all our examples. For Cases 2 and 3, the capillary
pressure is negative in equilibrium and so:
a L
Sw∗V = − ln(1 − ) , L≤a (5.18)
L 2a
a
Sw∗V = 1 − (1 − ln 2) , L>a (5.19)
L
The numerical values are shown in Table 5.2
5.2.3 Results
For Case 1, the initial pressure difference between fracture and matrix at the bottom
of the block is zero. The capillary pressure pushes the water into the matrix to overcome
gravity. This means that the final recovery is lower than for the base case. The water
imbibition stops once it reaches gravity and capillary equilibrium.
Fig. (5.6) shows the predicted and simulated recovery for a block of height 1 m. The
recovery is less than 0.5, which is the end of imbibition point if there is no gravita-
tional force. In this case, imbibition stops when the capillary pressure is equal to the
gravitational potential.
106
0.5

0.4
0.3
Simulation
0.2
0.1
Kazemi et al.
0.0
1 100 10000
Time (s)
Figure 5.6: Case 1 for a block height of 1 m. The ultimate recovery is slightly lower than
for the base case - see Fig. (5.2). The general transfer function gives good prediction for
the recovery at early time, but the conventional model under estimates the recovery.
Fig. (5.7) shows the recovery for a block of height 10 m, capillary and gravitational
forces are of similar magnitude and the recovery is significantly lower and slower than
recovery in the base case, since the gravitational forces are significant.
The general transfer function gives good predictions since it considers the capillary
pressure and gravitational forces separately. The conventional model does not give the
correct ultimate recovery and rate, since it computes transfer at a rate governed by a
difference in average potential.
107
0.5
0.4

0.3
0.2 Simulation
0.1
Kazemi et al.
0.0
1000 100000 10000000
Time (s)
Figure 5.7: Case 1 for a block height of 10 m. It has much slower recovery than the 1 m
case in Fig. (5.4), since the capillary pressure has to overcome large gravitational forces
in this case. The conventional, Kazemi, model seriously under-estimates recovery since
it computes transfer at an average potential difference.
In Case 2, the pressure difference between the fracture and matrix at the base is (ρw −
ρo )gh (3924 P a for the 1 m case and 39240 P a for the 10 m case). The final recovery is
higher than the base case and Case 1, since gravity acts with capillary forces together,
so there is forced water displacement at a negative capillary pressure.
Fig. (5.8) shows the results for a 1 m height matrix block. The general transfer function
gives accurate predictions, but the conventional model underestimates the recovery at
early time and the final recovery is not correct.
108
0.6
0.5

Simulation
0.4
0.3
0.1
Kazemi et al.
0.0
10 1000 100000
Time (s)
Figure 5.8: Case 2 for a block height of 1 m. In this example buoyancy aids recovery,
giving a higher final recovery than the base Case - Fig. (5.2). The general transfer
function gives excellent predictions, but the conventional model’s result is rather poor,
with an erroneous estimate of ultimate recovery.
Fig. (5.9) gives the results for the 10 m height matrix block. The general transfer
function can predict the oil recovery in the matrix precisely. The conventional model
does not necessarily give the right results.
109
1.0
0.8
0.6
0.4
Simulation
0.2
Kazemi et al.
0.0
1000 100000 10000000
Time (s)
Figure 5.9: Case 2 for a block height of 10 m. Note again that recovery is slower than for
a smaller block -Fig. (5.6). In comparison with Fig. (5.5) for flow against gravity, where
the final recovery is better, but the recovery rate is slower.
Note, however, that the recovery rate is lower than when gravitational forces oppose
each other. This is because the typical mobilities are slightly lower in this case. These
examples give accurate agreement with the predictions from the simulator. The general
transfer function can give good predictions for vertical models with different boundary
conditions.
In Case 3, we consider two-dimensional displacement with capillary and gravitational
forces acting together. The capillary pressure will dominate the recovery process by
water imbibition from the bottom and both sides of the model by overcoming the gravi-
tational forces. However once the capillary pressure becomes negative, gravity drainage
will dominate the recovery process.
Fig. (5.10) shows the oil recovery in the matrix for the 1 m case. The recovery is more
rapid than the corresponding one-dimensional Cases( Fig. (5.5) - (5.8)), since water can
invade from three sides of the block. The general transfer function gives a good agreement
with the simulation, but the conventional model underestimates the recovery.
110
0.6
0.4
Simulation

0.2
Kazemi et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.10: Case 3 for a block height of 1 m. Compared to a one-dimensional displace-

ment, recovery is faster because water can imbibe from the sides of the matrix block.
When we have a tall matrix block, Fig. (5.11), we see a remarkable recovery trend.
There is a period of rapid imbibition, controlled largely by the large fracture/matrix
surface along the sides, to a saturation of around 0.5 that corresponds to the saturation at
zero capillary pressure. There subsequently follows a slower recovery to a final saturation
of around 0.83, this is a gravity drainage displacement, where water displaces oil against
capillary forces, driven by the density difference between the phases.
The general transfer function is able to make excellent predictions of the recovery trend
over six orders of magnitude in time with no adjustable parameters. The conventional
model cannot give good predictions either at early or late times. To predict this behaviour,
it is necessary to have a model that splits the contributions of horizontal and vertical
displacement.
111
1.0
Simulation
0.8
0.6
0.4
Kazemi et al.
0.2
0.0
10 1000 100000 10000000
Time (s)
Figure 5.11: Case 3 for a block height of 10 m. Notice the remarkable recovery profile:
initially there is rapid imbibition from three sides of the block to a saturation of around 0.5
where the capillary pressure is zero, followed by a slower drainage displacement against
capillary forces to the final saturation of around 0.83. The general transfer function gives
a good prediction of recovery over six orders of magnitude in time.
5.2.4 Comparison with other models
For a more exhaustive comparison of our method, the results for gravity-controlled
displacement - where we see some errors - were compared with other models in the
literature.
There are three different transfer functions in the Eclipse model we used: Kazemi
(defaulted which has been tested already); other two are used for fluid exchange due to
gravity modeled by including either keywords GRAVDR (Sonier,1986) or GRAVDRM
(Quandalle and Sabathier, 1989).
The Kazemi model is used as the default to calculate the transmissibility calculations,
while the Sonier (1986) and Quandalle and Sabathier (1989) equations are used for the
fluid exchange between the fracture and matrix due to gravity. by introducing a non-zero
value of DZMTRX in the GRID section (Dual porosity, page 152-158, Eclipse Manual)
Three cases were tested individually to compare the differences and tried to challenge
the general transfer function. Here some example results are presented. The general
transfer function gives more robust results than other equations tested for the case stud-
112
ied.
Fig. (5.12) shows that the results of the 1 m Case1. We adjusted the shape factor
in the Sonier model to match either the late or early time behaviour - the method was
unable to capture the recovery for all times.
0.5
Sonier-1
0.4
General transfer function Sonier-2

0.3
Simulation
0.2
0.1 Kazemie et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.12: Case 1 for a block height of 1 m. Sonier-1 uses 0.4 times sigma, and Sonier-2
uses 0.13 times sigma, where sigma is the Kazemi shape factor.
Fig. (5.13) shows the results from Quandalle and Sabathier equation (key word: GRAV-
DRM). Three different curves show with different tuning parameters for the shape factors
by QS-1 (0.4 times sigma); QS-2 (0.1 times sigma); QS-3 (0.05 times sigma). It shows
that it could match the initial or final recovery by tuning the shape factors, but it can
not capture the full recovery behaviour.
113
0.5

0.4
0.3 Simulation
Saturation
QS-1 QS-3
0.2
QS-2
0.1
Kazemi et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.13: Case 1 for a block height of 1 m.QS-1, QS-2 and QS-3 represent that sigma
times 0.4, 0.1 and 0.05 separately.
Fig. (5.14) shows Case2 of 1 m tested using the Sonier equation. It gives accurate
early time and final recovery by a tuned shape factor. Sonier-1, Sonier-2 and Sonier-3
represent same sigma/same DZMTRX (1 m); same sigma/different DZMTRX (3.2 m)
and 0.13 times sigma/different DZMTRX (3.2 m) separately. Again, it can capture the
early or final recovery by tuning parameters; but the general transfer function gives better
predictions without adjustable parameters.
114
0.6
0.5 Sonier-2
Sonier-1
0.4
0.3
Sonier-3
0.2
Simulation
0.1
Kazemi et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.14: Case 2 for a block height of 1 m. Sonier-1 is same sigma, same DZMTRX;
Sonier-2 is same sigma, different DZMTRX (3.2 m); Sonier-3 is 0.13 times sigma, different
DZMTRX (3.2 m).
Fig. (5.15) shows Case2 of 1 m tested using the Quandalle and Sabathier equation. It
gives accurate early time and final recovery by a tuned shape factor. QS-1, QS-2 and QS-
3 represent same sigma/same DZMTRX (1 m); same sigma/different DZMTRX (3.2 m)
and 0.13 times sigma/different DZMTRX (3.2 m) separately. Again, it can capture the
early or final recovery by tuning parameters, but once again the general transfer function
gives better results.
115
0.6
0.5
QS-2
QS-1
0.4
QS-3
0.3
Simulation
0.2
0.1
Kazemi et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.15: Case 2 for a block height of 1 m. QS-1 is same sigma/same DZMTRX (1 m);
QS-2 is same sigma/different DZMTRX (3.2 m); QS-3 is 0.13 times sigma/ different
DZMTRX (3.2 m).
Fig. (5.16) shows Case3 of 1 m tested using Sonier equation. It gives accurate early
time or final recovery by tuning the DZMTRX. Sonier-1 and Sonier-2 represent same
sigma/same DZMTRX (1 m); same sigma/different DZMTRX (3 m) separately. Again
we can match either the early or the late time behaviour, but not both.
116
0.6
0.5
Sonier-2
0.4
Sonier-1
Simulation
0.3
0.2
0.1
Kazemi et al.
0.0
1 100 10000
Time (s)
Figure 5.16: Case 3 for a block height of 1 m. Sonier-1 uses the same sigma, same
DZMTRX (1 m); Sonier-2 is same sigma, different DZMTRX (3 m).
Fig. (5.17) gives Case3 of 1 m tested using the Quandalle and Sabathier equation. It
gives accurate early time and final recovery by tuning the DZMTRX. QS-1, QS-2 and QS-
3 represent same sigma/same DZMTRX (1 m); same sigma/different DZMTRX (3 m)
and 0.5 times sigma/different DZMTRX (3 m) separately. Once again only the general
transfer function gives reliable results for all time.
117
0.6
0.5
QS-2
0.4
QS-1
0.3 Simulation
0.2
QS-3 General transfer function
0.1
Kazemi et al.
1 100 10000 1000000

Time (s)
Figure 5.17: Case3 for a block height of 1 m.QS-1, QS-2 and QS-3 represent same
sigma/same DZMTRX (1 m); same sigma/different DZMTRX (3 m) and 0.5 times
sigma/different DZMTRX (3 m) separately.
I also run many other cases. Overall, the general transfer function shows better perfor-
mance than the other equations. As a last example, Fig. (5.18) shows a 10 m block for
case3. The three equations (Kazemi et al., Sonier et al. and Quandalle ane Sabathier)
cannot capture the two transfer rate behaviours. The general transfer function gives
accurate predictions.
118
1.0
0.8
QS
0.6
Simulation
0.4
Sonier
0.2
Kazemi et al.
0.0
1 100 10000 1000000 100000000
Time (s)
Figure 5.18: Case3 for a block height of 1 m.
5.2.5 Sensitivity studies for the capillary pressure
Capillary pressure is one of the main forces to dominate the recovery in fractured
reservoir. It affects the final recovery and the recovery trend in different ways. So far
only a few test cases have been presented. In this section a more exhaustive set of
simulations is presented to see where the general transfer function works well, where it
fails and how it could be improved.
Recovery study for different capillary pressure curves
The shape of the capillary pressure curve will affect the transfer rate during oil produc-
tion. So I set up several different curves to compare the prediction by different models
for the 1 m mode of Case 1-3.
We double and quadruple the capillary pressure for positive values - see Fig. (5.19).
This will make capillary forces stronger and will increase the imbibition rate.
119
275000
225000
175000
Capillary pressure increase

125000
75000
25000
-25000
Water saturation
Figure 5.19: Capillary pressure curves used to test the general transfer function. The red
curve is the base case, Eq. (5.15). The light and dark blue shows a capillary pressure
that is doubled and quadrupled for its positive values. The negative part is the same
giving the same final recovery.
Fig. (5.20) and (5.21) show the oil recovery curves for Case 1 ( no gravity ) with
block lengths of 1 m and 10 m respectively. The general transfer function gives accurate
recovery predictions, but the conventional model cannot capture the correct behaviour,
as before.
0.5
Simulation
0.4
0.3
0.2
0.1
Kazemi et al.
0.0
1 100 10000 1000000
Time (s)
Figure 5.20: Case 1 for a block length of 1 m. Oil was recovered more rapidly as the
capillary pressure increases. The general transfer function gives good predictions.
120
0.5
0.4

0.3
0.2
Kazemi et al.
0.1 Simulation
0.0
10 1000 100000 10000000
Time (s)
Figure 5.21: Case 1 for a block height of 10 m. The oil was recovered rapidly since
the capillary pressure increases. The final recovery increase since the higher capillary
pressure pushes more oil out. Again, the general transfer function gives good prediction.
Case 2 was tested using the same capillary pressures. Fig. (5.22) and (5.23) show that
the transfer rate increases since the capillary pressure is higher. The general transfer
function can predict the oil recovery in the matrix well. The conventional model again
cannot give good predictions.
0.6
0.5
0.4 Capillary pressure increase

0.3
Simulation
0.1
Kazemi et al.
0.0
10 1000 100000
Time (s)
Figure 5.22: Case 2 for a block height of 1 m. Compared to a one-dimensional displace-

ment, the recovery is faster than Fig. (5.7) because water can imbibe into the matrix
since capillary forces are larger.
121
1.0
0.8 Capillary pressure increase

Simulation
0.6
0.4
0.2
Kazemi et al.
0.0
1000 100000 10000000
Figure 5.23: Case 2 for a block height of 10 m. The imbibition process is slower than in
Fig. (5.15) since the matrix block size is larger.
For Case3, compared to a one-dimensional displacement, the recovery is faster because

water can imbibe into the matrix through the sides. As before, the general transfer
function gives better predictions than the conventional model ( Fig. (5.24)).

Simulation
Kazemi
Time (s)
Figure 5.24: Case 3 for a block height of 1 m with different capillary pressures.
For the 10 m tall block for Case 3 we run a more extensive series of tests, since this is
the most interesting and challenging test, involving the interplay of both capillary and
gravitational forces.
We study four different curves: (1) increasing the positive portion of the capillary
122
pressure curve a factor of two without changing the negative values - see Fig. (5.12);
(2) increasing the positive portion of the capillary pressure by a factor of four without
changing the negative values - see Fig. (5.12); (3) decreasing the negative part of the
capillary pressure by a factor of two without changing the positive portion - see Fig.
(5.25); (4) decreasing the negative portion by a factor of four keeping the positive value
the same - see Fig. (5.24). The first two conditions have been described previously. We
will compare the results to test how the general transfer function works to probe the
strengths and weakness of the model.
250000
Capillary pressure decrease

200000
150000
100000
50000
-50000 Water saturation
-100000
Figure 5.25: Different capillary pressures used. The red represents the base case Fig.
(5.12), while light blue has the negative portion of the curve multiplied by a factor of
two, and for the dark blue line the capillary pressure is four times lower. The positive
portion of the curves remains the same.
Fig. (5.26) shows that the imbibition processes is affected by different shapes of the
capillary pressure curve when the capillary pressure is positive. At early time, increasing
the capillary pressure makes imbibition more rapid. At late time the results are relatively
insensitive to the positive part of the capillary pressure curve, since here the behaviour
is controlled by a gravity drainage process. The conventional model does not give the
right prediction and misses the characteristic double-rate behaviour, as discussed earlier.
The general transfer function gives almost exactly the same predictions as the simulator.
The model is robust for this set of boundary conditions. The final oil recovery will not
123
be affected, since the negative portion of the capillary pressure is the same.
1.0
Kazemi et al.
0.8
0.6
0.4
simulation
0.2
0.0
1 100 10000 100000 100000000
Time (s)
Figure 5.26: Case 3 for a block height of 10 m. The different colours correspond to
different capillary pressure curves shown in Fig. (5.12). The general transfer function
gives accurate predictions in all cases.
Fig. (5.27) shows the cases for the same imbibition process with different capillary
pressure curves. the picture shows that the general transfer function is accurate for the
imbibition parts as before, but at the late time when gravitational forces dominate, the
predictions are less accurate with the general transfer function slightly over-predicting the
recovery. Here the early-time spontaneous imbibition recovery is unaffected by changes
in the negative region of the capillary pressure. At late time, however, where there is
gravity drainage against the capillary force, the recovery is lower as the capillary pressure
is made more neglected.
The results show that the general transfer function gives satisfactory predictions for
the imbibition process, but it is less reliable the gravity dominates the recovery forces.
124
1.0
0.8

Simulation
0.6
Pressure decrease
0.4
0.2 Kazemi et al.
0.0
1 100 10000 1000000 100000000
Time (s)
Figure 5.27: Case 3 for a block height of 10 m. The different colours correspond to the
different capillary pressure curves shown in Fig. (5.25). The general transfer function
gives accurate predictions for the imbibition process, but there is some discrepancy at
late times when gravitational forces dominate.
Relative permeability effects
The exponent of the oil relative permeability affects mobility and transfer rates. We
designed different cases with: (a) krw = Sw2 ; kro = So2 ; (b) krw = Sw4 ; kro = So4 ; (c)
krw = Sw8 ; kro = So8 - see Fig. (5.28).
125
1.0
0.8 (a)
(b)
Relative permeability
0.6
(c)
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Water saturation
Figure 5.28: Matrix relative permeabilities with different exponents:

(a) krw = Sw2 ; kro = So2 ; (b) krw = Sw4 ; kro = So4 ; (c) krw = Sw8 ; kro = So8 .
Fig. (5.29) shows the predictions using the general transfer function, conventional
model and the simulator for the different relative permeability curves for Case 3 with
a 10 m block height. The general transfer function can capture the main trend of the
recovery behaviour but it is not as accurate as the previous cases. This suggests that the
definition of the typical mobility could be refined.
1.0
Exponenent increase
0.8
0.6
0.4
0.2 Simulation
0.0
10 1000 100000 10000000 1000000000
Time (s)
Figure 5.29: The curves for different relative permeability exponents for the 10 m block.
The red colour represents the case we tested before where the exponent is 2, the light
blue has the exponent of 4, and the dark blue represents an exponent of 8.
126
To explore this in more detail we now define the relative permeability with an initial
water saturation in the matrix. Fig. (5.30) gives the relative permeabilities and capil-
lary pressure for this test case. This case can tell if the typical saturation we define is
appropriate for different boundary conditions.
1
725000
0.8
475000
0.6
(Pa)
K rw
Kr
K ro
Pc
Pc
0.4
225000
0.2
0 -25000
0 0.2 0.4 0.6 0.8 1
Sw
Figure 5.30: The relative permeability and capillary curves used for the mixed-wet system;
krw = s2w ; kro = (1 − sw )2 .
Fig. (5.31) shows the prediction for Case 3 with a 10 m block with an irreducible water
saturation. The results are in reasonable agreement with simulation. The conventional
model gives the wrong predictions. We can say that the general transfer function’s
prediction is good, but there is a possibility to improve it by reconsidering the typical
saturation.
127
1.0

0.6
Simulation
0.4
Kazemi et al.
0.2
10 1000 100000 10000000
Time (s)
Figure 5.31: Case 3 with a matrix block of 10 m using relative permeabilities in Fig.
(5.28). The general transfer function gives the correct trend, but shows some disagreement
with simulation.
5.2.6 Sensitivity studies for gas gravity drainage
The previous section has shown that in some cases the general transfer function may
give slightly inaccurate predictions when gravity drainage is the dominant displacement
mechanism. This will be further tested in this section for gas/oil systems with a significant
density contrast.
I used Case 4 in Chapter 6 to do more sensitivity studies for the gas gravity drainage
case. The sensitivity studies can frame the validation of the transfer function, so it will
give the directions for future work.
Oil viscosity
The recovery will be slow if the oil is viscous. I still use the one-dimensional model
presented in Chapter four (Case 4). The datum pressure is defined at the bottom of
model. The pressure difference between the gas and oil phase at the top of the model is
(ρo − ρg )gh (34335 P a). The capillary pressure is 1414 P a. So the gas will drain down
because of the density difference. The ultimate oil recovery will be 0.919 calculated by
Eq. (2.19). In all cases the gas viscosity is 5 × 10−4 P a.s while the oil viscosity is varied.
128
Fig. (5.32) shows that when the oil viscosity is twenty times the gas viscosity, the
gas moves down slower than shown in Fig. (5.31), since the viscous oil provides more
resistance to flow, and the transfer rate is smaller ( Eq. (2.17)). The general transfer
function shows better results, in this case, although it over-predicts recovery at early
time. This is a physically more realistic test case.
1.0
0.8
0.6
Kazemi et al.

0.4
Simulation
0.2
0.0
100 100000 1000000000
Time (s)
Figure 5.32: The oil viscosity is 10−3 P a.s. The general transfer function gives accurate
results at late time and now over-predicts recovery at early times.
Fig. (5.33) gives the oil recovery for a more viscous oil (µo = 10−2 P a.s). The general
transfer function gives the correct ultimate recovery and a similar trend, but the transfer
rate is high at early times.
Fig. (5.34) shows the oil recovery in the matrix when the oil viscosity is 0.1 P a.s. The
transfer rate is decreasing since the oil viscosity increases. Again the general transfer
function gives better predictions than the conventional model, but with too high a rate
at early time.
Fig. (5.35) gives the oil recovery in the matrix with an oil viscosity of 1 P a.s, when
the oil viscosity is much higher than the gas viscosity. The transfer rate is now very low,
but with the same ultimate recovery. Again the general transfer function over-predicts
recovery at early time.
In these cases the transfer rate is inversely proportional to the oil viscosity unless the
129
1.0
0.8
0.6
Kazemi et al. General transfer function
0.4
Simulation
0.2
0.0
1000 1000000 1000000000
Time (s)
Figure 5.33: The oil viscosity is 10−2 P a.s. The general transfer function can give similar
recovery trends to simulators, but the transfer rate is still high at early times.
gas and oil viscosities are similar.
130
1.0
0.8
0.6 Kazemi et al.
0.4
0.2 Simulation
0.0
1000 1000000 1000000000
Time (s)
Figure 5.34: The oil viscosity is 0.1 P a.s, and the gas moves even slowly than in Fig.
(5.27) since the oil mobility is low.
Matrix block height
We studied gravity drainage with block sizes of 10 m, 15 m, 20 m, 25 m, 50 m models.

Fig. (5.36) - (5.40) show that the higher the model is, the slower the recovery. The Kazemi
et al. model does not give the right behaviour for all the models. The general transfer
function gives similar trends to the simulation, but the early-time rate is again over-
estimated with the correct final recoveries. In these cases the recovery rate approximately
scales as 1/L, where L is the block height.
The matrix block height sensitivity studies suggest that the general transfer function
can give a reasonable ultimate recovery prediction for gas gravity drainage for different
matrix block sizes, but over-estimates the early time recovery. The predictions could be
improved by modifying the typical mobility for gravity-controlled processes.
131
1.0
0.8
0.6
Kazemi et al.
0.4
Simulation
0.2
0.1
10000 10000000 10000000000
Time (s)
Figure 5.35: The oil viscosity is 1 P a.s, 20, 000 times the gas viscosity. Gas gravity
drainage becomes very slow in this case.
1.0
0.8 Kazemi et al.

0.6
0.2
Simulation
0.0
1000 100000 10000000
Time (s)
Figure 5.36: The matrix block is 10 m high. This is the base case for studing the effect
of block height on recovery. The general transfer function gives good predictions at late
time, but over-estimates the transfer rate at early times.
132
1.0
0.8
0.6
General transter function
Kazemi et al.
0.4
Simulation
0.2
0.0
1000 100000 10000000 1000000000
Time (s)
Figure 5.37: The matrix block is 15 m high. The general transfer function gives good
predictions at late time, but over-estimates the transfer rate at early times. The recovery
is slower than the recovery in Fig. (5.29).
1.0
0.8
0.6
Kazemi et al. General transfer function
0.4
0.2 Simulation
0.0
1000 100000 10000000 1000000000
Time (s)
133
1.0
0.8
Kazemi et al.
0.6
0.4
Simulation

0.2
0.0
1000 100000 10000000 1000000000
Time (s)
1.0
0.8
Kazemi et al.
0.6

0.4
Simulation
0.2
0.0
10000 1000000 100000000
Time (s)
134
5.2.7 Correct factor of the gas gravity drainage
For imbibition we introduced a boost or correction factor to give the correct early-time
recovery. Our results suggest that a correction factor is required for gravity drainage,
however, it will be a suppression factor to decrease the rate initially.
The gas gravity drainage sensitivity study shows that the general transfer function can
give reasonable predictions, but it does not necessary give accurate results at early time,
where it over-estimates oil recovery. We suggest therefor that we write instead of Eq.
(5.6):
V V ∗V
Tgos = Bgd βgo (Sgim + F (Sgf )(Sgm − Sgim ) − Sgm )b ; H
Tgos =0 (5.20)
Bgd is a correction factor to give the correct early and late-time behavior. Di Donato
et al.(2006) showed, though a combination of numerical, analytical and experimental
analysis, that at early time gravity drainage has linear recovery with time, changing to a
power-law form later. To obtain the same early-time behaviour derived before (Di Donato
et al., 2006) we suggest:
∗V
Sgm − Sgm
Bgd = 1 − (1 − δ) ∗V
(5.21)
Sgm − Sgim
where:
∗V
Km λVgm λVom ∆ρog gSgm
δ= V V ∗V − S b
(5.22)
βgo λtm φm H(Sgm gim )
Bgd is close to 1 at late time, but gives the correct linear scaling of recovery at early
time. I used the new equation to test the gas gravity drainage cases we studied before.
The results are much better especially at early time.
Test cases and results
I compared the results on the basis of the previous work on gas gravity drainage,
presented in the last section, with variation of both block height and oil viscosity. The
135
improvement is significant and the correction factor gives accurate predictions for the gas
gravity drainage at the early time ( Figs. (5.41) - (5.45)).
1.0
0.8
0.6
Before correction After correction
0.4
0.2 Simulation
0.0
1000 100000 10000000
Time (s)
Figure 5.41: Gravity drainage with an oil viscosity of 10−3 P a.s. The matrix block is
5 m high. This figure shows that the prediction is improved using the correction factor
Eq. (5.20), although the late-time predictions are now slightly worse.
1.0
0.8
Simulation
0.6
Before correction After correction
0.4
0.2
0.0
1000 100000 10000000
Time (s)
Figure 5.42: Gravity drainage with an oil viscosity of 10−3 P a.s.The matrix block is 10 m
high.
136
1.0
0.8
Simulation
0.6
After correction
0.4
Before correction
0.2
0.0
1000 100000 10000000 1000000000
Time (s)
Figure 5.43: Gravity drainage with an oil viscosity of 10−2 P a.s. The matrix block is
5 m high.
1.0
0.8
0.6
Simulation
After correction
0.4
Before correction
0.2
0.0
1000 100000 Time (s) 10000000
Figure 5.44: Gravity drainage with an oil viscosity of 0.1 P a.s.The matrix block is 5 m
high.
137
1.0
0.8
0.6 Simulation
After correction
0.4
Before correction
0.2
0.0
1000 100000 10000000
Time (s)
Figure 5.45: Gravity drainage with an oil viscosity of 0.1 P a.s. The matrix block is 5 m
high.
138
Chapter 6
Conclusions and future work
6.1 Conclusions
6.1.1 Development of a general transfer function
The main work of this thesis has been the development and testing of a general transfer
function for dual permeability and dual porosity simulation. This started with simple
linear models for capillary imbibition and gravity drainage, and has culminated in a
general formulation that accounts for expansion and mixed-wet systems.
A spreadsheet has been developed to make predictions using the general transfer func-
tion and Kazemi et al. model. This provides a convenient tool for the user to explore the
behaviour of one and two-dimensional models. It offers insights into the different physical
affects of fluid flow between matrix and fractures. It could be used for teaching, research
and rapid analysis.
6.1.2 Developed a multi-rate model for displacement processes
in fractured reservoirs
At the field scale I have used a dual porosity streamline-based simulator. This model
has been extended to include multi-rate transer functions with block by block variation
139
in the rate used. I use transfer functions for imbibition and gravity drainage based on
analytical solutions to the transport equations, and develop a multi-rate model that is
coupled to the fracture fractional flow. This approach suggests a rigorous method of
upscaling fracture flow using a combination of theoretical analysis, experimental results
and discretized fracture network simulation.
I applied the methodology to an extensively fractured mixed-wet dolomite oil field in
China. I demonstrated that using a single average rate based on the average fracture
density would over-estimate recovery compared to the use of a model that captured the
heterogeneity of the fracture network within each grid block. Coupling to the fracture
fractional flow predicted even lower recovery. For this field, gas gravity drainage was
more favorable than waterflooding, since it had a higher sweep efficiency and ultimate
recovery.
I also presented a preliminary study of the Clair Field in the North Sea. This is
a fractured sandstone reservoir with significant fracture and matrix permeability. My
results suggested that for this field a fine-grid single porosity simulation approach with
an effective permeability that accounts for both fracture and matrix would be appropriate.
6.1.3 Sensitivity studies of the transfer function
The sensitivity studies to test the accuracy of predictions made by the general transfer
function came to the following conclusions. The general transfer function gives good
predictions for different capillary pressure curves for one-dimensional displacements. It
is a significant improvement over predictions using the widely used Kazemi’s transfer
function and has no adjustable parameters. This indicates that the general transfer
function can capture the recovery process properly. For two-dimensional models, the
predictions are also good, although there are some differences with simulation at late
time.
Changing the relative permeability curves indicated that the model gives good predic-
tions for quadratic matrix relative permeabilities. The prediction for other cases are not
140
so accurate, although the recovery trend is successfully matched.
Overall, the general transfer function gives much better predictions than the coventional
Kazemi model. The model works for capillary-dominated displacement.
For gas gravity drainage, an early-time correction is needed to capture the initial be-
haviour. Overall the results are good, but it may be that further refinements in the
definition of typical mobility are required for some cases.
6.2 Future work
There are many ways in which this work can be extended. This includes:
• Combining the multi-rate and general transfer function models to allow a general
multi-rate formulation. This is relatively straight forward to do, although somewhat
algebraically complex.
• Implementing the general transfer function into the streamline-based simulator.
While compressible flow cannot be accommodated in our code, the current linear models
could be extended to incorporate the Vermeulen boost factor and rate constants that
account for capillary and buoyancy forces.
• Field studies using the general transfer function. The two field studies presented in
this thesis were somewhat preliminary in their conclusions and it would be instructive
to show, as part of a more comprehensive study, how the incorporation of an improved
fracture model impacted reservoir management,
•| Comparison with other conventional models (Kazemi et al., Sonier et al., Quandalle
and Sabathier) shows that the general transfer functions give better predictions for the
1-D and 2-D geometries.
• Further tests of the transfer function. While I have run many tests, I have still not
considered three-dimensional models, nor situations with anything other than a simplistic
fracture geometry, nor have I run cases where the fracture saturation changes. All these
cases are examples where different aspects of the current model could be challenged.
141
6.2.1 Upscaling strategy
The main outcome of this research is to integrate my ideas into a fracture-to-field

upscaling strategy. It is difficult to translate the geological and hydraulic description
of fracture networks into reservoir simulation parameters [45]. The dual-porosity model
requires the determination of equivalent fracture permeability and equivalent matrix block
dimensions or shape factors. Matthai has developed the Complex Systems Modeling
Platform (CSP) to model fractured reservoirs using a discrete fracture model, where the
fracture network is explicitly represented [45]. This can be used to find the parameters
for field-scale simulation.
As part of a larger project on fractured reservoirs, we are working on how to upscale
the results of sector or grid-block scale discrete fracture simulations to derive large-scale
properties of the flow, such as the average permeability, relative permeability, mobility
and fracture to matrix transfer. So far, in our work we have only considered very ideal-
ized fracture geometries and relatively simple physical processes. We would like to study
more realistic displacement scenarios with geologically representative fracture networks
to compute effective transport properties for field-scale simulation. These effective prop-
erties could then be used to make predictions of overall recovery with more confidence
than the current empirical models.
As an example of this approach, Fig. (6.1) shows the water saturation distribution from
a discrete fracture model simulation [45]. Fig. (6.2) shows the corresponding average
fracture relative permeability that - in a larger scale model - represents the breakthrough
behaviour. Note that the relative permeabilities are very unlike the linear forms we
usually assume in fractured reservoir simulation. We intend to extend this approach by
also computing the rate at which water transfers from fracture to matrix and inputting
this rate into field-scale simulation. By using fracture networks that are representative
of the field, we hope to be able to develop a fracture-to-field upscaling methodology that
considerable enhances our ability to predict reservoir performance and to design optimal
recovery strategies.
142
(a)
(b)
Figure 6.1: Simulations using a discrete fracture model at water breakthrough after 136
days (a) and at t=462 days during a primary drainage run (b). Iso-surfaces are used to
display fluid pressure (gray translucent) and the Buckley-Leverett shock (green-yellow)
and other saturation values. Note that the water front in (a) is more irregular than the
oil front in (b) [45].
FRAC100
Kri
Water saturaion
Figure 6.2: Relative permeability curves determined from waterflood simulation with the
model above (n = non-wetting = oil) [45].
143
Appendix A
Barenblatt et al.’s analytical solution

for counter-current imbibition
Here we review the analytic solution for counter-current imbibition in one dimension
derived by Barenblatt et al. [5]. Conservation of water volume in one dimension with no
overall flow can be expressed as follows:

∂Sw ∂ λw λo ∂Pc ∂Sw
φ + K =0 (A.1)
∂t ∂x λt ∂Sw ∂x
Eq. (A.1) is written in terms of dimensionless variables: the normalized saturation, S:
Sw − Swi
S= (A.2)
1 − Sor − Swi
and a dimensionless length defined by xD = x/L. The boundary conditions for flow in
the region ≤ xD ≤ 1 are as follows. Continuity of capillary pressure at the inlet face
(Pc = 0) gives S = S ∗ at xD = 0, where S ∗ is the normalized maximum water saturation
after spontaneous imbibition (S ∗ < 1). The outlet is a no-flow boundary, which implies
that ∂S/∂xD = 0 at xD = 1. The conservation equation, assuming uniform absolute
permeability, is then:
1
!
∂S K ∂ λw λo ∂Pc ∂S
+ 2
· · · =0 (A.3)
∂t φ(1 − Swi − Sor )L ∂xD λt ∂S ∂xD
Rather than attempt a solution of the non-linear Eq. (A.3) directly, we will construct
a solution of the weak or integral form of the equation. Integrating Eq. (A.3) over the
domain and noting that the flux vanishes at xD = 1 yields:
!
∂S K λw λo ∂Pc ∂S
= · · (A.4)
∂t φ(1 − Swi − Sor )L2 λt ∂S ∂xD

xD =1
where the mean saturation S is defined by
Z 1
S= SdxD (A.5)
0
Following Eq. (2.9):
r
∂Pc φ 0 1 ∂Pc
= −σow J = · (A.6)
∂Sw K 1 − Swi − Sor ∂S

xD =0 S=S ∗
Then using Eqs. (A.4) and (A.6):
s !
∂S K σow J 0 λ o λw ∂S
=− · (A.7)
∂t φ L2 λt ∂xD

xD =0
Next, we define a dimensionless time tD :
s !
K σow λo λw
tD = t (A.8)
φ L2 λt
S=S ∗
and Eq. (A.8) take the form:

∂S ∂S
= −J 0 (A.9)

∂tD ∂xD

xD=0
2
Early-time solution
First we derive an early-time solution, valid until the advancing water front reaches the
far boundary. We assume a quadratic form for the saturation profile:

S ∗ − A(tD )xD + B(tD )x2D , xD ≤ x0D ,


S(xD , tD ) = (A.10)
x0D

0
 xD ≥
where x0D (tD ) is the distance that water has moved into the block at time tD . This form
automatically obeys the boundary condition at xD = 0. By definition, the quadratic
function in Eq. (A.9) must vanish at x0D (tD ), so that the saturation profile is continuous.
We also require that the saturation gradient is continuous at x0D (tD ). These two conditions
allow us to find the time-dependent coefficients A and B.
" ! !2 #
xD xD
S(xD , tD ) = S ∗ 1−2 0 + (A.11)
xD x0D
and S = 0 for x0D ≤ xD ≤ 1. It is then possible to derive:
p
x0D (tD ) = 12J 0 tD (A.12)
r
4J 0
S(tD ) = S ∗ tD (A.13)
3
Both the penetration distance and the mean saturation grow as the square root of
the dimensionless time. This solution is valid for tD ≤ tD1 , where tD1 is the time at
which the water first reaches the far (closed) boundary. This time is found by setting
1
x0D (tD = tD1 ) = 1 , which gives tD1 = 12J 0
.
Late-time solution
At late times, tD ≥ tD1 , we again assume a second-order polynomial for the saturation
profile, Eq. (A.10). The no-flux condition at the far end implies that the saturation
3
gradient vanishes at xD = 1, which leads to the condition 2B(tD ) = A(tD ). The average
saturation is:
S(tD ) = S ∗ − A(tD )/3 (A.14)
We then use Eq. (A.9) to derive a differential equation for A that is solved to find:
A(t) = A(tD1 )e−η(tD −tD1 ) (A.15)
where η = 3J 0 and A(tD1 ) is a constant whose value is determined by requiring that

the early-time and late-time solutions coincide when tD = tD1 . Equating the early time
solution and the late time solution at tD = tD1 leads to A(tD1 ) = 2S ∗ . Hence:
" #
2
S(tD ) = S ∗ 1 − e−η(tD −tD1 ) (A.16)
3
For simplicity, when we derive a transfer function, we ignore the early time behaviour,
although later we introduce a correction factor in Section 5.2.6. Identifying the average
saturation with the recovery, we find Eq. (2.7) where the constant A is now found to
ensure that S(tD = 0) = R(0) = 0 .
The crucial issue in this derivation is the boundary condition at the inlet. We assume
that the system is not strongly water-wet and so the oil mobility is finite at the end of
imbibition. It the system is strongly water-wet and S ∗ = 1, the functional form of the
recovery is very different [62; 66]. This is a subtlety overlooked in the original analysis
[5].
4
Appendix B
Numerical implementation
We first solve for an intermediate saturation in the fractures ignoring transfer using
single point upstream weighting:
int n ∆t
Swf j = Swf j − (fwf (snwf j ) − fwf (Swf
n
j−1 )) (B.1)
∆τ
where j labels the grid block and n labels the time level. ∆t is the time-step size and is
chosen to ensure that solutions to Eq. (B.1) are stable and accurate, as in single porosity
simulation. In streamline simulation many time-steps may be taken to transport fluid
along a streamline before the pressure field is recomputed [7].
Single-rate model
n+1 n+1 n n+1

For T given by Eq. (2.14), if Swmj = 0 then T = 0 and Swmj = Swmj ; Swf j = 0.
n+1
If Swf j > 0 we can solve for the saturation at the next time level analytically from its
previous value using Eq. (2.12):
n+1
Swmj = Sw∗ − (Sw∗ − Swmj
n
)e−β∆t (B.2)
n+1 int φm n+1 n

Swf j = Swf j − (S − Swmj ) (B.3)
φf wmj
5
n+1 n+1
If Swf j < 0 set Swf j = 0 and find the matrix saturation that is consistent with this
and mass balance:
n+1 n φf int
Swmj = Swmj + S (B.4)
φm wf j
Single-rate model for gravity drainage
int n+1 n n+1

For T given by Eq. (2.23), again if Sgf j = 0 then T = 0 and Sgf j = Sgmj ; Sgf j = 0.
int
If Sgf j > 0 we can solve for the saturation at the next time level analytically using Eq.
(2.21):
1
! a−1
n+1
Sgmj = Sg∗ − (Sg∗ − Sgmj
n
)1−a + α∆t (B.5)
n+1 int φm n+1 n

Sgf j = Sgf j − (S − Sgmj ) (B.6)
φf gmj
n+1 n+1
If Sgf j < 0 set Sgf j and find the matrix saturation that is consistent with this and
mass balance:
n+1 n φf int
Sgmj = Sgmj + S (B.7)
φm gf j
Simple multi-rate model
int n+1 n+1

Again if Swf j = 0 then T = 0 and Swmkj ;Swf j = 0. If we use expressions similar to
Eq. (B.2)
∗ ∗
n+1
Swmkj = Swk − (Swk n
− Swmkj )e−βk ∆t (B.8)
N
n+1 int
X φmk n+1 n
Swf j = Swf j − (Swmkj − Swmkj ) (B.9)
k=1
φf
n+1 n+1
If Swf j < 0 set Swf j = 0 and find the matrix saturations that are consistent with this
6
and mass balance. There are a number of ways to do this, we choose a simple approximate
approach. Define:
N
X
∗ n −βk ∆t
∆Swmkj = (Swk − Swmkj )(1 −e ) ; φf ∆SwmjT = φmk ∆Swmkj (B.10)
k=1
∆SwmjT
F = int
; F >1 (B.11)
Swf j
Then:
n+1 n ∆Swmkj
Swmkj = Swmjk + (B.12)
F
Simple multi-rate model for gravity drainage
int n+1 n n+1 int

Again if Sgf j = 0 then T = 0 and Sgmkj = Sgmkj ; Sgf j = 0. If Sgf j > 0 we use
expressions similar to Eq. (B.5):
! a 1−1
k
n+1 ∗ ∗ n
∆Sgmkj = Sgk − (Sgk − Sgmkj )1−ak + αk ∆t (B.13)
N
n+1 int
X φmk n+1 n
Sgf j = Sgf j − (Sgmkj − Sgmkj ) (B.14)
k=1
φf
n+1 n+1
If Sgf j < 0 set Sgf j = 0 and find the matrix saturation that are consistent with this
and mass balance. Again:
! a 1−1
k
∗ ∗
∆Sgmkj = Sgk − sngmkj − (Sgk n
− Sgmkj )1−a + αk ∆t (B.15)
N
X
φf ∆SgmjT = φmk ∆Sgmkj (B.16)
k=1
∆SgmjT
F = int
; F >1 (B.17)
Sgf j
7
Then:
n+1 n ∆Sgmkj
Sgmkj = Sgmkj + (B.18)
F
Fracture fractional flow model
The numerical implementation is more involved, since we need to check each fracture
int n+1
saturation Swk and invoke mass balance. Again if Swf j = 0 then T = 0 and Swmkj =
n n+1
Swmkj ; Swf j = 0 . For a general model we use an iterative approach to solving the
transport equations. We find the fracture saturation at time level n+1 that is consistent
with Eqs. (2.28) - (2.30):

 n+1
n
Swmkj , Swf j ≤ Swf k−1j





S n+1 −Swf k−1j

 n h i
wf j ∗ −S
n+1 Swmkj +βk ∆t (Swk wmi )+Swmi
Swmkj = Swf kj −Swf k−1j
n+1
, Swf kj > Swf (B.19)

 1+βk ∆t j > Swf k−1j


 Sn ∗
+βk ∆tSwk
 wmkj n+1
j ≥ Swf kj


1+β ∆t
k
, Swf
N
n+1 int
X φmk n+1 n
Swf j = Swf j − (Swmkj − Swmkj ) (B.20)
k=1
φk
n+1
Our initial estimate for the fracture saturation at time level n + 1 in Eq. (B.19) is Swf j
n+1 n+1
. We then use this to find another value of using Eq. (B.20). If Swf j ≤ 0 then set Swf j
n+1
to be half its previously estimated value. The updated value of Swf j is then put into
Eq. (B.19) to find a new estimate. We continue to iterate until we reach a converged
solution. Occasionally we find convergence problems. If the solution fails to converge
then we halve the time step.
Fracture fractional flow model for gravity drainage
int n+1 n n+1

As before if Sgf j = 0 then T = 0 and Sgmkj = Sgmkj ; Sgf j = 0 . We find the fracture
saturation at time level n + 1 that is consistent with Eqs. (2.30) - (2.31):
8
n+1
Sgf j − Sgf k−1j
f= (B.21)
Sgjkj − Sgf k−1j

n n+1

Sgmkj , Sgf j ≤ Sgf k−1j






n+1 ∗
Sgk 1
Sgmkj = − 1 ∗
(Sgk n
− f Sgmkj )1−ak + αk f ∆t ak −1 n+1
, Sgf kj > Sgf
 f f j > Sgf k−1j



∗ ∗ n
1 n+1
)1−ak + αk ∆t ak −1

Sgk
 − (Sgk − Sgmkj , Sgf j ≥ Sgf kj
(B.22)
N
n+1
X φmk
Sgf j = int
Sgf j − (sn+1 n
gmkj − Sgmkj ) (B.23)
k=1
φk
Our initial estimate for the fracture gas saturation at time level n + 1 in Eq. (B.21) is
n+1 int n+1 n+1
Sgf j = Sgf j . We find another value of Sgf j using Eqs. (B.22) and (B.23). If Sgf j ≤ 0
n+1 n+1
then set Sgf j to be half its previously estimated value. The updated value of Sgf j is
then put into Eq. (B.21) and we iterate until we reach a converged solution.
9
Appendix C
Downscaling permeability data
To generate a finer fracture permeability dataset (with 4, 488, 000 grid blocks) from
SP E 10th and (with 13, 000 active blocks) from the Liu7 oil field, a downscaling technique
that preserves the geometric average of the permeability was employed. Each block
was divided into four parts, by dividing the block in half in the x and y directions.
Four random numbers (x1 , x2 , x3 , x4 ) were generated uniformed in the range 0 − 1. The
permeabilities of the downscaled grid block were then computed as follows:
Ki = K · eα(xi −x) , i = 1 , 2, 3, 4 (C.1)
1
x = (x1 + x2 + x3 + x4 ) (C.2)
4
where K is the original permeability of the block before downscaling and Ki is the new
permeability for each new part. α is a parameter to be determined. Here it is set to 2.5.
For grid blocks with a permeability K = 1 mD (the minimum allowed fracture perme-
ability), Ki = 1 mD. This method ensures that the geometric average permeability is the
same as the original field. The parameter a is a measure of the small-scale randomness
in the permeability.
10
Appendix D
Calculation of the average saturation

in gravity/capillary equilibrium
1
725000
0.8
475000
0.6
(Pa)
K rw
Kr
K ro
Pc
Pc
0.4
225000
0.2
0 -25000
0 0.2 0.4 0.6 0.8 1
Sw
Figure D.1: The capillary pressure curves used for the mixed-wet system, S∗ represents
the final water saturation.
r
φm 1
Pcowm = 0.3σow ( − 2) (D.1)
Km S w
When it reaches the gravity and capillary equilibrium, water saturation in the matrix
11
reach its final value, then
r
φ 1
(ρw − ρo )gz = 0.3 σow ( − 2) (D.2)
K Sw
a equation obtained from Eq.D2:
s
1 ∆ρow gz K
=2+ (D.3)
Sw σow 0.3 φ
1
Sw = (D.4)
az + 2
q
∆ρow g K
where a = 0.3σow φ
, so the final average water situation is:
Different boundary condition will put into different cases,
Case 1, it reaches the final saturation before Pc = 0, so the capillary pressure is positive:
Z L
1 1 1 L 1 aL
Sw∗V = dz = ln(az + 2) 0 = ln( + 1) (D.5)
L 0 az + 2 aL aL 2
Case 2 and Case 3, it reaches the final saturation when the capillary pressure is negative:

−L
Sw∗V = − 1 −L 1 1 1 aL
 R

L 0 az+2
dz = − aL ln(az + 2) 0 = − aL ln(1 − 2
) , L≤a
(D.6)
Sw∗V = 1 − a (1 − ln2)


L
, L>a
12
Appendix E
Input deck of the Base Case

(mixed-wet) for Eclipse model
RUNSPEC
TITLE Base Case for the General Transfer Function Comparison
START 1 ’APR’ 2006 /
LAB
FMTOUT
UNIFOUT
OIL
WATER
MONITOR
RSSPEC
NOINSPEC
DIMENS
42 1 1/
EQLDIMS
2 100 100 1 20 /
REGDIMS
2 1 0 0/
13
TABDIMS
2 1 21 20 2 20 20 1/
WELLDIMS
1 1 1 1/
GRID
GRIDFILE
2/
INIT
GRIDUNIT
– – Grid data units –
’CM’ /
MAPAXES
– – Grid Axes wrt Map Coordinates
0 0 0 0 0 0/
DX
0.1 0.1 0.1 0.112951338 0.127580047 0.14410337 0.162766684 0.183847146 0.207657811 0.234552275
0.264929933 0.299241903 0.337997732 0.38177296 0.431217665 0.487066121 0.550147699 0.621399185
0.701878692 0.792781372 0.895457164 1.011430845 1.14242467 1.290383946 1.457505929 1.646272443
1.859486747 2.100315154 2.372334062 2.679583058 3.026624908 3.41861332 3.861369476 4.361468476
4.926336986 5.564363524 6.285023034 7.099017591 8.018435331 9.056929968 10.22992355 11.5548355 /
DY
42*100 /
DZ
42*100 /
TOPS
42*10 /
PERMX
5000 5000 40*100/
PERMY
5000 5000 40*100 /
PERMZ
5000 5000 40*100 /
PORO
2*1 40*0.2 /
14
MULTPV
2*10E+18 40*1 /
PROPS
– OFFICE-PVTN-HEADER-DATA – – Dead Oil PVT
Properties
– PVCDO – – Dead Oil PVT Properties –
10.85616 1 0.0001013253 1 1* /
PVTW
– – Water PVT Properties –
10.85616 1 0.00010132535 1 1* /
DENSITY
– – Fluid Densities at Surface Conditions –
0.6 1 0/
ROCK
– – Rock Properties –
0.1 0/
SWOF
– – Water/Oil Saturation Functions –
0 0 1 0
1 1 0 0/
– Water/Oil Saturation Functions –
0.0000 0.0000 1.000000 104.260304
0.0031 0.0000 0.993760 66.575857
0.0063 0.0000 0.987539 33.07857
0.0125 0.0002 0.975156 16.329927
0.0250 0.0006 0.950625 7.955606
0.0500 0.0025 0.902500 3.768445
0.1000 0.0100 0.810000 1.674864
0.2000 0.0400 0.640000 0.628074
0.3000 0.0900 0.490000 0.279144
0.4000 0.1600 0.360000 0.104679
0.5000 0.2500 0.250000 0.000000
0.6000 0.3600 0.160000 -0.069786
0.7000 0.4900 0.090000 -0.119633
15
0.8000 0.6400 0.040000 -0.157019
0.9000 0.8100 0.010000 -0.186096
0.9500 0.9025 0.002500 -0.198339
0.9750 0.9506 0.000625 -0.20399
0.9875 0.9752 0.000156 -0.206708
0.9938 0.9875 3.9e-005 -0.208041
0.9969 0.9938 1.0e-005 -0.208702
1.0000 1.0000 0.000000 -0.209358/
REGIONS
FIPNUM
1 1 40*2 /
EQLNUM
2*1 40*2 /
SATNUM
2*1 40*2 /
SOLUTION
EQUIL
– – Equilibration Data Specification –
60 10.8561184343306 0 1* 1* 1* 1* 1* 1* /
60 10.8561184343306 10000000 1* 1* 1* 1* 1* 1* /
SUMMARY
FOPR
FWPR
RGPR /
ROPR /
RORME /
ROSAT /
RPR /
RPTONLY
RWPR /
RWSAT /
SEPARATE
WOPR /
WWPP /
16
SCHEDULE
RPTSCHED
’PRES’ ’SOIL’ ’RESTART=2’ /
TSTEP 2.77777777777778e-012 /
TSTEP 2.77777777777778e-012 /
TSTEP 2.77777777777778e-012 /
TSTEP 2.77777777777778e-012 /
TSTEP 2.77777777777778e-012 /
TSTEP 2.75e-010 /
TSTEP 2.77750002222222e-006 /
TSTEP 5.6388885e-006 /
TSTEP 8.5858328e-006 /
TSTEP 1.1621185e-005 /
TSTEP 1.4747598e-005 /
TSTEP 1.7967801e-005 /
TSTEP 2.1284612e-005 /
TSTEP 2.4700927e-005 /
TSTEP 2.8219731e-005 /
TSTEP 3.1844098e-005 /
TSTEP 3.5577195e-005 /
TSTEP 3.9422288e-005 /
TSTEP 4.338273e-005 /
TSTEP 4.7461988e-005 /
TSTEP 5.1663617e-005 /
TSTEP 5.5991295e-005 /
TSTEP 6.0448805e-005 /
TSTEP 6.504004e-005 /
TSTEP 6.9769005e-005 /
TSTEP 7.4639851e-005 /
TSTEP 7.9656813e-005 /
TSTEP 8.4824285e-005 /
TSTEP 9.0146787e-005 /
TSTEP 9.5628959e-005 /
TSTEP 0.0001012756 /
17
TSTEP 0.00010709163 /
TSTEP 0.00011308215 /
TSTEP 0.00011925239 /
TSTEP 0.00012560774 /
TSTEP 0.00013215374 /
TSTEP 0.00013889612 /
TSTEP 0.00014584078 /
TSTEP 0.00015299376 /
TSTEP 0.00016036135 /
TSTEP 0.00016794995 /
TSTEP 0.00017576624 /
TSTEP 0.000183817 /
TSTEP 0.00019210928 /
TSTEP 0.00020065033 /
TSTEP 0.00020944761 /
TSTEP 0.00021850881 /
TSTEP 0.00022784184 /
TSTEP 0.00023745485 /
TSTEP 0.00024735625 /
TSTEP 0.00025755473 /
TSTEP 0.00026805911 /
TSTEP 0.00027887867 /
TSTEP 0.00029002281 /
TSTEP 0.00030150128 /
TSTEP 0.0003133241 /
TSTEP 0.00032550161 /
TSTEP 0.0003380444 /
TSTEP 0.00035096347 /
TSTEP 0.00036427012 /
TSTEP 0.00037797593 /
TSTEP 0.00039209292 /
TSTEP 0.00040663345 /
TSTEP 0.00042161017 /
TSTEP 0.00043703619 /
18
TSTEP 0.00045292496 /
TSTEP 0.00046929042 /
TSTEP 0.00048614683 /
TSTEP 0.00050350896 /
TSTEP 0.00052139192 /
TSTEP 0.00053981139 /
TSTEP 0.00055878342 /
TSTEP 0.0005783246 /
TSTEP 0.00059845205 /
TSTEP 0.00061918324 /
TSTEP 0.00064053643 /
TSTEP 0.00066253019 /
TSTEP 0.0006851838 /
TSTEP 0.00070851704 /
TSTEP 0.00073255028 /
TSTEP 0.00075730454 /
TSTEP 0.00078280142 /
TSTEP 0.00080906326 /
TSTEP 0.00083611289 /
TSTEP 0.00086397404 /
TSTEP 0.000892671 /
TSTEP 0.00092222891 /
TSTEP 0.00095267349 /
TSTEP 0.00098403147 /
TSTEP 0.0010163301 /
TSTEP 0.0010495977 /
TSTEP 0.0010838635 /
TSTEP 0.0011191572 /
TSTEP 0.0011555097 /
TSTEP 0.0011929527 /
TSTEP 0.0012315193 /
TSTEP 0.0012712427 /
TSTEP 0.0013121578 /
TSTEP 0.0013543004 /
19
TSTEP 0.0013977074 /
TSTEP 0.0014424163 /
TSTEP 0.0014884666 /
TSTEP 0.0015358984 /
TSTEP 0.0015847532 /
TSTEP 0.0016350736 /
TSTEP 0.0016869035 /
TSTEP 0.0017402883 /
TSTEP 0.0017952747 /
TSTEP 0.0018519106 /
TSTEP 0.0019102456 /
TSTEP 0.0019703307 /
TSTEP 0.0020322183 /
TSTEP 0.0020959626 /
TSTEP 0.0021616193 /
TSTEP 0.0022292456 /
TSTEP 0.0022989006 /
TSTEP 0.0023706453 /
TSTEP 0.0024445422 /
TSTEP 0.0025206562 /
TSTEP 0.002599054 /
TSTEP 0.0026798034 /
TSTEP 0.0027629754 /
TSTEP 0.0028486426 /
TSTEP 0.0029368799 /
TSTEP 0.0030277641 /
TSTEP 0.0031213749 /
TSTEP 0.003217794 /
TSTEP 0.0033171058 /
TSTEP 0.0034193969 /
TSTEP 0.0035247568 /
TSTEP 0.0036332775 /
TSTEP 0.0037450539 /
TSTEP 0.0038601835 /
20
TSTEP 0.003978767 /
TSTEP 0.0041009081 /
TSTEP 0.0042267134 /
TSTEP 0.004356293 /
TSTEP 0.0044897599 /
TSTEP 0.004627231 /
TSTEP 0.0047688261 /
TSTEP 0.0049146693 /
TSTEP 0.0050648879 /
TSTEP 0.0052196132 /
TSTEP 0.0053789802 /
TSTEP 0.0055431281 /
TSTEP 0.0057122009 /
TSTEP 0.0058863456 /
TSTEP 0.0060657146 /
TSTEP 0.0062504648 /
TSTEP 0.0064407573 /
TSTEP 0.0066367583 /
TSTEP 0.0068386397 /
TSTEP 0.0070465771 /
TSTEP 0.0072607528 /
TSTEP 0.0074813534 /
TSTEP 0.0077085723 /
TSTEP 0.0079426076 /
TSTEP 0.0081836646 /
TSTEP 0.0084319524 /
TSTEP 0.008687689 /
TSTEP 0.0089510977 /
TSTEP 0.0092224097 /
TSTEP 0.0095018605 /
TSTEP 0.009789696 /
TSTEP 0.010086166 /
TSTEP 0.01039153 /
TSTEP 0.010706055 /
21
TSTEP 0.011030016 /
TSTEP 0.011363696 /
TSTEP 0.011707386 /
TSTEP 0.012061387 /
TSTEP 0.012426008 /
TSTEP 0.012801567 /
TSTEP 0.013188393 /
TSTEP 0.013586824 /
TSTEP 0.013997208 /
TSTEP 0.014419904 /
TSTEP 0.014855281 /
TSTEP 0.015303719 /
TSTEP 0.015765609 /
TSTEP 0.016241357 /
TSTEP 0.016731378 /
TSTEP 0.017236099 /
TSTEP 0.017755959 /
TSTEP 0.018291418 /
TSTEP 0.018842941 /
TSTEP 0.019411009 /
TSTEP 0.019996118 /
TSTEP 0.02059878 /
TSTEP 0.021219522 /
TSTEP 0.021858888 /
TSTEP 0.022517433 /
TSTEP 0.023195734 /
TSTEP 0.023894386 /
TSTEP 0.024613997 /
TSTEP 0.025355197 /
TSTEP 0.026118632 /
TSTEP 0.02690497 /
TSTEP 0.027714899 /
TSTEP 0.028549124 /
TSTEP 0.029408377 /
22
TSTEP 0.030293405 /
TSTEP 0.031204984 /
TSTEP 0.032143909 /
TSTEP 0.033111006 /
TSTEP 0.034107111 /
TSTEP 0.035133101 /
TSTEP 0.036189873 /
TSTEP 0.037278347 /
TSTEP 0.038399473 /
TSTEP 0.039554231 /
TSTEP 0.040743634 /
TSTEP 0.041968714 /
TSTEP 0.043230549 /
TSTEP 0.044530235 /
TSTEP 0.045868915 /
TSTEP 0.047247756 /
TSTEP 0.04866796 /
TSTEP 0.05013077 /
TSTEP 0.051637463 /
TSTEP 0.05318936 /
TSTEP 0.054787811 /
TSTEP 0.056434214 /
TSTEP 0.058130015 /
TSTEP 0.059876688 /
TSTEP 0.061675761 /
TSTEP 0.063528813 /
TSTEP 0.065437451 /
TSTEP 0.067403354 /
TSTEP 0.069428228 /
TSTEP 0.071513847 /
TSTEP 0.073662043 /
TSTEP 0.075874679 /
TSTEP 0.078153685 /
TSTEP 0.080501065 /
23
TSTEP 0.08291886 /
TSTEP 0.085409187 /
TSTEP 0.087974228 /
TSTEP 0.090616219 /
TSTEP 0.093337469 /
TSTEP 0.096140355 /
TSTEP 0.099027328 /
TSTEP 0.1020009 /
TSTEP 0.1050637 /
TSTEP 0.1082184 /
TSTEP 0.1114677 /
TSTEP 0.1148145 /
TSTEP 0.1182617 /
TSTEP 0.1218123 /
TSTEP 0.1254694 /
TSTEP 0.1292362 /
TSTEP 0.133116 /
TSTEP 0.1371123 /
TSTEP 0.1412284 /
TSTEP 0.145468 /
TSTEP 0.1498348 /
TSTEP 0.1543326 /
TSTEP 0.1589653 /
TSTEP 0.163737 /
TSTEP 0.1686519 /
TSTEP 0.1737142 /
TSTEP 0.1789284 /
TSTEP 0.184299 /
TSTEP 0.1898307 /
TSTEP 0.1955284 /
TSTEP 0.201397 /
TSTEP 0.2074416 /
TSTEP 0.2136676 /
TSTEP 0.2200804 /
24
TSTEP 0.2266855 /
TSTEP 0.2334889 /
TSTEP 0.2404963 /
TSTEP 0.2477139 /
TSTEP 0.2551481 /
TSTEP 0.2628052 /
TSTEP 0.2706921 /
TSTEP 0.2788156 /
TSTEP 0.2871829 /
TSTEP 0.2958011 /
TSTEP 0.3046779 /
TSTEP 0.313821 /
TSTEP 0.3232383 /
TSTEP 0.3329383 /
TSTEP 0.3429292 /
TSTEP 0.3532198 /
TSTEP 0.3638192 /
TSTEP 0.3747365 /
TSTEP 0.3859813 /
TSTEP 0.3975635 /
TSTEP 0.4094932 /
TSTEP 0.4217807 /
TSTEP 0.4344369 /
TSTEP 0.4474728 /
TSTEP 0.4608997 /
TSTEP 0.4747294 /
TSTEP 0.488974 /
TSTEP 0.503646 /
TSTEP 0.5187581 /
TSTEP 0.5343236 /
TSTEP 0.5503561 /
TSTEP 0.5668695 /
TSTEP 0.5838782 /
TSTEP 0.6013973 /
25
TSTEP 0.6194419 /
TSTEP 0.6380279 /
TSTEP 0.6571714 /
TSTEP 0.6768892 /
TSTEP 0.6971986 /
TSTEP 0.7181173 /
TSTEP 0.7396635 /
TSTEP 0.7618561 /
TSTEP 0.7847144 /
TSTEP 0.8082585 /
TSTEP 0.832509 /
TSTEP 0.8574869 /
TSTEP 0.8832142 /
TSTEP 0.9097133 /
TSTEP 0.9370073 /
TSTEP 0.9651202 /
TSTEP 0.9940764 /
TSTEP 1.023901 /
TSTEP 1.054621 /
TSTEP 1.086262 /
TSTEP 1.118853 /
TSTEP 1.152421 /
TSTEP 1.186996 /
TSTEP 1.222608 /
TSTEP 1.259289 /
TSTEP 1.29707 /
TSTEP 1.335985 /
TSTEP 1.376067 /
TSTEP 1.417352 /
TSTEP 1.459875 /
TSTEP 1.503673 /
TSTEP 1.548786 /
TSTEP 1.595252 /
TSTEP 1.643112 /
26
TSTEP 1.692408 /
TSTEP 1.743183 /
TSTEP 1.795481 /
TSTEP 1.849348 /
TSTEP 1.904831 /
TSTEP 1.961978 /
TSTEP 2.02084 /
TSTEP 2.081467 /
TSTEP 2.143914 /
TSTEP 2.208234 /
TSTEP 2.274483 /
TSTEP 2.34272 /
TSTEP 2.413003 /
TSTEP 2.485396 /
TSTEP 2.55996 /
TSTEP 2.636761 /
TSTEP 2.715866 /
TSTEP 2.797344 /
TSTEP 2.881267 /
TSTEP 2.967707 /
TSTEP 3.05674 /
TSTEP 3.148445 /
TSTEP 3.2429 /
TSTEP 3.340189 /
TSTEP 3.440397 /
TSTEP 3.543611 /
TSTEP 3.649922 /
TSTEP 3.759421 /
TSTEP 3.872206 /
TSTEP 3.988374 /
TSTEP 4.108028 /
TSTEP 4.231271 /
TSTEP 4.358211 /
TSTEP 4.48896 /
27
TSTEP 4.623631 /
TSTEP 4.762342 /
TSTEP 4.905215 /
TSTEP 5.052374 /
TSTEP 5.203948 /
TSTEP 5.360068 /
TSTEP 5.520872 /
TSTEP 5.6865 /
TSTEP 5.857098 /
TSTEP 6.032813 /
TSTEP 6.213799 /
TSTEP 6.400215 /
TSTEP 6.592223 /
TSTEP 6.789992 /
TSTEP 6.993693 /
TSTEP 7.203506 /
TSTEP 7.419614 /
TSTEP 7.642204 /
TSTEP 7.871472 /
TSTEP 8.107619 /
TSTEP 8.350849 /
TSTEP 8.601377 /
TSTEP 8.859421 /
TSTEP 9.125206 /
TSTEP 9.398965 /
TSTEP 9.680937 /
TSTEP 9.971367 /
TSTEP 10.27051 /
TSTEP 10.57863 /
TSTEP 10.89599 /
TSTEP 11.22287 /
TSTEP 11.55956 /
TSTEP 11.90635 /
TSTEP 12.26355 /
28
TSTEP 12.63146 /
TSTEP 13.01041 /
TSTEP 13.40072 /
TSTEP 13.80275 /
TSTEP 14.21683 /
TSTEP 14.64334 /
TSTEP 15.08264 /
TSTEP 15.53513 /
TSTEP 16.00118 /
TSTEP 16.48122 /
TSTEP 16.97566 /
TSTEP 17.48494 /
TSTEP 18.00949 /
TSTEP 18.54977 /
TSTEP 19.10627 /
TSTEP 19.67946 /
TSTEP 20.26984 /
TSTEP 20.87794 /
TSTEP 21.50428 /
TSTEP 22.14941 /
TSTEP 22.81389 /
TSTEP 23.49831 /
TSTEP 24.20326 /
TSTEP 24.92936 /
TSTEP 25.67724 /
TSTEP 26.44756 /
TSTEP 27.24099 /
TSTEP 28.05822 /
TSTEP 28.89997 /
TSTEP 29.76697 /
TSTEP 30.65997 /
TSTEP 31.57977 /
TSTEP 32.52716 /
TSTEP 33.50298 /
29
TSTEP 34.50807 /
TSTEP 35.54331 /
TSTEP 36.6096 /
TSTEP 37.70789 /
TSTEP 38.83912 /
TSTEP 40.00429 /
TSTEP 41.20442 /
TSTEP 42.44055 /
TSTEP 43.71376 /
TSTEP 45.02517 /
TSTEP 46.37591 /
TSTEP 47.76718 /
TSTEP 49.20018 /
TSTEP 50.67618 /
TSTEP 52.19645 /
TSTEP 53.76233 /
TSTEP 55.37519 /
TSTEP 57.03643 /
TSTEP 58.74751 /
TSTEP 60.50991 /
TSTEP 62.3252 /
TSTEP 64.19494 /
TSTEP 66.12077 /
TSTEP 68.10438 /
TSTEP 70.14749 /
TSTEP 72.2519 /
TSTEP 74.41944 /
TSTEP 76.65202 /
TSTEP 78.95156 /
TSTEP 81.3201 /
TSTEP 83.75969 /
TSTEP 86.27247 /
TSTEP 88.86063 /
TSTEP 91.52644 /
30
TSTEP 94.27222 /
TSTEP 97.10036 /
TSTEP 100.0134 /
TSTEP 103.0137 /
TSTEP 106.1041 /
TSTEP 109.2872 /
TSTEP 112.5658 /
TSTEP 115.9428 /
TSTEP 119.421 /
TSTEP 123.0037 /
TSTEP 126.6937 /
TSTEP 130.4945 /
TSTEP 134.4093 /
TSTEP 138.4416 /
TSTEP 142.5948 /
TSTEP 146.8727 /
TSTEP 151.2788 /
TSTEP 155.8172 /
TSTEP 160.4917 /
TSTEP 165.3064 /
TSTEP 170.2656 /
TSTEP 175.3736 /
TSTEP 180.6348 /
TSTEP 186.0538 /
TSTEP 191.6354 /
TSTEP 197.3844 /
TSTEP 203.3059 /
TSTEP 209.4051 /
TSTEP 215.6872 /
TSTEP 222.1578 /
TSTEP 228.8225 /
TSTEP 235.6871 /
TSTEP 242.7577 /
TSTEP 250.0404 /
31
TSTEP 257.5415 /
TSTEP 265.2677 /
TSTEP 273.2257 /
TSTEP 281.4224 /
END
32
Bibliography
[1] Al-Huthali, A. and Datta-Gupta, A. Streamline simulation of counter-current imbibition in naturally
fractured reservoirs. SPEJ, 2004, Vol: 43(3-4), Pages: 271-300.
[2] Aronofsky, J. S., Masse, L., and Natanson, S. G. A Model for the Mechanism of Oil Recovery from
the Porous Matrix Due to Water Invasion in Fractured Reservoirs. Trans. AIME, 1958, Vol: 213,
Pages: 17-24.
[3] Baker, R. O., Kuppe, F., Chugh, S., Bora, R., Stojanovic, S., and Batycky, R. Full-Field Modeling
Using Streamline-Based Simulation: Four Case Studies. SPEREE, 2002, Vol: 5(2), Pages: 126-134.
[4] Barenblatt, G. I., Zheltov, I. P., and Kochina, I. N. Basic concepts in the theory of seepage of
homogeneous liquids in fissured rocks. J. Appllied Mathematics and Mechanics, English Translation,
1960, Vol: 24, Pages: 1286-1303.
[5] Barenblatt, G. I., Entov, V. M., and Ryzhik, V. M. Theory of Fluid Flows Through Natural Rocks.
Kluwer Academic Publishers, Dordrecht, 1990.
[6] Barr D., Savory K., Fowler S., Arman K., McGarrity J. Fracture-Matrix Interaction and its Impli-
cations for Reservoir and Well Performance in The Clair Field, West Of Shetland. Press, 2005.
[7] Batycky, R. P., Blunt, M. J., and Thiele, M. R. 3D Field-Scale Streamline-Based Reservoir Simula-
tor. SPERE, November 1997, Vol: 11, Pages: 246-254.
[8] Behbahani, H. and Blunt, M. J. Analysis of Imbibition in Mixed-Wet Rocks Using Pore-Scale
Modeling. SPEJ, 2005, Vol: 10(4), Pages: 466-474.
[9] Behbahani, H., Di Donato, G. and Blunt, M. J. Simulation of counter-current imbibition in water-
wet fractured reservoirs. SPEJ, 2006, Vol: 50, Page: 21-39.
[10] Belayneh, M. and Cosgrove, J. W. Fracture pattern variations around a major fold and their im-
plications regarding fracture prediction using limited data: an example from the Bristol Channel
Basin, UK. In: Cosgrove, J. W. and Engelder, T. (eds.) The initiation, propagation and arrest of
33
joints and other fractures. Geological Society, London, Special Publications, 2004, Vol: 231, Page:
89-102.
[11] Birks, J. A Theoretical Investigation into the Recovery of Oil from Fissured Limestone Formulation
by Water-Drive and Gas Cap Drive. Proceedings of the Fourth World Petroleum Congress, 1955,
II/F, Pages: 425-440.
[12] Bommer, M.P. and Schechter, R.S. Mathematical Modeling of In-Situ Uranium Leaching. SPEJ,
March 1962, Pages: 1-8.
[13] Bratvedt, F., Gimse T., Tegnander C. Streamline Computations for Porous Media Flow Including
Gravity. Transport in Porous Media, October 1996, Vol: 25, Pages: 63-78(16).
[14] Chang, M. Deriving the Shape Factor of a Fractured Rock Matrix. Technical Report NIPER-696
(DE93000170), NIPER, Bartlesville OK , 1993.
[15] Chen, J., Miller, M. A. and Sepehrnoori, K. An Approach for Implementing Dual Porosity Models in
Existing Simulators. SPE 28001, proceedings of the University of Tulsa/SPE Centennial Petroleum
Engineering Symposium, Tulsa, Oklahoma, 29-31, August 1994.
[16] Christie, M.A. and Blunt M.J. Tenth SPE Comparative Solution Project: A Comparison of Upscal-
ing Techniques. SPEREE, August 2001, Vol: 4, Pages: 308-317.
[17] Civan, F. and Rasmussen, M. L. Asymptotic analytical solutions for imbibition waterflood in frac-
tured reservoirs. SPEJ, 2001, Vol: 6, Page: 171-181.
[18] Clifford, P. J., O’Donovan, A. R., Savory, G. Smith, and D. Barr. Clair Field-Managing Uncertainty
in the Development of a Waterflooded Fractured Reservoir. SPE 96316, proceedings of Offshore
Europe 2005 held in Aberdeen, Scotland, U. K. 6-9, September 2005.
[19] Coats, K.H. Implicit compositional simulation of single-porosity and dual-porosity reservoirs, SPE
18427, proceedings of the 10th SPE Symposium on Reservoir Simulation, Houston, Texas, Feburary
1989.
[20] Daly, C. and Mueller, D. Characterization and Modeling of Fractured Reservoirs: Static Model.
proceedings of the ninth European Conference on the Mathematics of Oil Recovery, Cannes, France,
September 2004.
[21] De Swaan, A. Theory of waterflooding in fractured reservoirs. SPEJ, April, 1978, Vol: 18, Pages:
117-122.
34
[22] Dean, R.H. and Lo, L. L. Simulations of Naturally Fractured Reservoirs. SPERE, May 1988, Pages:
638-48.
[23] Di Donato, G., Huang, W., and Blunt, M. J. Streamline-Based Dual Porosity Simulation of Frac-
tured Reservoirs. SPE 84036, proceedings of the SPE Annual Technical Meeting and Exhibition,
Denver, Colorado, October 2003.
[24] Di Donato, G. and Blunt, M. J. Streamline-based dual-porosity simulation of reactive transport and
flow in fractured reservoirs. Water Resources Research, 40, W04203, doi: 10.1029/2003WR002772
(2004).
[25] Di Donato, G. Tavassoli, Z. and Blunt, M. J. Analytical and numerical analysis of oil recovery by
gravity drainage. SPEJ, 2006, Vol: 54, Pages: 55-69.
[26] Di Donato, G., Lu, H., Tavassoli, Z., Blunt, M. J. Multi-rate transfer dual porosity modeling of
gravity drainage and imbibition. SPEJ, 2007, Vol: 12, Pages: 77-88.
[27] Ding, Y. Upscaling Fracture Networks for Simulation of Horizontal Wells using a Dual-Porosity
Reservoir Simulator. SPE 92774, proceedings of the SPE Reservoir Simulation Symposium, Houston
Texas, 31 January - 2 February, 2005.
[28] Gelhar, L. W. and Collins M. A. General analysis of longitudinal dispersion in nonuniform flow.
Water Resources Research, 1971, Vol: 7(6), Pages: 1511-21.
[29] Gilman J.R., and Kazemi, H. Improvement in simulation of naturally fractured reservoirs. SPEJ,
August 1983, Vol: 23, Pages: 695-707.
[30] Grinestaff, G. H., and Caffrey, D. J. Waterflood Management: A Case Study of the Northwest
Fault Block Area of Prudhoe Bay, Alaska, Using Streamline Simulation and Traditional Waterflood
Analysis. SPE 63152 proceedings of the 2000 SPE Annual Technical Conference and Exhibition,
Dallas, 1-4 October, 2000.
[31] Hagoort, J. Oil Recovery by Gravity Drainage. SPEJ, June 1980, Vol: 20, Pages: 139-150.
[32] Higgins, R. V. and Leighton, A. J. A Computer Method to Calculate Two-Phase Flow in Any
Irregularly Bounded Porous Medium. JPT, June 1962, Pages: 1048-1054.
[33] Hill, A.C. and Thomas, G.W. A New Approach for Simulating Complex Fractured Reservoirs. SPE
13537, proceedings of the SPE Middle East Oil Technical Conference and Exhibition, Bahrain, 1985.
35
[34] Horie, T., Firoozabadi, A. and Ishimoto, K. Laboratory Studies of Capillary Interaction in Frac-
ture/Matrix Systems. SPERE , August 1990, Vol: 4, Pages: 353-360.
[35] Huang, W., Di Donato, G. and Blunt, M. J. Comparison of streamline-based and grid-based dual
porosity simulation. SPEJ, 2004, Vol: 43, Page: 129-137.
[36] Kazemi, H., Merrill, L. S., Porterfield, K. L. and Zeman, P. R. Numerical Simulation of Water-Oil
Flow in Naturally Fractured Reservoirs. SPEJ, Dec. 1976 Vol: 16, Page: 318-326.
[37] Kazemi, H., Gilman, J. R., and Eisharkawy, A. M. Analytical and Numerical Solution of Oil Re-
covery From Fractured Reservoirs with Empirical Transfer Function. SPERE, May 1992, Vol: 6,
Pages: 219-227.
[38] King, M. J., and Datta-Gupta, A. Streamline Simulation: A Current Perspective. In Situ, 1998,
Vol: 22(1), Pages: 91-140.
[39] Lake, L. W., Johnston, J. R., and Stegemeier, G. L. Simulation and performance. prediction of a
large-scale surfactant/polymer project: SPEJ, 1981, Vol: 6(21), Pages: 731-739.
[40] Lim, K. T. and Aziz, K. Matrix-Fracture Transfer Shape Factors for Dual Porosity Simulators.
SPEJ, 1995, Vol: 13, Pages: 169-178.
[41] Litvak, B.L. Simulation and Characterization of Naturally Fractured Reservoirs. Proceedings of the
Reservoir Characterization Technical Conference, Dallas, 1985.
[42] Lu H., Di Donato,G., Blunt M. J. General Transfer Functions for Multiphase Flow. SPE 102542,
proceedings of the SPE Annual Technical Conference and Exhibition held in San Antonio, Texas,
U.S.A., 24-27 September, 2006.
[43] Lu H., Blunt M. J. General Fracture/Matrix Transfer Functions for Mixed-Wet Systems. SPE
107007, proceedings of the SPE Europec/EAGE Annual Conference and Exhibition held in London,
United Kingdom, 11-14 June, 2007.
[44] Ma, S., Morrow, N.R. and Zhang, X. Generalized Scaling of Spontaneous Imbibition Data for
Strongly Water-Wet Systems. SPEJ, 1997, Vol: 18, Pages: 165-178.
[45] Matthai, S. K., Mezentsev, A., Belayneh, M. Finite Element C Node-Centered Finite Volume
Two-Phase Flow Experiments with Fractured Rock Represented by Unstructured Hybrid Element
Meshes. SPEREE, 2007, Vol: 9, Pages: 1-18
36
[46] Mattax, C. C. and Kyte, J. R. Imbibition Oil Recovery from Fractured, Water-Drive Reservoirs.
SPEJ, June 1962, Pages: 177-184; Trans., AIME, Pages: 225.
[47] Moreno, J., Kazemi, H. and Gilman, J. R. Streamline Simulation of Counter-Current Water-Oil
and Gas-Oil Flow in Naturally Fractured Dual-Porosity Reservoirs. SPE 89880, proceedings of the
SPE Annual Meeting, Houston, Texas, 26-29 September, 2004.
[48] Morrow, M. R and Mason, G. Recovery of oil by spontaneous imbibition. Current Opinion in Colloid
and Interface Science 6, 2001, Pages: 321-337.
[49] Muskat, M. and Wyckoff, R. D. Approximate Theory of Water Coning in Oil Production. Trans
AIME, 1935, Vol: 114, Pages: 144-163.
[50] Ponting, D. Characterization and Modeling of Fractured Reservoirs: Flow Simulation. Proceedings
of the ninth European Conference on the Mathematics of Oil Recovery, Cannes, France, September
2004.
[51] Pollock, D. W. Semianalytical Computation of Path Lines for Finite-Difference Models. Ground
Water, December 1988, Vol: 6, Pages: 743-750.
[52] Por, G. J., Boerrigter, P., Maas, J. G. and de Vries, A. A Fractured Reservoir Simulator Capable
of Modeling Block-Block Interaction. SPE 1980, proceedings of the SPE Annual Technical Meeting
and Exhibition, San Antonio, Texas, October 1989.
[53] Pruess, K. and Narasimhan, T. N. A Practical Method for Modeling Fluid and Heat Flow in
Fractured Porous Media. SPEJ, 1985, Vol: 25(1), Pages: 14-26.
[54] Quandalle, P. and Sabathier, J. C. Typical Features of a Multipurpose Reservoir Simulator. SPE
Reservoir Engineering, 1989, Vol: 4, Pages: 475-480.
[55] Sarma, P. and Aziz, K. New Transfer Functions for Simulation of Naturally Fractured Reservoirs
With Dual Porosity Models. SPEJ, 2006, Vol: 11, Pages: 328-340.
[56] Samier, P., Quettier, L., and Thiele, M. Applications of Streamline Simulations to Reservoir Studies.
SPEREE, 2002, Vol: 5(4), Pages: 324-332.
[57] Sonier, F., Souillard, P., Blaskovich, F. T. Numerical Simulation of Naturally Fractured Reservoirs.
SPE 15627 proceedings of the Annual Technical Conference and Exhibition of the Society of the
Petroleum Engineers, New Orleans, LA, 5-8 October, 1986.
37
[58] Sonier, F. Discussion of Improved Calculations for Viscous and Gravity Displacement in Matrix
Blocks in Dual-Porosity Simulators. JPT, June 1988, Pages: 784.
[59] Sahni, A. Measurements of three phase relative permeability during gravity drainage using CT
scanning. Thesis, 1998, Stanford University, USA
[60] Thiele, M. R., Batycky R. P., Iding, M. and Blunt, M. J. Extension of Streamline-Based Dual
Porosity Flow Simulation to Realistic Geology. proceedings of the ninth European Conference on
the Mathematics of Oil Recovery, Cannes, France, September 2004.
[61] Terez, I. E. and Firoozabadi, A. Water injection in water-wet fractured porous media: experiments
and a new model with modified Buckley-Leverett Theory. SPEJ, 1999, Vol. 4(2), 135-141.
[62] Tavassoli, Z., Zimmerman, R. W. and Blunt, M J. Analytic Analysis for Oil Recovery During
Counter-Current Imbibition in Strongly Water-Wet Systems. Transport in Porous Media, 2005,
Vol: 58, Pages: 173-189, doi:10.1007/s11242-004-5474-4.
[63] Thomas, L.K., Dixon, T.N. and Pierson, R.G. Fractured Reservoir Simulation. SPEJ, February
1983, Pages: 42-54.
[64] Udea, Y., Murata, S., Watanabe, Y. and Funatsu,K. Investigation of the Shape Factor Used in the
Dual Porosity Reservoir Simulator. SPE 19469, proceedings of the Asia Pacific Conference, Sydney,
Australia, Septemter 1989.
[65] van Heel A.P.G. and Boerrigter P.M. Shape-factor in Fractured Reservoir Simulation. SPE 1024711,
proceedings of the SPE Annual Technical Conference and Exhibition held in San Antonio, Texas,
U.S.A., 24-27 September,2006.
[66] Vermeulen, T. Theory of irreversible and constant-pattern solid diffusion. Industrial & Engineering
Chemistry, 1953, Vol: 45, Pages: 1664-1670.
[67] Warren, J. E., and Root, P. J. The Behaviour of Naturally Fractured Reservoirs. SPEJ, Septemter
1963, Vol: 3, Pages: 245-255; Trans. AIME, Pages: 228.
[68] Wood Mackenzie Research Website: www.woodmacresearch.com
[69] Wu, Y.-S. and Pruess, K. A Multiple-Porosity Method for Simulation of Naturally Fractured Pe-
troleum Reservoirs. SPERE, 1988, Vol: 3, Pages: 327-336.
[70] Zhang, X., Morrow, N. R., and Ma, S. Experimental Verification of a Modified Scaling Group for
Spontaneous Imbibition SPERE, 1996, Vol: 11, Pages: 280-285.
38
[71] Zhou, Hongbo, Li, Zhipin, Li, Yun. A Super-Dual-Porosity Simulation Model for Water-Channeling
in Naturally Fractured Gas and Oil Reservoirs. SPE 50927, proceedings of the SPE International
Oil and Gas Conference and Exhibition, Beijing, China, 2-6 November, 1997.
[72] Zhou, D., Jia, L., Kamath, J., and Kovscek, A. R. Scaling of counter-current imbibition processes
in low-permeability porous media. Journal of Petroleum Science and Engineering, 2002, Vol: 33,
Pages: 61-74.
[73] Zhou, X., Morrow, N.R., Ma, S. Interrelationship of Wettability, Initial Water Saturation, Aging
Time and Oil Recovery by Spontaneous Imbibition and Waterflooding. SPEJ, June 2000, Vol: 5(2),
Pages: 199-207.
[74] Zimmerman, R. W. and Bodvarsson, G. S. Integral method solution for diffusion in spherical blocks.
J. Hydrology, 1989, Vol: 111, Pages: 213-224.
[75] 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 Research, 1993, Vol:
29(7), Pages: 2127-2137.
39

Investigation of Recovery Mechanisms in Fractured Reservoirs

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Investigation of Recovery Mechanisms in Fractured Reservoirs

Caricato da

Copyright:

Formati disponibili

Investigation of Recovery

Mechanisms in Fractured Reservoirs

A dissertation submitted in fulfillment of the requirements for

My work focuses on developing a physical-motivated approach to modeling displacement

Publication as a results of this research is as follows:

2 Streamline-based dual porosity simulation 16

3 Oil field case studies 32

4 General Transfer Functions for Multiphase Flow 72

5 General Fracture/Matrix Transfer Functions for Mixed-Wet Systems 95

6 Conclusions and future work 139

A Barenblatt et al.’s analytical solution for counter-current imbibition 1

C Downscaling permeability data 10

D Calculation of the average saturation in gravity/capillary equilibrium 11

E Input deck of the Base Case (mixed-wet) for Eclipse model 13

1.1 Schematic of different recovery processes in fractured reservoirs . . . . . . 2

2.1 Oil production rate comparisons of grid-based and streamline-based model 22

3.1 The porosity distribution of North Sea reservoir model . . . . . . . . . . 33

4.1 Predicted and simulated oil transfer rate for Case 1 . . . . . . . . . . . . 87

5.1 Capillary pressure curve for the mixed-wet case . . . . . . . . . . . . . . 100

6.1 Simulations using a discrete fracture model at water breakthrough after

D.1 The capillary curves used for the mixed-wet system . . . . . . . . . . . . 11

3.1 The parameters used in dual porosity simulation . . . . . . . . . . . . . . 34

5.1 Parameters used for the mixed-wet case . . . . . . . . . . . . . . . . . . . 100

a inverse time constant

1.1 Recovery mechanisms in fractured reservoirs

Oil (c) Oil (d)

In order to improve recovery, it is necessary to maintain pressure - ideally above the

Oil recovery - % recoverable oil

１，１７ ｈｏｕｒｓ ２．５０ ｈｏｕｒｓ １０．５ ｈｏｕｒｓ １１７ ｈｏｕｒｓ

1.2 Dual porosity modelling and shape factors

One matrix block Idealized model reservoir

derived based on a direct material balance on a rectangular parallelpiped matrix block

where tD is the dimensionless time.

where ηth is the thermal diffusivity of the rock, and θ is temperature.

φpm − φpf = Pom − Pof − δv (p+

and the in the z + direction,

1.4 Streamline-based simulation

(a) Matrix (b) Streamline

polymer/surfactant displacement (Lake, 1981).

In this work I will present a physically-motivated approach to modeling displacement

Streamline-based dual porosity

Streamline-based simulation is now established as an attractive alternative to conven-

2.2 Streamline-based dual porosity formulation

Single porosity streamline simulation ( equivalent to setting T = 0; m = 0 in Eq. (2.1)

2.3 Single and multi-rate transfer functions

2.3.1 Capillary-controlled imbibition

Ma et al. presented an analysis of capillary-controlled imbibition, where water is im-

where R∞ is the ultimate recovery. The rate constant β is defined by:

where we assume Leverett J-function scaling of the capillary pressure:

where Swmi is the initial water saturation in the matrix. Thus:

Swm = Swmi + (Sw∗ − Swmi ) · (1 − e−βt ) (2.13)

then from Eq. (2.12):

Dual porosity model

Number of grid blocks

2.3.2 Gravity drainage

where the rate constant α is:

Then following the same approach as for capillary-controlled imbibition:

and hence writing the transfer as a function of gas saturation:

2.3.3 Extensions to the approach

2.3.4 Multi-rate transfer function

2.3.5 Multi-rate model

2.3.6 Coupling with the fracture fractional flow

ｌａｒｇｅ ｍｅｄｉｕｍ ｓｍａｌｌ

１，１７　ｈｏｕｒｓ２．５０　ｈｏｕｒｓ１０．５　ｈｏｕｒｓ１１７　ｈｏｕｒｓ

ｌａｒｇｅ　ｍｅｄｉｕｍ　ｓｍａｌｌ　