Enhanced Motor Protection With the Slip-Dependent Thermal ...
EFFICIENT SIMULATION OF THERMAL ENHANCED …fn674bf7126/Thesis...EFFICIENT SIMULATION OF THERMAL...
Transcript of EFFICIENT SIMULATION OF THERMAL ENHANCED …fn674bf7126/Thesis...EFFICIENT SIMULATION OF THERMAL...
EFFICIENT SIMULATION OF THERMAL ENHANCED OIL
RECOVERY PROCESSES
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF ENERGY
RESOURCES ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Zhouyuan Zhu
August 2011
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/fn674bf7126
© 2011 by Zhouyuan Zhu. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Margot Gerritsen, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Anthony Kovscek, Co-Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Marco Thiele
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
Abstract
Simulating thermal processes is usually computationally expensive because of the
complexity of the problem and strong nonlinearities encountered. In this work, we
explore novel and efficient simulation techniques to solve thermal enhanced oil recov-
ery problems. We focus on two major topics: the extension of streamline simulation
for thermal enhanced oil recovery and the efficient simulation of chemical reaction
kinetics as applied to the in-situ combustion process.
For thermal streamline simulation, we first study the extension to hot water flood
processes, in which we have temperature induced viscosity changes and thermal vol-
ume changes. We first compute the pressure field on an Eulerian grid. We then solve
for the advective parts of the mass balance and energy equations along the individual
streamlines, accounting for the compressibility effects. At the end of each global time
step, we account for the nonadvective terms on the Eulerian grid along with gravity
using operator splitting. We test our streamline simulator and compare the results
with a commercial thermal simulator. Sensitivity studies for compressibility, gravity
and thermal conduction effects are presented.
We further extended our thermal streamline simulation to steam flooding. Steam
flooding exhibits large volume changes and compressibility associated with the phase
behavior of steam, strong gravity segregation and override, and highly coupled energy
and mass transport. To overcome these challenges we implement a novel pressure up-
date along the streamlines, a Glowinski θ-scheme operator splitting and a preliminary
streamline/finite volume hybrid approach. We tested our streamline simulator on a
series of test cases. We compared our thermal streamline results with those computed
by a commercial thermal simulator for both accuracy and efficiency. For the cases
iv
investigated, we are able to retain solution accuracy, while reducing computational
cost and gaining connectivity information from the streamlines. These aspects are
useful for reservoir engineering purposes.
In traditional thermal reactive reservoir simulation, mass and energy balance equa-
tions are solved numerically on discretized reservoir grid blocks. The reaction terms
are calculated through Arrhenius kinetics using cell-averaged properties, such as av-
eraged temperature and reactant concentrations. For the in-situ combustion process,
the chemical reaction front is physically very narrow, typically a few inches thick. To
capture accurately this front, centimeter-sized grids are required that are orders of
magnitude smaller than the affordable grid block sizes for full field reservoir models.
To solve this grid size effect problem, we propose a new method based on a non-
Arrhenius reaction upscaling approach. We do not resolve the combustion front on
the grid, but instead use a subgrid-scale model that captures the overall effects of
the combustion reactions on flow and transport, i.e. the amount of heat released, the
amount of oil burned and the reaction products generated. The subgrid-scale model
is calibrated using fine-scale highly accurate numerical simulation and laboratory ex-
periments. This approach significantly improves the computational speed of in-situ
combustion simulation as compared to traditional methods. We propose the detailed
procedures to implement this methodology in a field-scale simulator. Test cases il-
lustrate the solution consistency when scaling up the grid sizes in multidimensional
heterogeneous problems. The methodology is also applicable to other subsurface
reactive flow modeling problems with fast chemical reactions and sharp fronts.
Displacement front stability is a major concern in the design of all the EOR
processes. Historically, premature combustion front break through has been an issue
for field operations of in-situ combustion. In this work, we perform detailed analysis
based on both analytical methods and numerical simulation. We identify the different
flow regimes and several driving fronts in a typical 1D ISC process. For the ISC
process in a conventional mobile heavy oil reservoir, we identify the most critical
front as the front of steam plateau driving the cold oil bank. We discuss the five
main contributors for this front stability/instability: viscous force, condensation, heat
conduction, coke plugging and gravity. Detailed numerical tests are performed to test
v
and rank the relative importance of all these different effects.
vi
Acknowledgements
First of all, I would like to express my appreciation to my advisors: Margot Gerritsen,
Anthony Kovscek, Marco Thiele and Louis Castanier. They provided me with great
experience, guidance and continuous support during my doctoral study. I want to
thank Professor Louis Durlofsky for serving on my committee. I would also like to
thank Professor Biondo Biondi for kindly chairing my defense.
I owe many thanks to my research group members in the SUPRI-C and SUPRI-A
group, including Alexandre Lapene, Wenjuan Lin, Murat Cinar, and Golnaz Alipour.
I would like to thank Mohammad Bazargan for providing the three reaction model. I
want to thank former and fellow students, including Morten Kristensen, Qing Chen,
Tianhong Chen, Jeremy Kozdon, Hui Zhou and many others. I would also like to
thank other alumnus Yuguang Chen, Tom Tang, Yuanlin Jiang, Yaqing Fan.
I would also like to thank all the other faculties, staff, and fellow students of
the Department of Energy Resources Engineering for making my doctoral study so
fruitful and enjoyable. Special thanks to Robert Lindblom, who is the coolest geologist
teacher I ever met.
I would like to thank my mentors and friends at BP, Steven Vittoratos and Chris
West, with special thanks to Steven who brought me into the Canadian heavy oil
community. I also would like to thank industry affiliates and mentors who provided
great help and advices during the course of my doctoral research work: Paul Naccache,
David Law, Paul Hammond, Franck Monmont, and many others.
This work was prepared with the support of the Stanford Graduate Fellowship
and financial aid from Schlumberger Ltd. The support of the Stanford University
Petroleum Research Institute (SUPRI-A and SUPRI-C) Industrial Affiliates is also
vii
acknowledged.
Finally, I would like to thank my family for their support during these years. I
want to especially thank my mother, Hui Jiang, who has devoted tremendous support
and encouragement for my education in the last twenty years. I would also like to
thank my cousin Mingyu Zhu and my uncle Hang Jiang who gave much support
during my PhD studies.
viii
Contents
Abstract iv
Acknowledgements vii
1 Introduction 1
1.1 Heavy Oil and Thermal Enhanced Recovery . . . . . . . . . . . . . . 1
1.2 Motivations and Objectives . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Literature Review 5
2.1 Streamline Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Streamline Simulation Concept and Basics . . . . . . . . . . . 5
2.1.2 Advantages and Disadvantages of Streamline Simulation . . . 6
2.1.3 Streamline Simulation for Complex Physics Processes . . . . . 9
2.2 In-situ Combustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 In-situ Combustion Process . . . . . . . . . . . . . . . . . . . 11
2.2.2 Field Applications of ISC . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Numerical Simulation of In-situ Combustion . . . . . . . . . . 15
2.2.4 Challenges in Combustion Reaction Modeling . . . . . . . . . 16
2.2.5 Alternative Approaches . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6 Thermal Front Stability for ISC . . . . . . . . . . . . . . . . . 18
3 Governing Equations 21
3.1 General Conservation Equations . . . . . . . . . . . . . . . . . . . . . 21
ix
3.2 Phase Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Fluid Property Calculations . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Chemical Reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5 Primary Equations and Variables . . . . . . . . . . . . . . . . . . . . 26
4 Thermal Streamline Simulation for Hot Water Flood 28
4.1 Streamline Simulation Framework . . . . . . . . . . . . . . . . . . . . 28
4.2 Streamline Formulation for Hot Water Flood . . . . . . . . . . . . . . 29
4.3 Specific Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3.1 Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3.2 Operator Splitting for Non-advective Processes . . . . . . . . . 32
4.3.3 Treatment of Compressibility . . . . . . . . . . . . . . . . . . 33
4.4 Hot Water Flood Results . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4.1 Incompressible Hot Water Flood . . . . . . . . . . . . . . . . . 34
4.4.2 Compressible Hot Water Flood . . . . . . . . . . . . . . . . . 39
4.4.3 Gravity Effects . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4.4 Heat Conduction Effects . . . . . . . . . . . . . . . . . . . . . 43
4.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5 Thermal Streamline for Steam Flood 45
5.1 Streamline Simulation Framework . . . . . . . . . . . . . . . . . . . . 45
5.2 Streamline Formulation for Steam Flood . . . . . . . . . . . . . . . . 46
5.3 Specific Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3.1 1D Pressure and Volumetric Flux Update Approach for Large
Compressibility . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3.2 Glowinski θ-Scheme Operator Splitting Approach for Non-advective
Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3.3 Preliminary Hybrid Simulation . . . . . . . . . . . . . . . . . 56
5.4 Steam Flood Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4.1 Heterogeneous Quarter-Five Spot Steam Flood . . . . . . . . . 58
5.4.2 Cyclic Steam Stimulation . . . . . . . . . . . . . . . . . . . . 62
5.4.3 Heterogeneous Multi-well Pattern Steam Flood . . . . . . . . 64
x
5.4.4 Vertical Cross Section Steam Flood . . . . . . . . . . . . . . . 69
5.5 Discussion and Applications . . . . . . . . . . . . . . . . . . . . . . . 72
5.5.1 Streamline Simulation as Fast Proxy: Cost Comparison . . . . 72
5.5.2 Optimization and Flux Patterns . . . . . . . . . . . . . . . . . 73
5.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6 In-situ Combustion Simulation 76
6.1 Kinetic Reaction Models . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2 1D Combustion Tube Simulation . . . . . . . . . . . . . . . . . . . . 79
6.3 Sensitivity Studies on 1D ISC Problem . . . . . . . . . . . . . . . . . 85
6.4 Grid Size Effects and their Cause . . . . . . . . . . . . . . . . . . . . 89
6.4.1 Grid Size Effects . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.4.2 The Cause of Grid Size Effects . . . . . . . . . . . . . . . . . . 91
6.4.3 Why not Adaptive Mesh Refinement? . . . . . . . . . . . . . . 92
6.4.4 The Need for Upscaling . . . . . . . . . . . . . . . . . . . . . 93
6.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7 Upscaling for In-situ Combustion Reactions 95
7.1 Non-Arrhenius Reaction Modeling . . . . . . . . . . . . . . . . . . . . 96
7.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.2.1 Pseudo Code . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.2.2 Implementation in Commercial Software . . . . . . . . . . . . 100
7.3 Upscaling Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.3.1 1D Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.3.2 2D Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3.3 3D Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.4 Discussion and Applications . . . . . . . . . . . . . . . . . . . . . . . 108
7.4.1 Valid Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.4.2 Sub-grid Scale Heterogeneity and its Effects on Reaction Up-
scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
xi
8 Front Stability Study for In-situ Combustion 121
8.1 1D Flow Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2 Contributors for Front Stability/Instability . . . . . . . . . . . . . . . 123
8.3 Numerical Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.3.1 Minimizing Numerical Errors . . . . . . . . . . . . . . . . . . 126
8.3.2 Small Sub-grid Scale Front Stability . . . . . . . . . . . . . . . 131
8.3.3 Large Field-scale Front Stability . . . . . . . . . . . . . . . . . 136
8.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
9 Future Directions 139
9.1 Thermal Streamline Simulation . . . . . . . . . . . . . . . . . . . . . 139
9.1.1 Non-advective Forces . . . . . . . . . . . . . . . . . . . . . . . 139
9.1.2 Guidelines for Commercial Code Development . . . . . . . . . 139
9.1.3 Thermal Streamline Simulation for SAGD? . . . . . . . . . . . 140
9.2 In-situ Combustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
9.2.1 History Matching of Field-scale ISC process . . . . . . . . . . 142
9.2.2 Calibrating the Operational Range for Sustaining ISC Combus-
tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.2.3 Grid Size Effects for Other Thermal or Non-Thermal EOR Pro-
cesses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.2.4 How to Use In-situ Combustion as a Gravity Drainage Process
in Fractured Carbonate Reservoir? . . . . . . . . . . . . . . . 145
10 Conclusions 150
A Simulation Inputs 154
B STARS Input File for Three Reaction Model 158
C Eclipse Input File for Three Reaction Model 166
D STARS Input File for Upscaled Three Reaction Model 179
E SAGD Input File 187
xii
F SAGD Input File for Grid Size Effects Study 193
Bibliography 203
xiii
List of Tables
4.1 Errors of different viscosity properties compared with reference result
(400× 400 STARS result) using L2 norm . . . . . . . . . . . . . . . . 37
4.2 Effects of thermal compressibility on the production (surface condition) 41
4.3 Heat Pe number and its influence on the simulation result. . . . . . . 44
5.1 Errors of thermal streamline and STARS compared to reference (360×180 STARS result) using the relative L2 norm. . . . . . . . . . . . . . 61
5.2 Sensitivity study on choice of θ. The simulation results close to each
other by using different θ. . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Errors of thermal streamline and STARS compared to reference (200×200 STARS result) using L2 norm. . . . . . . . . . . . . . . . . . . . 69
7.1 Explanation of major variables in the upscaling pseudo code. . . . . . 101
7.2 Explanation of major functions in the upscaling pseudo code. . . . . . 102
7.3 Summary of Sub-grid Scale Heterogeneity ISC Tests . . . . . . . . . . 118
A.1 Reservoir properties for hot water flood . . . . . . . . . . . . . . . . . 154
A.2 Viscosity relationships for hot water flood . . . . . . . . . . . . . . . 155
A.3 Fluid parameters for incompressible hot water flooding . . . . . . . . 155
A.4 Coefficients of density calculations in compressible hot water flood . . 155
A.5 Reservoir properties for steam flood . . . . . . . . . . . . . . . . . . . 155
A.6 Well control for steam flood . . . . . . . . . . . . . . . . . . . . . . . 156
A.7 Fluid parameters for steam flood . . . . . . . . . . . . . . . . . . . . 156
A.8 Reservoir properties for cyclic steam stimulation . . . . . . . . . . . . 156
xiv
A.9 Fluid properties in three reaction ISC model. . . . . . . . . . . . . . . 157
A.10 Rock properties in ISC simulation. . . . . . . . . . . . . . . . . . . . 157
A.11 Kinetics parameters for three reaction ISC model. . . . . . . . . . . . 157
A.12 Reaction stoichiometry for three reaction ISC model. . . . . . . . . . 157
xv
List of Figures
2.1 Streamline simulation framework, with the four major steps in a global
time step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Different zones in the field in-situ combustion process, courtesy of US
Department of Energy, Bartlesville, Oklahoma. . . . . . . . . . . . . . 13
2.3 Different zones in 1D in-situ combustion process. . . . . . . . . . . . 13
4.1 The permeability field in quarter five-spot hot water flood test case. . 34
4.2 The water saturation results in quarter five-spot hot water flood test
case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 The temperature results in quarter five-spot hot water flood test case. 35
4.4 The pressure results in quarter five-spot hot water flood test case. . . 35
4.5 The comparison between 1D fully implicit (FIM) transport solver and
1D single point upwind explicit (SPU) transport solver for thermal SL
simulation (50X50). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6 The surface cumulative production results for M=10 quarter five-spot
hot water flood test case. . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.7 The surface cumulative production results for M=100 quarter five-spot
hot water flood test case. . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.8 The surface cumulative production results for M=1000 quarter five-
spot hot water flood test case. . . . . . . . . . . . . . . . . . . . . . . 39
4.9 The saturation results in compressible quarter five-spot hot water flood
test case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.10 The temperature results in compressible quarter five-spot hot water
flood test case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
xvi
4.11 The surface cumulative production comparison results in compressible
quarter five-spot hot water flood test case. . . . . . . . . . . . . . . . 41
4.12 Saturation results in gravity test case 1 (quarter five-spot hot water
flood with dipping). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.13 Saturation results in gravity test case 2 (quarter five-spot hot water
flood with dipping). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.1 Steam condensation causes volume change and flux reduction right at
the steam front. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 Streamline segments constructed from the injector to the producer. . 50
5.3 1D streamline pressure and volumetric flux update approach. Trans-
missibility and block volume of each segment is calculated for the 1D
transport. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.4 Streamline (s, n, m) coordinates and cross sectional area along the
streamline. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.5 1D pressure update improvements in capturing transient pressure change
within a global time step, (M=1000 quarter five spot hot water flood
test case). With two BHP controlled wells, classical SL underestimates
the breakthrough. Both SL predict-correct [71] and SL with pressure
update improves the hot water breakthrough results, compared to fine-
scale reference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.6 Pressure transient behavior inside a global time step (M=1000 quarter
five spot hot water flood test case). The pressure stitched together
from all the 1D pressure solves (B) is similar to the pressure we get at
the beginning of the next global time step (C). . . . . . . . . . . . . 54
5.7 Schematic of coupled energy-mass transport. Changing the total en-
ergy of a grid block changes the fluid volume inside, thus causing fluid
transport to adjacent cells. . . . . . . . . . . . . . . . . . . . . . . . . 55
5.8 Glowinski θ-scheme operator splitting. Most of the transport is solved
along 1D streamlines [0 ∼ (1− θ)∆t]. Small amount of advective flux
is used to correct the volume changes in heat conduction/gravity step. 56
xvii
5.9 An equivalent representation of Glowinski θ-scheme operator splitting.
Most of the transport is solved along 1D streamlines. Small amount
of advective flux is used to correct the volume changes in heat conduc-
tion/gravity step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.10 Quarter five spot steam flood permeability field and initial streamline
shape. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.11 Viscosity curve for water and oil. . . . . . . . . . . . . . . . . . . . . 59
5.12 Quarter five spot temperature and gas saturation at 3000 days (pre
breakthrough). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.13 Quarter five spot water and oil saturation at 3000 days (pre break-
through). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.14 Sensitivity study on the choice of θ. The simulation results are close
to each other by using different θ. . . . . . . . . . . . . . . . . . . . . 62
5.15 Hybrid procedures in cyclic steam stimulation (SL → FV → SL) and
the initial streamline shape. . . . . . . . . . . . . . . . . . . . . . . . 63
5.16 Cyclic steam stimulation temperature result. . . . . . . . . . . . . . . 64
5.17 Cyclic steam stimulation oil saturation result. . . . . . . . . . . . . . 65
5.18 Cyclic steam stimulation production profile (SL and FV). . . . . . . . 65
5.19 2D multiple well test case permeability field. . . . . . . . . . . . . . . 66
5.20 2D multiple well test case initial streamline shape. . . . . . . . . . . . 66
5.21 Multiple well test case temperature and oil saturation at 1500 days
(pre breakthrough). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.22 Multiple well test case field production history (pre breakthrough). . 68
5.23 Multiple well test case temperature and oil saturation at 2100 days
(post breakthrough). . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.24 Multiple well test case field production history (post breakthrough). . 69
5.25 2D vertical cross section steam flood test case 1 (homogeneous perme-
ability field). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.26 2D vertical cross section steam flood case 2 (heterogeneous permeabil-
ity field). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.27 Flux pattern map, by Thiele and Batycky [106]. . . . . . . . . . . . . 74
xviii
6.1 Gas composition from ramped temperature oxidation experiments. . . 77
6.2 Relative permeability (water-oil and liquid-gas, respectively) and oil
viscosity curve used in the simulation of Hamaca oil 1D combustion
tube experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.3 Temperature, oil saturation and pressure profiles in 1D combustion
tube simulation after 498 min of air injection. . . . . . . . . . . . . . 81
6.4 Typical temperature profile history from a 1D combustion tube exper-
iment, from Lapene et al., [66] . . . . . . . . . . . . . . . . . . . . . . 81
6.5 Water, oil and gas production rates in 1D combustion tube simulation. 83
6.6 Produced gas composition analysis in 1D combustion tube simulation. 84
6.7 Oil bank and combustion front distance-time diagram from 1D com-
bustion tube simulation. . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.8 1D ISC simulation: peak temperature and combustion front velocity
as a function of the injection air flux. . . . . . . . . . . . . . . . . . . 85
6.9 1D ISC simulation: oil saturation profile as a function of the initial oil
saturation Soi. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.10 Temperature profiles as a function of the initial water saturation Swi
of 0.10, 0.24 and 0.50 from left to right. . . . . . . . . . . . . . . . . . 88
6.11 Combustion tube simulation of a mobile heavy oil with high initial oil
saturation Soi at 378min. . . . . . . . . . . . . . . . . . . . . . . . . 89
6.12 Characteristics of different fronts in 1D ISC problem: scenario 1 (bi-
tumen like oil viscosity and oil bank stays inside the steam plateau). . 90
6.13 Characteristics of different fronts in 1D ISC problem: scenario 2 (mo-
bile heavy oil viscosity and leading edge of steam plateau follows behind
the oil bank). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.14 Basic assumption for thermal reservoir simulation: instantaneous mix-
ing in each cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.15 Grid-size effects in 1D combustion tube simulation. We require at
least 200 grid blocks to achieve convergent solution in this 0.8m long
combustion tube simulation. . . . . . . . . . . . . . . . . . . . . . . . 92
xix
7.1 The critical physics for 1D ISC. When the reaction front sweeps through
the reservoir, a certain amount of oil becomes fuel and burns (usually
x = 5 ∼ 10% So). The rest of the mobile oil is displaced further
downstream. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.2 ISC upscaling work flow. . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.3 1D ISC upscaling tests: kinetic and upscaled model using different grid
resolutions. The coarse scale upscaled model matches the fine-scale
kinetics based reference, while the coarse scale kinetic model fails. . . 103
7.4 1D ISC upscaled model with different equivalent fuel amount Sofuel.
The front location has a direct relationship with the fuel amount Sofuel. 103
7.5 ISC upscaling in horizontal 2D 1/4 five spot case: O2 fraction in gas
phase with different grid resolution and both kinetic and upscaled model.105
7.6 ISC upscaling in horizontal 2D multi well case: coarse grid upscaled
model and fine grid kinetic model. . . . . . . . . . . . . . . . . . . . . 105
7.7 ISC upscaling in horizontal 2D heterogeneous case: coarse grid up-
scaled model and fine grid kinetic model. . . . . . . . . . . . . . . . . 106
7.8 ISC upscaling in 2D vertical case: coarse grid upscaled model and
fine-grid kinetic model. . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.9 3D field-scale case using the upscaled reaction model. Consistency is
achieved between different grid resolutions. . . . . . . . . . . . . . . 109
7.10 Permeability and oil saturation results for 2D case with uncorrelated
white noise permeability field. The combustion front propagates stably. 112
7.11 2D case with correlated heterogeneity (low perm in the middle). We
observe some oxygen bypass and change in equivalent fuel amount. . 113
7.12 2D case with correlated low permeability squares. We calculate oxygen
efficiency Eu = 0.90 and fuel amount Sofuel = 9.7%. . . . . . . . . . 114
7.13 2D case with horizontal layered permeability. We observe oxygen by-
pass and change in equivalent fuel amount. . . . . . . . . . . . . . . . 115
7.14 2D case with heterogeneity (layer 1 of SPE 10 [23]). The oxygen effi-
ciency is Eu = 0.91 and fuel amount Sofuel = 8.6%. . . . . . . . . . . 116
xx
7.15 2D case with channelized heterogeneity (layer 51 of SPE 10 [23]). The
oxygen efficiency is Eu = 0.85 and fuel amount Sofuel = 13%. . . . . 117
7.16 Different scales in measurements, geomodeling and reservoir simula-
tion, lecture notes of ERE 241 Seismic Reservoir Characterization. . . 120
8.1 Different fronts and flow regimes in 1D ISC process. . . . . . . . . . . 123
8.2 Perturbation length λ at leading edge of steam plateau, from [50]. . . 125
8.3 Trigger instability by changing the initial oil saturation at the boundary.127
8.4 Test of numerical errors with triggered instability. The effect of grid
size and 5-point versus 9-point scheme is illustrated. . . . . . . . . . . 129
8.5 Test of temporal numerical errors with automatic time-step selection
and restricted time step sizes. . . . . . . . . . . . . . . . . . . . . . . 130
8.6 Sensitivity to thermal conductivity. Heat conduction stabilizes the dis-
placement front in small lab-scale tests. Reduced thermal conductivity
cases show greater instability. . . . . . . . . . . . . . . . . . . . . . . 132
8.7 Nusselt number characterizes the heat conduction stabilizing effect,
compared to Fig. 8.6. . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.8 Pore blocking with the Kozeny-Carman correlation. Very small changes
are observed compared to base case. . . . . . . . . . . . . . . . . . . 134
8.9 Pore blocking with an exaggerated permeability reduction. The com-
bustion front slows down only when implementing pore blocking of
more than one order of magnitude reduction in permeability. . . . . 135
8.10 Unstable displacement in 2D larger scale ISC process. Heat conduction
is incapable of dissipating the energy of large wavelength perturbations. 137
9.1 Total phase velocity vectors in 2D SAGD process. Convection cell/loop
current occur when tracing streamlines. . . . . . . . . . . . . . . . . 141
9.2 Grid size effects in 2D SAGD simulation (early time 360 days). The
simulation achieves convergent solution when using grid size 0.5m ×0.5m in this case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
xxi
9.3 Grid size effects in 2D SAGD simulation (late time 1100 days). The
simulation achieves convergent solution when using grid size 0.5m ×0.5m in this case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.4 Concept of ISC assisted gas oil gravity drainage in fractured media.
Vent well is added to achieve hydrostatic gravity drainage condition in
the reservoir. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.5 Gravity stable ISC process for a single block of fractured media. Be-
cause of the gas diffusion, we are able to combust into the matrix.
Most of the flue gas is produced from upper right corner. . . . . . . . 149
xxii
Chapter 1
Introduction
1.1 Heavy Oil and Thermal Enhanced Recovery
A large part of the world oil resource exists in the form of heavy oil, which is usu-
ally defined as oil with API gravity less than 22 and viscosity typically larger than
100cp. Estimated original oil in place of more than 1.8 trillion barrels is present in
Venezuela, 1.7 trillion barrels in Alberta, Canada, and 20- 25 billion barrels on the
North Slope of Alaska. The development of such resources by traditional methods
(primary depletion, waterflood) is often inefficient due to the high viscosity of the
heavy oil. At such high viscosities, the oil flows extremely slowly or not at all. For
example, the bitumen resources in Athabasca oil sands typically have an extremely
large viscosity of about 106cp.
Thermal recovery processes rely on viscosity reduction of the oil through heat that
is injected (steam or hot water injection) or generated in-situ (in-situ combustion),
and are well suited for efficiently unlocking these heavy oil resources. Accordingly
to current U.S. Department of Energy data, thermal enhanced recovery techniques
account for about 50% of the domestic Enhanced Oil Recovery (EOR) production.
Steam flooding, cyclic steam stimulation and hot water flooding are widely used,
but other processes, such as in-situ combustion (ISC) and increasingly steam-assisted
gravity drainage (SAGD) are applied and are attractive to recover heavy oil resources.
1
CHAPTER 1. INTRODUCTION 2
1.2 Motivations and Objectives
Planning and management of thermal EOR processes generally make extensive use
of reservoir simulation. Nearly all the commercial and academic thermal simula-
tors are traditional finite volume (FV) based codes that use either a fully implicit
(FIM) time stepping method or an adaptive implicit (AIM) method [6, 37, 101]. The
computational costs for simulating thermal processes are usually high because of the
complexity of the problem and strong nonlinearity encountered. As a result, it is
time consuming to run optimization and/or sensitivity studies on grids with desirable
numerical resolution. For problems such as ISC and SAGD, the simulations show
extremely large computational costs with very strict requirements on the sizes of
grid blocks, due to the need to capture/resolve accurately the narrow thermal fronts
existing in these processes. Field-scale ISC simulation is still impractical, due to
large computational costs associated with accurately resolving the inch-sized reactive
combustion fronts. There is, therefore, an urgent need to develop fast and efficient
numerical simulation methodologies for thermal EOR problems. Predictive mathe-
matical models and efficient simulators are needed to improve our understanding of
these thermal EOR processes and enable cost-effective design of these projects.
In this work, we focus on three major problems: the extension of streamline
simulation (SL) to thermal problems (hot water flood and steam flood), the efficient
simulation of field-scale ISC process through reaction upscaling, and the analysis of
thermal front stability in ISC. The detailed objectives are as follows:
1. For the thermal streamline simulation, we are seeking a fast and effective reser-
voir simulator that gives sufficient accuracy for use in reservoir simulation stud-
ies, such as ranking, optimization and history matching. This is a first time that
streamline simulation is extended to complex thermal problems such as steam
flood. Problems such as fluid compressibility, strong coupling and gravity effects
need to be addressed.
2. For the ISC simulation, our main objective is to find an efficient simulation
technique to upscale the reaction kinetics for full field-scale ISC simulation. To
CHAPTER 1. INTRODUCTION 3
achieve this goal, the key problem relies on how to capture accurately the thin
reactive combustion front. We design upscaled reaction models that are tailored
to field-scale ISC process, which take into account the multi-scale nature of the
process.
3. Driving front stability is an important part for designing EOR processes. Ther-
mal front stability criteria needs to be extended to the ISC process. The stabil-
ity analysis is based on both analytical analysis and numerical tests. The front
stability is essential for guiding the design of field-scale ISC process, in which
different injection/production or wet/dry combustion scenarios are evaluated.
1.3 Thesis Outline
This thesis is organized in six parts:
Chapter 2 summarizes the literature on the main research topics pertinent to
this work, thermal reservoir simulation, streamline simulation, in-situ combustion
processes and thermal front stability.
Chapter 3 gives the basic equations for thermal compositional problems, including
the basic mass and energy transport equations, phase equilibrium, fluid property
calculations and kinetic chemical reactions.
Chapter 4 shows the thermal streamline simulation for hot water flood, in which
we include small compressibility, heat conduction and gravity effects.
Chapter 5 discusses the further extension of streamline simulation to steam flood-
ing. We extend streamline simulation to problems with strong compressibility, com-
plex phase behaviors, nonlinear coupling, and large gravity segregation effects. We im-
plement several novel techniques, include 1D pressure/flux update along the stream-
line, Glowinski θ-scheme operator splitting and hybrid finite volume/streamline sim-
ulation. The results show thermal streamline simulation capable of retaining solution
accuracy while reducing computational cost.
Chapter 6 shows the formulation of general reactive thermal reservoir simulation
for in-situ combustion. We then analyze the actual cause for the grid-size effects in
CHAPTER 1. INTRODUCTION 4
in-situ combustion simulation. Based on this, we conclude that ISC is a multi-scale
process such that the reaction kinetics and flow transport exist in two completely
different scales.
Chapter 7 introduces the concept of upscaling the chemical reaction kinetics for
field-scale in-situ combustion simulations. We present the reaction upscaling method-
ology and workflow. We show test cases to demonstrate the consistency achieved by
reaction upscaling and the capability for handling field-scale problems.
Chapter 8 shows the reaction front stability study for in-situ combustion process.
Multiple fronts and flow regimes are identified for the ISC process. We further define
the five major factors affecting the front stability, viscous force, condensation, heat
conduction, coke plugging and gravity. Numerical simulation tests are performed to
explore and construct the stability criteria and regions.
In Chapter 9, we propose the future research directions for thermal streamline
simulation and in-situ combustion process simulation.
Chapter 2
Literature Review
2.1 Streamline Simulation
2.1.1 Streamline Simulation Concept and Basics
Most of the academic and industrial reservoir simulators are based on Eulerian meth-
ods, in which the grids are considered stationary and fluids are transported between
discrete cells [6]. All these simulators treat the pressure equation implicitly. They
usually use either fully implicit and adaptive implicit time stepping, or implicit pres-
sure explicit saturation/composition (IMPES/IMPEC) method [18]. The computa-
tional costs are often very high, when simulating large full field reservoir models
with large number of simulation grid blocks. For thermal enhanced recovery simula-
tions in particular, the computational costs are even greater compared to black oil or
compositional problems, because of the large number of unknowns, complex thermal
phase behavior, strong nonlinearities in the system and highly coupled nature of the
governing equations. This creates great difficulty for running optimization and/or
sensitivity studies on grids with desirable numerical resolution.
Streamline-based flow simulation is a fundamentally different simulation technique
than traditional Eulerian methods [71, 7]. Streamline is defined as the curve that is
instantaneously tangent to the velocity vector of the flow. Streamline simulation is
based on the observation that in heterogeneous reservoirs the time scale with which
5
CHAPTER 2. LITERATURE REVIEW 6
fluids flow along streamlines is often faster than the time scale at which the streamline
locations change significantly. This allows decoupling of the costly 3D transport
problem into a set of 1D advection problems along the streamlines. Streamline based
reservoir simulation is conducted based on a series of global time steps and a dual grid
approach [7]. During each global time step, the pressure equation is initially solved
implicity on the Eulerian grid, followed by streamline tracing on the corresponding
fixed total phase velocity field [92]. Flow transport is then calculated along each
streamline, by solving the corresponding fractional flow transport problems along
the 1D streamline grid. A coordinate transformation is conducted to transfer the
flow problem from physical arc-length coordinate to 1D time of flight (TOF) grid
[7]. The time of flight is the time required to reach a distance, s, along a streamline
based on the velocity field along the streamline. Eventually, the solution is mapped
back to the original Eulerian grid, which concludes a global time step. Streamline
methods were originally applied to incompressible, multi-phase flow problems without
gravity, capillary or diffusive terms in the transport equations. Gravity, capillary and
diffusive terms have since been included through the operator splitting techniques
[14, 13], usually at the end of a global time step. The entire work flow for a global
time step in streamline simulation is shown in Fig. 2.1
2.1.2 Advantages and Disadvantages of Streamline Simula-
tion
Streamline simulation is deemed as an effective and complementary technology to
traditional Eulerian finite volume based simulation. This is because streamline simu-
lation is particularly effective in solving large, geologically complex and heterogeneous
systems, where flow is dictated by well rates and positions, reservoir structure, per-
meability and porosity, fluid mobility and gravity. These are the problems for which
traditional finite volume simulation has difficulties. A thorough summary and review
of streamline simulation theory and application is provided by Thiele [105].
The most important advantage of streamline simulation is its inherent efficiency.
Streamline simulation has been successfully applied to reduce computational costs
CHAPTER 2. LITERATURE REVIEW 7
Figure 2.1: Streamline simulation framework, with the four major steps in a globaltime step.
CHAPTER 2. LITERATURE REVIEW 8
whilst retaining accuracy. The computational efficiency is achieved because the trans-
port problem is decoupled from the 3D grid and solved along each 1D streamline,
which can use well-established mathematical techniques. This follows an inherent di-
vide and conquer strategy, in which a high dimension fully coupled system is divided
into a series of smaller 1D systems and solved individually. Furthermore, the number
of streamlines typically increase linearly with the number of active cells. And the
streamlines only need to be updated infrequently, which guarantees the usage of large
global time steps and a smaller number of pressure updates. This is different from
IMPES/IMPEC type methods, whose time step sizes are constrained by the well-
known Courant-Friedricks-Lewy (CFL) stability restriction. This is why streamline
simulation can exhibit a close to linear scaling in run times as a function of active
cells in the model, according to Thiele [105].
The bulk of the computational work in streamline simulation is typically associated
with solving the relevant transport equations along the 1D independent streamlines for
compositional/thermal problems. This makes the streamline simulation architecture
inherently parallelizable. Recent work on a research streamline simulator provides
promising results in parallel speed-up [70]. Parallelization of a commercial streamline
simulator to multi-core architectures based on the OpenMP programming model and
its performance on various field examples have also been presented [9].
The computational efficiency of streamline simulation is especially suitable for
work flows involving many simulations as might be the case in ranking, history match-
ing and optimization problems. Commercial streamline simulation is applied for such
purposes [1]. One example is the reservoir history matching studies for Iraq’s South
Rumaila field that first used simple material-balance approach, followed by fast reser-
voir simulations with streamlines, and eventually full finite volume simulation in se-
quence to aid understanding of reservoir-flow behaviors [59].
A powerful application of streamline simulation is the capability to define dynamic
well allocation factors (WAFs) and injection efficiencies (IEs) using streamline derived
information [106]. Because WAFs and IEs are derived directly from streamlines, the
data reflect all the complexities impacting the dynamic behavior of the reservoir
model. IE quantifies how much oil can be recovered at the offset producing wells for
CHAPTER 2. LITERATURE REVIEW 9
every unit of water injected by an injector. Improved water flood management can
be implemented by reallocating injection water from low-efficiency to high-efficiency
injectors.
Streamline simulation is a type of Euler-Lagrange method, in which the pressure
and transport equations are treated and solved in different stages. Compared to fi-
nite volume simulation, streamline simulation implements a dual grid and dual time
stepping approach. Therefore, one disadvantage is that it is not mass conservative,
because of the mappings between the Eulerian grid and 1D streamline grid. Improved
mapping methods have been proposed to control the mass balance errors [72]. De-
spite the lack of exact mass conservation, studies have shown that the mass balance
errors are usually relatively small and streamline methods predict the global sweep
in heterogeneous reservoirs effectively.
2.1.3 Streamline Simulation for Complex Physics Processes
Streamline/streamtube simulation was initially applied successfully to problems such
as water flooding, first contact miscible flooding and tracer flows [76, 8, 7]. Streamline
simulation results have shown advantages in retaining simulation accuracy, reducing
numerical dispersion and achieving speed-up against finite volume simulations. Ac-
cordingly, commercial streamline simulators are designed to handle black oil, first
contact miscible injection, and tracer flows [1].
Since then, streamline simulation has also been applied for gas injection pro-
cesses and enhanced gas condensate processes, which includes full equation of state
(EOS) compositional effects [107, 29, 45, 56, 99]. For multi-contact miscible gas
injection, high resolution 1D solutions are constructed either by Method of Charac-
teristics (MOC) analytical solutions or higher order accurate numerical discretization
schemes, such as total variation diminishing schemes (TVD) and essentially non-
oscillatory schemes (ENO) [71, 100]. These techniques more accurately represent
the composition paths in phase space for 1D displacements, and hence, better pre-
dict local displacement efficiency than conventional first order numerical simulation.
While the MOC analytical solutions are attractive, they can only be used along
CHAPTER 2. LITERATURE REVIEW 10
SL that have constant initial and boundary conditions and cases where the stream-
lines are fixed over all time. This is the major limitation for mapping analytical
solutions along the streamline. Streamline simulation enables the usage of the high
accuracy 1D solution techniques for solving the 3D reservoir processes, because the
flow transport is solved along 1D streamlines. The simulation results have shown
improvements in capturing the viscous fingering effects and reducing the numerical
smearing/dispersion in simulating the 3D miscible/immscible gas injection and gas
condensate processes. Streamline simulation effectively controls the numerical dis-
cretization errors and makes reliable predictions of the local displacement efficiency
and global sweep efficiency.
Chemical EOR processes, for example polymer flooding, can also be modeled
effectively by streamline simulation [104]. Because the flow transport is solved along
each streamline, it is convenient to include accurate 1D solutions using locally refined
grids blocks, which can potentially help to control the grid size effects (numerical
errors due to using large grid blocks) encountered in chemical EOR simulations [111].
Streamline simulation has also been extended to model fractured media dual
porosity single permeability problems [36]. The matrix-fracture interactions are in-
cluded in this work for two-phase transport in fractured systems. Fluid transfer be-
tween fracture and matrix enters as source/sink terms in the 1D transport equations
along the streamlines.
Traditional streamline simulation has been based on the incompressible flow as-
sumption [7], in which the incompressible total phase velocity field is fixed during
a global time step. This becomes a strong restriction in cases when we need to in-
clude fluid and rock compressibility. Consequently, streamline simulation has been
extended to account for compressible fluid and rock properties [88, 12]. In the first
work [21, 88], the total phase velocity field is still fixed during a global time step. The
divergence of the velocity field is added into the 1D transport equation as source/sink
terms along each streamline. In this way, the changes in the fluid volume with pres-
sure is captured, accounting for the compressibility effects. In the second work [12],
a compressible formulation for streamline simulation is presented, that also considers
API gravity tracking. Compared to the previous work by Cheng [21], cumulative
CHAPTER 2. LITERATURE REVIEW 11
volume is used in the 1D transport, instead of time of flight, which reduces mass-
conservation errors and further allows the update of fluid phase velocities along each
individual streamline. Test cases have shown its capability to handle significant liquid
compressibility.
To the best of the author’s knowledge, very little work has been conducted in
extending streamline simulation to heavy-oil or thermal enhanced recovery problems.
Streamline simulation is investigated as a potential method for heavy-oil waterflood
performance forecasting [87]. The authors show that streamline simulation is a pow-
erful tool for quick screening of heavy-oil waterflood projects, estimating oil recovery,
and water breakthrough time. The overall streamline results are similar to that of
FV simulation, with reduced grid orientation effects. In the recent work by Pasarai
and Arihara [89], the authors propose an extension of streamline methods to hot
water flood, which solves the energy equation along streamlines. The algorithm first
computes the mass transport equations based on FV simulation and the current tem-
perature. It then solves the energy equation along the 1D streamlines. The work may
be further improved by fully solving both mass and energy transport along streamlines
and rigorously accounting for the compressibility effects and non-advective forces.
There is an urgent need for fast approximate solvers for thermal simulation prob-
lems. The essence of this work is to explore the extension of streamline simulation
to thermal enhanced recovery processes, which include hot water flooding and steam
flooding. Our aim is to design a fast and effective simulator that gives sufficient
accuracy for use in reservoir simulation studies, such as ranking, well placement op-
timization and history matching.
2.2 In-situ Combustion
2.2.1 In-situ Combustion Process
Among all the thermal recovery processes, ISC has been a source of interest for a
long time. In-situ combustion (ISC) is the process of injecting air (or air enriched
with oxygen) into oil reservoirs to oxidize part of the crude oil and enhance recovery
CHAPTER 2. LITERATURE REVIEW 12
through the heat and pressure produced [93, 96]. In contrast to other thermal recovery
processes, the main energy required to displace the oil in ISC is generated in the
reservoir from the heat released by combustion reactions between oxygen and fractions
of the crude oil. In-situ combustion has been successfully applied to both light oil
and heavy oil recovery.
A typical field-scale forward ISC process is shown in Fig. 2.2. The oil is initially
ignited around the injection well. Continuous air injection keeps the combustion front
propagating towards the production well. As air is injected and the combustion front
moves in the reservoir, several zones are identified. Starting from the injector, the
first zone is the burned zone in which the reservoir volume has already been swept
by the combustion front. The burned zone typically contains the injected air and
almost no hydrocarbons. The combustion front has the highest temperature and
this is the place where most of the chemical reactions are taking place. Most of the
heat is generated in this narrow zone, typically several inches thick. The injected
oxygen reacts with the hydrocarbons, generating flue gas and steam. Hydrocarbons
contacted by the leading edge of the high temperature zone undergo thermal cracking
and vaporization. Mobilized light components are transported downstream where
they mix with the original crude. The heavy residue, which is normally referred to
as solid coke, is deposited on the core matrix and is the main fuel source for the
combustion process. Downstream of the vaporization zone is the steam plateau that
is formed from water of combustion and vaporization of formation water. Further
downstream, the steam condenses into a hot water bank when the temperature drops
below the steam saturation temperature. The leading edge of the hot water bank
is the primary area of oil mobilization where the oil is banked by the hot water. A
schematic of the 1D ISC process is also shown in Fig. 2.3.
2.2.2 Field Applications of ISC
Since its discovery in the early 20th century, ISC as an enhanced oil recovery process
has experienced a checkered history [96]. It was initially successfully applied as an
effective thermal recovery technique in California heavy-oil fields in the 1950s and
CHAPTER 2. LITERATURE REVIEW 13
Figure 2.2: Different zones in the field in-situ combustion process, courtesy of USDepartment of Energy, Bartlesville, Oklahoma.
Figure 2.3: Different zones in 1D in-situ combustion process.
CHAPTER 2. LITERATURE REVIEW 14
1960s, such as the West Newport ISC project [96]. The success achieved in California
ISC projects brought the process to Canada during the 1970s, 1980s and 1990s. Some
successful projects were demonstrated in the Lloydminster heavy-oil accumulations
in Canada, such as Murphy Oil’s Eyehill thermal project [81], Amoco’s Morgan cyclic
combustion project [75] and Husky’s ISC projects [65]. ISC has also been successfully
used as a follow-up process to oil sand cyclic steam stimulation in BP Canada’s Wolf
Lake bitumen thermal project [48]. This project is well known as the pressure-up
blowdown combustion process in a channeled reservoir with existing thermal com-
munication with adjacent wells. In spite of these successes, there were many more
failures. For example, in the Viking Kinsella Wainwright B project, the combustion
front failed to sustain and switched over to the undesirable low temperature oxidation
(LTO) mode, when the air injection rate was decreased [17]. ISC processes in Cana-
dian heavy oils and oil sands rarely reached its theoretical potential. Comprehensive
review of the field trials, successes and failures, and operational issues in Canada are
summarized [110, 80].
Currently the successful heavy oil ISC commercial projects in operation include
the Suplacu de Barcau field in Romania and the Balol/Santhal fields in India. The
Suplacu field is a conventional shallow heavy-oil field, that started in-situ combustion
process in 1965. The project implements a line-drive configuration and now has an
approximately 11 km long combustion front that is gradually driven down-dip within
the reservoir. As the combustion front moves close to the down-dip row of producers,
the producers are converted to new injectors and the old injector rows are converted
to water injectors. The details of the Suplacu field operations are summarized in [91].
The Balol/Santhal ISC projects started in 1990. They are heavy/viscous oil reservoirs
in the heavy-oil belt of Mehsana, India. Similarly, they implemented air injection from
up-dip of the reservoir and drive the combustion front down-dip towards the producer.
Both dry and wet combustion were used in this project [94].
In-situ combustion or air injection has also been applied for light-oil reservoirs,
which is often known as the High Pressure Air Injection (HPAI) process [46, 79, 42].
One of the most well known early light oil in-situ combustion project is the double
displacement process in Amoco’s West Hackberry project [41]. HPAI has also been
CHAPTER 2. LITERATURE REVIEW 15
widely used as an enhanced oil recovery technique for carbonate reservoirs. It is the
distant second most widely used EOR technique to CO2 injection for carbonate reser-
voirs in the United States [73]. Most of the light-oil HPAI processes are operated at
the low temperature oxidation mode without high temperature combustion. These
processes mostly generate an immiscible flue gas drive, or serve as a pressure main-
tenance technique during the gravity drainage process. In this work, we will focus
mainly on the heavy-oil in-situ combustion process, where the reactions are intended
to operate in the exothermic high temperature oxidation (HTO) mode. The high
temperature oxidation and low temperature oxidation modes are discussed in detail
in section 6.1.1.
2.2.3 Numerical Simulation of In-situ Combustion
In traditional thermal reservoir simulation, conservation equations are solved numer-
ically on discretized reservoir grid blocks [27, 112, 61]. The governing mass and
energy transport equations for reactive thermal compositional problems include ex-
tra source/sink terms to represent the chemical reaction effects. The reaction source
and sink terms are calculated cell by cell using Arrhenius kinetics during each time
step, as shown in Eq. 3.16. This is the common reaction modeling technique in most
commercial simulators [37, 101].
Different Arrhenius kinetic reaction models have been developed for ISC simula-
tions. All these reaction models are based on multiple oil pseudo components and on
grouping the chemical reactions into a series of pseudo reactions. Parameters in these
models are usually derived from laboratory experiments of kinetic cells [24]. One of
the earliest and most commonly used models is the so-called Crookston model, also
known as the minimal model [30]. It comprises two oil components (heavy oil HO
and light oil LO) and four chemical reactions. Both HO and LO directly combust
with oxygen. The component HO pyrolyzes generating coke. Eventually, coke burns
completely with oxygen. Other models such as the Belgrave model [11] and SARA
model [43] group pseudo components of the crude oil based on the SARA fractions
(Saturate, Aromatic, Resin and Asphaltene). These methods also employ a series of
CHAPTER 2. LITERATURE REVIEW 16
reactions, including coking (pyrolysis), low temperature oxidations, and high temper-
ature oxidations. The reactions generate some intermediate compound products and
eventually fully combust to generate heat. Another more simplified three reaction
model has also been proposed, which includes only one oil component and three se-
quential kinetic reactions (pyrolysis, LTO and HTO) [33]. The details of LTO and
HTO reaction modes are provided in section 6.1.1.
2.2.4 Challenges in Combustion Reaction Modeling
Most ISC simulation studies are performed on small laboratory scale problems, such
as history matching of kinetic cells and combustion-tube experiments. It is well
known that when increasing the length scale to full field reservoir models, severe
grid-size effects are encountered in in-situ combustion simulations [61, 112, 28, 47, 53].
This means the simulation results, especially the combustion front speed and peak
temperature, become functions of the grid block sizes.
To mitigate the grid-size effects in ISC simulations, various empirical and heuristic
techniques have been developed. In many cases, the reaction kinetics parameters are
changed or tuned to force the oil to burn in a large field-scale grid block [47]. This
can be achieved, for example, by adjusting the activation energy to small or even zero
values, which lowers the threshold crude-oil burning energy requirement. In some
cases, the temperature values in Arrhenius kinetics are also adjusted to make the
oil much easier to burn [28]. In another work [53], the oil phase flux has also been
modified/enhanced in an attempt to force more oil to flow out of the grid blocks
where combustion reactions are taking place. All these parameter adjustments have
to be repeated every time the grid size is changed. A comprehensive summary of all
these techniques is presented in [74].
2.2.5 Alternative Approaches
Alternatives to the Arrhenius kinetics based reservoir simulation first include the
various types of analytical solutions. Empirical and analytical models, such as the
Ramey and Gates model, were developed for prediction of ISC recovery performance
CHAPTER 2. LITERATURE REVIEW 17
[44]. The Ramey and Gates model is essentially based on the material balance cal-
culations that estimates the amount of oil and water displaced from the volume of
reservoir swept by the high temperature oxidation front. Many variants have been
proposed since then [64], with the material balance main concept unchanged. A re-
view of these approaches and corresponding economic considerations for ISC process
is summarized by Nodwell [85]. Another set of analytical solutions were derived by
Akkutlu [3]. The authors proposed an analytical solution based on a constant pre-laid
fuel (coke) amount, and coke burning reactions with the injected oxygen.
There also exist many alternatives to Arrhenius kinetics in reservoir simulation.
A thin-flame technique is suggested for in-situ combustion simulation in a 1D system
[32]. In this work, the reaction front velocity is assumed to be constant. A coordinate
transformation is conducted to track the front at the same speed, thus changing the
governing equations to a steady-state boundary value problem. Furthermore, a trial
for field-scale history match is also performed for the Suplacu field pilot in Romania
[91]. Constant coke deposition and fuel amount is assumed in this simulation study,
that greatly reduces the grid-size effects. The fuel burning reaction is still based on
Arrhenius kinetics, with temperature modified according to method by Coats [28].
Finally, a field-scale in-situ combustion simulator is developed using a moving front
representation by Hwang [52]. The front is viewed not only as a moving heat source,
but also as a displacement pump enhancing oil flow. In this simulator, detailed
kinetics and thermodynamic relations are not used. Instead it approximates the
overall effects of combustion with a few measurable parameters. All these approaches
share one thing in common, that is to simplify the traditional kinetics based simulation
and introduce approximations valid at larger scale.
Finally, specialized numerical techniques such as adaptive mesh refinement [22, 84]
and operator splitting [61] have also been proposed or applied to ISC modeling. These
methods lead to some improvements. For example, simulation accuracy is clearly
improved through the use of operator splitting technique to capture the fast kinetic
reactions separately from the flow transport using higher order ordinary differential
equation (ODE) solvers [61]. The dynamic gridding option has been implemented in
commercial thermal simulator CMG STARS [101]. The gridding identifies the moving
CHAPTER 2. LITERATURE REVIEW 18
front through large gradients of specific properties (temperatures, fluid saturations
and compositions). In the front vicinity, it will de-amalgamate the original coarse
cells, and later on re-amalgamate them once the front has passed through. Though
test cases of ISC process, it is shown that dynamic gridding is capable to divide the
CPU time of thermal simulations by a factor of 2 to 3 [22]. The simulator, however,
still lacks the capability to scale up the grid sizes several orders of magnitude to coarse
grid full field reservoir simulation. A field-scale predictive ISC simulator is therefore
urgently needed.
2.2.6 Thermal Front Stability for ISC
Front stability is an important issue when designing an EOR processes. In the classical
steam flooding problems, the thermal front stability criterion has been well established
[15]. In a 2D horizontal problem, the main factors affecting steam flood front stability
include: pressure gradient (viscous force), steam condensation, and heat conduction.
Among them, the pressure gradient contrast serves as a destabilizing effect, because
of the adverse mobility ratio of steam driving the hot water and the viscous oil
bank. The condensation of steam into hot water serves as a stabilizing effect. A
detailed formulation of front-stability criterion is presented for the idealized problem
of steam driving hot water in the porous media in [15]. Heat conduction also plays an
important role in stabilizing the steam front, which is dependent on the perturbation
wavelength λ [78]. For small scale perturbations, heat conduction quickly dissipates
the heat of the steam finger, thus stabilizing the front. On the other hand, for
larger scale perturbations, the stabilizing effects is less significant. This actually
serves as one of the main mechanisms for using steam injection as EOR method for
highly heterogeneous reservoirs such as fractured carbonate and fractured diatomite
formations [50]. A combined study of analytical methods and thermal simulation to
examine frontal stability in high-porosity diatomite rocks is presented in [50].
One of the early work of examining the front stability for wet/dry in-situ combus-
tion process is by Armento et al., [5]. An idealized model is introduced to analyze the
combustion zone stability. On the upstream size, it is the burned zone with air and
CHAPTER 2. LITERATURE REVIEW 19
water. In the downstream side, it is the region with combustion generated steam/flue
gas mixture. The similar analytical approach for steam front stability [78] is utilized
here for studying the combustion zone stability. The drawback of this study is that
it does not take into account the existence of heavy oil in the in-situ combustion
process. Furthermore, multiple fronts besides the combustion zone exist in the ISC
process, as it will be shown later in chapter 8. A more detailed study is required to
fully understand the front stabilities.
In the recent work by Javad et al., [54], a 2D vertical physical experiment was im-
plemented in order to study the combustion front stability in a top down combustion
configuration. Top-down combustion is a process in which air is injected at the top of
the reservoir through a horizontal well or a number of vertical wells and fluids are pro-
duced from the horizontal producers placed near the bottom [26]. It was found that
the combustion front first spreads horizontally then moves down dip stably. These
experiments shows that gravity also helps with the front stability. This is consistent
and analogous with the classical vision of gravity stabilizing effects in steam based
gravity drainage process [5]. Furthermore, a concept of front self-correction is de-
scribed in this work by two effects: pore blocking and local drop in air flux [54]. It
was mentioned ”pore blocking: when an air finger develops, displacement of liquid
phase by the combustion front will build up a localized fluid bank which will in turn
reduce the gas mobility to such an extent that air will be diverted to other locations
on the combustion front” and ”local drop in air flux: finger formation in the com-
bustion front, enlarges front area locally and drops air flux in the finger. Lower flux
slows down combustion reactions in the finger and postpones its progress”. Further
study needs to be performed to study the detailed mechanism whether ISC front is
self-correcting or not.
Historically, premature combustion front breakthrough has also been an impor-
tant issue for field operations of in-situ combustion. For example, it is very often
advised that the combustion process should start from the top of the reservoir and
use line drive configuration to push everything downward towards production wells
[85]. This is mainly due to the fact that in a dipping reservoir, gravity greatly con-
tributes to stabilizing the ISC fronts. This technique has been applied to both heavy
CHAPTER 2. LITERATURE REVIEW 20
oil and light oil ISC process, such as Suplacu de Barcau field in Romania and the
West Hackberry project [41, 91]. Furthermore, cyclic combustion technique has been
applied extensively in field operations in Canada in order to overcome premature
break through in the highly heterogeneous channel sand in the Lloydminster region
[74]. The BP Canada’s Wolf Lake pressure-up blowdown combustion process is also
designed to over come the over early breakthrough in heated channels from previous
cyclic steam stimulation [48]. Unfortunately, the early breakthrough still causes well
failures in part of the project. This shows the importance of understanding the front
stability in ISC process.
Overall, thermal front stability criteria needs to be extended to the ISC process.
The stability analysis is based on both analytical analysis and numerical tests in
this work. The front stability is essential for guiding the design of field-scale ISC
process, in which different injection/production or wet/dry combustion scenarios will
be evaluated for field operations.
Chapter 3
Governing Equations
3.1 General Conservation Equations
The governing transport equations for general thermal, compositional porous media
flow include Nc mass conservation equations and one energy conservation equation.
The mass balance equation for the i-th component is given by
∂
∂t
np∑j=1
φyijρjSj +∇ ·np∑j=1
yijρjuj = q̃i, (3.1)
where np is the number of total phases, φ is the porosity of the porous medium, yij is
the mole fraction of component i in phase j (Nc∑i=1
yij = 1), ρj is the phase molar density,
Sj the volumetric phase saturation (np∑j=1
Sj = 1), uj the velocity of phase j, and q̃i the
mass source or sink term.
The energy conservation equation is given by
∂
∂t
(1− φ) Ur +np∑j=1
φρjSjUj
+∇ ·np∑j=1
hjρjuj +∇ · (−α∇T ) = q̃h, (3.2)
where Ur is the internal energy of rock, Uj is the internal energy of phase j, hj is the
enthalpy of phase j and α the heat conductivity, and q̃h is the heat source or sink
term. Ut = Ur +np∑j=1
φρjSjUj is the total energy of the grid block.
21
CHAPTER 3. GOVERNING EQUATIONS 22
The volume conservation expresses that the fluid must fill the pore space, that is,
Vp =np∑j=1
Vj, (3.3)
where Vj is the volume of phase j, and Vp is the volume of the pore space.
Total fluid volume balance leads to the pressure equation
ct∂P
∂t+
nc∑i=1
∂Vt
∂Ni
∇ · np∑j=1
yijρjuj
+∂Vt
∂Ut
∇ · np∑j=1
hjρjuj
=∂Vt
∂Ut
q̃h +nc∑i=1
∂Vt
∂Ni
q̃i.
(3.4)
Here, ct is the total compressibility and q̃i and q̃h are the source and sink terms for
composition and energy. The partial derivative ∂Vt
∂Niand ∂Vt
∂Utexpress the dependency
of the total fluid volume on composition and total internal energy, respectively.
For simplicity, we first consider the case without capillary effects. The multi-phase
extension of Darcy’s law gives
uj = −krj
µj
k (∇P − ρjg∇D) , (3.5)
with P the pressure, k the permeability tensor, krj the relative permeability of phase
j, µj the phase viscosity of phase j, ρj the mass density and D the depth of the
reservoir.
3.2 Phase Equilibrium
We group all the components into two categories: water-like (aqueous) and oil-like
(oleic). Aqueous components exist in both gas and water phase, while oleic compo-
nents exist in the oil and gas phase. In addition to the conservation equations, we
assume the system is in instantaneous thermodynamic equilibrium. This is expressed
by the equality between chemical potentials, or equivalently fugacities,
f oi = f g
i ,
fwk = f g
k ,(3.6)
CHAPTER 3. GOVERNING EQUATIONS 23
with f oi and f g
i the fugacity of the i-th oleic component in the oil phase and gas phase,
fwk and f g
k the k-th aqueous component in the water and gas phase. The fugacity
equality equation Eq. 3.6 is usually solved using the cubic equation of state (EOS)
approach. The phase equilibrium calculations are also known as flash calculations,
with details found in [77]. We introduce a simplified ideal mixture approach in which
phase equilibrium is modeled using the K-value approach [101],
yi,o ·Ki = yi,g,
yk,w ·Kk = yk,g,(3.7)
with Ki and Kk being the K-value for i-th and k-th component. The K-value is
pressure-temperature dependent and calculated by the Crookston correlation [101],
Ki =KV 1
Pexp(
KV 4
T −KV 5
). (3.8)
with KV 1, KV 4, KV 5 being the constants in Crookston K-value correlation.
For most heavy oil applications, the crude oil has significant amount of heavy
components. Because heavy oil reservoirs are usually shallow in depth, the operating
pressure is often very low. Therefore, the fluid system is relatively far away from
its critical point. The K-value approach is a valid approximation for most of the
heavy oil applications, and is also implemented in most commercial thermal reservoir
simulators [101, 37]
CHAPTER 3. GOVERNING EQUATIONS 24
3.3 Fluid Property Calculations
For each component the fluid enthalpy and latent heat calculation are performed in
the same way as in [101], that is,
hig(T ) =T∫
Tref
Cgi(T ) · dT,
hiv(T ) = HV Ri · (Tc − T )0.38,
hil(T ) = hig(T )− hiv(T ).
(3.9)
hiv is the latent heat for the i-th component. hil and hig are the liquid phase enthalpy
and gas phase enthalpy for each component, respectively. HV Ri is a constant for
latent heat calculations.
We calculate the liquid phase (aqueous and oleic phase) molar densities for each
component using the commonly used relationship
ρil (T, P ) = ρscil exp [cil (P − Psc)− ail (T − Tsc)] , (3.10)
where cil and ail are the compressibility and thermal expansion coefficients for i-th
component. The Redlich-Kwong EOS is used to calculate the gas phase density for
each component [101]. It is given by,
ρig = P/(RTZi),
Z3i − Z2
i + (A−B2 −B)Zi − AB = 0,
A = 0.42748 · ( PPc
) · (Tc
T)2.5,
B = 0.08664 · ( PPc
) · (Tc
T).
(3.11)
We assume ideal mixing when calculating the phase density and phase enthalpy:
1/ρj =Nc∑i=1
(yij/ρij)
hj =Nc∑i=1
yij · hij
(3.12)
CHAPTER 3. GOVERNING EQUATIONS 25
The viscosities of the i-th component in the water and oil phases are given by the
correlation [93]
µil = Ail exp(Bil/T ), (3.13)
where Ail and Bil are the empirical parameters for the temperature dependent vis-
cosities. Later, the water and oil phase viscosities are calculated by the logarithmic
mixing rule,
ln(µl) =nc∑i=1
yil · ln(µil). (3.14)
The viscosities of gas phase is calculated simply as a function of temperature T as,
µg = 0.0136 + 3.8 · 10−5 · T, (3.15)
with the units for µg in cp, and the unit for T in oC.
3.4 Chemical Reactions
When chemical reactions are present, such as in the case of ISC process, the governing
mass and energy transport equations become,
∂Ci
∂t+∇ · qi = Qwell
i + Qreaci ,
∂Ut
∂t+∇ ·
(qh,adv + qh,cond + qh,loss
)= Qh,well + Qh,reac,
(3.16)
where Ci is the mass composition of the i-th component, Ut is the total internal
energy, qi and qh,adv are the advective mass and energy fluxes, qh,cond and qh,loss
are the terms for heat conduction and heat losses, and Qwelli and Qh,well are the well
terms. The flux terms qi, qh,adv and qh,cond are calculated as previously. The phase
equilibrium and the fluid property calculations are the same as before.
The extra terms for chemical reactions are the source/sink terms Qreaci and Qh,reac.
We assume all chemical reactions to be kinetically driven. The reactions are mod-
eled using standard Arrhenius kinetics, as in most commercial reservoir simulators
[101, 37]. Both homogeneous and heterogeneous reactions are allowed. Nr chemical
reactions are assumed in the simulation. We denote the stoichiometry matrix for
CHAPTER 3. GOVERNING EQUATIONS 26
these Nr chemical reactions by A, with Air the stoichiometry coefficient for compo-
nent i in the r-th reaction. We use negative values of Air for participating reactants,
and positive values for reaction products. The mass conservation of the r-th chemical
reaction gives ∑i
Air ·Mi = 0 (r = 1, · · · , Nr), (3.17)
where Mi is the molecular weight of the i-th component.
By using Arrhenius kinetics to express the rate of each reaction, the reaction
source/sink term for the i-th component is written as
Qreaci =
Nr∑r=1
Air · ar · exp(−Ear
Rg · T) ·∏l
(Cl)nrl , (3.18)
where Ear is the activation energy for the r-th reaction, ar is a constant known
as the pre-exponential frequency factor for the r-th reaction, Rg is the ideal gas
constant, T is the local absolute temperature, nrl is the reaction order and Cl is the
l-th reactant concentration. Generally, two main types of chemical reactions exist in
ISC: cracking/pyrolysis and full/partial oxidation reactions. For oxidation reactions,
usually the partial pressure of oxygen PO2 is used in the rate expressions, instead of
reactant concentrations in Eq. 3.18. The heat source/sink term is also calculated
using Arrhenius kinetics, and is written as,
Qh,reac =Nr∑r=1
∆Hr · ar · exp(−Ear
Rg · T) ·∏l
(Cl)nrl , (3.19)
where ∆Hr is the reaction enthalpy for the r-th reaction, that is the heat generated
from this reaction.
3.5 Primary Equations and Variables
We solve the governing equation numerically using fully implicit discretization [6].
The natural choice of primary equations are the mass balances Eq. 3.1, the energy
balance Eq. 3.2, and the volume balance Eq. 3.3. There are two main categories
CHAPTER 3. GOVERNING EQUATIONS 27
of choosing primary variables: natural variables and overall quantity variables. Nat-
ural variables are based on saturations and mole fractions, whereas overall quantity
variables are based on total component moles and total energy/enthalpy. The main
feature of using natural variables is that we need to switch variables when the phase
state changes in a grid block. Using overall quantities as primary variables, we avoid
complicated variable switching for the phase changes. In this work, we implement
the overall quantity variables as primary variables. They include: Nc overall compo-
nent concentrations Ci, and total energy Ut and pressure P . For the calculation of
phase saturation, mole fractions and other fluid and rock properties, we implement
the isenthalpic/isothermal flash calculation method [103].
Chapter 4
Thermal Streamline Simulation for
Hot Water Flood
4.1 Streamline Simulation Framework
We first give the general framework of our thermal streamline simulator for hot water
flood, which is also illustrated in Fig. 2.1. Further details on the formulation and
individual stages are discussed in subsequent sections.
1. Given boundary conditions (well conditions) and current spatial distributions
of pressure and saturations, the pressure equation Eq. 3.4 is solved (implicitly)
for a global time step on the three-dimensional Eulerian grid. With the pressure
known, the velocity is explicitly computed using Darcy’s law Eq. 3.5.
2. Given the total phase velocity field
ut =np∑j=1
uj, (4.1)
tracing is performed with Pollock’s analytical tracing method [92].
3. The advective parts of the mass balance equations and energy equation Eq. 4.8
are solved along the 1D streamlines using an appropriate numerical method and
28
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD29
for the same global time step used in the pressure solve. During this global time
step, the streamlines locations, the velocity and pressure field are held fixed.
This is slightly different than the approach we implement later on for steam
flood, in which we only fix the streamline locations and update the velocity and
pressure along each 1D streamline. We will discuss this in detail in chapter 5.
4. At the end of the global time step, the newly computed solution variables are
mapped back from the streamlines to the Eulerian grid. The non-advection
effects gravity and heat conduction are accounted for on the Eulerian grid using
an operator splitting approach. This finishes a global time step. The process is
now restarted from step 1.
The streamline simulation uses two distinct time steps: a global time step between
pressure updates, and local time steps used when solving for the advective processes
along the 1D streamlines and the non-advection steps on the Eulerian grid.
4.2 Streamline Formulation for Hot Water Flood
We simplify the hot water flood problem to two phases immiscible water-oil system.
The mass and energy conservation can be simplified as
∂∂t
φρwSw +∇ · ρwuw = q̃w,∂∂t
φρoSo +∇ · ρouo = q̃o,
∂∂t
((1− φ) Ur +
∑j=w,o
φρjSjUj
)+∇ · ∑
j=w,ohjρjuj +∇ · (−α∇T ) = q̃h,
(4.2)
and the pressure equation simplifies to
ct∂P
∂t+∇ · (ρouo)
ρo
+∇ · (ρwuw)
ρw
+∂Vt
∂Ut
∇ · ∑j=w,o
hjρjuj
=q̃w
ρw
+q̃o
ρo
+∂Vt
∂Ut
q̃h. (4.3)
If compressibility effects and thermal expansion of water and oil are sufficiently
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD30
small to be neglected, the pressure equation becomes
∇ · (kλt∇P ) = 0, (4.4)
where λt = krw
µw(T )+ kro
µo(T )is the total fluid mobility, k is the permeability tensor, and
the fluid viscosities µw (T ) and µo (T ) are functions of temperature, according to Eq.
3.13. The mass transport can be further simplified as
φ∂Sj
∂t+∇ · uj = 0. (4.5)
Both the compressible and incompressible formulations are implemented in this
work. We initially start with the simplified incompressible hot water flood and then
include the compressibility effects. The major thermal effects for thermal enhanced
recovery process include reduced oil viscosity, volume expansion, wettability changes
and oil and water interfacial tension changes [93]. The viscosity reduction is gener-
ally the primary mechanism for hot water flood, followed by volume expansion. We
currently do not consider the wettability and interfacial tension effects. We assume
incompressible rock in this study for simplicity.
By rearranging the governing mass and energy equations and splitting the gravity
segregation flux from the pressure driven flux, we get the governing equations along
each 1D streamline, given by
φ ∂∂t
ρwSw + ut∂∂s
fwρw + fwρw (∇ · ut) + ∂Gw
∂z= 0,
φ ∂∂t
ρoSo + ut∂∂s
foρo + foρo (∇ · ut) + ∂Go
∂z= 0,
∂∂t
(Ut) + ut∂∂s
∑j=w,o
fjρjhj +∑
j=w,ofjρjhj (∇ · ut) + ∂Gh
∂z+∇ · (−α∇T ) = 0,
(4.6)
where g is gravity constant, fw and fo are the fractional flow functions, Gw, Go and Gh
are the gravity segregation terms for mass and energy transport. Spatial derivatives
in these equations are taken with respect to the streamline arc length s. Following
the traditional streamline formulation, we can instead use the time-of-flight (TOF)
τ , given by
τ =∫ φ
ut
ds (4.7)
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD31
Operator splitting technique is implemented to solve the advective part of the trans-
port equations and to leave the gravity segregation and conduction terms to be solved
later on. The 1D transport equations along streamlines can then be written as
∂∂t
ρwSw + ∂∂τ
fwρw + fwρw(∇·ut)
φ= 0
∂∂t
ρoSo + ∂∂τ
foρo + foρo(∇·ut)
φ= 0
∂∂t
(Ut/φ) + ∂∂τ
∑j=w,o
fjρjhj +∑
i=w,ofjρjhj
(∇·ut)φ
= 0
(4.8)
In the compressible case, the nonzero divergence of the velocity is taken into account
as source or sink terms in the mass and energy transport equations, given respectively,
by fjρj(∇·ut)
φand
∑j=w,o
fjρjhj(∇·ut)
φ.
We use both a first order explicit and a first order implicit method to solve these
1D transport equations along the TOF grid Eq. 4.8 combined with the standard first
order upstream weighting scheme (SPU).
4.3 Specific Techniques
4.3.1 Mappings
Solution variables are mapped from the streamlines to the Eulerian grid at the end
of each global time step, and back from the Eulerian grid to the new streamlines
after the pressure update at the start of the next global time step. Care must be
taken when designing mappings as they can lead to smoothing of the solutions, as
well as mass balance errors. Mapping errors are not as great a concern for temperture
(or energy) as for saturations (or compositions): the temperature field is generally
smooth, whereas the saturation and composition may contain sharp gradients. In
order to alleviate the numerical diffusion that is often introduced by the mapping
from the pressure grid to the streamlines, we apply a piecewise linear interpolation of
the variables based on TVD (Total Variation Diminishing) slope limiting as proposed
by [72]. The same strategy is applied when mapping the solution variables from the
initial irregular TOF grid to the regularized TOF grid. When mapping back from
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD32
streamlines to the pressure grid, we employ the standard running sum approach [7]:
for any pressure grid cell we compute the solution variables as a weighted average
over the N streamlines crossing the cell, that is,
Ccell =N∑
i=1
αiCi, (4.9)
where Ci is a solution variable. The weights αi are chosen to reflect the relative
volumetric contribution of streamline i to the cell [7]. We choose αi = qiτi/(∑
qiτi),
with qi the volumetric flux of the streamline and τi the time of flight of the streamline
intersect that grid block [7]. In the incompressible case, qi is usually calculated as
a portion of the total volumetric flux from the injection well. It is also possible
to calculate qi from the production well. To calculate qi from the injector side or
the producer side is still open research topic. For compressible flow, the flux qi is
calculated according to q = qoe(∇·ut)τ
φ , which accounts for the compressibility of the
total phase velocity field [88].
4.3.2 Operator Splitting for Non-advective Processes
At the end of a global time step, we solve the conduction term of the energy equation
on the Eulerian grid, while holding the other solution variables, including pressure
and saturations/compositions, constant in each grid block. Gravity effects are also
accounted for at this time using operator splitting [14],
φ ∂∂t
ρwSw + ∂Gw
∂z= 0,
φ ∂∂t
ρoSo + ∂Go
∂z= 0,
∂∂t
Ut + ∂Gh
∂z+∇ · (−α∇T ) = 0.
(4.10)
After these corrections, the fluid volume will generally not fit the cell pore volume.
We estimate the magnitude of these volume discrepancies and include them as source
or sink terms in the 3D pressure equation in the subsequent global time step. The
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD33
discretized form of the pressure equation is now
ctP n+1 − P n
∆t=
∆V
∆t+∇ ·
(Tn∇ · P n+1
), (4.11)
where ∆V is the amount of volume mismatch between fluid volume and pore volume.
The operator splitting applied to the energy balance equations introduces a split-
ting error. This splitting error for the energy equation is acceptable provided the
processes considered are advection dominated. Sensitivity tests are conducted in the
next section to verify the applicability of operator splitting to hot water flooding.
4.3.3 Treatment of Compressibility
The streamline approach followed here can be interpreted as a streamtube approach,
in which each streamline has an associated volume. Streamlines are equivalent to
streamtubes if one considers the relationship between a streamtube and the central
streamline within a streamtube. As mass and energy are moved along the streamlines,
the fluid shrinks or expands because of temperature changes increasing or decreasing
the volumetric flux. We model the volume discrepancy using a dimensionless velocity
approach similar to the one suggested by [71] [35]. We set
∂∂t
(Ci) + ∂∂τ
(udFi) + 1φ
(∇ · ut) · Fi = 0,
ud,k+ 12
= ud,k− 12
+ ε∆τ∆t
(Vfluid/Vcell − 1) ,(4.12)
where Vfluid represents the fluid volume and Vcell is the block pore volume at time tn.
For hot water flooding, this dimensionless velocity approach is effective because
the volume discrepancy is relatively small. And the fluid properties are only weak
functions of absolute pressure.
4.4 Hot Water Flood Results
We test our thermal streamline simulator for hot water injection in a 2D quarter five-
spot configuration. The heterogeneous domain is 500× 500 m2. The permeability is
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD34
Figure 4.1: The permeability field in quarter five-spot hot water flood test case.
Figure 4.2: The water saturation results in quarter five-spot hot water flood test case.
shown in Fig. 4.1. We use the quadratic relative permeability curves krw = Sw2 and
kro = So2 for water and oil phases, respectively. The rock and reservoir parameters
are listed in Table A.1 in the appendix. We compare simulation results of the thermal
streamline simulator with a commercial thermal simulator.
4.4.1 Incompressible Hot Water Flood
Since the compressibility of water and oil liquid is usually small, we first assume the
liquids are incompressible. We use the viscosity relationship, given in Equation Eq.
3.13. We start the simulation with an initial oil saturation equal to 0.9. The initial
temperature of the oil in the reservoir is set to be 20oC, and the temperature of
injected hot water at 80oC. As we heat up the fluid, the oil viscosity will be reduced
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD35
Figure 4.3: The temperature results in quarter five-spot hot water flood test case.
Figure 4.4: The pressure results in quarter five-spot hot water flood test case.
dramatically, therefore making the mobility ratio M = µw/µo decrease from M = 10
at 20oC to M = 1 at 80oC. The other fluid properties for water and oil and well
settings are listed in Table A.2 and Table A.3.
The base pressure grid in the streamline simulator is chosen as 50 × 50. For
comparison, we show FV simulator results for grids varying from 50×50 to 400×400.
Comparisons of the water saturation, temperature, pressure field and cumulative
production are shown in Fig. 4.2, Fig. 4.3 and Fig. 4.4. Here, we used 90 streamlimes
and 20 global time steps for a total simulation time of 50000 days, equivalent to
0.29 pore volume injected (PVI). Because of the large heat capacity of the rock, the
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD36
Figure 4.5: The comparison between 1D fully implicit (FIM) transport solver and1D single point upwind explicit (SPU) transport solver for thermal SL simulation(50X50).
temperature front is moving slower than the leading edge of the water. We observe
water breakthrough first and the temperature front lags behind.
The results show that our streamline method is capable of generating results
of comparable accuracy with FV simulator. Furthermore, the streamline method
is especially capable of reducing numerical diffusion and accurately resolving the
channeling through the domain as we can see from the comparison in saturation of
FV simulation and streamline simulation. Also, since the fluid transport is calculated
along the optimal streamline direction, it can potentially help to alleviate the grid
orientation effects in FV simulation.
We note that we ran the simulations for both FIM and the explicit time step-
ping method along streamlines. The results are shown in 4.5. FIM leads to slightly
increased diffusion, but otherwise results are comparable.
The sensitivity of the solution to the number of global time steps is demonstrated
in Table 4.1. The table shows that the accuracy of the method depends on the
number of global time steps chosen. Too many pressure updates lead to increased
mapping errors. On the other hand, too few pressure updates degrade the accuracy
of the transport calculations since we are not capturing the temporal velocity field
sufficiently well.
We also tested the sensitivity of the streamline simulator accuracy to the viscosity
relationships µj = Aj exp(Bj/T ). The water viscosity is set as constant with respect
to temperature. Of course, as the viscosity increases, the reservoir needs a larger
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD37
Table 4.1: Errors of different viscosity properties compared with reference result(400× 400 STARS result) using L2 norm
M=10 (20oC) to M=1 (80oC) PVI= 0.32
Number of global time steps 2 5 10 20 40Streamline saturation error 18.1% 12.8% 10.1% 9.1% 11.4%Streamline temperature error 12.1% 8.0% 6.5% 7.1% 7.2%FV saturation error (50× 50) with reference 10.0%FV temperature error (50× 50) with reference 5.9%
M=100 (20oC) to M=1 (80oC) PVI= 0.18
Number of global time steps 2 5 10 20 40Streamline saturation error 25.0% 22.6% 19.6% 17.7% 18.4%Streamline temperature error 22.1% 16.0% 12.5% 10.7% 10.5%FV saturation error (50× 50) with reference 17.6%FV temperature error (50× 50) with reference 11.0%
M=1000 (20oC) to M=10 (80oC) PVI= 0.10
Number of global time steps 2 5 10 20 40Streamline saturation error 45.4% 35.6% 29.7% 26.2% 30.5%Streamline temperature error 35.7% 28.2% 22.5% 19.2% 17.5%FV saturation error (50× 50) with reference 18.8%FV temperature error (50× 50) with reference 13.6%
driving force and hence a higher pressure difference between injection and production
wells. We note that as the the exponential viscosity relationship becomes more non-
linear, FV simulator also requires smaller time steps. Both 50×50 thermal streamline
results and 50× 50 FV simulator results are compared with the reference result. The
relative error is calculated according to e = ‖δX‖‖X‖ . The surface cumulative production
is also plotted.
As we can see, the increasing exponents in the temperature dependent viscosity
relationship µj = Aj exp(Bj/T ) leads to larger errors in the streamline simulation
as the mobility ratio increases with temperature. Keeping the total velocity fixed
during a global time step in such highly viscous oil cases leads to more errors. Also
since we are using the initial velocity field un for the transport between time level
tn and tn+1, we are underestimating the water production in this hot water injection
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD38
Figure 4.6: The surface cumulative production results for M=10 quarter five-spot hotwater flood test case.
Figure 4.7: The surface cumulative production results for M=100 quarter five-spothot water flood test case.
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD39
Figure 4.8: The surface cumulative production results for M=1000 quarter five-spothot water flood test case.
case. We can improve the accuracy by increasing the number of pressure updates.
Another approach would be to account for total velocity changes by solving the 1D
pressure equation along the 1D streamline. We will show further improvements by
using pressure/flux update along 1D streamlines in later chapter.
4.4.2 Compressible Hot Water Flood
For hot water flooding problems with low compressibility, we use the dimensionless
velocity approach discussed previously. The initial oil saturation in this case is set at
1.0. All other variables are the same as in the previous test case. The temperature
dependent viscosity relationship is the same as the previous test case. The fluid
density calculations parameters are listed in Table A.4 in the appendix. The reference
pressure is Psc = 100kPa and the reference temperature is Tsc = 273.15K.
The results are shown below in Fig. 4.9, Fig. 4.10 and Fig. 4.11. The compressible
streamline simulation is able to account for the effect of volume changes and yields
results with similar accuracy as compared with the 400× 400 FV simulation results.
Again, we observe water breakthrough to the producer first and the temperature front
lags behind.
We have also explored the sensitivities of larger thermal compressibility for hot
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD40
Figure 4.9: The saturation results in compressible quarter five-spot hot water floodtest case.
Figure 4.10: The temperature results in compressible quarter five-spot hot water floodtest case.
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD41
Figure 4.11: The surface cumulative production comparison results in compressiblequarter five-spot hot water flood test case.
Table 4.2: Effects of thermal compressibility on the production (surface condition)
Small compressibility: cw = co = 1× 10−7/kPa, aw = ao = 1× 10−4/K
Oil Prod (104m3) Water Prod (104m3) Water Inj (104m3)SL (50× 50) 3.218 0.183 3.373STARS (50× 50) 3.121 0.193 3.299STARS (200× 200) 3.312 0.359 3.669
Large compressibility: cw = co = 5× 10−6/kPa, aw = ao = 5× 10−3/K
Oil Prod (104m3) Water Prod (104m3) Water Inj (104m3)SL (50× 50) 3.126 0.143 2.872STARS (50× 50) 3.068 0.166 2.852STARS (200× 200) 3.266 0.304 3.143
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD42
Figure 4.12: Saturation results in gravity test case 1 (quarter five-spot hot water floodwith dipping).
Figure 4.13: Saturation results in gravity test case 2 (quarter five-spot hot water floodwith dipping).
water flooding. Based on the previous case, we have increased the water and oil
compressibility to cw = co = 5× 10−6/kPa, aw = ao = 5× 10−3/K, which is a highly
compressible system. The production and injection results of SL and FV are shown in
Table 4.2. Once again, the thermal streamline simulation yields a comparable solution
to FV simulation. Further more, we can observe the thermal compressibility effects
in the oil production from our results. Compared with the small compressibility case,
the highly compressible system needs less amount of water injected to produce the
equivalent amount of oil at the producer, due to thermal expansion effects.
4.4.3 Gravity Effects
Gravitational effects are included through operator splitting [14]. We test the ap-
proach on a tilted (30 degree) reservoir for incompressible fluids. The results are
shown below in Fig. 4.12 and Fig. 4.13.
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD43
The result has shown by changing the depth of the reservoir, we have accelerated
and decelerated the flood. By using the operator splitting technique, we are able
to capture the gravity. Here the water and oil density difference is relatively small.
Therefore the gravity segregation effect between phases is weak.
4.4.4 Heat Conduction Effects
The streamline method is best suited for simulation of an advection dominated flow
process. Non-advective effects (gravity segregation, heat conduction and capillary)
are accounted for using operator splitting. As the heat Peclet number (Pe) decreases,
we expect the performance of the streamline simulator to worsen. The heat Peclet
number is the ratio between heat advection and conduction. We set Pe = L×u( α
Cv), where
L is the distance between the injector and producer, α the average heat conductivity,
Cv the average total heat capacity, and u the average velocity along the diagonal
direction. We have tested the original incompressible test problem for different Pe
numbers (Pe = 1.5, Pe = 15, Pe = 150) by arbitrarily changing the heat conductivity
of the fluid and rock. Again, both the 50× 50 streamline and 50× 50 FV simulation
results are compared with the refined 400 × 400 FV simulation results. The results
are listed in Table 4.3.
For cases with Pe = 15 and Pe = 150, the SL errors are comparable with the
FV simulation for both saturation and temperature fields because the flow is more
advection dominated. For the smaller Pe number (Pe = 1.5), the SL errors in the
temperature is larger than the FV temperature error (8.4% for SL compared to 5.3%
for FV). This is because of the splitting when solving the heat conduction and heat
convection on different fractional time steps. The saturation results show a weaker
relationship with the Pe number. This is because mass transport is purely advective
flow, thus, it is less sensitive to the operator splitting.
CHAPTER 4. THERMAL STREAMLINE SIMULATION FOR HOT WATER FLOOD44
Table 4.3: Heat Pe number and its influence on the simulation result.
Pe = 1.5Streamline saturation error 9.4%FV saturation error (50× 50) with reference 8.4%Streamline temperature error 8.4%FV temperature error (50× 50) with reference 5.3%
Pe = 15Streamline saturation error 9.4%FV saturation error (50× 50) with reference 8.7%Streamline temperature error 9.1%FV temperature error (50× 50) with reference 7.8%
Pe = 150Streamline saturation error 9.3%FV saturation error (50× 50) with reference 10.8%Streamline temperature error 11.9%FV temperature error (50× 50) with reference 11.1%
4.5 Concluding Remarks
In this chapter, we have presented a thermal streamline simulation framework for two
phase hot water flood with the thermal effects of temperature dependent viscosity
and thermal compressibility. Our method is based on an operator splitting approach
where the advective parts of the governing equations are solved separately along the
streamlines with the other effects solved on the original Eulerian grid. The streamline
results are compared with a commercial FV simulator. We have shown the method
is capable of producing comparable results at a lower computational cost. Various
sensitivity tests were performed to study the accuracy and robustness of the method.
Our results provide guidelines for implementing the thermal streamline simulation for
hot water flood as a fast proxie with reasonable accuracy.
Chapter 5
Thermal Streamline for Steam
Flood
5.1 Streamline Simulation Framework
In this chapter, we focus on the extension of streamline simulation to steam flooding
of heavy oils. Steam floods pose severe challenges to traditional streamline methods
because of the large volume changes and compressibility associated with the phase
behavior of steam, the coupled mass and energy transport, plus the gravity segre-
gation and override. To account for the volume changes and compressibility, which
are particularly strong when steam condenses to water, we implement a pressure and
flux update along the 1D streamlines. To more accurately capture the strong cou-
pling between energy and mass transport, we employ a Glowinski θ-scheme operator
splitting [109]. We also utilize a preliminary hybrid FV/SL approach to deal with
short simulation periods when strong non-advective forces dominates, for example the
soaking period in cyclic steam stimulation. All these new improvements to traditional
streamline simulation are presented and discussed in this chapter.
We again first give the general framework of our thermal streamline simulation
for steam floods. The key steps are as follows:
1. Given boundary conditions (well constraints) and initial saturation/pressure
45
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 46
distributions, the pressure equation accounting for fluid and rock compressibility
Eq. 5.3 is solved on the 3D static Eulerian grid. With the pressure known, the
total phase velocity is explicitly computed at each grid interface using Darcy’s
law Eq. 3.5.
2. With the total phase velocity ut = ug +uo +uw known, streamlines are traced
using Pollock’s method [92]. Solution variables are mapped from the static
Eulerian grid onto the streamlines.
3. The advective parts of the mass and energy equations Eq. 5.2 are solved along
the streamlines along the arc length coordinates. The solution is moved forward
for a global time step by taking streamline specific time steps in a FIM 1D solver.
The key is that pressure and flux along the 1D streamlines are updated at each
local time step to account for steam compressibility and condensation (volume
changes). We are in essence solving a series of 1D, full-physics steam flooding
problems during a global time step. This is different than the previous approach
for hot water flood and is discussed in more detail in later section.
4. At the end of the global time step, the newly computed solution variables are
mapped back from the streamlines to the Eulerian grid. The gravity and heat
conduction effects are accounted for on the Eulerian grid using a Glowinski
θ-scheme operator splitting approach, detailed below.
5. We now return to step 1 to start the next global time step.
5.2 Streamline Formulation for Steam Flood
Steam flooding is a general multi-component compositional thermal simulation prob-
lem. In this work, we implement 1 aqueous component (water) and no oleic compo-
nents. Water exists in aqueous and gas phases, while the no oleic components exist in
oil and gas phases. The mass and energy transport equations are listed in Eq. 3.1 and
Eq. 3.2. The volume conservation equation is Eq. 3.3. The fluid phase equilibrium
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 47
for this multi-component system is expressed as,
xw = 1,no∑i=1
xi = 1,
Kw · xw +no∑i=1
Ki · xi = 1,
(5.1)
where xw is the water mole fractions in aqueous phase, xi is the oleic component mole
fractions in oil phase. K-values Kw and Ki are calculated by Eq. 3.8. We implement
flash calculations using a modified Rachford-Rice equation to account for the water
component in three phase equilibrium [67]. Other properties such as density, enthalpy
and viscosity are calculated according to the previous expressions Eq. 3.9, Eq. 3.10,
Eq. 3.11, Eq. 3.13 and Eq. 3.15.
In this work, we also utilize a simplified steam flood model. We assume that we
only have one water (component 1) and one nonvolatile hydrocarbon (component 2).
This is a valid simplification when the oil is relatively non-volatile. The water exists
in both the water and gas phase (steam), while the hydrocarbon exists only in the
oil phase. A K-value approach is used to model the water/steam phase behavior, i.e.
yw = Kw · xw. We use the standard K-value for water in the commercial simulator
[101].
The mass and energy conservation can now be simplified to
∂∂t
φ (ρwSw + ρgSg) +∇ · (ρwuw + ρgug) = q̃1,∂∂t
φρoSo +∇ · ρouo = q̃2,
∂∂t
((1− φ) Ur +
∑j=w,o,g
φρjSjUj
)+∇ · ∑
j=w,o,ghjρjuj +∇ · (−α∇T ) = q̃h.
(5.2)
The pressure equation becomes
ct∂P∂t
+
(∂Vt
∂N1
(∇ · ∑
j=w,gρjuj
)+ ∂Vt
∂N2(∇ · ρouo)
)+ ∂Vt
∂Ut
(∇ · ∑
j=w,o,ghjρjuj
)= ∂Vt
∂Utq̃h +
∑i=1,2
∂Vt
∂Niq̃i,
(5.3)
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 48
where N1 represents the water composition, and N2 represents the non-volatile hy-
drocarbon composition.
We rearrange the governing mass and energy equations and split the gravity
segregation flux from the pressure driven flux. We define total phase velocity as
ut = uw + uo + ug, where the phase velocities uj is defined as previously in Eq. 3.5.
We also define the aligned components of the phase velocities as u′j = fjut, where fj
is the fractional flow function of the j-th phase. Consequently, the pressure driven ad-
vective fluxes (Fi for the i-th component and Fh for energy) and gravity segregation
fluxes (Gi for the i-th component and Gh for energy) can be written as,
Fi =∑
j=o,w,gρju
′jyij,
Gi =∑
j=o,w,gρj
(uj − u′
j
)yij,
Fh =∑
j=o,w,ghjρju
′j,
Gh =∑
j=o,w,ghjρj
(uj − u′
j
).
(5.4)
With the advective fluxes written along the 1D streamline: in the arc length
coordinate s, the governing equations could be rewritten as,
∂∂t
Ci + ∂∂s
Fi + ∂∂z
Gi = 0,∂∂t
Ut + ∂∂s
F h + ∂∂z
Gh +∇ · (−α∇T ) = 0,(5.5)
where Ci is the i-th component accumulation term, Fi and Gi are the advective and
gravity segregation fluxes of i-th component, Ut is the total energy accumulation term,
F h and Gh are the advective and gravity segregation enthalpy fluxes, and ∇·(−α∇T )
is the heat conduction term. We solve the advective fluxes Fi and F h along the 1D
streamlines in SL step 3. The non-advective fluxes Gi, Gh and heat conduction are
solved later on the original Eulerian grid in SL step 4.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 49
Figure 5.1: Steam condensation causes volume change and flux reduction right at thesteam front.
5.3 Specific Techniques
5.3.1 1D Pressure and Volumetric Flux Update Approach
for Large Compressibility
In previous work by [21], compressibility effects are accounted for by adding source/sink
terms along the streamlines. The volumetric flux along each 1D streamline is held
fixed during a global time step but it is no longer uniform as it would be for a strictly
incompressible case. A two/three phase fractional flow problem is then solved along
each streamline by accounting for these (fixed) source/sink terms. These terms are
updated at each new 3D pressure solution. In the work by [12], the transport along
1D streamlines is solved with the spatially varying volumetric flux accounted for ex-
plicitly along each streamline. The volumetric flux is also assumed to remain constant
over a global time step. However, for steam flooding the volumetric flux along the
streamline will reduce dramatically at the steam front due to condensation as shown
in Fig. 5.1. In other words, the volumetric flux can and will change dramatically dur-
ing the 1D transport step. Assuming a constant volumetric flux (constant source/sink
terms) over a single global time step as the work by [21] and [12] will necessarily mis-
calculate the speed of the steam front and/or require excessively small global time
steps. To retain the ability to take large global time steps, it is necessary to account
for the dependency of the volumetric flux during the 1D transport solve. To do so, we
have implemented a new approach to update pressure and temperature at each mini
time step along a streamline and consequently update the volumetric flux during the
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 50
Figure 5.2: Streamline segments constructed from the injector to the producer.
Figure 5.3: 1D streamline pressure and volumetric flux update approach. Transmis-sibility and block volume of each segment is calculated for the 1D transport.
transport calculations along each streamline.
In our new approach, we fix the streamline locations during a global time step,
but update the 1D pressure and volumetric flux to account for volume changes. We
must take into account the shrinking and expanding of the fluid as it moves along
the streamlines as illustrated in Fig. 5.2. To do so, we define the cross sectional area
A and block volume V =∫
A · ds along the streamlines, see Fig. 5.3. We rewrite the
divergence operation for total phase velocity ut as
∇ · ut =∂ut
∂s+
ut
A
∂A
∂s, (5.6)
where s is the local streamline arc length coordinate, illustrated in Fig. 5.4.
Thend(Aut)
Aut
=∫ ∇ · ut
ut
ds =∫∇ · utdτ ′, (5.7)
where τ ′ is the travel time, defined as τ ′ =∫ 1
utds. Note τ ′ does not include the
porosity φ , which is different than the definition of time of flight (TOF) in traditional
streamline simulation [8]. This treatment is intended to include rock compressibility.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 51
Figure 5.4: Streamline (s, n, m) coordinates and cross sectional area along the stream-line.
After integration, we get
Qi+ 12
Qi− 12
=Ai+ 1
2uti+ 1
2
Ai− 12uti− 1
2
= exp
∫ si+1
2
si− 1
2
∇ · utdτ ′
. (5.8)
The cross sectional areas can now be integrated one by one from i− 12
to i+ 12
according
to Eq. 5.8 as ut is known along the streamlines.
The block volume Vi are now calculated by
Vi =∫ s
i+12
si− 1
2
A · ds =∫ s
i+12
si− 1
2
A · ut · dτ ′ =∫ s
i+12
si− 1
2
d(Aut)
(∇ · ut). (5.9)
The transmissibilities between cells i and i + 1 are calculated by
Ti+ 12
=
(∫ si+1
s
1
(k · A)ds
)−1
, (5.10)
where k are the permeability values derived from the Eulerian background grid. If
anisotropic permeability fields are present, a coordinate transformation is needed to
calculate the permeability values along the streamline directions.
With volumes and transmissibilities known along each 1D streamline, we can now
solve the Nc mass transport equations, coupled with one energy transport and the
volume conservation equation along the streamline. We do this fully implicitly. The
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 52
equations are given by
Viφi
(Cn+1
i,k − Cni,k
)=(Fi− 1
2,k + Fi+ 1
2,k
)·∆tlocal,
Vi
(Un+1
ti − Unti
)=(F h
i− 12
+ F hi+ 1
2
)·∆tlocal,
Vi = Vwi + Voi + Vgi.
(5.11)
Vi is the block pore volume of segment i along the streamline. Vwi, Voi and Vgi are the
fluid volume of water, oil and gas phases in segment i. ∆tlocal is the local time step. φi
is the porosity. Ci is the mass composition in block i. Uti is the total energy density
of block i. Fi+ 12,k and F h
i+ 12
are the mass advective flux for k-th component and the
energy advective flux, respectively, between segments i and i+1. They are calculated
according to the transmissibility calculated previously Eq. 5.10. We typically merge
very small grid blocks to regularize the grid and improve the conditioning of the 1D
transport problem.
Note that rock compressibility is taken into account through the porosity φi term
in Eq. 5.11. The Nc mass compositions, the total energy and the pressure are chosen
as the primary variables aligned with the given Nc +2 governing equations. An exact
pressure-energy flash is performed to calculate the secondary variables, which are the
saturations Sj, the mole fractions yij and the phase densities ρj [103]. At the end of
1D transport solve, we map the 1D streamline pressure back onto the Eulerian grid.
This stitched pressure profile is only used as an initial guess of the pressure at the
beginning of the next global time step.
In the previous chapter, we have shown the larger errors in streamline simulation
when simulating a high initial mobility ratio water flooding problem. This is mainly
due to the fact that in traditional streamline simulation, we need to fix the velocity
field during a global time step. In some sense, it is similar as forward Euler type time
stepping that only utilizes the initial velocity/pressure field. Reducing the global
time step sizes may help to improve this problem to some extent. The accumulative
mapping errors due to large number of mappings, however, eventually makes reducing
global time step size impractical. Updating the 1D flux/pressure along streamlines,
on the other hand, helps to mitigate the errors and capture the velocity and pressure
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 53
transient behavior inside a global time step.
To demonstrate this, we show results for a quarter five-spot hot water flood with
an initial mobility ratio of M = 1000, as illustrated in Fig. 5.5. We have two bottom
hole pressure controlled wells. In a high mobility ratio water flood problem, the
averaged field fluid mobility increases, as more and more mobile hot water is injected.
If we implement two pressure controlled wells, this will lead to a larger total flow
rate between injector and producer over the time. Classical streamline simulation
underestimates this water breakthrough because it uses the initial velocity/pressure
field. We update the pressure/flux along 1D streamlines during each global time step,
and also test the predictor-corrector method described in [71]. The predictor-corrector
method [71] is essentially utilizing the velocity field at the middle, instead of the
beginning of a global time step. Eventually, as we can see in Fig. 5.5, by comparing
to the 200 × 200 reference, using both the 1D pressure update and predict-correct
help to improve the accuracy, while classical streamline method underestimates the
break through.
The updated transient pressure field along 1D streamlines reflects the mobility
changes in heavy/viscous oil flooding. And this gives a better understanding of the
actual pressure changes. We can also observe this from the pressure profiles inside a
global time step as in Fig. 5.6. Here the first plot (A) is the pressure at the beginning
of the global time step. At the end of 1D transport solve, we map the 1D streamline
pressure back onto the Eulerian grid to get the second pressure field (B). At the start
of the next global time step, we solve the 3D pressure equation and get the third
pressure field (C). As we can see, the pressure field stitched together from all the 1D
pressure profiles (B) is similar to the pressure field we get at the beginning of the next
global time step (C). Indeed, by updating the 1D pressure we are able to capture the
pressure transients within a global time step.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 54
Figure 5.5: 1D pressure update improvements in capturing transient pressure changewithin a global time step, (M=1000 quarter five spot hot water flood test case). Withtwo BHP controlled wells, classical SL underestimates the breakthrough. Both SLpredict-correct [71] and SL with pressure update improves the hot water breakthroughresults, compared to fine-scale reference.
Figure 5.6: Pressure transient behavior inside a global time step (M=1000 quarterfive spot hot water flood test case). The pressure stitched together from all the 1Dpressure solves (B) is similar to the pressure we get at the beginning of the next globaltime step (C).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 55
Figure 5.7: Schematic of coupled energy-mass transport. Changing the total energy ofa grid block changes the fluid volume inside, thus causing fluid transport to adjacentcells.
5.3.2 Glowinski θ-Scheme Operator Splitting Approach for
Non-advective Processes
At the end of a global time step, we solve for the heat conduction part of the energy
equation on the static Eulerian grid. This part is represented by the term ∇·(−α∇T )
in equation Eq. 3.2. In steam flood problems, the fluid flow and energy are highly
coupled. When the energy of the fluid inside a cell changes, the fluid volume can
change because of phase behavior (condensation and vaporization). This is especially
the case in step 4 in section 4.1 at the steam flood condensing front when we diffuse
and lose heat of the fluid system, shown in Fig. 5.7. Therefore, if we are not allowing
the fluid in each individual cell to flow to account for the volume changes, we will
make significant volume balance errors, which leads to a volume imbalanced fluid
system at the beginning of the next global time step. This causes great numerical
difficulty.
To reduce volume balance errors in this operator splitting step, we employ the
Glowinski θ-scheme [109] shown in Fig. 5.8. An equivalent form of the Glowinski
θ-scheme is also shown in Fig. 5.9. In the conventional operator splitting scheme,
all of the advective flux is accounted for in the 1D streamline advective solves (SL
step 3) and none in the Eulerian grid update (SL step 4). In Glowinski’s θ−scheme,
a small fraction (θ) of the advective flux is kept in the heat conduction step. This is
desirable when large volume changes occur because of phase behavior: the fluids can
redistribute themselves and volume balance errors are reduced before the next global
time step is started.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 56
Figure 5.8: Glowinski θ-scheme operator splitting. Most of the transport is solvedalong 1D streamlines [0 ∼ (1 − θ)∆t]. Small amount of advective flux is used tocorrect the volume changes in heat conduction/gravity step.
Gravity segregation effects are also accounted using this Glowinski θ-scheme oper-
ator splitting method Fig. 5.8. In the traditional streamline method, gravity segrega-
tion effects, resulting from density differences between phases, are taking into account
by solving for flow along gravity lines [7]. Again, gravity segregation, in which rising
steam can condense when it contacts colder areas of the formation, can lead to large
volume changes and fluid redistribution.
The choice of θ is process dependent and must be deduced heuristically. It is
dependent on the heat Peclet number Pe (the dimensionless number which represents
the heat advection effects against the heat conduction effects) and the gravity number
Ng (the dimensionless number which represents the gravity segregation forces against
the viscous forces). It will also be impacted by the details of the static geological
model. We typically use a large θ for processes with small Pe and large Ng, that is,
for problems with strong heat conduction and gravity segregation. We use small θ for
large Pe and small Ng when heat conduction and gravity segregation are relatively
weak. This is the case for most realistic steam flood problems when the steam gravity
override has been fully developed.
5.3.3 Preliminary Hybrid Simulation
Streamline simulation is based on an operator-splitting technique, in which we solve
the advective flow along the streamlines, and non-advectve flow on the Eulerian grid
later at the end of a global time step. Therefore, streamline method is well suited for
advection dominated flow problems, because the splitting error is relatively small. For
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 57
Figure 5.9: An equivalent representation of Glowinski θ-scheme operator splitting.Most of the transport is solved along 1D streamlines. Small amount of advective fluxis used to correct the volume changes in heat conduction/gravity step.
some particular thermal recovery processes, we do have periods dominated by non-
advective forces. For example, the initial steam circulation periods in SAGD process
are heat conduction dominated, with the purpose of creating thermal communication
between the two horizontal wells [19]. The soaking period in cyclic steam stimulation
process is also heat conduction dominant. The injected steam releases its heat to
the formation, thus making it easy for the oil to flow back into the producer in the
later production stage. In these scenarios, streamline simulation will have difficulty
handling the strongly non-advective flows.
Another type of problems challenging for streamline are cases with strong gravity
segregation effects. Streamline simulation combines the water, oil and gas phase
velocities together to solve the transport along total phase velocity streamlines [14, 7].
Gravity segregation between the three phases typically are accounted for at the end
of global time step using operator splitting. The splitting error may lead to larger
errors for problems that have strong segregation between different phases. Strong
gravity segregation is very often seen at the beginning of gas/steam flooding process,
but quickly diminishes as the injected low density gas/steam rises to the top of the
formation [57]. Therefore, it is these initial periods that often pose problems for the
accuracy of streamline simulation.
In recent work by Kumar [62], a SL/FV hybrid simulation approach has been
proposed. The hybrid approach consisted of using FV simulation when the cross flow
was strong and shifting to the streamline method when the cross flow was negligible.
We use a similar concept here for short simulation periods with strong non-advection
effects. A heuristic criterion was proposed for the crossover from initial FV to later
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 58
Figure 5.10: Quarter five spot steam flood permeability field and initial streamlineshape.
SL, based on the cross product of gas and oil phase velocities by Kumar [62]. As
we will see later, we also use FV simulation for the relatively short soaking periods
in cyclic steam stimulation and the initial steam rising periods in steam flooding
simulation. In this way, we can combine the relative advantages of SL and FV during
the course of the whole simulation to achieve both accuracy and efficiency.
5.4 Steam Flood Results
5.4.1 Heterogeneous Quarter-Five Spot Steam Flood
We first test our thermal streamline simulator for steam injection in a 2D quarter five-
spot heterogeneous reservoir. The reservoir properties are given in appendix Table
A.5 and the permeability field with associated streamlines in Fig. 5.10. The reservoir
is 150m by 150m, discretized by 50× 50 grid blocks. We have a bottom hole pressure
controlled injector and a bottom hole pressure controlled producer as described in
appendix Table A.6.
We implement the simplified two component steam flood model given in section
4.2, with the oil and water phase viscosity profiles shown in Fig. 5.11. At an initial
reservoir temperature of 50oC, the oil viscosity is µo = 105cp and the water viscosity
is µw = 0.5cp. At a steam chamber temperature of 276oC, µo = 0.18cp and µw =
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 59
Figure 5.11: Viscosity curve for water and oil.
0.075cp. The injected steam quality (the mass of steam divided by the total mass of
hot water and steam) is fs = 0.8. Other fluid properties are given in appendix Table
A.7. For each component the fluid enthalpy and latent heat calculation is performed
in the same way as previously, with Cg(T ) = CPG1 + CPG2 · T + CPG3 · T 2 +
CPG4·T 3. We simulate continuous steam injection for 3000 days, which corresponds,
approximately to a recovery factor (% of OOIP) of 18%. We use 30 global time steps,
with each global time step equal to 100 days. In this simulation, the finite volume
simulation uses 342 total time steps. For the operator splitting, θ is chosen as a
constant of 0.1, which works well for this advection-dominated case. We study the
sensitivity of θ in detail later. We use both our streamline simulator and a commercial
finite volume simulator [101]. A fine-grid (200 × 200) finite-volume solution is used
as reference solution. Results are shown in Fig. 5.12 and Fig. 5.13.
We compare the relative errors in the temperature and oil saturation profiles.
Both 50 × 50 thermal streamline results and 50 × 50 finite volume simulator results
are compared with the reference 200×200 solution for three different simulation times
Table 5.1. The relative error is calculated according to error = ‖δX‖‖X‖ , where x is the
solution variable, and the norm is the standard L2 norm.
The results show that our streamline simulator has comparable accuracy to the
finite volume simulator at the same grid resolution (50 × 50). We can observe the
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 60
Figure 5.12: Quarter five spot temperature and gas saturation at 3000 days (prebreakthrough).
Figure 5.13: Quarter five spot water and oil saturation at 3000 days (pre break-through).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 61
Table 5.1: Errors of thermal streamline and STARS compared to reference (360×180STARS result) using the relative L2 norm.
Simulation Time 1000 days 2000 days 3000 daysStreamline oil saturation error 3.5% 4.5% 4.2%Streamline temperature error 14.3% 11.8% 9.8%FV oil saturation error (50× 50) with reference 5.7% 6.2% 5.1%FV temperature error (50× 50) with reference 16.7% 10.8% 11.2%
shape of the steamed zone, the hot water condensate bank outside of the steamed
zone, and the swept zone in the oil saturation profile.
We perform further sensitivity studies on the choice of θ. We test the same
quarter five spot steam flood problem for 1000 days with 10 global time steps, using
θ = 0.05, 0.1, 0.2. We compare the simulation results in between as shown in Fig.
5.14. We found the simulation results are very close to each other. As we can see the
steam zone sizes and temperature profiles are very similar in Fig. 5.14. The relative
difference in temperature results based on L2 norm and Inf norm is also relatively
small (within∼ 1%), shown in Table 5.2. Note here our main purpose is to use thermal
streamline simulation as a proxy. Steam flood is an advection dominated process, in
which heat conduction serves as a second order term compared to advection flux.
Therefore, we find the splitting error to heat conduction and the choice of θ is not
giving rise to large changes in the streamline result. In practice, we want to keep θ as
low as possible to keep most of the computations in 1D solve along streamlines and
minimize the computational costs. On the other hand, if we take θ to be close to 0,
the scheme becomes close to classical operator splitting. As we discussed previously,
classical operator splitting will cause a volume imbalanced system, because of the
coupled mass and energy transport. Thus, we can not taken extremely small θ so
that the advection flux becomes insufficient to compensate volume changes caused by
conduction. In practice, it often leads to numerical convergence issues when solving
the non-advective flux (step C in Fig. 5.8).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 62
Figure 5.14: Sensitivity study on the choice of θ. The simulation results are close toeach other by using different θ.
Table 5.2: Sensitivity study on choice of θ. The simulation results close to each otherby using different θ.
(θ = 0.05)− (θ = 0.1) (θ = 0.2)− (θ = 0.1)Relative Difference (L2 norm) 0.3% 1.2%Relative Difference (Inf norm) 0.5% 1.4%
5.4.2 Cyclic Steam Stimulation
We test the thermal streamline method for single well cyclic steam stimulation prob-
lem. This case is intended to fully demonstrate the capability of handling compress-
ibility effects, since at any time we are only injecting or producing at a single well.
In this test, we have a 2D reservoir discretized by 51 × 51 grid blocks, of size 102m
by 102m. The reservoir properties are given in appendix Table A.8. We implement
homogeneous permeability in this case K = 500mD. We have a single well in the
middle that will serve as both steam injector and oil producer. We inject steam for
60 days, let it soak and release the heat for 100 days, and produce from the same
well for another 60 days. During the injection stage, we have a bottom hole pressure
controlled injector injects steam at 6000kPa. In the production stage, we control the
producer total bottom hole liquid production rate at 54m3/day as in appendix Table
A.6. We take 3 global time steps in the injection stage and 2 global time steps in
the production stage. We implement the same simplified two component steam flood
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 63
Figure 5.15: Hybrid procedures in cyclic steam stimulation (SL → FV → SL) andthe initial streamline shape.
model as in section 4.2.
As discussed previously, in the soaking period, streamline will have difficulty han-
dling the strongly heat conduction dominated flow. Therefore, we implement a SL-FV
hybrid approach to use SL for injection and production stages and FV for the soaking
period Fig. 5.15. When injecting or producing, the streamline will either start from
the injector well and end at the boundary of the reservoir, or start from the boundary
and converge to the central producer Fig. 5.15. This is due to the compressibility
effects. Our streamline simulator is able to fully capture this and stop the streamline
tracing at the certain locations when reaching the local source/sink generated from
the compressible velocity field.
The resulting temperature and oil saturation profiles are shown in Fig. 5.16 and
Fig. 5.17. We also perform the same problem with the commercial FV simulator
[101] (9 point stencil discretization scheme applied here to reduce the grid orientation
effects). The profiles show that our streamline simulator has comparable accuracy to
the FV simulator. We can observe a high temperature zone in the injection stage.
During soaking, the temperature diffuses and steam condenses to warm up the for-
mation around the well. When the producer is opened again, the oil flows out of the
reservoir with reduced viscosity, and the reservoir cools down. Streamline simulation
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 64
Figure 5.16: Cyclic steam stimulation temperature result.
is capable of handling the primary steam injection and primary production scenarios,
which are completely compressibility driven. Without the help of 1D pressure/flux
update, this will be an extremely difficult problem for traditional streamline methods.
5.4.3 Heterogeneous Multi-well Pattern Steam Flood
We also test the thermal streamline simulator on a heterogeneous steam flood problem
with multiple wells. We again use the two component simplified steam flood model.
The reservoir rock properties, the fluid properties, and well controls are the same as
before. We place two pressure controlled steam injectors and six pressure controlled
producers as shown in Fig. 5.19. The 2D reservoir is now 180m by 360m, and is
discretized into 60 × 120 grid blocks. Grid refinement to 180 × 360 is performed in
finite volume simulator to obtain a reference solution. The permeability field is shown
in Fig. 5.19.
We first run the steam flood for 1500 days, which corresponds to a recovery factor
of approximately 7%. In the streamline simulation, the global time step size is chosen
as a constant 100 days. Here, the streamline simulator uses 15 global time steps. For
the operator splitting, θ is again chosen as constant 0.1. The temperature profiles
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 65
Figure 5.17: Cyclic steam stimulation oil saturation result.
Figure 5.18: Cyclic steam stimulation production profile (SL and FV).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 66
Figure 5.19: 2D multiple well test case permeability field.
Figure 5.20: 2D multiple well test case initial streamline shape.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 67
Figure 5.21: Multiple well test case temperature and oil saturation at 1500 days (prebreakthrough).
and oil saturation profiles are shown in Fig. 5.21. The water and oil production
histories are plotted in Fig. 5.22. We continue the simulation to 2100 days, at
which time the steam has broken through at one of the producers. The streamline
simulator has 21 global time steps, while the finite volume simulation uses 207 FIM
time steps. The temperature and oil saturation profiles for this time are shown in
Fig. 5.23. The water and oil production histories are also plotted in Fig. 5.24.
After steam breakthrough, the water production increases quickly, indicating that
steam has connected the injector and producer. Oil is still produced together with
the steam through the steam drag mechanism. Heat conduction plays an important
role in steam drag, as discussed in the analytical model by Edmunds in [38]. In Fig.
5.23, we observe the steam zone around the post breakthrough producer is slightly
smaller than the FV simulation. This may be due to the fact that heat conduction
(non-advective force) becomes a more important effect in post breakthrough steam
drag process than pre breakthrough.
We compare the errors in the temperature and oil saturation for the coarser 120×
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 68
Figure 5.22: Multiple well test case field production history (pre breakthrough).
Figure 5.23: Multiple well test case temperature and oil saturation at 2100 days (postbreakthrough).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 69
Figure 5.24: Multiple well test case field production history (post breakthrough).
Table 5.3: Errors of thermal streamline and STARS compared to reference (200×200STARS result) using L2 norm.
Simulation Time 1500 days 2100 daysStreamline oil saturation error 1.7% 11.5%Streamline temperature error 5.4% 19.4%FV oil saturation error (50× 50) with reference 1.4% 2.1%FV temperature error (50× 50) with reference 3.9% 6.4%
60 grids relative to the 360×120 reference solution in Table 5.3. Again, our streamline
simulator generates results of similar accuracy to the finite volume simulator. The
predicted breakthrough times are also acceptable.
5.4.4 Vertical Cross Section Steam Flood
We test the thermal streamline simulator in a vertical cross section to demonstrate its
ability to handle gravity override. We also use the simplified steam flood model in this
case, with the same fluid and reservoir rock properties as before. The reservoir is 12m
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 70
Figure 5.25: 2D vertical cross section steam flood test case 1 (homogeneous perme-ability field).
thick and 120m long, discretized by 8 × 20 grid cells. Two pressure controlled wells
are perforated at the bottom of the reservoir with pressure of 6000kPa and 4500kPa
respectively, as shown in Fig. 5.25. We use a homogeneous field permeability in this
case, with K = 2000mD. The resulting temperature profiles at 150 days are shown
in Fig. 5.25. We start simulation from the initial condition shown in Fig. 5.25.
Comparison is made to the commercial FV simulator [101]. We use a 9 point stencil
scheme to reduce grid orientation effects. The use of 9 point stencil is a temporary
solution. Rigorous multi-D upwinding [60] is more desired to help controlling grid
orientation effects. As we can see, the thermal streamline simulation is capable of
generating results similar to that of the FV simulator for this test case. We can
observe the shape of the steam chamber and the steam gravity override.
We further test a heterogeneous 2D vertical reservoir. The size of the vertical
reservoir is 15m thick and 120m long, discretized by 10 × 20 grid cells. A reference
solution is achieved with further grid refinement 20 × 40. We implement the same
pressure controlled wells at the bottom of the reservoir. The heterogeneous perme-
ability field is shown in Fig. 5.26. The resulting temperature and oil saturation
profiles at 150 days are also shown in Fig. 5.26. We again compare the thermal SL
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 71
Figure 5.26: 2D vertical cross section steam flood case 2 (heterogeneous permeabilityfield).
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 72
results with the commercial FV simulator [101]. The thermal streamline method gen-
erates comparable results with the FV simulator. We can clearly observe the gravity
override, which is very similar as the previous case.
5.5 Discussion and Applications
5.5.1 Streamline Simulation as Fast Proxy: Cost Comparison
One advantage of streamline simulation over traditional finite volume simulation is
its inherent computational efficiency [105]. To demonstrate this potential for ther-
mal simulation study, we present the profiling results for the multi-well steam flood
example shown previously.
For commercial FV simulator, we have 7,200 grid blocks. The commercial simu-
lator takes 207 time steps, thus 207 fully coupled solves with all the 7200× (Nc + 2)
unknown variables involved (Nc is the number of components).
The thermal streamline simulation uses 7,200 grid blocks on the Eulerian grid. In
this case, we currently use θ = 0.1 for the operator splitting (a smart selection of the
choice of θ to reduce spending on operator splitting is our ongoing work). Thus, for
the operator splitting part we assume 10% comparable cost as the FV simulation,
since we are solving for θ ·∆t time period. Compared to the rest 90% FV simulation
cost, we have 21 global time steps in streamline simulation, which implies 21 pressure
solves (much cheaper than FIM solves with all the unknowns). Inside each global
time step, we also have a set of independent inherently parallel 1D solves along the
streamlines. We included 410 streamlines, each with only 68 grid blocks on average.
Note these 1D solves are very low dimensional 1D problems, which can be solved
and parallelized very effectively [9]. In previous works, we have shown the speed-up
we achieve against FV simulations when simulating advection only problems such
as water flooding [9, 105]. If we assume a conservative 5× speed-up for this 90%
calculations, we will achieve about 3.6× overall speed-up. As we move to simulation
of full field reservoir model with many wells and millions cells, the large FV simulation
solves become intractable because of the increasing number of the unknowns and the
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 73
large condition numbers. By using streamline simulation, we can significantly reduce
the size of the problem by solving the transport along individual 1D streamlines.
Thus we are expecting more improvements for realistic full field reservoir simulation.
5.5.2 Optimization and Flux Patterns
One application of streamline simulation is its capability of outlining the flow patterns
[105]. Streamlines offer an immediate snapshot of the flow field clearly showing where
the flow is coming from and where the flow is going to. This can be see from the
flux pattern map Fig. 5.27 generated from commercial streamline simulator 3DSL
[1]. Same flux pattern can be easily constructed for thermal problems, i.e. hot water
flood and steam flood.
Another application of streamline method is the pattern balancing and flood man-
agement through streamline derived information [106]. Through streamline tracing,
well allocation factor and injection efficiency can be calculated, which provide quan-
titative measure of the connectivity in the reservoir. The well allocation factor is
calculated according to the flux from the injector to each corresponding producer.
The injection efficiency is calculated by the total offset oil production divided by the
injection rate for each injector. The same strategy can be used here for thermal flood-
ing problems. General well management will allocate more steam/hot water injection
rate to injectors with higher injection efficiencies.
If one already has a model based on a thermal FV simulator and just wants
to explore the connectivity and improve flood efficiency via the streamline-derived
information, we can also directly trace streamlines on the velocity field output from
the FV simulator. We will implement the same Pollock’s tracing method on the
velocity field calculated in FV simulation [92]. The same logic of well allocation factor
and injection efficiency based flood management will be implemented accordingly to
thermal problems.
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 74
Figure 5.27: Flux pattern map, by Thiele and Batycky [106].
5.6 Concluding Remarks
We have extended streamline simulation to thermal recovery steam floods. We imple-
mented three techniques that address the large volume changes and strong couplings
inherent to steam floods: a 1D pressure update along the streamline, a Glowinski
θ-scheme operator splitting approach and preliminary hybrid SL-FV simulation. Our
test cases, which include a quarter five spot steam flood, a steam solvent flood, a
cyclic steam stimulation, a vertical cross section steam flood with gravity override,
and a multi-wells example, show that this steam flood streamline simulation is a vi-
able and accurate alternative to traditional finite volume approaches. We also show
for particular numerical issues such as grid orientation effects, SL can perform better
than normal FV simulation based on two point flux since we take into account the
upstream directions for the transport.
As in other streamline simulation application, the reduction of the 3D trans-
port problem to a series of 1D streamline systems is computationally attractive, in
particular because the streamline problems can be parallelized effectively [9]. The re-
sults presented in this paper are particularly encouraging as they show that thermal
streamline simulation may serve as a fast and effective proxy for reservoir simulation
studies, such as well placement, optimization and history matching. For example,
a reliable, fast proxy is particularly useful in the context of work flows associated
with the uncertainty of multiple geological models. Being able to screen earth models
CHAPTER 5. THERMAL STREAMLINE FOR STEAM FLOOD 75
efficiently is a key component of modern reservoir engineering work flows and by ex-
tending streamline simulation to thermal floods the computational cost of such work
flows can be substantially reduced. The unique streamline-derived dynamic connec-
tivity data remains a key element and can be used to improve pattern conformance
and improved steam injection in the same as streamlines are used to improve water
floods [106]. It is important to emphasize that our work is not to be understood as a
substitute to traditional simulation approaches. Instead, we see thermal streamline
simulation as complimentary tool to traditional finite-volume approaches by extend-
ing well-established uncertainty work flows to thermal problems as well as introducing
dynamic, streamline-based connectivity for improved oil recovery.
Chapter 6
In-situ Combustion Simulation
In thermal reactive reservoir simulation, conservation equations are solved numerically
on discretized reservoir grid blocks [27, 61]. Crude oil consists of thousands of complex
chemical species. When the crude oil reacts, we expect tens of thousands of chemical
reactions. When modeling in a numerical reservoir simulator, we need to lump the
components into several pseudo components and group the reactions into a limited
number of pseudo reactions. Accordingly, different kinetic reaction models have been
developed for ISC simulations.
6.1 Kinetic Reaction Models
The parameters for proposed kinetic models are usually derived from kinetic cell
laboratory experiments of ramped temperature oxidation (RTO) [24, 40]. In such
experiments, a small sample mixture of oil, water, sand, and clay is placed in the
kinetic cell that is subjected to controlled linear ramped heating while oxygen is in-
jected at a constant rate from the bottom of the cell. The produced gas composition
is obtained using a gas analyzer. A typical produced gas and ramped temperature
profile for a heavy oil is shown in Fig. 6.1. We observe two peaks in oxygen consump-
tion that correspond to two increases in the temperature profiles deviating from the
linear profile. Starting from the temperature of about 300oC, we initially have the
76
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 77
Figure 6.1: Gas composition from ramped temperature oxidation experiments.
pyrolysis/cracking reaction that creates solid compound coke and some light hydro-
carbon components. Coke, that has very high molecular weight, serves as the main
source of fuel for the later combustion reactions. The first peak in temperature indi-
cates the low temperature oxidation (LTO) reactions, in which the coke is partially
oxidized, generating small amounts of heat. This is followed by a negative tempera-
ture gradient region, where the reaction rate decreases as the temperature increases.
Eventually, the coke fully combusts at higher temperatures, a process that is referred
to as high temperature oxidation (HTO). HTO is represented by the second peak
in the temperature profile. In the HTO region, the amount of oxygen consumption
is comparable to the production of carbon oxides. This implies a full combustion
reaction that generates CO2, CO, water, and a large amount of heat [24, 40].
One of the most commonly used kinetic reaction models is the Crookston model,
also known as the minimal model [30], given by
HO + O2 → CO + CO2 + H2O, (Heavy oil combustion)
LO + O2 → CO + CO2 + H2O, (Light oil combustion)
HO → Coke + LO, (Cracking/Pyrolysis)
Coke + O2 → CO + CO2 + H2O. (Coke combustion)
(6.1)
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 78
It comprises two oil components (heavy oil HO and light oil LO) and four chemical
reactions. Both HO and LO directly combust with oxygen to generate heat. The
component HO also pyrolyzes, generating coke and LO. This LO generation represents
an oil upgrading effect. Eventually, coke burns completely with oxygen, generating
heat. One disadvantage of the Crookston model is that it does not explicitly represent
the two peaks of LTO and HTO reactions as seen in RTO experiment gas analysis.
This is because we only have one step coke combustion, plus LO and HO direct
combustion. With its simple form and the early appearance, however, the Crookston
model is still very widely used in the literature.
We also introduce another three reaction model [33]. We create three reactions to
represent the pyrolysis, the low temperature oxidation and high temperature oxida-
tion, given by,
Oil + O2 → Coke1, (Pyrolysis)
Coke1 + O2 → Coke2 + CO + CO2 + H2O, (LTO)
Coke2 + O2 → CO + CO2 + H2O, (HTO)
(6.2)
where Coke1 and Coke2 are distinct fuel species. The advantage of the three reaction
model is that it captures fully the distinct LTO and HTO peaks. The disadvantage is
that we have only one dead oil component, thus no compositional effects or oil upgrad-
ing through combustion. We could combine the advantages of both the Crookston
and the three reaction model by adding an LTO intermediate step in the Crookston
model, or by adding light end products in the three reaction model’s pyrolysis re-
action. In the original Crookston and three reaction model, however, they are not
included [30, 33].
Other models such as the Belgrave model [11] and SARA model [43] group crude
oil pseudo components based on the SARA fractions (Saturate, Aromatic, Resin and
Asphaltene). These methods also employ a series of reactions, including coking (py-
rolysis), low temperature oxidation (LTO), and high temperature oxidation (HTO).
The main purpose of these models is to group the complex chemical reactions accord-
ing to the SARA fractions. This is because the different SARA fractions go through
different types of reactions when oxidized. Suggested by Freitag et al.[43], the SARA
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 79
model may be extended to different crude oils based on their SARA fractions. The
disadvantage of such models is that eventually when implementing in numerical sim-
ulators, the PVT property of the crude oil is not trivial to derive because the pseudo
component lumping is based on SARA fractions. This is because the SARA fractions
include hydrocarbons with wide range of carbon numbers, from the most light ends
all the way to very heavy hydrocarbons.
In our current ongoing research [25], a new kinetic reaction model is also proposed.
The reactions include: precipitation of coke, LTO reaction, change of coke active sites,
and HTO reaction. In the work by Lapene [68], a new kinetic reaction model with full
PVT characterization of Zuata crude oil is presented. It lumps the crude into eight
pseudo components (C1, C2−11, C12−16, C17−21, C22−27, C28−35, C36−49 and C50+). The
chemical reactions include the typical pyrolysis, LTO and HTO reactions.
We can choose different kinetic reaction models, depending on the complexity
we require and the experimental information we have. For example, for a bitumen
like heavy oil such as Hamaca, the oil composition is very heavy. The use of three
reaction model may be applied. For Athabasca bitumen, if we desire more detailed
intermediate steps in pyrolysis, LTO and HTO reactions, the Belgrave model serves
as a suitable choice.
In all the kinetics models, the common feature is that the crude oil is composed of
several components, and the chemical reactions are grouped into a series of reactions.
These components react, form intermediate products (oxidized compound or coke)
and eventually combust completely with oxygen. Each of these reactions has its own
kinetic parameters, that include stoichiometry coefficients, activation energy, pre-
exponential factor and reaction order. All these reactions follow the same Arrhenius
kinetics law as in Eq. 3.18 and Eq. 3.19.
6.2 1D Combustion Tube Simulation
We perform numerical simulation for a 1D dry combustion tube problem to demon-
strate the typical combustion behavior of an 1D ISC process. The three reaction
model is used in this study. This is because we have much experience with this
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 80
Figure 6.2: Relative permeability (water-oil and liquid-gas, respectively) and oil vis-cosity curve used in the simulation of Hamaca oil 1D combustion tube experiment.
model. The parameters are also well calibrated to the Hamaca oil experimental re-
sults [66]. The simulation domain is 2.16m, discretized into 600 grid blocks, each
with size 0.36cm. The oil sample is Hamaca oil from Venezuela, with 8.5 API gravity.
The corresponding fluid and rock properties are shown in Table A.9 and Table A.10
in Appendix A. The relative permeability and viscosity curve of the oil are shown
in Fig. 6.2. The kinetic parameters for the three reaction model are listed in Table
A.11, Table A.12 in Appendix A. We inject air from the left side and produce from
the right side. Initial ignition heating is introduced close to injector. The injector is
set at a constant air injection rate of 3000cm3/min at standard conditions, while the
producer is set at a pressure control of 780kPa. We perform simulation runs without
lateral heat losses for simplicity. Initially, we have a water saturation of Swi = 0.24
and an oil saturation of Soi = 0.26 in the tube. We simulate the problem for a total
of 1200 minutes. The resulting temperature, pressure and oil saturation profiles at
498 minutes are shown in Fig. 6.3.
We identify several different flow regimes in this 1D ISC problem. The leading edge
of the high temperature front is the combustion zone. Temperature ramps up very
steeply to about 800oC. This indicates a very narrow reaction zone typically several
inches thick. Note here that the peak temperature is greater than in the physical
1D combustion tube experiment. This is because we currently do not include heat
losses in our model. A typical combustion tube experiment temperature profile is
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 81
Figure 6.3: Temperature, oil saturation and pressure profiles in 1D combustion tubesimulation after 498 min of air injection.
Figure 6.4: Typical temperature profile history from a 1D combustion tube experi-ment, from Lapene et al., [66]
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 82
shown Fig. 6.4, by Lapene et al., [66]. Upstream of the reaction zone is the burned
zone. In this numerical experiment we see no residual oil saturation, because the high
temperature combustion reaction consumes all the fuel. The rest of the oil has been
displaced further downstream. In the burned zone, the temperature remains quite
high, because of the heat generated previously by combustion. Downstream of the
reaction zone, we have the steam plateau. The temperature remains at around 120oC.
The oil banks in the steam plateau, and So is constant around 0.8. In front of the
steam plateau is the unaffected zone with an initial oil saturation of 0.26. Along this
combustion tube, the overall pressure drop is relatively small, within 100kPa. In this
case, we combust with a low initial oil saturation of 0.26. The fuel amount for most
ISC processes is around 5 ∼ 10% So. Therefore, the size of the oil bank is relatively
small in this case. Furthermore, the oil is an 8.5 API bitumen with an initial viscosity
of 2 × 105cp at 15oC. According to the viscosity-temperature table, the oil still has
about 100cp viscosity at the steam plateau temperature. Thus, the oil bank remains
inside the steam plateau. If a less viscous oil sample (for example, Lloydminster type
conventional heavy oil) is simulated with oil viscosity reduced to < 10cp at steam
plateau temperature, the oil bank is likely to move in front of the steam plateau.
This illustrates that the flow regime behavior is very case dependent, depending on
the properties of the oil, the operating pressure, the initial oil saturation, and the
relative permeability.
We also show the production profiles in Fig. 6.5 and Fig. 6.6. After the initial
ignition period, the gas production reaches the relatively constant value of about
3000cm3/min. That behavior corresponds to the air injection rate. Most of the gas
produced is N2, with some CO2, and a very small amounts of CO and bypassed O2.
This indicates a good burn with combustion reactions in the HTO mode. The HTO
reactions consume most of the O2 and generate large amounts of CO2 as shown in Fig.
6.1. Water is constantly produced after the initial ignition, because the initial water
saturation is above the connate water saturation Swc in the relative permeability
curve. At about 750 minutes, the oil bank eventually reaches the producer. We
observe a sudden increase in the oil production rate. The oil bank produces until
about 1000 minutes, when the combustion front reaches the producer and O2 breaks
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 83
Figure 6.5: Water, oil and gas production rates in 1D combustion tube simulation.
through. This ends the 1D combustion process.
We further show the measured combustion front and the leading edge of the oil
bank movement in Fig. 6.7. In this case, we use a constant air injection rate. For
both front locations, we observe a close to linear increase with respect to time. This
means the combustion front has reached a pseudo steady state propagation in this 1D
problem. Combined with the previous stable production gas analysis, shown in Fig.
6.6, we conclude that when the combustion front is propagating steadily through the
1D system, the front is consuming the same amount of fuel, ”x”, per unit reservoir
volume. The rest of the oil flows into the oil bank, which makes the oil bank grow
linearly with time. This behavior has also been called a ”bulldozing” effect in the
literature [110]. This constant fuel consumption is also consistent with the basic
assumption in the classical Gates and Ramey analytical approach [44].
There is a fundamental difference between ISC and a steam flood process [96].
In ISC, we generate heat by the exothermic reactions in the combustion zone. This
generates non-condensable flue gas. This flue gas is constantly flushed further down-
stream and produced. The flue gas and steam mixture pushes the oil bank towards
the producer. The steam condenses at the leading edge of steam plateau and releases
heat to warm up the formation. In ISC, we need to keep supplying oxygen to the
combustion front in order to sustain the reactions. A good venting/circulation of the
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 84
Figure 6.6: Produced gas composition analysis in 1D combustion tube simulation.
Figure 6.7: Oil bank and combustion front distance-time diagram from 1D combustiontube simulation.
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 85
Figure 6.8: 1D ISC simulation: peak temperature and combustion front velocity as afunction of the injection air flux.
oxygen/flue gas from injector to producer is crucial to complete the whole process.
In a steam flood process, however, steam carries heat directly into the cold reser-
voir. Once the steam reaches the leading edge of the steam zone, it releases its latent
heat to the formation and condenses to hot water. We do not need to keep the oxy-
gen/flue gas circulation as in the ISC process. In a reservoir with undersaturated oil
(above the bubble point), we will not have free gas production until the live steam
breaks through to the producer.
6.3 Sensitivity Studies on 1D ISC Problem
In this section, we perform sensitivity studies on the previous 1D combustion tube
simulation. We test the effects of boundary conditions (air flux rate) and initial
conditions (Soi and Swi). Again, we use the same three reaction kinetic model [33]. We
have extensive experience using this model. It is also well calibrated to the Hamaca
oil experimental data here in Stanford University [66]. The previous simulation input
in section 6.1 is used as the base case for this sensitivity study. We increase the
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 86
Figure 6.9: 1D ISC simulation: oil saturation profile as a function of the initial oilsaturation Soi.
length of the 1D combustion tube to 10m to better observe the front propagation
behavior. The same grid block size (0.36cm in x direction) is used here. A more
detailed sensitivity study on the key input parameters for the three reaction model
is also presented in [86]. That study is based on an experimental design technique to
create the response surface map for different initial conditions, boundary conditions,
and input parameters.
We first perform tests for different air flux rates. The results are shown in Fig.
6.8. We observe a close to linear relationship between the air flux and combustion
front velocity. This implies that a constant amount of fuel, ”x” is consumed, again
consistent with the classical Gates and Ramey analytical approach [44]. The peak
temperature, as expected, shows a slightly increasing trend with increasing air injec-
tion rates, starting from 750oC at 750cc/min to 950oC at 7500cc/min.
We also test the effect of initial oil saturation, Soi, in Fig. 6.9. The combustion
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 87
fronts move to very similar locations, implying a consistent fuel amount ”x” per
reservoir volume burned. The Soi does impact the size of the oil banks that are
determined by the difference Soi − Sofuel, in which Sofuel is the saturation amount
of oil burned as fuel. The consistent fuel amount is likely due to the fact that once
steady state propagation is achieved, the combustion front sees the same oil bank
(So = 0.8 here) coming towards it, if we consider a frame of reference that moves at
the same speed as the combustion zone. Consequently the oil goes through similar
temperature increase in the combustion zone. The trailing edge of the oil bank moves
at more or less the same speed, regardless of Soi. The only difference is the different
oil bank lengths in front of the combustion zone. In the special case of Soi = 0.05,
the fuel amount Sofuel is higher than Soi. There is no oil bank.
For varying initial water saturation Swi, we observe a weak sensitivity on the
combustion front speed and fuel amount, as shown in Fig. 6.10. The reaction front
moves to similar locations in all these cases. This may be explained by the fact that
liquid initial water is vaporized further downstream away from the combustion zone
through the steam plateau. Thus the chemical reactions are not quite affected by
Swi.
We also study the sensitivity of oil in-situ viscosity. The previous simulation study
is for a very viscous bitumen type crude oil. We also study the 1D combustion process
of a mobile heavy oil (typical in-situ viscosity ∼ 103cp). We use the three reaction
models [33] with modifications to a conventional heavy-oil combustion problem (Lloy-
dminster type) by changing the viscosity-temperature relationship. This is because
the three reaction model is simple and capable of capturing the first order physics
of pyrolysis, LTO and HTO reactions. Initially, we have saturations Soi = 0.75,
Swi = 0.15 and Sgi = 0.10. The original oil viscosity is µo = 4600cp, with a initial
temperature of Ti = 15oC. We show the temperature, oil saturation and pressure
profile in Fig. 6.11. We observe a relatively small-sized steam plateau in this dry
forward combustion simulation. We also see a relatively large pressure drop over the
cold oil bank, in which the gas saturation has been reduced to a certain value that
is very close to critical gas saturation in the gas-liquid relative permeability. In the
case of wet combustion, the steam plateau size is greater, because of the additional
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 88
Figure 6.10: Temperature profiles as a function of the initial water saturation Swi of0.10, 0.24 and 0.50 from left to right.
heat carried forward from the burned zone by the injected water.
From these numerical simulation sensitivities studies, we observe the same be-
havior as the classical Gates and Ramey assumption [44]. Furthermore, we plot the
characteristics of different fronts in the 1D ISC process. These fronts include: com-
bustion front, leading edge of the oil bank, leading edge of the steam plateau. We
have observed two basic scenarios. Scenario 1 is the case when we have a low initial oil
saturation and a bitumen-like oil with very high in-situ oil viscosity, typically > 104cp.
Scenario 2 is the case when we have a less viscous conventional heavy oil (initial oil
viscosity in the 103cp range). In scenario 1, because of the high oil viscosity, the oil
bank stays inside the steam plateau. The leading edge of the steam plateau moves
ahead to warm up the formation downstream of the high saturation oil bank. This
is the case for the simulation studies of sections 6.1 and 6.2. In scenario 2, the oil is
more mobile such that the banked oil moves downstream of the steam plateau. The
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 89
Figure 6.11: Combustion tube simulation of a mobile heavy oil with high initial oilsaturation Soi at 378min.
steam plateau heat front drives the banked oil downstream. This is the typical case
in the field when we perform ISC for a mobile heavy-oil reservoir [41, 75, 81].
6.4 Grid Size Effects and their Cause
6.4.1 Grid Size Effects
In traditional Arrhenius kinetics, the reaction term Qreaci is highly dependent on the
local temperature T and reactant concentration Ci and/or O2 partial pressure. The
basic assumption for first order finite volume simulation is instant equilibrium and
instantaneous mixing in each cell, as shown in Fig. 6.14. This means that all physical
properties are homogenized (averaged) inside each grid block. Thus, we have only
averaged temperature and reactant concentrations that determine the kinetic reaction
terms in Eq. 3.18.
In reality, the reaction front is typically several centimeters thick at most, in both
lab experiments and field operations [96]. In order to capture accurately the detailed
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 90
Figure 6.12: Characteristics of different fronts in 1D ISC problem: scenario 1 (bitumenlike oil viscosity and oil bank stays inside the steam plateau).
Figure 6.13: Characteristics of different fronts in 1D ISC problem: scenario 2 (mobileheavy oil viscosity and leading edge of steam plateau follows behind the oil bank).
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 91
Figure 6.14: Basic assumption for thermal reservoir simulation: instantaneous mixingin each cell.
reaction kinetics inside the combustion front, we need centimeter sized grid blocks.
We tested a typical 1D combustion tube problem (0.8m long tube) using different
numbers of grid blocks. In this simulation study, we used the Crookston model for
kinetic reactions, with the same reaction parameters and settings independent of grid
size. The number of grid cells ranges from 10 to 800, with corresponding grid cell sizes
from 8cm to 0.1cm. Results are shown in Fig. 6.15. The temperature profiles and the
combustion front locations are strongly dependent on grid size. To achieve a spatially
convergent solution in this case (based on ”eye ball norm” of temperature profile), we
need at least 200 grid blocks, or grid blocks less than 1cm. For other kinetic models
such as the three reaction model, we have observed a similar grid-size requirement.
One important observation is that coarsening the grid typically leads to a greater peak
temperature and a lower front propagation speed. When the front sweeps through
a unit reservoir volume, the total amount of hydrocarbons burned increases with
grid-block size. This observation of excessive coke formation and slower front speed
on coarse grids is also well documented in the literature [22, 47, 61, 112]. In many
engineering applications, the reaction kinetics parameters are changed or tuned to
overcome this effect in a large field-scale grid block [47].
6.4.2 The Cause of Grid Size Effects
When the combustion front approaches a grid block, part of the oil inside the block
pyrolyzes and generates solid coke. This coke is the main portion of the fuel for later
exothermic oxidation reactions. The rest of the oil moves out of this cell by either
viscosity reduction or by distillation effects. This is a physically continuous process.
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 92
Figure 6.15: Grid-size effects in 1D combustion tube simulation. We require at least200 grid blocks to achieve convergent solution in this 0.8m long combustion tubesimulation.
When we heat up the oil inside a block, the oil cracks. In the mean time, the oil also
flows out of the cell. If we use a larger simulation grid block, the block has larger
heat capacity, which generates relatively slower heating rate. This makes the oil takes
longer to pyrolyse/crack. The competing mechanism between pyrolysis/coking and
oil flowing out of the reaction cell does not scale with grid block sizes. This is caused
by the default instantaneous mixing or averaging assumption and discrete nature in
first order finite volume reservoir simulation Fig. 6.14. Consequently, less fraction of
oil moves to the next cell and more fraction of solid fuel is generated as a numerical
artifact. As a result, the front speed slows down. Eventually, the excessive amount
of fuel or coke also introduces higher peak temperatures.
6.4.3 Why not Adaptive Mesh Refinement?
Cartesian Adaptive Mesh Refinement (AMR) [84] was previously proposed for kinetics
based in-situ combustion simulation. The main challenge for using AMR for field-
scale ISC simulation is that it is practically very difficult to implement because of
the dramatic grid size contrast between areas near the front and away from it. The
typical grid block size for full field reservoir simulation is on the order of 10m. We have
observed < 1cm grid size requirement for kinetics based ISC simulations. In practice,
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 93
it is difficult to generate a high quality AMR grid when we have a narrow combustion
front of large spatial extent and large shrinkage requirements of the grid size. A multi-
level AMR grid may be necessarily needed for such dramatic changes of grid sizes.
Another practical concern is the choice of time step sizes. When implementing AMR
grid for kinetics based ISC simulation, the time step size is still restricted by the fast
chemical reactions right at the combustion zone. In current kinetics based simulation,
we observe time step size requirements in minutes. Therefore, if we intend to simulate
the field-scale ISC process for years, we will necessarily have many time steps and the
simulation becomes impractical to execute. Furthermore, the computational workload
per grid cell is not the same everywhere. In areas away from the combustion zone, the
workload is dramatically smaller than near the combustion front. When coarsening
the grid far from the front in AMR, the cost saving is relatively small. Because of
the intense workload in the near front region, the computational cost is still intense.
Actually, the implementation of a two-level dynamic gridding in commercial thermal
simulator is shown in [22, 95]. The speed-up gained (reported 2× to 3×) is still
limited in 3D realistic cases. This is the main reason that we instead seek the reaction
upscaling approach as shown previously.
6.4.4 The Need for Upscaling
The computationally affordable grid block sizes for full field models are typically tens
to hundreds of meters laterally. There are orders of magnitude differences between
the affordable and the desired grid block sizes for kinetics based in-situ combustion
simulation. The cost of kinetics based simulation hinders the application of numerical
simulation for field-scale simulations.
In-situ combustion is indeed a multi-scale multi-physics process bringing together
porous media flow, chemical reaction kinetics, and phase equilibria. First, the chem-
ical reactions happen very fast, in the time-scale of minutes or hours. The advection
of fluid flow, however, is a much slower process that has a typical time scale of years.
The Suplacu field ISC process has been more than 40 years [91]. Second, the chemical
reactions are localized inside the centimeter-sized reaction front, while the advective
CHAPTER 6. IN-SITU COMBUSTION SIMULATION 94
flow takes place in the reservoir, typical length scale in kilometers. Therefore, both
the temporal and spatial scales of reaction kinetics and advection are very different in
the ISC process. An important dimensionless number that characterizes this differ-
ence is the Damkohler number (Da). It represents the ratio of the characteristic time
scale for the fluid flow to that of the kinetic reactions. In a lab-scale combustion tube
experiment, the Da number is usually of the order of 103. In field-scale ISC, the Da
number is typically around 107, because the advection time scale is in years rather
than hours. To alleviate this scaling problem, we must find grid-insensitive reaction
models. Pursuing such an upscaled reaction model is the goal of this work.
6.5 Concluding Remarks
In traditional thermal reactive reservoir simulation, mass and energy balance equa-
tions are solved numerically on discretized reservoir grid blocks. The reaction terms
are calculated through Arrhenius kinetics using cell-averaged properties, such as aver-
aged temperature and reactant concentrations. The chemical reaction front is phys-
ically very narrow, typically a few inches thick. To capture accurately this front,
centimeter-sized grids are required that are orders of magnitude smaller than the af-
fordable grid block sizes for full field reservoir models. Therefore, the reaction kinetics
is a subgrid-scale phenomenon for field-scale simulation. The kinetics temporal and
spatial scales are significantly smaller than that of advection. It is extremely diffi-
cult to solve flow and kinetically determined reactions together simultaneously. We
need to upscale the reaction kinetics to the similar temporal/spatial scale of flow and
transport to solve this problem on the full field scale. This upscaling is explained in
detail in the next chapter.
Chapter 7
Upscaling for In-situ Combustion
Reactions
In this chapter, we focus on reaction upscaling of In-situ Combustion reactions. The
critical physics for the ISC process is shown in Fig. 7.1. When the reaction front
sweeps through the reservoir, a certain amount of oil becomes fuel and burns (usually
x = 5 ∼ 10% So). The rest of the mobile oil gets pushed further downstream
through either viscosity reduction or distillation. This process is also envisioned as
a bulldozing effect [47]: the reaction front burns some fuel and bulldozes everything
forward downstream. In order to simulate field-scale ISC, we need to capture this
process through upscaling of the reactions.
Figure 7.1: The critical physics for 1D ISC. When the reaction front sweeps throughthe reservoir, a certain amount of oil becomes fuel and burns (usually x = 5 ∼10% So). The rest of the mobile oil is displaced further downstream.
95
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 96
7.1 Non-Arrhenius Reaction Modeling
We know from previous analysis that the main cause for grid-size effects is the rela-
tively larger amount of fuel generated as grid block volume increases. We also know
for ISC, the reaction terms Qreaci in Eq. 3.18 need to be captured accurately in
particular.
Our main idea is to consider the equivalent reaction effects on a large field-scale
grid block. The most critical reaction effects should match the reaction effects on a
fine-scale model during a simulation time step from tn to tn+1. A certain amount of
oil becomes fuel and burns, consumes oxygen, generates light ends through pyrolysis,
heat and flue gas, and the rest of the oil flows further downstream. Because the
size of the reaction front is much smaller than the reservoir grid block, we do not
calculate the source and sink terms using cell averaged properties and Arrhenius
kinetics. Instead, we treat the reaction front as a sub-grid scale phenomenon and
calibrate the reaction effects through fine-scale simulations and lab experiments. We
only replace the reaction terms in the governing equations, with transport solved as
before. Equivalent reaction effects (fuel amounts) are calibrated here through 1D
fine-scale simulation as in Chapter 6, or collected experimentally using combustion
tube runs. Once we calibrate the equivalent reaction effects (fuel amount) from 1D
fine-scale simulation, we then infer reaction effects for a 3D ISC process accordingly.
A similar concept has been applied to idealized convection-reaction problems in [49].
Tracking the fuel amount concept is very similar to the classical Gates and Ramey
analytical model [44].
We follow these steps:
1. We first achieve a reasonable history match of the kinetic cell and combustion
tube using a chosen kinetics based model [10, 68].
2. We run fine-scale 1D simulations (typically 5 to 10 meters long), with various
uniform initial conditions and constant air-flux boundary conditions. As shown
in section 6.2, after the initial ignition the reaction front very quickly reach a
steady-state propagation with almost constant speed (usually within the first
0.1m). This implies that the front consumes a constant amount of fuel ”x” per
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 97
unit reservoir volume, with a constant oxygen utilization ratio. We define the
amount of fuel as the amount of hydrocarbon components that are oxidized
when the reaction front sweeps through. The fuel amount is collected in a
table over the range of conditions tested (usually the fuel amount is 5 ∼ 10%
So). This amount of fuel will be a certain value for each oleic hydrocarbon
component [C1burn, C2burn · · ·Cnburn]. According to the local conditions such
as saturations and pressure, we decide the fuel amount for each grid block in
the 3D reservoir, when the combustion front sweeps through. Calibration tests
for fuel amount and oxygen efficiency are performed for different initial and
boundary conditions, including different pressure, different initial saturation,
different air flux and petrophysical conditions. Petrophysical properties such
as type of porous media, mineralogy and clay content, also affect kinetics and
the ISC fuel amount that are currently not supported in Arrhenius kinetics
based commercial simulators. We are able to include these effects into our new
upscaled reaction modeling framework.
3. We assume that with enough initial heating and proper air flux, the combustion
ignites around the injection wells. When the reaction front starts propagating,
we calculate the reaction source and sink terms. During each time step, we
flag the cells that currently have combustion reactions. Inside each combusting
cell, we have a reasonable estimate of the quantity of the fuel based on tests
performed under step 2. We now perform equilibrium combustion reactions
according to the supplied air flux and the remaining fuel. If all fuel is consumed,
the cell is un-flagged. When the reaction front moves into a new cell, we flag
that cell as one of the combustion cells.
In this method, we only track how much fuel has been burned during a time step.
The equivalent reaction effects are used for source and sink term calculations on a large
field-scale grid block. This approach also removes the stiffness encountered in kinetics
based simulation. We can let the transport decides the time step sizes. Typically this
results in time step sizes ∆t = tn+1 − tn of the order of days. Thus reaction effects
and flow advection are effectively modeled at the same temporal and spatial scales.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 98
Figure 7.2: ISC upscaling work flow.
As we discussed earlier in section 6.3, traditional kinetics based simulation is strongly
limited by the grid size requirement. This change makes the numerical simulation
much more efficient.
The complete upscaling work flow is shown in Fig. 7.2. First, we start the kinetic
cell experiments to understand the oxidation reactions for the crude. Combustion
tube experiments are also conducted to test the 1D combustion process. Then, we
perform numerical simulation to match the lab experiments. At this stage, we use
Arrhenius kinetics to model the reactions and history match lab experiments. We
use a large number of grid blocks with enough resolution to ensure convergence and
accuracy at lab scale. Based on this, we extract the most important parameter
for in-situ combustion design: the fuel amount ”x”. When the reaction front is
propagating through the reservoir, we directly calculate the reaction source and sink
terms according to this fuel amount ”x”. In this way, we achieve consistency when
scaling up the simulation from lab to field.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 99
7.2 Implementation
7.2.1 Pseudo Code
The pseudo code implementation is shown here. The explanations for major variables
and major functions are given in Table 7.1 and Table 7.2. The main idea is to track
the remaining fuel amount for each component Ci in each reaction grid block. Instan-
taneous equilibrium oxidation reaction is assumed. The fuel is generated when the
reaction front moves into a virgin cell. When the fuel has been completely consumed,
the reaction front moves onto the next cell. We use special treatment for the reaction
cell oil phase flux to ensure complete sweep Sor = 0, i.e. all the oil flows out after the
front has completely swept through that cell.
Within each Newton-Rhapson Iterations: Loop (i,j,k)
If (Flag_reaction_cell(i,j,k)==TRUE)
CALL Calculate_Oil_Flux(Flux_O2(i,j,k),Fuel_amount(i,j,k),Ci(i,j,k))
CALL Calculate_Reac_Term (Flux_O2(i,j,k),Fuel_amount(i,j,k))
Else
If (Flux_O2(i,j,k)>0 AND So(i,j,k)>0)
Flag_reaction_cell(i,j,k)=TRUE
If (So(i,j,k)>So_burn)
Fuel_amount(i,j,k)=Ci_burn
Ci(i,j,k)=Ci(i,j,k)-Ci_burn
!Remaining oleic components eventually be swept out by ISC front
Else
Fuel_amount(i,j,k)=Ci(i,j,k)
Ci(i,j,k)=0
!If So(i,j,k)<So_burn, all the oil will be consumed as fuel
!Nothing flows out
End If
CALL Calculate_Oil_Flux(Flux_O2(i,j,k),Fuel_amount(i,j,k),Ci(i,j,k))
CALL Calculate_Reac_Term (Flux_O2(i,j,k),Fuel_amount(i,j,k))
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 100
End If
End If
End Loop
Function Calculate_Reac_Term(Flux_O2(i,j,k),Fuel_amount(i,j,k))
If (Oxygen Limited) !More fuel than Oxygen
Reac_Term=Flux_O2(i,j,k)*Stoichiometry
Fuel_amount(i,j,k)=Fuel_amount(i,j,k)-Flux_O2(i,j,k)*Stoichiometry
Else If (Fuel Limited) !More Oxygen than fuel
Reac_Term=Fuel_amount(i,j,k)*Stoichiometry Fuel_amount(i,j,k)=0
Flag_reaction_cell(i,j,k)=FALSE
End If
End Function
Function
Calculate_Oil_Flux(Flux_O2(i,j,k),Fuel_amount(i,j,k),Ci(i,j,k))
If (Oxygen Limited) !More fuel than oxygen
Temp_ratio=Flux_O2(i,j,k)*Stoichiometry/Fuel_amount(i,j,k)
!Ratio of fuel be consumed, over the total fuel initially in this cell
Flux_Oil=Ci(i,j,k)*Blksize*Temp_ratio
Else If (Fuel Limited) !More oxygen than fuel
Flux_Oil=Ci(i,j,k)*Blksize
End If
End Function
7.2.2 Implementation in Commercial Software
To test our upscaling methodology, we have preliminarily implemented our concept
in commercial reservoir simulators [37, 101] using a simple technique as follows.
All current commercial simulators are based on Arrhenius kinetics. Therefore,
we have developed some techniques to implement the reaction upscaling. All the test
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 101
Table 7.1: Explanation of major variables in the upscaling pseudo code.
Major Variables ExplanationFlag_reaction_cell Flag for cell (i,j,k) that are undergoing combustion reactions.Fuel_amount The remaining fuel in cell (i,j,k) for Nc oleic components.Flux_O2 Total amount of supplying O2 flux flowing into cell (i,j,k).So_burn The oil saturation that will be burned as fuel, if initial oil
saturation Soi>So_burn. If initial oil saturation Soi<So_burn,then all the Soi will be burned as fuel.
Ci_burn Amount of the i-th component that will be consumed as fuel,if Soi>So_burn.
Ci Amount of the i-th component currently in cell (i,j,k).
problems are small scale problems with uniform initial conditions, i.e. constant initial
saturations Soi, Swi, Sgi and pressure Pi. The original kinetic reaction model we use
is the three reaction model shown in Eq. 6.2. The corresponding upscaled model is
based on two pseudo kinetic reactions given by,
1. 1.0 Oil1 −→ (1.0− x) Oil2(inert) + x Coke (at the start of simulation)
2. Coke + O2 −→ CO + CO2 + H2O (during the simulation)
(7.1)
Oil2 is an inert component that does not engage in any reactions. Both reactions
have very large constant reaction rates (equilibrium reaction). The fuel amount is
pre-determined by the value x in the first reaction, that takes place at the beginning
of the simulation. In this way, we make combustion reactions insensitive to grid size.
Because we lay down the fuel at the start of the simulation, the relative permeability
is adjusted accordingly to account for this loss of oil saturation. This adjustment is
similar to the method described in [98]. A sample STARS input file for upscaled three
reaction model is provided in Appendix D. We compare the original kinetics based
simulation and the upscaled model results.
We have tested 1D, 2D, and 3D cases with varying grid sizes. We do not history
match any particular experiments or field history, but show the consistency of results
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 102
Table 7.2: Explanation of major functions in the upscaling pseudo code.
Major Functions ExplanationCalculate_Reac_Term Function to calculate the reaction source and sink terms
for cell (i,j,k). The reaction terms are calculated accordingto the supplying oxygen flux flowing into the cell Flux_O2,and the remaining fuel amount Fuel_amount(i,j,k).If (O2 flux > remaining fuel), fuel all consumed, (i,j,k) unflagged.If (O2 flux < remaining fuel), fuel=Flux_O2*Stoichiometry.
Calculate_Oil_Flux Special treatment for oil flux flowing out of a reaction cell:when the fuel all consumed, remaining oil in this reactioncell completely flows out.
between the fine-scale kinetics based reference and the coarse scale upscaled model.
The major inputs of the kinetics of the three reaction model are provided in Appendix
A. The reference kinetics based simulation cases are performed in small physical
sizes, because we are limited by the computational costs associated with the fine-
scale Arrhenius kinetics based simulations. In all test cases, we have a large grid-size
upscaling ratio to show the consistent results achieved by the upscaled model.
7.3 Upscaling Test Cases
7.3.1 1D Tests
The first example is a 2.1m long 1D combustion tube problem shown in Fig. 7.3.
We inject air from the top and produce from the bottom. To achieve an accurate
reference solution, we first use the kinetics reaction model Eq. 6.2 with N=600 grid
blocks. Then, we simulate the same problem with N=10 grid blocks using both the
kinetics based model and our upscaled reaction model. The fuel amount in this case
is calibrated as Sofuel = 8%.
As we can see, the N=10 upscaled model is capable of predicting a similar reaction
front velocity and peak temperature as the fine-scale reference. If we instead use the
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 103
Figure 7.3: 1D ISC upscaling tests: kinetic and upscaled model using different gridresolutions. The coarse scale upscaled model matches the fine-scale kinetics basedreference, while the coarse scale kinetic model fails.
Figure 7.4: 1D ISC upscaled model with different equivalent fuel amount Sofuel. Thefront location has a direct relationship with the fuel amount Sofuel.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 104
original kinetic model and N=10 grid blocks, we fail to predict the correct front
location. We observe ”over coking” in this case, as explained in section 6.3. The
reaction front moves more slowly and we generate an excessive amount of coke in the
pyrolysis reaction. The excessive amount of coke also leads to greater heat generation.
This is consistent with the discussions in section 6.3.
We further show the results by making adjustments to the equivalent fuel amount
Sofuel value in Fig. 7.4. We show that by directly changing the Sofuel value in the
upscaled model we are able to adjust how fast the front propagates to match a certain
fine scale solution. There is a reciprocal relationship between fuel amount and front
speed, providing the same amount of injected air.
7.3.2 2D Tests
We test a 2D quarter five spot problem, Fig. 7.5. The size of the 2D problem is
0.72m× 0.72m with homogeneous permeability and porosity. We inject air from one
corner and produce from the other corner. We first run the simulation using Arrhenius
kinetics with 200×200 grid blocks. This is used as the fine-scale reference result. We
then use coarser grid blocks (20×20 and 5×5) and compare the results. We plot the
oxygen fraction in the gas phase to show the exact location of the ISC reaction fronts.
For grid resolutions 20×20 and 5×5, the kinetic simulation underestimates the front
speed (shown by the red dashed line). The coarse 5× 5 kinetic simulation introduces
so much averaging that we have an initial ignition problem. We have much oxygen
bypass initially because of the low temperature in the near well region. Furthermore,
the peak temperature is also greater because of the excessive amount of fuel (coke)
generated at later times in the larger grid blocks. On the other hand, the upscaled
model predicts front locations very similar to the fine-scale reference. This establishes
that a similar amount of fuel is combusted.
We also test a 2D problem with multiple wells, as shown in Fig. 7.6. The size of
the 2D domain is 0.72m × 0.36m with homogeneous permeability and porosity. We
inject from two air injectors on the left and produce from the right. Again, a fine-
scale kinetics based simulation (200 × 100 grid) is performed to achieve a reference
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 105
Figure 7.5: ISC upscaling in horizontal 2D 1/4 five spot case: O2 fraction in gas phasewith different grid resolution and both kinetic and upscaled model.
Figure 7.6: ISC upscaling in horizontal 2D multi well case: coarse grid upscaled modeland fine grid kinetic model.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 106
Figure 7.7: ISC upscaling in horizontal 2D heterogeneous case: coarse grid upscaledmodel and fine grid kinetic model.
solution. We upscale the model to a 20× 10 grid and run the simulation again with
our upscaled model. In this case we show the oil saturation contour. The upscaled
model successfully predicts a similar front location (the red dashed line) and also the
size and shape of the oil bank.
We further test a simple 2D heterogeneous case, as shown in Fig. 7.7. We the
permeability field (60 × 20) show in Fig. 7.7 with physical size 1.08m × 0.36m. We
further downscale the 60 × 20 permeability field to 300 × 100 using first order flat
reconstruction for fine scale simulation. We inject air uniformly from the left side
and produce from the right side. The oxygen mole fraction in the gas phase is again
used to show the reaction front movement. We observe clearly that the reaction front
locations are consistent between fine-scale kinetic model (300× 100) and coarse-scale
upscaled model (60× 20).
A 2D vertical reservoir cross section case is also performed to test gravity effects,
as shown in Fig. 7.8. We inject air from a vertical injector on the left and produce
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 107
Figure 7.8: ISC upscaling in 2D vertical case: coarse grid upscaled model and fine-gridkinetic model.
from a vertical producer on the other side. Again we show the oxygen mole fraction
contour. Gravity override is clearly observed in this case. Reaction front movement is
consistent between the fine-scale kinetic model and the coarse scale upscaled model.
7.3.3 3D Tests
Our final test case is a 3D large field-scale forward in-situ combustion process. The
reservoir is homogeneous with the size 150m×63m×15m. We have a vertical air
injector on the left and a producer on the right, as shown in Fig. 7.9. We have
implemented two different grid resolution here, discretizing the reservoir by 10×11×5,
25× 11× 10 and 50× 21× 10 grid blocks. This is to show the consistency achieved
by the upscaled model with different grid resolutions. We have only implemented the
upscaled model in this case, because of the lack of computational power for achieving
a fine-scale kinetics based reference solution with centimeter sized grid. If we use
the 0.36cm× 0.36cm× 0.36cm grid as in kinetics based simulation, the total number
of grid blocks is 3.04 × 1012 in this case. The results are shown in Fig. 7.9. We
observe the combustion front moving towards the producer. For the grid resolution
25× 11× 10 and 50× 21× 10, gravity override is also clearly captured. When using
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 108
10×11×5 grid, the simulation has some problem in representing the gravity override
because of the coarse grid in the Z direction. The simulation results are consistent
with these three different grid sizes. This demonstration purpose example indicates
the potential capability of the upscaled reaction model for field-scale simulation.
We also compare the average time step sizes taken in this 3D field-scale case
(50 × 21 × 10 grid) in Fig. 7.9 with the previous small scale 1D kinetics based
combustion tube simulation in Fig. 7.3. In this 3D field-scale with upscaled reaction
model, we take average time steps of 0.25day. The average time step size in kinetics
based small scale simulation is 0.002day. In the kinetics based model, we need to
capture the fast transient chemical reactions in every small grid block with combustion
reactions. As a result, the time step size in the simulation is highly restricted. On
the other hand, the upscaled model captures the cumulative reaction effects over a
much longer time period and also a larger spatial scale. Reaction upscaling helps to
overcome the stiffness in the governing equations, that makes the simulation capable
to take larger time steps (in days) and simulate for reservoir scale time periods.
7.4 Discussion and Applications
7.4.1 Valid Range
Our upscaling methodology replaces the Arrhenius kinetics terms with upscaled terms
that have, at the scale of the coarse grid block, the equivalent overall effects on mass
and energy transport. Other than that, the governing equations are not changed. In
the preliminary test cases shown, the upscaled model gives similar results on coarse
grids as the original kinetics based model on fine grids. This is because the fuel
amount, determined by ”x” in Eq. 7.1, is carefully calibrated. This also ensures the
consistency in the reaction term calculations, when using different grid block sizes
on all levels of resolution. The methodology is also different than simple parameter
tweaking [28, 47, 53], that is dependent on the grid block sizes. Because the very
stiff kinetics have been removed from the governing equations, we are able to take
dramatically larger time steps. Accordingly, we believe that the efficiency of our
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 109
Figure 7.9: 3D field-scale case using the upscaled reaction model. Consistency isachieved between different grid resolutions.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 110
upscaled method is much greater than traditional kinetics based simulation. We are
strongly encouraged by the accuracy obtained by our upscaling methodology.
Assumptions and restrictions for our ISC reaction upscaling methodology are laid
out next.
• In our upscaling approach, it is assumed that the combustion reaction has been
sustained in the HTO mode. In many field operations, failure to keep the
combustion in the HTO mode is the actual cause for many unsuccessful ISC
projects [85]. A further improvement that we suggest is the minimal air flux
approach. An operational window is defined for combustion to sustain in the
HTO mode [85]. This improvement is discussed in detail in the future direction
section 9.2.2.
• The upscaling approach is designed specifically for heavy oil combustion prob-
lems, where the reaction is sustained in the high temperature oxidation mode
with a very thin combustion zone. A light oil high pressure air injection (HPAI)
process operated in LTO mode could have wide extended reaction zone, instead
of the thin combustion front for heavy oil [46]. Combustion operated in LTO
mode is beyond the scope of the current work [42, 46, 79].
• We only consider the forward drive-type in-situ combustion process, in which
the different fluid phases flow almost co-currently. We currently do not consider
cases with counter-current combustion gravity drainage, because such cases do
not appear in practice at this time.
• We consider a single porosity reservoir. For a fractured dual porosity reservoir,
we need to understand the matrix-fracture heat and mass transfer. A specialized
upscaling technique is required, which is an ongoing research topic [39].
• The upscaling technique is for chemical reactions only. More research is needed
for sub-grid scale phenomena such as fluid fingering or unstable displacement
fronts. We discuss this in detail in the next section.
• We have either complete or partial oxygen utilization. In the case of partial
oxygen utilization, part of the oxygen bypasses the reaction front and is not
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 111
fully combusted. This may be due to local small-scale heterogeneity, that is
discussed in detail in the next section. We need to define an extra parameter:
oxygen utilization efficiency Eu(0 ∼ 1), that describes the fraction of oxygen
consumed in the reaction zone. This parameter is also calibrated through fine-
scale simulation or experiment.
7.4.2 Sub-grid Scale Heterogeneity and its Effects on Reac-
tion Upscaling
In most of the previous simulation studies, we implement small-scale problems with
moderately heterogeneous permeability fields (usually within 5 orders of magnitude).
In all these cases, the reaction front is topologically continuous. The domain is
divided into the burned zone and the unburned zone. Almost all the injected oxygen
is consumed at the thin reaction front, moving from injector towards the producer.
We do not observe oxygen or residual oil bypassing. One immediate question is what
if we have a more heterogeneous porous medium. Related questions arise. Is the
combustion front still topologically continuous and sweeping the whole area? Do we
get discontinuous fronts? Is there oil left behind? Do we have oxygen bypass? How
is fuel amount Sofuel obtained?
We perform tests of some sub-grid scale heterogeneous systems to test the oxygen
and residual oil bypassing and answer the above questions. The test cases are per-
formed using Arrhenius kinetics (three reaction model). We present a series of cases
here, that include white noise permeability, systems with square size low permeabil-
ity, layered systems, upper and lower layer of SPE 10 permeability [23]. The first
example is a 2D problem (0.072m × 0.54m) with uncorrelated white noise, 4 orders
of magnitude permeability contrast from 100mD to 1000D in Fig. 7.10. In this case,
we do not observe much oxygen bypass. The residual oil in the burned zone has been
fully swept forward by the continuous combustion front. Using material balance, we
calculate the equivalent fuel amount as Sofuel = 8.1%, which is very close to the fuel
amount of homogeneous case Sofuel = 8.2%.
The second example we tested is a 2D problem with a square low permeability
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 112
Figure 7.10: Permeability and oil saturation results for 2D case with uncorrelatedwhite noise permeability field. The combustion front propagates stably.
zone in the middle, shown in Fig. 7.11. The simulation results show three contin-
uous snapshots 1 to 3. The red dashed line shows the approximate location of the
high temperature front of the main combustion zone. Because of the strongly het-
erogeneous permeability field, some oil is initially bypassed and not fully combusted.
Eventually, by heat conduction and oxygen diffusion in the gas phase, the bypassed
oil is fully combusted after the main reaction front moves through. We also observe
oxygen bypassing the reaction front because of the strongly correlated heterogeneity.
Through the use of material balance, we further compare equivalent fuel amount to
the same 2D problem with 10D homogeneous permeability. For the 10D homoge-
neous case, the fuel amount is Sofuel = 8.2%. With the low permeability square, the
fuel amount becomes Sofuel = 9.0%. The fuel amount has been changed because of
the heterogeneity. When the combustion front moves in, the oil in the lower perme-
ability zone is more difficult to flow further downstream. This causes more oil in this
zone to coke, thus forming more fuel for later combustion reactions. Heterogeneity
causes oxygen bypassing and formation of a greater amount of coke relative to the
homogeneous case. We also calculate the cross sectional oxygen utilization efficiency
Eu = (Oxygen Consumed)(Oxygen Injected)
through material balance. For this low permeability square
case, the overall cumulative oxygen utilization efficiency is Eu = 0.85.
The third example is a 2D problem with its permeability field shown in Fig. 7.12.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 113
Figure 7.11: 2D case with correlated heterogeneity (low perm in the middle). Weobserve some oxygen bypass and change in equivalent fuel amount.
We randomly generate low permeability zones in the 2D problem. We again combust
from left to the right. A small amount of oxygen bypass is detected. We calculate the
overall oxygen utilization efficiency and fuel amount as Eu = 0.90 and Sofuel = 9.7%,
respectively.
The fourth example we show is a 2D horizontal layered system in Fig. 7.13. We
implement two permeability contrasts K1/K2 = 2 and K1/K2 = 10. We observe
some oxygen bypass and residual oil bypass. After material balance calculations,
we find oxygen utilization efficiency as Eu = 0.81 for K1/K2 = 2 and Eu = 0.66
for K1/K2 = 10. The overall fuel amount for K1/K2 = 2 is Sofuel = 8.1% and
Sofuel = 8.7% for K1/K2 = 10.
The fifth example we show is a 2D problem with layer 1 permeability of SPE 10
[23] in Fig. 7.15. In this case, we do not observe residual oil bypass. We also observe
very little oxygen bypassing the main reaction front. We calculate the fuel amount
and oxygen utilization efficiency as Sofuel = 8.6% and Eu = 0.91.
The sixth example we show is a 2D problem with highly heterogeneous channel-
ized permeability (layer 51 of SPE 10 [23]) in Fig. 7.15. In this case, we have the
temperature front moving relatively stably because of heat conduction. Again, some
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 114
Figure 7.12: 2D case with correlated low permeability squares. We calculate oxygenefficiency Eu = 0.90 and fuel amount Sofuel = 9.7%.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 115
Figure 7.13: 2D case with horizontal layered permeability. We observe oxygen bypassand change in equivalent fuel amount.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 116
Figure 7.14: 2D case with heterogeneity (layer 1 of SPE 10 [23]). The oxygen efficiencyis Eu = 0.91 and fuel amount Sofuel = 8.6%.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 117
Figure 7.15: 2D case with channelized heterogeneity (layer 51 of SPE 10 [23]). Theoxygen efficiency is Eu = 0.85 and fuel amount Sofuel = 13%.
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 118
Table 7.3: Summary of Sub-grid Scale Heterogeneity ISC Tests
Case Number and Description Eu Sofuel
1. White Noise 0.88 8.1%2. Low Permeability Middle Square 0.85 9.0%3. Low Permeability Squares 0.90 9.7%4a. Layered System (K1/K2 = 2) 0.81 8.1%4b. Layered System (K1/K2 = 10) 0.66 8.7%5. Layer 1 of SPE 10 [23] 0.91 8.6%6. Layer 51 of SPE 10 [23] 0.85 13%
oil is initially bypassed. We also observe oxygen bypassing the main reaction front.
We calculate the fuel amount and oxygen utilization efficiency as Sofuel = 13% and
Eu = 0.85.
In this section, we tested ISC front propagation in heterogeneous systems, cover-
ing a range of different permeability fields. A list of the summary of the test cases
are shown in Table 7.3. We have included some extreme cases here, such as the
channelized lower layer of the SPE 10. From the previous test studies, we see oxygen
utilization and residual oil bypassing events are case dependent. In cases with more
correlated heterogeneous permeability fields, we observe more different oxygen uti-
lization (oxygen bypass) and its equivalent fuel amount. We observe fuel amount and
oxygen bypass generally increases for more correlated heterogeneity. These numbers,
however, are still varying mostly within a small range, as shown in Table 7.3. The
fuel amount Sofuel varies from 8.1% to 13%, and oxygen utilization Eu from 0.66 to
0.91.
The different scales in measurements, geomodeling and reservoir simulation are
shown in Fig. 7.16. The typical geomodelling grid block size for a reservoir model is
on the order of 100ft×100ft×1ft. Between core/well log data and geo-cellular grid,
there is a ”missing scale”. This means the core/well log data is directly assigned to
the entire large geo-cellular grid in modern reservoir engineering. The scale of ISC
combustion front is actually inside this ”missing scale”. The simulation cases we have
shown here are small-scale test cases, typically less than 1m × 1m. We are actually
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 119
testing the potential impact of sub-grid scale heterogeneities in these cases. The
previous material balance calculations for Sofuel and Eu serve as a first step upscaling
methodology if ideally we are able obtain this detailed sub-grid scale heterogeneity.
For example, when we scale up the 2D domain in these tests to one geo-cellular
grid block in full field reservoir model, we calibrate the fuel amount and oxygen
utilization efficiency. In Fig. 7.16, we show the ”missing scale” from core/well log to
geomodelling grid (uncertainty in permeability, porosity, wetting, clay contents, etc.).
Considering the lack of sub-grid scale heterogeneity information through well log,
seismic and well tests in this ”missing scale”, we need to bridge the gap between the
core/well log and geo-cellular grid. We suggest performing uncertainty study based
on different realizations of the sub-grid scale heterogeneity. The simulations and the
sensitivities with respect to permeability distributions is a first step in understanding
the spread of the fuel amount and oxygen utilization. When history matching full field
reservoir models, the equivalent fuel amount and oxygen utilization coefficient are also
allowed to be perturbed. We hope this helps to represent heterogeneity in all scales.
This is very analogous to the concept of sand to shale net-to-gross ratio (NTG), or
the mixing parameter ω for sub-grid scale heterogeneity induced dispersion in Todd-
Longstaff miscible flooding formulation [108]. This enables us to account for these
sub-grid scale heterogeneities that cause oxygen bypass and change in fuel amount.
From geomodelling to reservoir simulation grid, we suggest the same classical flow
transport upscaling technique [20]. The fuel amount and oxygen utilization efficiency
are upscaled similarly as porosity φ or NTG.
7.5 Concluding Remarks
In this work, we propose a new upscaling method for heavy-oil ISC reaction modeling.
We no longer calculate the reaction terms using Arrhenius kinetics and cell-averaged
values. We instead find and calibrate the equivalent reaction effects, i.e. the amount
of heat released, the volume of fuel burned and the reaction products generated. This
calibration is done through fine-scale numerical simulation or from lab combustion
tube experiments. Test cases have shown the consistency achieved between fine-scale
CHAPTER 7. UPSCALING FOR IN-SITU COMBUSTION REACTIONS 120
Figure 7.16: Different scales in measurements, geomodeling and reservoir simulation,lecture notes of ERE 241 Seismic Reservoir Characterization.
reference and coarse-scale upscaled model. This approach significantly improves the
computational speed for in-situ combustion simulation. It is also applicable for other
subsurface reactive flow modeling problems with fast chemical reactions and sharp
reaction fronts.
Chapter 8
Front Stability Study for In-situ
Combustion
8.1 1D Flow Regimes
Very limited work has analyzed front stability during ISC. This topic is important for
the design of ISC processes, because an unstable front and viscous fingering can lead
to premature front break through. If we do have viscous fingering due to unstable
frontal displacement, we probably also need to implement some upscaling approach
for the viscous fingers, similar to the pseudo relative permeability technique used in
miscible flooding [51]. We need to answer these questions as we move on to field-scale
ISC simulation, and after implementing the chemical reaction upscaling technique
mentioned earlier. In this chapter, we intend to explore this issue.
We start by studying the different flow regimes and their behaviors in a 1D ISC
process, and then move to 2D and 3D problems. We focus on the study of the scenario
of the ISC process in a conventional mobile heavy oil reservoir (in-situ oil viscosity
µoi ∼ 103cp), instead of highly viscous bitumen reservoirs (µoi ∼ 106cp). This is due to
the fact that most of the past successful ISC trials are within heavy-oil reservoirs, such
as in California and the Lloydminster region in Canada. Conventional ISC processes
with vertical wells in bitumen reservoirs have well-known injectivity problems, that
may require extremely small well spacing (< 1 acre) or extensive heating with cyclic
121
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 122
steam stimulation, as discussed in reference [55, 48].
The conditions in field operations are different than the laboratory experiments
and also the corresponding combustion tube simulations. For example, most combus-
tion tube lab tests performed at Stanford University are at low initial oil saturation
Soi ∼ 0.25. Small pressure drop is observed through the combustion tube [66]. Field
operations usually involve much higher Soi, that may induce the potential flow re-
strictions (choking) problems, that has been noticed in many Canadian projects [85].
The predictability of flow restriction problems is essential to the design of the ISC
process, especially when deciding the well spacing and sizing air injection capacity.
We identify three different fronts in the typical 1D dry/normal wet forward com-
bustion process shown in Fig. 8.1: the combustion zone 1, the leading edge of the
steam plateau 2, and the leading edge of the oil bank 3. These three fronts also
divide the 1D process into several flow regimes, shown as A, B, and C in Fig. 8.1.
In the area A, we only have air and water flowing (wet combustion process). Across
combustion zone 1, we assume a constant fuel amount Sofuel, as in the classical Gates
and Ramey method [44]. This fuel consumes the supplied oxygen, generating steam
and non-condensable flue gas, that flows further down into the area B. At the leading
edge of the steam plateau 2, steam condenses, releasing latent heat and mobilizing
the viscous oil into the oil bank. Flue gas keeps moving through into the area C,
the cold oil bank. In area C, both oil, water and flue gas flows forward. The high
pressure drop is due to the fact that the gas saturation (fractional flow of gas) has
been reduced to a very small value.
These three fronts influence each other in a 3D forward combustion process. By
inspection, we identify the leading edge of steam plateau 2 as the most unstable front.
This is due to the high pressure gradient contrast between region B and C, shown in
Fig. 8.1, which is also analogous to the miscible flood front instability. This will be
the most critical front that instability will first occur.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 123
Figure 8.1: Different fronts and flow regimes in 1D ISC process.
8.2 Contributors for Front Stability/Instability
We identify five contributors for front stability/instability: viscous force, condensa-
tion, heat conduction, coke plugging and gravity. We do not consider here chemical
reaction instability in this study. This is because we have Da ·Pe � 1 the product of
Damkohler number and Peclet number much larger than 1 (Da � 1 is because of the
fast chemical reaction kinetics in ISC, Pe � 1 is because ISC process is advection
dominated). Thus, the problem is flow transport limited [34]. Some of these factors
can be partially quantified by dimensionless numbers. Some of them are not easy to
be described merely by dimensionless numbers. They are discussed in detail later.
The five contributors to front stability/instability are discussed next.
• Viscous force: Viscous force is quantified by the mobility ratio M between the
displacing and displaced fluid. Complex multi-phase flow exists both upstream
and downstream of the steam plateau front. This complicates the mobility ratio
calculations. We suggest the pressure gradient contrast to be used instead. The
pressure gradient contrast between the steam plateau and the cold oil bank is,
∇PB
∇PC
=λtButB
λtCutC
, (8.1)
where λtB and λtC are the total phase mobilities in region B and C, and utB and
utC are the total phase velocities. The mobility contrast of the displacing (steam
and flue gas mixture) and displaced fluid (heavy oil) makes λtB � λtC . The
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 124
magnitude of total phase velocities utB and utC are close to each other. Thus
the overall pressure gradient contrast ∇PB/∇PC is � 1. This viscous force will
cause the front to be unstable. For example, in the mobile heavy oil ISC process
in section 6.2, the calculated pressure gradient contrast is ∇PB/∇PC ≈ 110.
• Steam condensation: In classical front stability analysis of steam flooding hot
water (no heat conduction included), the criterion is stated as,
(1 + 1−φ
φcv(Ts−Tr)
ρs(Hs−Hw)
)· µs(
1 + 1−φφ
cv(Ts−Tr)ρw(Hs−Hw)
)· µw
> 1, (8.2)
where cv is the volumetric heat capacity of rock, Ts and Tr are the steam sat-
uration temperature and initial temperature in the reservoir, ρs and ρw are the
densities of steam and water, Hs and Hw are the enthalpy of steam and water.
In this expression, the term (1−φ)cv provides extra stabilizing force. This is due
to the fact that steam condensation reduces the downstream magnitude of hot
water flux, thus reducing the pressure gradient contrast, also shown in Fig. 5.1.
The greater the rock porosity and heat capacity cv, the stronger the stabilizing
effect [50]. Condensation is intrinsic property of steam, thus no dimensionless
number derived here.
• Heat conduction: In the classical steam flooding hot water problem by Miller
[78], with the heat conduction included, the stability criterion becomes,
2π
λ
(Vsµs
ks
− Vwµw
kw
)+
(µs
ρsks
+µw
ρwkw
)·
Keγ (γr − γ)
(Hs −Hw)− cv,w
ρw(Ts − Tw) (γr − γ) λ
2π
> 0
(8.3)
where Ke is the effective thermal conductivity, and λ is the average wavelength
of a perturbation to the displacement front, shown in Fig. 8.2. Displacement
front stability improves with the added heat conduction, compared with the
previous analysis without conduction Eq. 8.2. The heat conduction stabilizing
effect is also dependent on the wave length of perturbation λ. The smaller
the λ, the more stable the front. This is also consistent with physical intuition,
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 125
Figure 8.2: Perturbation length λ at leading edge of steam plateau, from [50].
because heat conduction dissipates the heat of smaller wave length steam fingers.
The heat conduction stabilizing effect is partially characterized by the Nusselt
number Nu = F h·Lα
, where α is the heat conductivity, L is the characteristic
length of the problem and F h is the convective heat flux. The dimensionless
number, however, does not take into account the perturbation wave length λ in
Eq. 8.3.
• Coke plugging: Inside the ISC combustion zone, we have the pyrolysis reac-
tion that generates solid fuel coke that is deposited onto the porous media
surface. This deposited coke influences the porosity and permeability of the
porous medium. The coke reduces the permeability of the sand through this
geomechanical effect. Whenever an unstable front develops, this permeabil-
ity reduction improves the stability by creating flow resistance behind the fast
advancing part of the front. The permeability reduction is modeled in the com-
mercial thermal simulator CMG STARS through the option of solid component
blocking (*Blockage key word) [101]. The magnitude of permeability reduction
is highly dependent on the locations where solid coke is deposited. For exam-
ple, the deposition of coke around the pore throat and pore body will have very
different effects on pore blocking. Deposition at the pore throat causes much
more reduction in permeability than at the pore body. This is very similar to
the clay swelling problem in steam injection process [16]. In order to model the
pore blocking, we have implemented different blockage magnitude. We have
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 126
used the classical Kozeny-Carman correlation of porosity and permeability:
k = αφi
3
(1− φi)2 . (8.4)
We have also performed sensitivity on stronger blockage by coke deposition near
pore throats. Detailed sensitivity studies follow.
• Gravity: Consider the displacement of steam and flue gas mixture invading the
cold oil bank from top to bottom. The gravity difference between the steam
flue gas mixture and the liquid oil tends to stabilize the interface between them.
Whenever a steam finger is formed, the density difference slows down the finger
growth. To achieve an efficient gravity stable process, the injection rate (the
viscous force) should be controlled within a certain level and has been discussed
in detail in the gravity drainage section in [16]. In many field projects, we
tend to combust from up dip in the reservoir and move to down dip. This
injection configuration improves the sweep efficiency of the combustion front
[91]. For two phase oil and gas flow, gravity effect is usually described by the
gravity number Ng = (ρo−ρg)KφµovT
g, where K is the permeability, ρo and ρg are oil
and gas phase density, vT is the total phase velocity. Gravity number is one
parameter that characterizes the gravity stabilizing effect. In the ISC process,
factors such as temperature dependent viscosity, phase behavior and three phase
flow complicate the gravity number calculation and makes this dimensionless
number highly variable. It is not easy to estimate a single gravity number for a
particular problem.
8.3 Numerical Tests
8.3.1 Minimizing Numerical Errors
In this section, we show the numerical simulation we tested for ISC front stability.
The first work we performed is to probe the numerical errors (grid orientation effects
and temporal error in particular) in our simulations. Grid orientation effects are well
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 127
Figure 8.3: Trigger instability by changing the initial oil saturation at the boundary.
known as having strong influence on thermal and miscible flood EOR simulations
[60, 101]. We check the simulation results with different grid block sizes, different
numerical schemes (five point stencil and nine point stencil in CMG STARS [101]) and
different time step sizes. We draw conclusions about the behavior of the instability
if that behavior is very much the same across these variations. Because only then
we know it is the physical process that controls stability/instability. We perform
sensitivity studies to make sure the stable/unstable fronts we observe are not due to
numerical artifacts. We always trigger the instability at the same place by putting in
small perturbations in the initial condition (usually involving several grid blocks), as
shown in Fig. 8.3. We use a homogeneous permeability field so that we do not have
the problem of upscaling/downscaling when changing grid block sizes. We check here
whether the finger (instability) dimensions and growth rate are grid size, time step
size and numerical scheme (five point or nine point scheme) dependent.
We first show the simulation test case of a ISC process in 2D horizontal plane by
different grid block sizes and numerical schemes in Fig. 8.4. FIM time stepping is used
here. The time step sizes are selected automatically by the heuristics in commercial
simulator [101]. The size of the problem is 10m by 30m. We start the combustion
from the left side and propagate to the right side. The upscaled non-kinetic reaction
model is used here for this larger size problem. We use the same mobile heavy-oil
property as in section 6.2. The permeability is 10D and porosity is φ = 0.36. The
total simulation time is 7 days, which corresponds to an approximate front velocity of
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 128
about 2ft/day in this 2D problem. The instability is triggered initially at the center
of the left boundary by changing the initial oil saturation Soi, as shown in Fig. 8.3.
The simulation results are very dependent on the choice of grid sizes. At larger
grid block sizes, the results of five and nine point stencil diverge from each other.
When the grid block sizes are reduced to 0.222m×0.222m and 0.167m×0.167m, the
five and nine point stencil results start to converge.
This behavior is explained by the relative magnitude of numerical and physical
diffusion as follows. We start with an idealized model of a simple linear conservation
equation in 2D space,
ct + γcs = 0, (8.5)
where t is time, s is the coordinate along the velocity direction, and γ =√
u2 + v2 is
the magnitude of velocity (u and v are velocities along x and y directions). We show
the modified equation written in the flow aligned coordinate system (s, n) [60],
ct + γcs = γh (α1css + α2cnn + α3csn) , (8.6)
where h is the dimension of the grid size, α1, α2 and α3 are constants depending on
the numerical schemes used. The terms on the right hand side of the equation are
numerical diffusion terms. The numerical diffusion magnitude is proportional to the
grid size h. And the ratio between transverse cnn, longitudinal css, and cross term
csn diffusion are grid-size independent. This difference between numerical diffusion
terms css, cnn and csn are the cause of grid orientation effects. As seen in the work
by Kozdon [60], the grid orientation effects do not disappear as we refine the grid. In
adverse mobility ratio flooding problems, this is often seen as the five point stencil
tends to break through along the grid lines, while the nine point stencil tends to break
through in diagonal directions [101, 60].
In our thermal ISC process, we also have heat conduction effects, that can be
idealized as an extra diffusion term to this linear conservation equation. Now the
modified equation becomes,
ct + γcs = α (css + cnn) + γh (α1css + α2cnn + α3csn) , (8.7)
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 129
Figure 8.4: Test of numerical errors with triggered instability. The effect of grid sizeand 5-point versus 9-point scheme is illustrated.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 130
Figure 8.5: Test of temporal numerical errors with automatic time-step selection andrestricted time step sizes.
where α is the constant for physical diffusion term. If we reduce the grid block size
h, the numerical diffusion term magnitude decreases. At a certain point, the physical
diffusion term dominates. At such refined grid resolution, we draw conclusions that
the stability/instability is caused by a physical process, not numerical artifacts. The
previous test case is for such purposes. The observations are consistent with the
analysis here. We have conducted similar tests for all the the simulation studies later
on to ensure control of the numerical errors within a reasonable range.
We also test the choice of time step sizes to probe the temporal errors in our
simulations. We use the same 10m by 30m 2D problem, with grid block size 0.222m×0.222m. The same instability is triggered in the middle of the left boundary. We
compare the base case with the case having highly restricted maximum time step
sizes. We show one test case here in Fig. 8.5. The average time step size of the
base case is 23min. The restricted maximum time step size is 2min. The simulation
results are very similar to each other in this case. Spatial discretization error is more
dominant than temporal error as in this case. Similar tests, however, are performed
for the following test cases to ensure the temporal numerical error is controlled within
a certain range.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 131
8.3.2 Small Sub-grid Scale Front Stability
We first perform tests of 2D lab-scale simulations to study frontal stability. The
size of the 2D problem is 0.36m× 1.08m. We test dry forward combustion with the
same oil. Again we inject from the left side and produce from the right side. The
simulation time is 4 hours. This corresponds to an approximate front velocity of
6ft/day. We show the base case front propagation in Fig. 8.6. The permeability and
porosity field is homogeneous (K = 10D and φ = 0.36). We trigger the instability by
perturbing the initial condition on the left injection side, shown in Fig. 8.3. For all
the test cases we show, we have performed the previous test to ensure that physical
conduction/diffusion is dominating the numerical diffusion.
We have found at small scale that heat conduction plays an important role in
stabilizing the front. The original conductivity value we use is 2.58J/(cm ·min ·o C).
For simplicity we use a uniform conductivity value for the entire domain without
considering the fluid saturation effects. We demonstrate this by arbitrarily changing
the heat conductivity. We also show the temperature and oil saturation profiles with
50% conductivity (1.29J/(cm·min·oC)) and 30% conductivity (0.774J/(cm·min·oC))
in Fig. 8.6. As we decrease the heat conductivity values, unstable fronts appear. The
smaller the heat conductivity, the faster the steam finger becomes unstable and grows.
As seen from both the oil saturation and temperature profile, the instability starts
at the leading edge of the steam plateau where it is invading the oil bank. The
combustion zone follows right after the unstable steam plateau. This is consistent
with the previous analysis. We further calculate the heat Nusselt numbers Nu = F h·Lα
.
The heat Nusselt numbers are Nu = 4 for 100% heat conductivity case, Nu = 8 and
Nu = 13.3 for 50% and 30% heat conductivity cases, respectively. This means the
flow transport in this case is advection dominated.
We also repeat a simulation test with twice the front speed (air injection rate),
but keeping the original 100% heat conductivity. The calculated Nusselt number is
Nu = 13.3, the same as the 50% heat conductivity case. We show the oil saturation
and temperature profile, that corresponds to the same unstable front in Fig. 8.7.
This shows that the Nusselt number serves as one parameter characterizing the heat
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 132
Figure 8.6: Sensitivity to thermal conductivity. Heat conduction stabilizes the dis-placement front in small lab-scale tests. Reduced thermal conductivity cases showgreater instability.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 133
Figure 8.7: Nusselt number characterizes the heat conduction stabilizing effect, com-pared to Fig. 8.6.
conduction stabilizing effect. Furthermore, the conduction stabilizing effect also de-
pends on the perturbation wavelength, shown in Eq. 8.3 by Miller [78]. Because of
the confined geometry here, this is currently not taken into account.
We test the pore blocking effects by introducing blockage in the simulation. The
first example we show is the same lab-scale case as before with reduced 50% heat
conduction (1.29J/(cm · min ·o C)). The blockage is adjusted to be similar to the
Kozeny-Carman correlation Eq. 8.4. We show the temperature, oil saturation and
also the blockage coefficient profile Fig. 8.8. The simulation time is 5 hours. The
maximum blockage coefficient (permeability reduction factor) achieved is around 1.5
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 134
Figure 8.8: Pore blocking with the Kozeny-Carman correlation. Very small changesare observed compared to base case.
to 2.0. The steam plateau finger growth is very close to the base case without any
blockage. Tests show that mild pore blocking has very limited effect on self-correcting
the front. This is also consistent with the elliptic nature of the pressure equation.
Mild permeability change in one place does not give significant change to the overall
pressure distribution, and thus will not strongly affect the flow transport.
We further test the scenario with even stronger blockage with exaggerated perme-
ability reductions in Fig. 8.9. We again show the temperature profile, oil saturation
and permeability reduction factor. In this case, the permeability reduction is as great
as a factor of 30. We find that blockage deters the finger growth and stabilizes the
front somewhat at these conditions. To conclude, we observe coke plugging stabilizing
the front when we have at least an order of magnitude reduction in the permeability.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 135
Figure 8.9: Pore blocking with an exaggerated permeability reduction. The combus-tion front slows down only when implementing pore blocking of more than one orderof magnitude reduction in permeability.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 136
8.3.3 Large Field-scale Front Stability
We further test the ISC front stability in larger field-scale cases. We use the upscaled
reaction model for all the tests here because of the large dimensions. We show the
first example in Fig. 8.10. It is a 2D problem with the size 13.2m × 26.4m, with
the grid block size 0.22m × 0.22m. The initial oil saturation is Soi = 0.75 and
the fuel amount in the upscaled model is set at Sofuel = 7.5%. The same mobile
heavy oil property is used in this case. Heat conductivity is set at a constant value
of 2.58J/(cm ·min ·o C). We again introduce perturbation at the left boundary by
changing initial saturations. Because we use the upscaled reaction model as described
in section 7.2.2, we also plot the remaining fuel amount to show the area that has been
burned. We simulate this problem for 21 days, with an approximate front speed of
1.5ft/day. We observe unstable displacement, as shown in Fig. 8.10. Heat conduction
is incapable of dissipating the energy of these large wavelength perturbations (in the
order of several meters).
8.4 Concluding Remarks
In this chapter, we study the front stability behavior during the ISC process. We
first performed detailed analysis of a 1D ISC process with high initial oil saturation
mobile heavy oil. We identified three main fronts in the process: combustion zone,
leading edge of the steam plateau and leading edge of the oil bank. Among them,
we find the leading edge of the steam plateau as the most critical front because of
the large pressure gradient contrast. This is the place where instability first occurs.
Numerical test cases are performed for small lab-scale forward combustion problems.
We observe heat conduction plays an important role in stabilizing the front, because
of the small length scale of perturbations. Coke plugging (pore blockage) is found to
have very minor effects in stabilizing the front, unless permeability reductions of up to
one order of magnitude are introduced. We have not observed significant ”front self-
correction”, unless with such large permeability reductions [54]. We also test front
stability in larger field-scale cases. The upscaled reaction model is used for these
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 137
Figure 8.10: Unstable displacement in 2D larger scale ISC process. Heat conductionis incapable of dissipating the energy of large wavelength perturbations.
CHAPTER 8. FRONT STABILITY STUDY FOR IN-SITU COMBUSTION 138
studies. We find unstable fronts in these simulations because the heat conduction is
incapable of dissipating large wavelength perturbations.
Chapter 9
Future Directions
9.1 Thermal Streamline Simulation
9.1.1 Non-advective Forces
The original idea of solving transport along the streamline direction makes streamline
simulation an ideal solution for strong advection dominated flow problems [7]. In ther-
mal EOR processes, often we have non-advective forces, such as heat conduction, heat
losses to over and under burden, gravity segregation, etc. The incorporation of these
non-advective effects, while at the same time retaining accuracy and improving simu-
lation speed is crucial to the success of thermal streamline simulation. In this thesis,
we show the usage of Glowinski θ-scheme operator splitting for solving non-advective
effects. We show the accuracy achieved and improved computational efficiency. The
current approach of θ-scheme operator splitting may not be the unique solution.
Other techniques such as putting some heat conduction along the streamlines may
also be applicable. We propose further study related to the efficient incorporation of
non-advective effects into the thermal streamline simulation framework.
9.1.2 Guidelines for Commercial Code Development
In this thesis, we have solved some of the numerical problems for applying the stream-
line method to thermal flood problems. Even though most of the test cases we show
139
CHAPTER 9. FUTURE DIRECTIONS 140
are in 2D space (horizontal and vertical), the same methodology is applicable to full
3D streamline simulator. Difficulty in 3D implementation, however, still exists.
It is recommended the 3D thermal streamline simulator to be developed on the
basis of an existing thermal reservoir simulator. In the 1D transport part of the
streamline global time step, we solve a series of full 1D problems. We can directly
utilize the existing thermal simulator to provide the 1D solution [29]. Multiple stream-
line solves are combined into one 1D problem. This is very similar to the work by
Crane et al., [29].
The implementation of complex well models is a potential challenge for thermal
streamline simulators. In streamline 1D transport, we usually specify a certain bound-
ary condition according to the physical properties in the well grid block, or certain
upstream location [71]. In thermal problems, we have complex multi-tubing string
wells with complicated flow and heat exchanges. Thermal problems such as steam
flooding also often have back flow problems. This means that the total phase velocity
is pointing backwards into the well grid block. We need further study for incorporat-
ing the detailed flows inside the well block into the streamline simulation framework.
Complex well models may be actually difficult to implement.
Specific issues such as mappings occurred inside the streamline global time step,
control of mass balance error, streamline tracing, selection of global time step sizes,
etc., are the common numerical issues for general streamline simulation. They have
been studied extensively in previous works by various parties [71, 72, 92, 58], which
are out of the scope of this work. Similar techniques are applicable here to thermal
streamline simulation.
9.1.3 Thermal Streamline Simulation for SAGD?
We have shown in this work the use of thermal streamline simulation for hot water
flood and steam flood problems. One interesting question here is can we also use
streamline simulation for SAGD simulation. The SAGD process is different than
conventional steam flooding in that we create a slow drainage process that fully takes
advantage of the gravity segregation [16]. Therefore, gravity may serve as the main
CHAPTER 9. FUTURE DIRECTIONS 141
Figure 9.1: Total phase velocity vectors in 2D SAGD process. Convection cell/loopcurrent occur when tracing streamlines.
driving force in SAGD process, instead of viscous force.
We show one typical SAGD simulation example for bitumen reservoir in Fig. 9.1.
The simulation input file is attached in Appendix E. We show the temperature profile
and also plot the total phase velocity vectors in this 2D SAGD problem. As we can
see, the steam rises upwards inside the steam chamber. Once it reaches the cap rock
ceiling, it starts moving sideways. At the edge of the steam chamber, the steam
condenses into hot water. The hot water and oil flows downwards by the gravity
force. If we trace the total phase velocity in this SAGD problem, we will get flow
patterns similar as the convection cells or loop currents. We can have streamlines
start from one cell and eventually trace all the way back to the same cell. This will
require special numerical treatment for the boundary conditions in these streamline
loops, which is also noted in [7]. The numerical challenges described here remain to
be solved for applying streamline simulation to SAGD process.
CHAPTER 9. FUTURE DIRECTIONS 142
9.2 In-situ Combustion
9.2.1 History Matching of Field-scale ISC process
History match of field-scale ISC process has been tried previously by different parties
for fields such as Suplacu [91] and West Heidelberg [63]. Most of these simulation
trials do not take into account the grid size effects for reaction kinetics modeling,
and more or less implement some heuristics to deal with the numerical artifact of
excessive coke formation. The simulation usually only involve a small number of
wells with relatively large grid blocks. A qualitative match has been achieved in
these early reservoir simulation studies.
The successful history match of a ISC field or part of the field is the subject of our
future work. Detailed laboratory measurements such as ramped temperature oxida-
tion kinetic cell and 1D combustion tube should be performed first to understand the
burning characteristics of the crude oil. After history matching the experiment, we
will upscale the kinetics based model to field-scale simulation. When history match-
ing the field production data, a first step Ramey and Gates analytical approach [44]
will be conducted to get the first order estimation. The Gates and Ramey method
and its variants provide the correlation of injected air volume, burned reservoir vol-
ume and the corresponding field production. They serve very similar roles as the
classical material balance approach used for modern reservoir engineering. A detailed
characterization of the reservoir will be followed. Eventually, we will try to match the
geologic model and tuning other fluid and rock properties to achieve the final match.
The proposed entire work flow from lab experiment to field-scale simulation has been
discussed by Bazargan et al., [10].
Through this history matching process, we have uncertainties associated within
each stage. In the lab-scale kinetics based simulation, we have uncertainty in the
reaction model and its parameters. The pseudo reactions are based on simplifica-
tion/lumping of the thousands of real chemical reactions. The parameters for each
pseudo reaction is based on matching the limited information in the RTO experiment
(mainly gas production data). When we scale up to non-kinetic upscaled reaction
CHAPTER 9. FUTURE DIRECTIONS 143
model, we have uncertainty in the sub-grid scale heterogeneity. As we discussed pre-
viously, this affects the fuel amount and oxygen utilization efficiency. During the
upscaling from geo-cellular grid to simulation, we also have errors in flow transport
upscaling. Eventually, when simulating full field problem, we also have uncertainty
in the geologic realizations of the field. As we can see, uncertainty is carried forward
within each stage. Our ultimate goal is to provide eventually ranking for general
reservoir engineering purposes.
9.2.2 Calibrating the Operational Range for Sustaining ISC
Combustion
In field ISC operations,sufficient air must be supplied to maintain the propagation of
combustion front in the desired HTO mode. Otherwise unfavorable LTO reactions
consume oxygen and immobilize the oil. To improve our current upscaling approach
that assumes HTO combustion mode, we propose to use a minimal air flux approach
to define the operational window for sustaining ISC combustion. We check the air
flux behind the combustion front if it is enough to sustain the HTO mode. This min-
imal air flux can be calibrated through three different ways: engineering knowledge,
physical experiment, and numerical simulation. Various parties have proposed empir-
ically the minimal air flux or minimal combustion front velocity based on engineering
field knowledge [85]. For example, a minimal front front velocity of 0.125ft/day is
proposed for successful operation of ISC project [83]. Recently, a novel conical com-
bustion tube design was built as it enables continuous air flux reductions without
having to adjust the air injection rate in a combustion tube experiment [4]. The
conical design is expected to define experimentally the minimal air flux. Numerical
simulation has also been used to define operational window for ISC. For example, the
ignition/extinction regimes are defined as a function of both air injection rate and
magnitude of heat loss in the kinetic cell experiment [61]. A similar technique can
be used here to calibrate the ignition/extinction behavior of 1D ISC process once we
achieve a history match using the Arrhenius kinetics based simulation.
CHAPTER 9. FUTURE DIRECTIONS 144
9.2.3 Grid Size Effects for Other Thermal or Non-Thermal
EOR Processes
The choice of simulation grid block sizes has always been an issue for reservoir sim-
ulation accuracy, especially in modeling EOR processes. This is due to the fact that
many EOR processes are often associated with large property changes within a small
distance. Capturing these property changes and their flow response usually requires
certain sized grid blocks. Most of the commercial simulators use first-order numerical
schemes, that calculate both the flux terms and source/sink terms in the governing
equations using cell averaged properties. This is the main cause for grid-size effects
in thermal or non-thermal EOR processes.
Compositional gas injection simulation is well known for its deteriorated accu-
racy when using large size grid blocks due to numerical dispersion effects [71]. Thus,
higher order numerical schemes such as TVD (Total Variation Diminishing) and ENO
(Essentially Non-Oscillatory) have been successfully applied for the accurate capture
of multiple fronts/contacts in multi-contact miscible flooding [71]. When simulat-
ing multi-dimensional first-contact miscible flooding, pseudo relative permeability
approach is also utilized to capture the sub-grid scale viscous fingering [51].
In this work, we propose the upscaling approach for capturing the sub-grid scale
chemical reaction kinetics in ISC process. For other thermal or non-thermal heavy
oil EOR processes, grid size effects have also been noticed, for example SAGD [22],
ESSAGD (Expansion Solvent Steam Assisted Gravity Drainage) [82] and VAPEX
(Vapor Extraction) processes [31, 95]. We test a simple SAGD process in a bitumen
reservoir with different grid block sizes, shown in Fig. 9.2 and Fig. 9.3 for both early
time and late time. The simulation input file is attached in Appendix F. We have
homogeneous reservoir and show half the plane of the steam chamber temperature
profile. We observe that we need 0.5m×0.5m size grid blocks to achieve the convergent
solution in this test case. This is a very demanding requirement for large 3D field-
scale simulation, especially when simulating multiple pads with many well pairs. This
phenomenon is due to the fact that the viscosity of bitumen is a strong function of
the local temperature (up to 106cp at reservoir temperature, and down to 10cp at
CHAPTER 9. FUTURE DIRECTIONS 145
Figure 9.2: Grid size effects in 2D SAGD simulation (early time 360 days). Thesimulation achieves convergent solution when using grid size 0.5m × 0.5m in thiscase.
the steam chamber temperature). In classical Butler’s analytical approach, the oil
drainage rate is calculated through an exact integration from the interface of the steam
chamber into the cold bitumen zone [16]. However, in numerical simulation, the total
oil drainage rate is calculated by the oil fluxes inside each discretized grid block. The
temperature is represented as a step-wise function, instead of continual variation. The
flux terms are calculated using the cell averaged temperature and its corresponding
viscosity value. This is the main cause for the grid size problems in SAGD. We propose
further work to explore the improvements for SAGD simulation accuracy. Techniques
such as higher order numerical schemes (similar as in compositional gas injection),
or upscaling the drainage flux (finding the equivalent drainage rate, instead of using
cell averaged mobility terms) may be applied in this area.
9.2.4 How to Use In-situ Combustion as a Gravity Drainage
Process in Fractured Carbonate Reservoir?
Significant volumes of heavy oil resources reside in the fractured carbonate reservoirs
worldwide. For instance, large amount of bitumen resource exists in the ”carbonate
triangle” area in Alberta, Canada [16]. Some of these reservoirs may be good candi-
dates for thermal EOR. For example, currently a major steam injection and gravity
drainage project is underway in the Qarn Alam field in Oman [90]. The purpose of
CHAPTER 9. FUTURE DIRECTIONS 146
Figure 9.3: Grid size effects in 2D SAGD simulation (late time 1100 days). Thesimulation achieves convergent solution when using grid size 0.5m × 0.5m in thiscase.
steam injection in this project is to promote the thermally assisted gas-oil gravity
drainage mechanism.
A considerable percentage of Mexican oil production is from naturally fractured
heavy oil carbonate reservoirs, such as the Cantarell field and the Ku-Maloob-Zaap
field in the Gulf of Mexico [2]. The field has been under extensive nitrogen injection
to promote the gas-oil gravity drainage in these thick reservoirs (more than 1000m
thick pay zone for Cantarell). Chemical EOR methods such as surfactant injection
are currently being evaluated as a possible option for Cantarell field [2]. The major
disadvantage of chemical EOR is its large cost of the injectant and the transportation
logistics in the offshore environment. Thermal EOR (steam injection or ISC) may be
a viable option for the Cantarell field and the Ku-Maloob-Zaap field. The reservoirs,
however are often deemed too deep for steam injection because of the high pressure
and heat losses through the injector well bore.
Is it possible to use in-situ combustion as a thermal EOR method for fractured
carbonate heavy oil reservoirs? Previous work by Fadaei et al., [39] has shown through
numerical simulation that oxygen diffusion plays an important role in the successful
propagation of the combustion front in fractured media. The measurement of oxy-
gen diffusion in carbonate rocks could be performed by the experiment described by
Stokka et al., [102]. Physical experiments have also been conducted to test the com-
bustion front propagation in fractured carbonate core [97]. The combustion front is
CHAPTER 9. FUTURE DIRECTIONS 147
successfully sustained in this lab experiment. The authors conclude that it is possible
to perform ISC in the fractured carbonate.
Furthermore, gravity drainage is a promising recovery mechanism for fractured
carbonate reservoirs. Classical in-situ combustion processes, however, are often deemed
as a forward drive process, instead of a drainage process [85]. Steam injection, on
the other hand, has been applied both as a flood process (steam flood) and gravity
drainage process (SAGD) [16]. One interesting problem is whether or not can we also
use in-situ combustion as a gravity drainage process. As discussed before in Chapter
6, the fundamental difference between steam injection and ISC is that in steam in-
jection steam carries heat into the reservoir and condenses into hot water, while ISC
continuously generates heat and non-condensable flue gas through the combustion
reaction. One key question to address is where the generated flue gas will be vented.
In classical forward ISC process with vertical injectors and producers, the flue gas
passes through the oil bank and exists at the producer. In a gravity drainage process,
we wish to keep the fluid close to hydrostatic condition so that the gas and liquid
segregates and maximizes the drainage production.
In the early 1990s, a task force was assembled to test the idea of Combustion
Overhead Split Horizontal (COSH) [69]. In this process, the concept of offset venting
wells is introduced. Instead of moving concurrently with the oil towards the producer,
the flue gas moves horizontally towards the offset venting wells, while the oil drains
downward into the horizontal producer placed at the bottom of the reservoir. The
process, however, never got advanced to a field pilot. COSH does bring new insight
into the traditional thinking of ISC as a drive process. The concept of vent wells
may also serve as a potential solution to using ISC as a drainage process for fractured
carbonate reservoirs.
Inspired by the concept of venting wells in COSH process [69], we also propose
the use of ISC to generate thermal gravity drainage in thick fractured carbonate
reservoirs. The schematic of the process is shown in Fig. 9.4. In this process,
we have three sets of wells, air injectors, chimney (vent) wells and producer wells.
The air injectors and vent wells are perforated in the upper part of the reservoir at
the similar depth level, while the producers are placed at the bottom to efficiently
CHAPTER 9. FUTURE DIRECTIONS 148
Figure 9.4: Concept of ISC assisted gas oil gravity drainage in fractured media. Ventwell is added to achieve hydrostatic gravity drainage condition in the reservoir.
drain the reservoir. We continuously inject air into the reservoir to propagate the
combustion front from the injector towards the venting well. The oxygen diffuses into
the matrix to react and combust with the crude. This generates heat, steam, hot
water and flue gas. Once warmed up, the carbonate matrix wettability is changed
to assist the oil to flow into the fracture. By the density difference, the flue gas
flows horizontally towards the vent well through the fracture network. The hot water
and oil flow downwards by gravity drainage. This process is analogous to the steam
injection based thermal assisted gas-oil gravity drainage process [90]. We perform
some preliminary numerical simulation, combusting a single block of fractured media.
The simulation input is the same as the three reaction model in Appendix B, with
the geometry and permeability field changed accordingly. The diffusivity of O2 in the
gas phase is set as 1m2/day. We show the results in Fig. 9.5. The the combustion
reaction is able to extend into the matrix of the fractured media. By the density
difference, most of the gas is produced on the upper right corner, while most of the
liquid is produced at the lower right corner.
CHAPTER 9. FUTURE DIRECTIONS 149
Figure 9.5: Gravity stable ISC process for a single block of fractured media. Becauseof the gas diffusion, we are able to combust into the matrix. Most of the flue gas isproduced from upper right corner.
Chapter 10
Conclusions
This dissertation addressed the problem of efficient simulation of thermal enhanced
oil recovery processes and proposed two major solutions: the usage of streamline
simulation for thermal recovery problems and chemical reaction kinetics upscaling for
in-situ combustion processes.
For the thermal streamline simulation, we first introduce the extension of stream-
line simulation to hot water flood. We tested our streamline simulator and compared
the results with a commercial thermal simulator. The main conclusions are summa-
rized next.
• We show our streamline simulator is capable of handling a thermal hot water
flood problem. We successfully included the effects of temperature dependent
viscosity changes, compressibility, gravity and heat conduction.
• Thermal streamline simulation is able to handle temperature induced viscosity
changes. For the small compressibility effects encountered in hot water flood,
the dimensionless velocity approach is able to capture the volume changes dur-
ing the 1D transport. We also solve heat conduction and gravity segregation on
separate steps using operator splitting. Thermal streamline simulation achieves
comparable accuracy compared to finite volume simulation, with reduced com-
putational cost.
Steam flooding poses a larger challenge for the use of the streamline method,
150
CHAPTER 10. CONCLUSIONS 151
because of the large volume changes, coupled mass and energy transport and gravity
effects. We implemented several new techniques to tackle these issues. A research
streamline simulator is developed for the steam flooding problem. We tested our
streamline simulator on a series of test problems and again compared the results with
a commercial thermal simulator. Specific conclusions are summarized next.
• The 1D pressure update approach is a major improvement to streamline simula-
tion. It makes streamline simulation capable of handling large volume changes
due to condensation/vaporization during the 1D transport. By implementing
the Glowinski θ-scheme operator splitting, we are also able to handle successfully
the coupling between mass and energy transport when solving heat conduction
and gravity in the operator splitting step. For problems which are dominated
by non-advection forces such as the soaking period in cyclic steam stimulation,
we also successfully implement the preliminary SL/FV hybrid approach.
• For the cases investigated, we are able to retain solution accuracy compared to
finite volume simulation.
• We analyze the computational complexity of both streamline simulation and
finite volume simulation, and show the reduced computational cost. For the
studied case, we are able to decouple the large solving system (7200 grid blocks)
into a series of smaller problems (410 streamlines, each with 68 grid blocks on
average) to achieve speedup.
• We also show the potential applications of thermal streamline simulation as
a fast tool for optimization studies, history matching, flow visualization and
achieving the well connectivity information.
Traditional simulation technique for the in-situ combustion process is limited by
the requirement of using extremely small grid blocks to capture the centimeters-
sized combustion front. We analyze the cause of grid size effects and propose a new
method based on a non-Arrhenius reaction upscaling approach. We try to capture
the equivalent effects of the combustion reactions. The upscaled reaction model is
CHAPTER 10. CONCLUSIONS 152
calibrated using fine-scale numerical simulation. We show the detailed procedures to
implement this methodology in a thermal reservoir simulator. The conclusions are
summarized next.
• Field scale ISC process usually has a large Damkohler number, which means
the chemical reaction kinetics is much faster than flow advection. An upscaled
equilibrium type reaction model is more appropriate for field scale ISC modeling.
• Various test cases have illustrated the consistency achieved in the upscaled
reaction model when scaling up the grid sizes in multidimensional ISC problems.
The reaction upscaling is not sensitive to grid block sizes. We do not need to
change reaction parameters when changing grid block sizes.
• We show this method significantly improves the computational efficiency of ISC
simulation by the capability of using larger grid blocks.
• The limitation and valid range of the reaction upscaling approach is summa-
rized. A minimal air flux criterion is suggested to model the extinction/transition
to LTO behavior.
• We further study the sub-grid scale heterogeneity effects on ISC reaction up-
scaling. We implement material balance approach to upscale the final reaction
effects to fuel amount and oxygen utilization efficiency. We find that sub-grid
scale heterogeneity induces changes in the fuel amount and oxygen utilization
efficiency. When history matching an ISC field, the sub-grid scale heterogeneity
needs to be taken into account.
Driving front stability in ISC process has been studied in the final part of this
thesis. We identified the different flow regimes in the 1D ISC process. Then we
analyzed the different contributors for ISC front stability based on analytical analysis.
Numerical test cases are presented to show the relative importance of these factors.
Specific conclusions drawn are summarized next.
CHAPTER 10. CONCLUSIONS 153
• The ISC process involves multiple fronts. We show that the most critical front
for stability is the leading edge of the thermal front driving the cold oil bank.
This is the front that has the highest mobility/pressure gradient contrast.
• We found coke plugging has a relatively small effect on front stability. To achieve
stabilization, we need a very large permeability reduction (two orders of magni-
tude in the case studied). Heat conduction plays an important role in stabilizing
the front with small wavelength perturbations. With large scale perturbations,
however, unstable front and pre-mature break through of combustion front may
occur. We have not observed significant ”front self-correction” in our numerical
test cases.
Appendix A
Simulation Inputs
The simulation input parameters for all the test cases are listed here.
Table A.1: Reservoir properties for hot water flood
Reservoir dimension [m2] 500× 500Thickness [m] 1φ 0.4Rock thermal capacity [kJ/m3 ·K] 2300Rock thermal conductivity [kJ/(m · day ·K)] 302
154
APPENDIX A. SIMULATION INPUTS 155
Table A.2: Viscosity relationships for hot water flood
Viscosity Coefficient (µj = Aj exp(Bj/T )) Aj [cp] Bj [K]Water 0.5 0.0Oil 0.6541× 10−5 3969
Table A.3: Fluid parameters for incompressible hot water flooding
Injection well pressure Pinj [kPa] 24000Production well pressure Pprod [kPa] 14000Injection well water temperture Tinj [oC] 80Initial oil temperature Tinit [oC] 20Water heat conductivity [kJ/(m · day ·K)] 51.8Oil heat conductivity [kJ/(m · day ·K)] 51.8Water density [kg/m3] 1004.26Oil density [kg/m3] 981.76Water heat capacity [kJ/kg] 4.19Oil heat capacity [kJ/kg] 2.02
Table A.4: Coefficients of density calculations in compressible hot water flood
Coefficients of density calculations ρscj [kg/m3] cj [1/kPa] aj [1/K]
Water 998 1.0× 10−7 1.0× 10−4
Oil 972 1.0× 10−7 1.0× 10−4
Table A.5: Reservoir properties for steam flood
Reservoir Thickness [m] 30 Porosity φ at standard condition 0.35Rock compressibility [1/kPa] 1e-5 Rock thermal expansion [1/K] 1e-4Rock heat capacity [kJ/m3 ·K] 2000 Rock conductivity [kJ/(m · day ·K)] 200
APPENDIX A. SIMULATION INPUTS 156
Table A.6: Well control for steam flood
Injection well pressure Pinj [kPa] 6000 Production well liquid rate BHL [m3] 54Injection well temperature Tinj [oC] 276 Initial reservoir temperature Tinit [oC] 50Steam Quality fs 0.8 Initial reservoir temperature Tinit [oC] 50
Table A.7: Fluid parameters for steam flood
Water density at standard conditions [kg/m3] 998Oil density at standard conditions [kg/m3] 972Water and oil compressibility [1/kPa] 1.0e-7Water and oil thermal expansion [1/K] 1.0e-4Water vaporization enthalpy HV R [kJ/mol] 4820Water critical temperature Tc [K] 647.3Oil vaporization enthalpy HV R [kJ/mol] 8569Oil critical temperature Tc [K] 767Water heat capacity CPG1 [J/mol ·K] 32.2Water heat capacity CPG2 [J/mol ·K2] 1.92E-3Water heat capacity CPG3 [J/mol ·K3] 1.06E-5Water heat capacity CPG4 [J/mol ·K4] -3.60E-9Oil heat capacity CPG1 [J/mol ·K] -22.4Oil heat capacity CPG2 [J/mol ·K2] 1.94Oil heat capacity CPG3 [J/mol ·K3] -1.12E-3Oil heat capacity CPG4 [J/mol ·K4] 2.53E-7Oil molecule weight [g/mol] 282
Table A.8: Reservoir properties for cyclic steam stimulation
Reservoir Thickness [m] 30 Porosity φ at standard condition 0.35Rock compressibility [1/kPa] 5e-5 Rock thermal expansion [1/K] 1e-4Rock heat capacity [kJ/m3 ·K] 3000 Rock conductivity [kJ/(m · day ·K)] 300
APPENDIX A. SIMULATION INPUTS 157
Table A.9: Fluid properties in three reaction ISC model.
Components Oil CO CO2 N2 O2 Coke1 Coke2Molecular Weight 537.6 28 44 28 32 673.04 179.8Tcrit (oC) 1472 -140.3 31.1 -147.0 -118.6Pcrit (kPa) 1168 3496 7376 3394 5046CPG1 (J/gmol ·K) 1138.8 30.0 16.9 30.3 27.63CPG2 (J/gmol ·K) 0 -1.54e-3 0.1063 -2.572e-3 6.437e-3ρSTD
o (kg/cm3) 1.008e-3
Table A.10: Rock properties in ISC simulation.
Rock Heat Capacity (J/cm3 ·K) 1.206 + 0.00236 · T (T is temperature)Rock Heat Conductivity (J/cm ·min ·K) 2.5833Porosity 0.36Permeability (mD) 10
Table A.11: Kinetics parameters for three reaction ISC model.
Reactions Pyrolysis LTO HTOActivation Energy (kJ/gmol) 22.6 67.5 87.6Pre-exponential Factor (1/(kPa ·min)) 0.01 250 220Reaction Enthalpy (kJ/gmol) 1.60× 103 1.28× 104 4.85× 103
Table A.12: Reaction stoichiometry for three reaction ISC model.
Pyrolysis 1.0Oil + 4.24O2 → 1.0Coke1LTO 1.0Coke1 + 33.8O2 → 1.0Coke2 + 4.75CO + 23.77CO2 + 22.0H2OHTO 1.0Coke2 + 12.29O2 → 1.73CO + 8.56CO2 + 8.0H2O
Appendix B
STARS Input File for Three
Reaction Model
**----------------------INPUT-OUTPUT CONTROL----------------------------
*TITLE1 ’Simulation of Dry Combustion Tube Test’
*INUNIT *LAB
*OUTUNIT *LAB
*WPRN *GRID *TIME
*WPRN *ITER *TIME
*OUTPRN *GRID *PRES *SW *SO *SG *TEMP *Y *X *MASS *SOLCONC
*OUTPRN *WELL *WELLCOMP
*OUTPRN *ITER *BRIEF
*OUTSRF *WELL *MASS *COMPONENT *ALL
*OUTSRF *GRID *PRES *SW *SO *SG *TEMP *Y *X *MASS *SOLCONC *VISO *VISW
*INTERRUPT *STOP
**-------------------------RESERVOIR DESCRIPTION------------------------
*GRID *CART 1 302 1
*KDIR *DOWN
*DI *CON 6.2
*DJ *CON 0.36
158
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 159
*DK *CON 6.2
*POR *CON 0.36
*PERMI *CON 10000
*PERMJ *CON 10000
*PERMK *CON 10000
*END-GRID
**------------------------OTHER RESERVOIR PROPERTIES--------------------
*CPOR 0
*CTPOR 0
*ROCKCP 1.20572 0.00236
*THCONR 2.5833
*THCONO 0.0
*THCONG 0.0
*THCONS 0.0
**--------------------------COMPONENT PROPERTIES------------------------
*MODEL 8 6 2
*COMPNAME ’WATER’ ’OIL’ ’CO’ ’CO2’ ’N2’ ’ O2’ ’COKE1’ ’COKE2’
*CMM 0 0.5376 0.028 0.044 0.028 0.032 0.67304 0.17976
*TCRIT 0 1472 -140.25 31.05 -146.95 -118.55
*PCRIT 0 1168 3496 7376 3394 5046
*KV1 0 0
*KV2 0 0
*KV3 0 0
*KV4 0 0
*KV5 0 0
*PRSR 101
*TEMR 15
*PSURF 101
*TSURF 15
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 160
*SURFLASH W O G G G G
** ’WATER’ ’OIL’ ’CO’ ’CO2’ ’N2’ ’ O2’ ’COKE1’ ’COKE2’
*CPG1 0 1138.8 29.987 16.864 30.288 27.627
*CPG2 0 0 -1.542e-3 0.1063 -2.572e-3 6.437e-3
*CPG3 0 0 0 0 0 0
*CPG4 0 0 0 0 0 0
*HVR 0 0
*EV 0 0
*MASSDEN 0 1.008E-3
*CP 0 0
*CT1 0 0
*CT2 0 0
*SOLID_DEN ’COKE1’ 0.0326265 0 0
*SOLID_CP ’COKE1’ 17.0 0
*SOLID_DEN ’COKE2’ 0.0088 0 0
*SOLID_CP ’COKE2’ 17.0 0
*AVISC 0 1E-4
*BVISC 0 5087
** ’WATER’ ’OIL’ ’CO’ ’CO2’ ’N2’ ’ O2’ ’COKE1’ ’COKE2’
*STOREAC 0.0 1.0 0.0 0.0 0.0 4.24 0.0 0.0
*STOPROD 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0
*FREQFAC 0.1 **1/kPa-MIN
*RENTH 1.597E6 ** J/gmol
*RPHASE 0 2 0 0 0 3 0 0
*RORDER 0 1 0 0 0 1 0 0
*EACT 45560 ** J/gmol
*O2PP ’O2’
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 161
** ’WATER’ ’OIL’ ’CO’ ’CO2’ ’N2’ ’ O2’ ’COKE1’ ’COKE2’
*STOREAC 0.0 0.0 0.0 0.0 0.0 33.8 1.0 0.0
*STOPROD 22.0 0.0 4.75 23.77 0.0 0.0 0.0 1.0
*FREQFAC 250 **1/kPa-MIN
*RENTH 1.28E7 ** J/gmol
*RPHASE 0 0 0 0 0 3 4 0
*RORDER 0 0 0 0 0 1 1 0
*EACT 67525 ** J/gmol
*O2PP ’O2’
** ’WATER’ ’OIL’ ’CO’ ’CO2’ ’N2’ ’ O2’ ’COKE1’ ’COKE2’
*STOREAC 0.0 0.0 0.0 0.0 0.0 12.29 0.0 1.0
*STOPROD 8.0 0.0 1.73 8.56 0.0 0.0 0.0 0.0
*FREQFAC 220 **1/kPa-MIN
*RENTH 4.85E6 ** J/gmol
*RPHASE 0 0 0 0 0 3 0 4
*RORDER 0 0 0 0 0 1 0 1
*EACT 87575 ** J/gmol
*O2PP ’O2’
**----------------------------------ROCK~FLUID DATA---------------------
*ROCKFLUID
*RPT 1 *LININTERP
** Sw Krw Krow Pcow
*SWT
0.068 0.0000 1.0000 0
0.15 0.0077 0.8229 0
0.2 0.0200 0.7232 0
0.218 0.0259 0.6889 0
0.25 0.0381 0.6300 0
0.3 0.0619 0.5432 0
0.35 0.0915 0.4629 0
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 162
0.4 0.1268 0.3889 0
0.45 0.1679 0.3214 0
0.5 0.2148 0.2604 0
0.55 0.2674 0.2057 0
0.6 0.3258 0.1575 0
0.65 0.3899 0.1157 0
0.7 0.4598 0.0804 0
0.75 0.5354 0.0514 0
0.8 0.6168 0.0289 0
0.85 0.7040 0.0129 0
0.9 0.7969 0.0032 0
0.932 0.8956 0.0000 0
1 1 0.0000 0
*SLT
** Sl Krg Krog Pcgo
0.068 0.8975 0 0
0.1368 0.6845 1E-3 0
0.2 0.6233 0.0077 0
0.25 0.5429 0.0200 0
0.268 0.5154 0.0259 0
0.3 0.4681 0.0381 0
0.35 0.3989 0.0619 0
0.4 0.3352 0.0915 0
0.45 0.277 0.1268 0
0.5 0.2244 0.1679 0
0.55 0.1773 0.2148 0
0.6 0.1357 0.2674 0
0.65 0.0997 0.3258 0
0.7 0.0693 0.3899 0
0.75 0.0443 0.4598 0
0.8 0.0249 0.5354 0
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 163
0.85 0.0111 0.6168 0
0.9 0.0028 0.7040 0
0.932 1E-3 0.7969 0
0.98 0 0.8956 0
1 0 1 0
**----------------------------------INITIAL CONDIRION--------------------
*INITIAL
*PRES *CON 780
*TEMP *CON 15
*SW *CON 0.068
*SO *CON 0.66
*MFRAC_WAT ’WATER’ *CON 1.00
*MFRAC_OIL ’OIL’ *CON 1.00
*MFRAC_GAS ’N2’ *CON 1.00
*MFRAC_GAS ’CO2’ *CON 0.00
*MFRAC_GAS ’CO’ *CON 0.00
*MFRAC_GAS ’O2’ *CON 0.00
**-----------------------------------NUMERICAL CONTROL---------------------
*NUMERICAL
*MAXSTEPS 100000
**DTMAX 0.1
*DTMIN 0.000001
**-----------------------------------RECURRENT DATA-------------------------
*RUN
*TIME 0
*DTWELL 0.01
*WELL ’INJECTOR’
*WELL ’PRODUCER’
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 164
**--------------NITROGENE INJECTION ---
*PRODUCER ’PRODUCER’
*OPERATE *BHP 780
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’PRODUCER’
1 302 1 1.00
*INJECTOR ’INJECTOR’
*TINJW 20
*PINJW 780
*INCOMP *GAS 0 0 0 0 1 0
*OPERATE *STG 3000
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’INJECTOR’
1 1 1 1.00
**----------------HEATERS ON!----------------
*TIME 18
*HEATR *IJK 1 1:30 1 320 **
*HEATR *IJK 1 1:25 1 260 **
*TIME 78
*INJECTOR ’INJECTOR’
*TINJW 20
*PINJW 780
*INCOMP *GAS 0 0 0 0 0.79 0.21
*OPERATE *STG 3000
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’INJECTOR’
1 1 1 1.00
**----------------HEATERS OFF!----------------
*HEATR *IJK 1 1:30 1 0
*HEATR *IJK 1 1:25 1 0
*TIME 138
APPENDIX B. STARS INPUT FILE FOR THREE REACTION MODEL 165
*TIME 198
*TIME 258
*TIME 318
*TIME 378
*TIME 438
*STOP
Appendix C
Eclipse Input File for Three
Reaction Model
RUNSPEC
---------------------------------------------------------------
LAB
MESSAGES
8* 100000 /
DIMENS
1 1 602 /
COMPS
7 /
UNIFOUT
WATER
THERMAL
SOLID
HWELLS
WELLDIMS
1* 180 7* 180 /
REACTION
166
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 167
3 /
START
22 JAN 2010 /
HEATDIMS
30 /
ROCKDIMS
1 52 2 /
ACTDIMS
2 50 132 3 /
GRID
---------------------------------------------------------------
-- GRID GEOMETRY
TOPS
0 /
DX
602*6.2 /
DY
602*6.2/
DZ
602*0.36 /
PORO
602*0.36 /
PERMX
602*10000.0 / mD
PERMY
602*10000.0 / mD
PERMZ
602*10000.0 / mD
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 168
-- Thermal keywords:for over/under burden ROCKPROP
1 15 0. 1. /
/
-- Thermal keyword: THCONR
602*155 /
-- Thermal keyword: HEATCR
602*1.20572 /
HEATCRT
602*0.00236 /
EDIT
---------------------------------------------------------------
PROPS
---------------------------------------------------------------
SWOF
--SW KRW KRO PCOW
0.068 0.0000 1.0000 0
0.15 0.0077 0.8229 0
0.2 0.0200 0.7232 0
0.218 0.0259 0.6889 0
0.25 0.0381 0.6300 0
0.3 0.0619 0.5432 0
0.35 0.0915 0.4629 0
0.4 0.1268 0.3889 0
0.45 0.1679 0.3214 0
0.5 0.2148 0.2604 0
0.55 0.2674 0.2057 0
0.6 0.3258 0.1575 0
0.65 0.3899 0.1157 0
0.7 0.4598 0.0804 0
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 169
0.75 0.5354 0.0514 0
0.8 0.6168 0.0289 0
0.85 0.7040 0.0129 0
0.9 0.7969 0.0032 0
0.932 0.8956 0.0000 0 /
SGOF
--SG KRG KROG PCOG
0 0 0.8956 0
0.068 1E-3 0.7969 0
0.1 0.0028 0.7040 0
0.15 0.0111 0.6168 0
0.2 0.0249 0.5354 0
0.25 0.0443 0.4598 0
0.3 0.0693 0.3899 0
0.35 0.0997 0.3258 0
0.4 0.1357 0.2674 0
0.45 0.1773 0.2148 0
0.5 0.2244 0.1679 0
0.55 0.277 0.1268 0
0.6 0.3352 0.0915 0
0.65 0.3989 0.0619 0
0.7 0.4681 0.0381 0
0.732 0.5154 0.0259 0
0.75 0.5429 0.0200 0
0.8 0.6233 0.0077 0
0.8632 0.6845 1E-3 0
0.932 0.8975 0 0 /
-- saturation endpoints in INIT file, FILLEPS -- Stone’s Rel Perm
Model 2 STONE2
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 170
EOS RK /
-- Component data
------------------------------------------------------
CNAMES
"OIL" "COKE1" "COKE2" "O2" "CO2" "N2" "CO" /
CVTYPE
DEAD SOLID SOLID GAS GAS GAS GAS /
CVTYPES
DEAD SOLID SOLID GAS GAS GAS GAS /
MW
537.6 673.04 179.76 32 44 28 28 /
PCRIT
11.68 1* 1* 50.46 73.76 33.94 34.96 /
TCRIT
1745 1* 1* 154.6 304.2 126.2 132.9 /
DREF
1.008 1* 1* 1* 1* 1* 1* /
PREF
1.0 1* 1* 1* 1* 1* 1* /
TREF
288.15 1* 1* 1* 1* 1* 1* /
CREF
0.0 1* 1* 1* 1* 1* 1* /
THERMEX1
0.0 1* 1* 1* 1* 1* 1* /
SDREF
1* 32.626 8.8 1* 1* 1* 1* /
SPECHA
2.1183 1* 1* 1* 1* 1* 1* /
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 171
SPECHB
0.0 1* 1* 1* 1* 1* 1* /
SPECHG
1* 1* 1* 0.8633 0.3833 1.082 1.0710 /
SPECHH
1* 1* 1* 2.0116e-4 0.0024 -9.186e-5 -5.507e-5 /
SPECHS
1* 0.0253 0.0946 1* 1* 1* 1* /
HEATVAPS
0.0 0.0 0.00 0.00 0.00 0.0 0.0 /
OILVISCT
5 1636262 0.01 0.01 0.01 0.01 0.01 0.01
15 229374 0.01 0.01 0.01 0.01 0.01 0.01
25 54485.9 0.01 0.01 0.01 0.01 0.01 0.01
30 30000 0.01 0.01 0.01 0.01 0.01 0.01
48 10600 0.01 0.01 0.01 0.01 0.01 0.01
65 3000 0.01 0.01 0.01 0.01 0.01 0.01
100 290 0.01 0.01 0.01 0.01 0.01 0.01
150 32 0.01 0.01 0.01 0.01 0.01 0.01
200 7.5 0.01 0.01 0.01 0.01 0.01 0.01
250 2.7 0.01 0.01 0.01 0.01 0.01 0.01
300 1.0 0.01 0.01 0.01 0.01 0.01 0.01
325 0.75 0.01 0.01 0.01 0.01 0.01 0.01
350 0.60 0.01 0.01 0.01 0.01 0.01 0.01
375 0.50 0.01 0.01 0.01 0.01 0.01 0.01
400 0.47 0.01 0.01 0.01 0.01 0.01 0.01
425 0.41 0.01 0.01 0.01 0.01 0.01 0.01
450 0.36 0.01 0.01 0.01 0.01 0.01 0.01
475 0.315 0.01 0.01 0.01 0.01 0.01 0.01
500 0.28 0.01 0.01 0.01 0.01 0.01 0.01
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 172
525 0.25 0.01 0.01 0.01 0.01 0.01 0.01
550 0.22 0.01 0.01 0.01 0.01 0.01 0.01
2000 0.12 0.01 0.01 0.01 0.01 0.01 0.01
/ GASVISCT
0 0.0136 0.0136 0.0136 0.0136 0.0136 0.0136 0.0136
137 0.0188 0.0188 0.0188 0.0188 0.0188 0.0188 0.0188
210 0.0216 0.0216 0.0216 0.0216 0.0216 0.0216 0.0216
615 0.0370 0.0370 0.0370 0.0370 0.0370 0.0370 0.0370
1000 0.0516 0.0516 0.0516 0.0516 0.0516 0.0516 0.0516
2000 0.0896 0.0896 0.0896 0.0896 0.0896 0.0896 0.0896
/
STCOND --Temp Pressure
15 1.0 /
-- Chemical reaction data
----------------------------------------------
STOREAC -- "OIL" "COKE1" "COKE2" "O2" "CO2" "N2"
"CO" "Water"
1.0 0.0 0.0 4.24 0.0 0.0 0.0 0.0 / REACTION 1
0.0 1.0 0.0 33.8 0.0 0.0 0.0 0.0 / REACTION 2
0.0 0.0 1.0 12.29 0.0 0.0 0.0 0.0 / REACTION 3
STOPROD -- "OIL" "COKE1" "COKE2" "O2" "CO2" "N2"
"CO" "Water"
0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 / REACTION 1
0.0 0.0 1.0 0.0 23.77 0.0 4.75 22.0 / REACTION 2
0.0 0.0 0.0 0.0 8.56 0.0 1.73 8.0 / REACTION 3
REACRATE
60 1.5E6 1.32E6 /
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 173
REACACT
22560 67525 87575 /
REACENTH
1.597E6 1.28E7 4.85E6 /
REACPHA --
"OIL" "COKE1" "COKE2" "O2" "CO2" "N2" "CO" "Water"
OIL 1* 1* GPP 1* 1* 1* 1* / REACTION 1
1* SOLID 1* GPP 1* 1* 1* 1* / REACTION 2
1* 1* SOLID GPP 1* 1* 1* 1* / REACTION 3
REACCORD -- "OIL" "COKE1" "COKE2" "O2" "CO2" "N2" "CO""Water"
1 0 0 1 0 0 0 0 / REACTION 1
0 1 0 1 0 0 0 0 / REACTION 2
0 0 1 1 0 0 0 0 / REACTION 3
-------------------------------------------------------------------------------
SOLUTION
-------------------------------------------------------------------------------
TEMPI
602*15 / Initial Temperature C
PRESSURE
602*7.8 /
SOIL
602*0.26 /
SWAT
602*0.24 /
SSOLID
602*0 /
------specify initial oil composition in cells
XMF
602*1.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 174
/
---------specify initial gas composition in cells
YMF
602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*1.0 602*0.0 602*0.0
/
SMF
602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0 602*0.0
/
RPTSOL
SOLVD TEMP PRESSURE SSOLID SOIL SGAS SWAT MLSC TEMP HOIL HSOL
XMF YMF /
RPTRST
PRESSURE SSOLID SOIL SGAS SWAT MLSC TEMP REAC HOIL HSOL XMF
YMF ZMF VOIL VGAS VWAT KRO KRW KRG PCOW PCOG FLOC1 FLOC2 FLOC4 FLOC5
FLOC6 FLOWAT FLOE DENW DENO DENG /
SUMMARY
---------------------------------------------------------------
WBHP
INJE PROD /
WTEMP INJE PROD /
-- Rates
FOPR FWPR FGPR FGIR FREAC /
-- Totals
FOPT FOIP FWPT FWIP FGPT FGIT FREAT /
PERFORMA
RUNSUM
SCHEDULE
---------------------------------------------------------------
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 175
RPTSCHED
FIP=1 /
-- Convergence & time step criteria
------------------------------------ --
TSCRIT
CVCRIT
.005 10 4* .00005 0.05 / Need tight convergence criteria for small cells
TSCRIT
1.0E-8 1.0E-9 /
7* 10.0 / Need small time steps
WELSPECS INJE
1* 1 1 1* GAS /
PROD 1* 1 1 1* OIL /
/
COMPDAT --
WELL I J K1 K2
INJE 1 1 1 1 OPEN 1* 1E5 /
PROD 1 1 602 602 OPEN 1* 1E5 /
/
WCONPROD
PROD OPEN BHP 5* 7.8 /
/
WELLSTRE -- HEAVY LIGHT COKE O2 CO2 N2
AIR 0.0 0.0 0.0 0.21 0.0 0.79 /
NIT 0.0 0.0 0.0 0.00 0.0 1.00 /
/
WCONINJE -- Rate BHP
INJE GAS OPEN RATE 1.8e+5 1* /
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 176
/
WINJGAS
INJE STREAM NIT /
/
WINJTEMP -- T P
INJE 1* 20 7.8 /
/
TSTEP 1*0.3 /
HEATER
HEAT1 1 1 2 17400 1*/
HEAT1 1 1 3 17400 1*/
HEAT1 1 1 4 17400 1*/
HEAT1 1 1 5 17400 1*/
HEAT1 1 1 6 17400 1*/
HEAT1 1 1 7 17400 1*/
HEAT1 1 1 8 17400 1*/
HEAT1 1 1 9 17400 1*/
HEAT1 1 1 10 17400 1*/
HEAT1 1 1 11 17400 1*/
HEAT1 1 1 12 17400 1*/
HEAT1 1 1 13 17400 1*/
HEAT1 1 1 14 17400 1*/
HEAT1 1 1 15 17400 1*/
HEAT1 1 1 16 17400 1*/
HEAT1 1 1 17 17400 1*/
HEAT1 1 1 18 17400 1*/
HEAT1 1 1 19 17400 1*/
HEAT1 1 1 20 17400 1*/
HEAT1 1 1 21 17400 1*/
HEAT1 1 1 22 17400 1*/
HEAT1 1 1 23 17400 1*/
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 177
HEAT1 1 1 24 17400 1*/
HEAT1 1 1 25 17400 1*/
HEAT1 1 1 26 17400 1*/
HEAT1 1 1 27 9600 1*/
HEAT1 1 1 28 9600 1*/
HEAT1 1 1 29 9600 1*/
HEAT1 1 1 30 9600 1*/
HEAT1 1 1 31 9600 1*/
/
TSTEP
1*1 /
WCONINJE -- Rate BHP
INJE GAS OPEN RATE 1.8e+5 1* /
/
WINJGAS
INJE STREAM AIR /
/
WINJTEMP -- T P
INJE 1* 20 7.8 /
/
HEATER
HEAT1 1 1 2 0 1*/
HEAT1 1 1 3 0 1*/
HEAT1 1 1 4 0 1*/
HEAT1 1 1 5 0 1*/
HEAT1 1 1 6 0 1*/
HEAT1 1 1 7 0 1*/
HEAT1 1 1 80 1*/
HEAT1 1 1 9 0 1*/
HEAT1 1 1 10 0 1*/
APPENDIX C. ECLIPSE INPUT FILE FOR THREE REACTION MODEL 178
HEAT1 1 1 11 0 1*/
HEAT1 1 1 12 0 1*/
HEAT1 1 1 13 0 1*/
HEAT1 1 1 14 0 1*/
HEAT1 1 1 15 0 1*/
HEAT1 1 1 16 0 1*/
HEAT1 1 1 17 0 1*/
HEAT1 1 1 18 0 1*/
HEAT1 1 1 19 0 1*/
HEAT1 1 1 20 0 1*/
HEAT1 1 1 21 0 1*/
HEAT1 1 1 22 0 1*/
HEAT1 1 1 23 0 1*/
HEAT1 1 1 24 0 1*/
HEAT1 1 1 25 0 1*/
HEAT1 1 1 26 0 1*/
HEAT1 1 1 27 0 1*/
HEAT1 1 1 28 0 1*/
HEAT1 1 1 29 0 1*/
HEAT1 1 1 30 0 1*/
HEAT1 1 1 31 0 1*/
/
TSTEP
8*1 /
END
Appendix D
STARS Input File for Upscaled
Three Reaction Model
In this input file, the fuel amount in this sample input file is set at Sofuel = 7.9%.
The oxygen utilization efficiency is set at Eu = 0.917.
**----------------------INPUT-OUTPUT CONTROL----------------------------
*TITLE1 ’Simulation of Dry Combustion Tube Test’
*INUNIT *LAB
*OUTUNIT *LAB
*WPRN *GRID *TIME
*WPRN *ITER *TIME
*OUTPRN *GRID *PRES *SW *SO *SG *TEMP *Y *X *MASS *SOLCONC
*OUTPRN *WELL *WELLCOMP
*OUTPRN *ITER *BRIEF
*OUTSRF *WELL *MASS *COMPONENT *ALL
*OUTSRF *GRID *PRES *SW *SO *SG *TEMP *Y *X *MASS *SOLCONC *VISO *VISW
*INTERRUPT *STOP
**-------------------------RESERVOIR DESCRIPTION------------------------
*GRID *CART 1 602 1
*KDIR *DOWN
179
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL180
*DI *CON 6.2
*DJ *CON 0.36
*DK *CON 6.2
*POR *CON 0.36
*PERMI *CON 10000
*PERMJ *CON 10000
*PERMK *CON 10000
*END-GRID
**------------------------OTHER RESERVOIR PROPERTIES--------------------
*CPOR 0
*CTPOR 0
*ROCKCP 1.20572 0.00236
*THCONR 2.5833
*THCONO 0.0
*THCONG 0.0
*THCONS 0.0
**--------------------------COMPONENT PROPERTIES------------------------
*MODEL 10 8 3
*COMPNAME ’WATER’ ’OIL1’ ’OIL2’ ’CO’ ’CO2’ ’N2’ ’O2’ ’O2_B’ ’COKE1’ ’COKE2’
*CMM 0 0.5376 0.5376 0.028 0.044 0.028 0.032 0.032 0.5376 0.17976
*TCRIT 0 1472 1472 -140.3 31.05 -146.95 -118.6 -118.6
*PCRIT 0 1168 1168 3496 7376 3394 5046 5046
*KV1 0 0 0
*KV2 0 0 0
*KV3 0 0 0
*KV4 0 0 0
*KV5 0 0 0
*PRSR 101
*TEMR 15
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL181
*PSURF 101
*TSURF 15
** ’WATER’ ’OIL1’ ’OIL2’ ’CO’ ’CO2’ ’N2’ ’O2’ ’O2_B’ ’COKE1’ ’COKE2’
*CPG1 0 1138.8 1138.8 29.987 16.864 30.288 27.627 27.627
*CPG2 0 0 0 -1.542e-3 0.1063 -2.572e-3 6.437e-3 6.437e-3
*CPG3 0 0 0 0 0 0 0 0
*CPG4 0 0 0 0 0 0 0 0
*HVR 0 0 0
*EV 0 0 0
*MASSDEN 0 1.008E-3 1.008E-3
*CP 0 0 0
*CT1 0 0 0
*CT2 0 0 0
*SOLID_DEN ’COKE1’ 1.008E-3 0 0
*SOLID_CP ’COKE1’ 1138.8 0
*SOLID_DEN ’COKE2’ 1.008E-3 0 0
*SOLID_CP ’COKE2’ 1138.8 0
*VISCTABLE
** Tempe Water Oil1 Oil2
5 0 1636262 1636262
15 0 229374 229374
25 0 54485.9 54485.9
30 0 30000 30000
48 0 10600 10600
65 0 3000 3000
100 0 290 290
150 0 32 32
200 0 7.5 7.5
250 0 2.7 2.7
300 0 1.0 1.0
325 0 0.75 0.75
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL182
350 0 0.60 0.60
375 0 0.50 0.50
400 0 0.47 0.47
425 0 0.41 0.41
450 0 0.36 0.36
475 0 0.315 0.315
500 0 0.28 0.28
525 0 0.25 0.25
550 0 0.22 0.22
2000 0 0.12 0.12
** ’WATER’ ’OIL1’ ’OIL2’ ’CO’ ’CO2’ ’N2’ ’O2’ ’O2_B’ ’COKE1’ ’COKE2’
*STOREAC 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
*STOPROD 0.0 0.0 0.605 0.0 0.0 0.0 0.0 0.0 0.395 0.0
*FREQFAC 1E5 **1/kPa-MIN
*RENTH 0 ** J/gmol
*RPHASE 0 2 0 0 0 0 0 0 0 0
*RORDER 0 1 0 0 0 0 0 0 0 0
*EACT 0 ** J/gmol
*O2PP ’O2’
** ’WATER’ ’OIL1’ ’OIL2’ ’CO’ ’CO2’ ’N2’ ’O2’ ’O2_B’ ’COKE1’ ’COKE2’
*STOREAC 0.0 0.0 0.0 0.0 0.0 0.0 38.04 0.0 1.0 0.0
*STOPROD 22.0 0.0 0.0 4.75 23.77 0.0 0.0 0.0 0.0 1.0
*FREQFAC 1E5 **1/kPa-MIN
*RENTH 1.44E7 ** J/gmol
*RPHASE 0 0 0 0 0 0 3 0 4 0
*RORDER 0 0 0 0 0 0 1 0 1 0
*EACT 0 ** J/gmol
*O2PP ’O2’
** ’WATER’ ’OIL1’ ’OIL2’ ’CO’ ’CO2’ ’N2’ ’O2’ ’O2_B’ ’COKE1’ ’COKE2’
*STOREAC 0.0 0.0 0.0 0.0 0.0 0.0 12.29 0.0 0.0 1.0
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL183
*STOPROD 8.0 0.0 0.0 1.73 8.56 0.0 0.0 0.0 0.0 0.0
*FREQFAC 1E5 **1/kPa-MIN
*RENTH 4.85E6 ** J/gmol
*RPHASE 0 0 0 0 0 0 3 0 0 4
*RORDER 0 0 0 0 0 0 1 0 0 1
*EACT 0 ** J/gmol
*O2PP ’O2’
**----------------------------------ROCK~FLUID DATA---------------------
*ROCKFLUID
*RPT 1 *LININTERP
** Sw Krw Krow Pcow
*SWT
0.074447121 0 1 0
0.16422159 0.0077 0.8229 0
0.21896212 0.02 0.7232 0
0.23866871 0.0259 0.6889 0
0.273702649 0.0381 0.63 0
0.328443179 0.0619 0.5432 0
0.383183709 0.0915 0.4629 0
0.437924239 0.1268 0.3889 0
0.492664769 0.1679 0.3214 0
0.547405299 0.2148 0.2604 0
0.602145829 0.2674 0.2057 0
0.656886359 0.3258 0.1575 0
0.711626889 0.3899 0.1157 0
0.766367418 0.4598 0.0804 0
0.821107948 0.5354 0.0514 0
0.875848478 0.6168 0.0289 0
0.930589008 0.704 0.0129 0
0.985329538 0.7969 0.0032 0
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL184
1 0.8956 0 0
*SLT
** Sl Krg Krog Pcgo
** -- --- ---- ----
0.054959492 0.6845 1.00E-03 0
0.124151522 0.6233 0.0077 0
0.178892052 0.5429 0.02 0
0.198598642 0.5154 0.0259 0
0.233632582 0.4681 0.0381 0
0.288373111 0.3989 0.0619 0
0.343113641 0.3352 0.0915 0
0.397854171 0.277 0.1268 0
0.452594701 0.2244 0.1679 0
0.507335231 0.1773 0.2148 0
0.562075761 0.1357 0.2674 0
0.616816291 0.0997 0.3258 0
0.671556821 0.0693 0.3899 0
0.726297351 0.0443 0.4598 0
0.78103788 0.0249 0.5354 0
0.83577841 0.0111 0.6168 0
0.89051894 0.0028 0.704 0
0.925552879 1.00E-03 0.7969 0
1 0 0.8956 0
**----------------------------------INITIAL CONDITION--------------------
*INITIAL
*PRES *CON 780
*TEMP *CON 15
*SW *CON 0.24
*SO *CON 0.20
*MFRAC_WAT ’WATER’ *CON 1.00
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL185
*MFRAC_OIL ’OIL’ *CON 1.00
*MFRAC_GAS ’N2’ *CON 1.00
*MFRAC_GAS ’CO2’ *CON 0.00
*MFRAC_GAS ’CO’ *CON 0.00
*MFRAC_GAS ’O2’ *CON 0.00
**-----------------------------------NUMERICAL CONTROL---------------------
*NUMERICAL
*MAXSTEPS 100000
**DTMAX 0.1
*DTMIN 0.000001
**-----------------------------------RECURRENT DATA-------------------------
*RUN
*TIME 0
*DTWELL 0.01
*WELL ’MB-INJECTOR’
*WELL ’MB-PRODUCER’
*PRODUCER ’MB-PRODUCER’
*OPERATE *BHP 780
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’MB-PRODUCER’
1 1 602 1.00
*INJECTOR ’MB-INJECTOR’
*TINJW 20
*PINJW 780
*INCOMP *GAS 0 0 0 0 0 1 0 0
*OPERATE *STG 3000
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’MB-INJECTOR’
1 1 1 1.00
APPENDIX D. STARS INPUT FILE FOR UPSCALED THREE REACTION MODEL186
*TIME 18
*HEATR *IJK 1 1 1:30 320
*HEATR *IJK 1 1 1:25 260
*TIME 78
*INJECTOR ’MB-INJECTOR’
*TINJW 20
*PINJW 780
*INCOMP *GAS 0 0 0 0 0 0.79 0.1926 0.0174
*OPERATE *STG 3000
*GEOMETRY *K 0.50 1 1 0
*PERF *TUBE-END ’MB-INJECTOR’
1 1 1 1.00
*HEATR *IJK 1 1 1:30 0
*HEATR *IJK 1 1 1:25 0
*TIME 138
*TIME 198
*TIME 258
*TIME 318
*TIME 378
*TIME 438
*STOP
Appendix E
SAGD Input File
RESULTS SIMULATOR STARS
*interrupt *stop
*INUNIT *SI
*OUTPRN *WELL *ALL
*OUTPRN *GRID *NONE
*OUTPRN *ITER *NEWTON
**restart
*WPRN *GRID 200
*WPRN *ITER 200
*PRNTORIEN 2 0
*WPRN *SECTOR 100
*WSRF *SECTOR 100
*OUTSRF *SPECIAL
MATBAL WELL ’OIL’
MATBAL WELL ’WATER’
*OUTSRF *GRID *PRES *SW *SO *SG *TEMP *QUALBLK *CCHLOSS
*VISO *VISW *VISG *MOLDENO
*MASDENO *MASDENW *MASDENG *THCONDUCT *VPOROS *KRW *KRO *KRG
** ============== GRID AND RESERVOIR DEFINITION =================
*GRID *CART 45 1 30
187
APPENDIX E. SAGD INPUT FILE 188
*KDIR *DOWN
*DI *CON 1 **
*DJ *CON 30 **
*DK *CON 1 **
*DEPTH 1 1 1 1000
*POR *CON 0.35
*PERMI *CON 2000
*PERMJ *EQUALSI
*PERMK *EQUALSI
*NINEPOINT *IK
*END-GRID
*PRPOR 100
*ROCKTYPE 1
*CPOR 1e-5
*CTPOR 1e-4
*ROCKCP 1500E3
*THCONR 3E5
*thconw 0
*thcono 0
*thcong 0
*THTYPE *CON 1 ** Assigns the properties of the rocktype 1 to whole res.
** ============== FLUID DEFINITIONS =================================
MODEL 2 2 2 1
COMPNAME ’WATER’ ’OIL’
** -------- --------
CMM 0.018 0.282
PCRIT 22048 1115
TCRIT 374.15 493.85
KV1 1.1860E7 0.0
KV2 0.00000 0.000E+0
KV3 0.00000 0.000E+0
APPENDIX E. SAGD INPUT FILE 189
KV4 -3816.44 -4680.46
KV5 -227.02 -132.05
MASSDEN 0.998E3 0.972E3
CP 1E-7 1E-7
CT1 1E-4 1E-4
** T, deg C ’Water’ ’OIL’
*avisc 0.0047352 1E-8
*bvisc 1515.7 10000
** ’WATER’ ’OIL’
*CPG1 32.243 -22.383
*CPG2 1.924E-3 1.939
*CPG3 1.055E-5 -1.117E-3
*CPG4 -3.596E-9 2.528E-7
*HVR 4820 8569
*EV 0.38 0.38
PRSR 100 ** reference pressure, corresponding to the density
TEMR 20 ** reference temperature, corresponding to the density
PSURF 1E2 ** pressure at surface, for reporting well rates, etc.
TSURF 20 ** temperature at surface, for reporting well rates, etc.
** ============== ROCK-FLUID PROPERTIES ======================
rockfluid
swt ** Water-oil relative permeabilities
** Sw Krw Krow
** ---- -------- --------
0.00 0 1
0.05 0.0025 0.9025
0.10 0.01 0.81
0.15 0.0225 0.7225
0.20 0.04 0.64
0.25 0.0625 0.5625
APPENDIX E. SAGD INPUT FILE 190
0.30 0.09 0.49
0.35 0.1225 0.4225
0.40 0.16 0.36
0.45 0.2025 0.3025
0.50 0.25 0.25
0.55 0.3025 0.2025
0.60 0.36 0.16
0.65 0.4225 0.1225
0.70 0.49 0.09
0.75 0.5625 0.0625
0.80 0.64 0.04
0.85 0.7225 0.0225
0.90 0.81 0.01
0.95 0.9025 0.0025
1.00 1.0000 0.0000
slt **NOSWC ** Liquid-gas relative permeabilities
** Sl Krg Krog
** ---- -------- --------
0.00 1 0
0.05 0.9025 0.0025
0.10 0.81 0.01
0.15 0.7225 0.0225
0.20 0.64 0.04
0.25 0.5625 0.0625
0.30 0.49 0.09
0.35 0.4225 0.1225
0.40 0.36 0.16
0.45 0.3025 0.2025
0.50 0.25 0.25
0.55 0.2025 0.3025
0.60 0.16 0.36
APPENDIX E. SAGD INPUT FILE 191
0.65 0.1225 0.4225
0.70 0.09 0.49
0.75 0.0625 0.5625
0.80 0.04 0.64
0.85 0.0225 0.7225
0.90 0.01 0.81
0.95 0.0025 0.9025
1.00 0.0000 1.0000
** ============== INITIAL CONDITIONS ======================
*INITIAL
*PRES *CON 6000
*SW *CON 0.0
*SO *CON 1.0
*SG *CON 0.0
*TEMP *CON 55
*mfrac_oil ’OIL’ *CON 1.0
*mfrac_gas ’WATER’ *CON 1.0
*mfrac_wat ’WATER’ *CON 1.0
** ============== NUMERICAL CONTROL ======================
*NUMERICAL
*DTMAX 2
*NEWTONCYC 15
*NORM *PRESS 100 *SATUR 0.1 *TEMP 10 *Y 1 *X 1
*MINPRES 0.01
*CONVERGE *PRESS 0.1 *SATUR 0.1 *TEMP 0.1
*MATBALTOL 0.00001
*PIVOT *ON ** Pivoting stabilization
*RANGECHECK OFF
*RUN
** ============== RECURRENT DATA ========================================
*DATE 2002 01 01
APPENDIX E. SAGD INPUT FILE 192
*DTWELL 1
*WELL ’Injector1’
*WELL ’Producer1’
*INJECTOR ’Injector1’
**--------------’WATER’ ’C2 toC14’
*INCOMP WATER 1.0 0.0
*TINJW 275.86** degree C
*QUAL 1.0
*OPERATE *BHP 6000 **IWELLBORE 1 *PWELLBORE 1
**---------- rad geofac wfrac skin
*GEOMETRY *K 2.0 0.249 1.0 0.0
*PERF GEO ’Injector1’ ** i j k wi.
2 1 23 1.0
*PRODUCER ’Producer1’
*OPERATE *BHP 6000 **IWELLBORE 1 *PWELLBORE 1
**---------- rad geofac wfrac skin
GEOMETRY *K 2.0 0.249 1.0 0.0
*PERF GEO ’Producer1’ ** i j k wi.
2 1 29 1.0
*TIME 100
*TIME 200
*TIME 300
*TIME 400
*TIME 500
*TIME 600
*STOP
Appendix F
SAGD Input File for Grid Size
Effects Study
RESULTS SIMULATOR STARS
*INTERRUPT *STOP
** ============== INPUT/OUTPUT CONTROL ======================
dim mdiclu 200000
*INUNIT *SI
*OUTUNIT *SI
*WRST *TIME
**RESTART 185
**REWIND XXXX
*OUTPRN *GRID *PRES *SW *SO *SG *TEMP *OBHLOSS *VISO
*OUTPRN *WELL *ALL
*OUTSRF *GRID *PRES *SW *SO *SG *TEMP *VISO
*WPRN *GRID *TIME
*PRNTORIEN 2 0
*RANGECHECK *OFF
wprn iter 1
outprn iter *newton
**restart 154 **186
193
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 194
** ========= DEFINITION OF FUNDAMENTAL CARTESIAN GRID ========
*GRID *VARI 50 3 47
*KDIR *DOWN
*DI *IVAR 50*1
*DJ *JVAR 3*100
*DK *KVAR 34*1 1 1 8*1 3*1
*DTOP 50*734.4 50*739.6 50*744.0
*WELLBORE 0.0365
*LAMINAR
wellinfo
*CIRCWELL 0.0875 50 3 36 0
*RANGE 50 1:3 36
*WELLBORE 0.0365
*LAMINAR
*CIRCWELL 0.0875 50 3 46 0
*RANGE 50 1:3 46
*POR *IJK
1:50 1 1:20 0.25368
1:50 1 21:47 0.253509
1:50 2 1:20 0.250318
1:50 2 21:47 0.280487
1:50 3 1:20 0.232101
1:50 3 21:47 0.301246
*PERMI *CON 1468
*PERMJ *EQUALSI * 1.00
*PERMK *EQUALSI * 0.3828
*END-GRID
*PRPOR 7600
*ROCKTYPE 1
*CPOR 1.8E-5
*ROCKCP 1.5E6
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 195
*THCONR 1.5E5
*THCONW 1.5E5
*THCONO 1.5E5
*THCONG 1.5E5
*HLOSSPROP *OVERBUR 1.0E6 8.0E4
*UNDERBUR 1.0E6 8.0E4
*ROCKTYPE 2 copy 1
*ROCKCP 2.35E6
*THCONR 1.5E5
*THCONW 1.5E5
*THCONO 1.5E5
*THCONG 1.5E5
**cpor 0.0
THTYPE CON 1 ** RESERVOIR
THTYPE WELLBORE 50 1:3 36 CON 2 **TOP WELLBORE
THTYPE WELLBORE 50 1:3 46 CON 2 **BOTTOM WELLBORE
** ============== FLUID DEFINITIONS ======================
*MODEL 2 2 2 ** Components are water and dead oil. Most water
** properties are defaulted (=0). Dead oil K values
** are zero, and no gas properties are needed.
*COMPNAME ’WATER’ ’OIL’
** ------- --------
CMM 0.018 0.45 ** kg/gmol
MOLDEN 0 2222.22 ** gmol/m3
CP 0 7.25E-7 ** oil cr
CT1 0 6.84E-4 **3.8e-4
PCRIT 0 0 **217.6
TCRIT 0 0 **374.15
CPG1 0 942.0 **2.09
AVG 0 0 **1.13e-5
BVG 0 0 ** 1.075
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 196
*VISCTABLE
** TEMP
47 0 122807.
60. 0 35541.
80. 0 6751.
100 0 1659.
140 0 186.
180 0 39.
260 0 6.
300 0 3.3
900 0 3.3
** REFERENCE CONDITIONS
*PRSR 101.3
*TEMR 20.0
*PSURF 101.3
*TSURF 20.0
** ============== ROCK-FLUID PROPERTIES ======================
*ROCKFLUID
*RPT 1
*SWT smoothend quad
** Sw Krw Krow
0.35 0.00000 1.00000
0.40 0.00638 0.73551
0.43 0.01290 0.59032
0.46 0.02080 0.45618
0.49 0.02987 0.33409
0.52 0.03997 0.22540
0.55 0.05100 0.13201
0.58 0.06290 0.05702
0.61 0.07560 0.00713
0.62 0.08000 0.00000
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 197
**1.00 0.14000 0.00000
*SLT
** Sl Krg Krog
0.73 1.000 0.000
0.76 0.838 0.037
0.79 0.686 0.105
0.82 0.544 0.192
0.85 0.414 0.296
0.88 0.296 0.414
0.91 0.192 0.544
0.94 0.105 0.686
0.97 0.037 0.838
1.00 0.000 1.000
** ROCK 1
*KRTEMTAB *SWR *SORW *SGR *SORG *KRWRO *KROCW *KRGCW
48 0.35 0.40 0.00 0.38 0.08 1.0 1.0
200 0.40 0.28 0.00 0.25 0.12 1.0 1.0
*RPT 2 **--- WELLBORE ---
*SWT
** Sw Krw Krow
0.0 0.00 1.00
1.00 1.00 0.00
*SLT
** Sl Krg Krog
0.0 1.00 0.00
1.00 0.00 1.00
** ============== INITIAL CONDITIONS ======================
*INITIAL
*VERTICAL *DEPTH_AVE
*REFPRES 7600.0
*REFDEPTH 760.0
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 198
**PRES *CON 7600.
*TEMP *CON 48.
** ============== NUMERICAL CONTROL ======================
*NUMERICAL
***AIM *OFF
*DTMAX 15.0
*ITERMAX 60 **12
*SDEGREE 2
*SORDER rcmrb ***REDBLACK
*NORM *PRESS 400 *TEMP 50
*SATUR .3 *Y .3 *X .2
*NEWTONCYC 17
**RANGECHECK *OFF
**PVTOSCMAX 4
*upstream klevel
**converge totres normal
** ============== RECURRENT DATA ======================
*RUN
*DATE 1996 01 01
*DTWELL 1.0e-1
*WELL 1 ’UPCASI’ *FRAC 0.5
*WELL 2 ’UPTUB’ *FRAC 0.5
*WELL 3 ’LWCASI’ *FRAC 0.5
*WELL 4 ’LWTUB’ *FRAC 0.5
*******************
** PREHEAT PHASE
*******************
*PRODUCER ’UPCASI’ ** UPCAS
*OPERATE *MIN *BHP 11700
*GEOMETRY *J 0.0875 0.249 1 0 ** 7" liner
*PERF *GEO ’UPCASI’
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 199
50 1 36 / 2 1 1 1.0
*INJECTOR *MOBWEIGHT ’UPTUB’ ** UPTUB
*TINJW 324.6
*QUAL 0.5
*INCOMP WATER 1.0 0.0
*OPERATE *MAX *STW 200
*GEOMETRY *J 0.0875 0.249 1 0 ** 7" liner
*PERF *GEO ’UPTUB’
50 1 36 / 1 1 1 1.0
*PRODUCER ’LWCASI’ ** LWCASI
*OPERATE *MIN *BHP 10700
*GEOMETRY *J 0.0875 0.249 1 0 ** 7" liner
*PERF *GEO ’LWCASI’
50 1 46 / 2 1 1 1.0
*INJECTOR *MOBWEIGHT ’LWTUB’ ** LWTUB
*INCOMP WATER 1.0 0.0
*TINJW 324.6
*QUAL 0.5
*OPERATE *MAX *STW 200
*OPERATE *MAX *BHP 11650
**STOP
*GEOMETRY *J 0.0875 0.249 1 0 ** 7" liner
*PERF *GEO ’LWTUB’
50 1 46 / 1 1 1 1.0
*DATE 1996 02 01
*DATE 1996 03 01
*DATE 1996 04 01
*DTWELL 1.0e-4
*******************
** HIGH PRESSURE SAGD
*******************
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 200
*SHUTIN ’UPCASI’
*SHUTIN ’LWTUB’
*INJECTOR *MOBWEIGHT ’UPTUB’
** UPTUB
*INCOMP WATER 1.0 0.0
*TINJW 324.6
*QUAL 0.9
*OPERATE *MAX *BHP 11700
*OPERATE *MAX *STW 200
*PRODUCER ’LWCASI’
*OPERATE *MAX *STL 50
*OPERATE *MAX *STEAM 1 cont
*WELL 5 ’LWTUBP’ *FRAC 0.5
*PRODUCER ’LWTUBP’
** LWTUBP
*OPERATE *MAX *STL 300
*OPERATE *MAX *STEAM 5 cont
*OPERATE *MIN *BHP 5200
*NEXTSEG
*GEOMETRY *J 0.0875 0.249 1 0 ** 7" liner
*PERF *GEO ’LWTUBP’
50 1 46 / 1 1 1 1.0
*DATE 1996 05 25
*DTWELL 1.0e-2
*******************
** DEPRESSURIZATION PHASE
*******************
*INJECTOR *MOBWEIGHT ’UPTUB’
*INCOMP WATER 1.0 0.0
*OPERATE *MAX *BHP 11700
*OPERATE *MAX *STW 100
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 201
*PRODUCER ’LWCASI’
*OPERATE *MIN *BHP 5500
*SHUTIN
*OPERATE *MAX *STL 50
*OPERATE *MAX *STEAM 1 cont
*PRODUCER ’LWTUBP’
** LWTUBP
*OPERATE *MIN *BHP 5200
*NEXTSEG
*OPERATE *MAX *STL 300
*OPERATE *MAX *STEAM 30 cont
*DATE 1996 07 15
*DTWELL 1.0e-3
*******************
** SAGD PHASE
*******************
*INJECTOR *MOBWEIGHT ’UPTUB’
*TINJW 270
*QUAL 0.9
*INCOMP WATER 1.0 0.0
*OPERATE *MAX *BHP 5500
*OPERATE *MAX *STW 500
*PRODUCER ’LWTUBP’
*OPERATE *MIN *BHP 5200
*OPERATE *MAX *STL 300
*OPERATE *MAX *STEAM 5 cont
*DATE 1996 09 30
*PRODUCER ’LWTUBP’
*OPERATE *MIN *BHP 5200
*OPERATE *MAX *STL 300
*OPERATE *MAX *STEAM 5 cont
APPENDIX F. SAGD INPUT FILE FOR GRID SIZE EFFECTS STUDY 202
*MONITOR *MIN *STO 5
*STOP
transient ’LWTUBP’ on
*DATE 1996 12 31
*DATE 1997 09 30
*PRODUCER ’LWTUBP’
*OPERATE *MIN *BHP 5200
*OPERATE *MAX *STL 500
*OPERATE *MAX *STEAM 5 cont
*MONITOR *MIN *STO 5
*STOP
*DATE 1997 12 31
*DATE 1998 12 31
*DATE 1999 12 31
*DATE 2000 12 31
*DATE 2001 12 31
*DATE 2002 12 31
*DATE 2003 12 31
*DATE 2004 12 31
*DATE 2005 12 31
*DATE 2006 12 31
*DATE 2007 12 31
*STOP
Bibliography
[1] 3DSL. 3DSL Version 2010 Software Manual. StreamSim Technologies, Inc.,
2010.
[2] M. Abbaszadeh, F.R. Garza, C. Yuan, and G.A. Pope. Single-well simulation
study of foam eor in gas-cap oil of the naturally-fractured cantarell field. In
SPE129867 presented at the SPE Improved Oil Recovery Symposium. Society
of Petroleum Engineers, April 2010.
[3] I.Y. Akkutlu and Y.C. Yortsos. The dynamics of in-situ combustion fronts in
porous media. Combustion and Flame, 134:229–247, 2003.
[4] A. Alamatsaz, R.G. Moore, S.A. Mehta, and M.G. Ursenbach. Experimental
investigation of in situ combustion at low air fluxes. In SPE144517 presented
at the SPE Western North American Region Meeting. Society of Petroleum
Engineers, 2011.
[5] M.E. Armento and C.A. Miller. Stability of moving combustion fronts in porous
media. SPE Journal, 17(6):423–430, December 1977.
[6] K. Aziz and A. Settari. Petroleum Reservoir Simulation. Elsevier Applied
Science Publishers, London, 1979.
[7] Roderick Panko Batycky. A Three-dimensional Two-phase Field Scale Stream-
line Simulator. PhD thesis, Stanford University, 1997.
[8] R.P. Batycky, M.J. Blunt, and M.R. Thiele. A 3d field-scale streamline-based
reservoir simulator. SPE Reservoir Engineering, 12(4):246–254, November 1997.
203
BIBLIOGRAPHY 204
[9] R.P. Batycky, M. Forster, M.R. Thiele, and K. Stuben. Parallelization of a
commercial streamline simulator and performance on practical models. SPE
Reservoir Evaluation & Engineering, 13(3):383–390, June 2010.
[10] M. Bazargan, B. Chen, M. Cinar, G. Glatz, A. Lapene, Z. Zhu, L. Castanier,
M. Gerritsen, and A.R. Kovscek. A combined experimental and simulation
workflow to improve predictability of in situ combustion. In SPE144599 pre-
sented at the SPE Western Regional Meeting. Society of Petroleum Engineers,
2011.
[11] J.D.M. Belgrave, R.G. Moore, M.G. Ursenbach, and D.W. Bennion. A compre-
hensive approach to in-situ combustion modeling. SPE Advanced Technology
Series, 1(1):37–58, April 1993.
[12] V.T. Beraldo, M.J. Blunt, and D.J. Schiozer. Compressible streamline-based
simulation with changes in oil composition. SPE Reservoir Evaluation & En-
gineering, 12(6):963–973, December 2009.
[13] I. Berre, H.K. Dahle, K.H. Karlsen, and H.F. Nordhaug. A streamline front
tracking method for two- and three-phase flow including capillary forces. Con-
temporary Mathematics, 295:49–61, 2002.
[14] F. Bratvedt, T. Gimse, and C. Tegnander. Streamline computations for porous
media flow including gravity. Transport in Porous Media, 25(1):63–78, October
1996.
[15] J. Burger, P. Sourieau, and M. Combarnous. Thermal Methods of Oil Recovery.
Paris, France, 1985.
[16] R.M. Butler. Thermal Recovery of Oil and Bitumen. 1991.
[17] M.L. Byl, R.G. Moore, and M.G. Ursenbach. Field observations of in situ
combustion in a waterflooded reservoir in the kinsella field. Journal of Canadian
Petroleum Technology, 32(7), July 1993.
BIBLIOGRAPHY 205
[18] Hui Cao. Deveopment of Techniques for General Purpose Simulators. PhD
thesis, Stanford University, 2002.
[19] Qing Chen. Assessing and Improving Steam-Assisted Gravity Drainage: Reser-
voir Heterogeneities, Hydraulic Fractures, and Mobility Control Foams. PhD
thesis, Stanford University, 2009.
[20] Yuguang Chen. Upscaling and Subgrid Modeling of Flow and Transport in Het-
erogeneous Reservoirs. PhD thesis, Stanford University, 2005.
[21] H. Cheng, I. Osako, A. Datta-Gupta, and M. King. A rigorous compress-
ible streamline formulation for two- and three-phase black-oil simulation. SPE
Reservoir Evaluation & Engineering, 11(4):407–417, December 2006.
[22] J.R. Christensen, G. Darche, B. Dechelette, H. Ma, and P.H. Sammon. Ap-
plications of dynamic gridding to thermal simulations. In SPE86969 presented
at the SPE International Thermal Operations and Heavy Oil Symposium and
Western Regional Meeting. Society of Petroleum Engineers, 2004.
[23] M.A. Christie, Heriot-Watt U., and M.J. Blunt. Tenth spe comparative solution
project: A comparison of upscaling techniques. SPE Reservoir Engineering,
4(4):308–317, August 2001.
[24] M. Cinar, L.M. Castanier, and A.R. Kovscek. Isoconversional kinetic analysis
of the combustion of heavy hydrocarbons. In SPE120995 presented at the SPE
Western Regional Meeting. Society of Petroleum Engineers, 2009.
[25] Murat Cinar. Kinetics of Crude-oil Combustion in Porous Media Interpreted
Using Isoconversional Methods. PhD thesis, Stanford University, 2011.
[26] R. Coates, S. Lorimer, and J. Ivory. Experimental and numerical simulations
of a novel top down in-situ combustion process. In SPE30295 presented at the
SPE International Heavy Oil Symposium. Society of Petroleum Engineers, 1995.
[27] K.H. Coats. In-situ combustion model. SPE Journal, 20(6):533–554, December
1980.
BIBLIOGRAPHY 206
[28] K.H. Coats. Some observations on field-scale simulation of the in-situ combus-
tion process. In SPE12247 presented at the SPE Reservoir Simulation Sympo-
sium. Society of Petroleum Engineers, 1983.
[29] M. Crane, F. Bratvedt, K. Bratvedt, and R. Olufsen. A fully compositional
streamline simulator. In SPE63156 presented at the SPE Annual Technical
Conference and Exhibition. Society of Petroleum Engineers, 2000.
[30] R.B. Crookston, W.E. Culham, and W.H. Chen. A numerical simulation model
for thermal recovery processes. SPE Journal, 19(1):37–58, February 1979.
[31] S. Das. Diffusion and dispersion in the simulation of vapex process. In SPE97924
presented at the SPE/PS-CIM/CHOA International Thermal Operations and
Heavy Oil Symposium. Society of Petroleum Engineers, 2005.
[32] R.J. Davies. The thin flame technique for in-situ combustion simulation. In
SPE19647 presented at the SPE Annual Technical Conference and Exhibitio.
Society of Petroleum Engineers, 1989.
[33] B. Dechelette, O. Heugas, G. Quenault, J. Bothua, and J.R. Christensen. Air
injection-improved determination of the reaction scheme with ramped temper-
ature experiment and numerical simulation. Journal of Canadian Petroleum
Technology, 45(1), January 2006.
[34] I. Diabira, L.M. Castanier, and A.R. Kovscek. Porosity and permeability evo-
lution accompanying hot fluid injection into diatomite. Petroleum Science and
Technology, 19:1167–1185, 2001.
[35] Birol Dindoruk. Analytical Theory of Multiphase Multicomponent Displacement
in Porous Media. PhD thesis, Stanford University, 1992.
[36] G.D. Donato and M.J. Blunt. Streamline-based dual-porosity simulation of
reactive transport and flow in fractured reservoirs. Water Resources Research,
40(W04203), 2004.
[37] Eclipse. Eclipse Version 2009 Software Manual. Schulumberger Ltd., 2009.
BIBLIOGRAPHY 207
[38] N.R. Edmunds. A model of the steam drag effect in oil sands. Journal of
Canadian Petroleum Technology, 23(5), Sep 1984.
[39] H. Fadaei, G. Debenest, A.M. Kamp, M. Quintard, and G. Renard. How the
in-situ combustion process works in a fractured system: 2d core- and block-scale
simulation. SPE Reservoir Evaluation & Engineering, 13(1):118–130, February
2010.
[40] Mohammad Reza Fassihi. Analysis of Fuel Oxidation in In-Situ Combustion
Oil Recovery. PhD thesis, Stanford University, 1981.
[41] M.R. Fassihi and T.H. Gillham. The use of air injection to improve the double
displacement processes. In SPE26374 presented at the SPE Annual Technical
Conference and Exhibition. Society of Petroleum Engineers, 1993.
[42] M.R. Fassihi, D.V. Yannimaras, and V.K. Kumar. Estimation of recovery factor
in light-oil air-injection projects. SPE Reservoir Engineering, 12(3):173–178,
August 1997.
[43] N.P. Freitag and D.R. Exelby. A sara-based model for simulating the pyrolysis
reactions that occur in high-temperature eor processes. Journal of Canadian
Petroleum Technology, 45(3), March 2006.
[44] C.F. Gates and H.J. Ramey. A method for engineering in-situ combustion oil
recovery projects. Journal of Petroleum Technology, 32(2):285–294, February
1980.
[45] M.G. Gerritsen, K. Jessen, B.T. Mallison, and J. Lambers. A fully adaptive
streamline framework for the challenging simulation of gas injection processes.
In SPE97270 presented at the SPE Annual Technical Conference and Exhibition.
Society of Petroleum Engineers, 2005.
[46] M. Greaves, S.R. Ren, R.R. Rathbone, T. Fishlock, and R. Ireland. Improved
residual light oil recovery by air injection (lto process). Journal of Canadian
Petroleum Technology, 39(1), January 2000.
BIBLIOGRAPHY 208
[47] D. Gutierrez, F. Skoreyko, R.G. Moore, S.A. Mehta, and M.G. Ursenbach. The
challenge of predicting field performance of air injection projects based on lab-
oratory and numerical modelling. Journal of Canadian Petroleum Technology,
48(4):23–33, April 2009.
[48] R.J. Hallam and J.K. Donnelly. Pressure-up blowdown combustion: A chan-
neled reservoir recovery process. SPE Advanced Technology Series, 1(1):153–
158, April 1993.
[49] H. Hassanzadeh, J. Abedi, and M. Pooladi-Darvish. Equivalent rate constant for
numerical simulation of linear convection-diffusion-reaction. Journal of Cana-
dian Petroleum Technology, 49(4):51–57, April 2010.
[50] B.T. Hoffman and A.R. Kovscek. Displacement front stability of steam in-
jection into high porosity diatomite rock. Journal of Petroleum Science and
Engineering, 46(4):253–266, April 2005.
[51] M.H. Hui, D. Zhou, X.H. Wen, and L.J. Durlofsky. Development and application
of a new technique for upscaling miscible processes. SPE Reservoir Evaluation
& Engineering, 8(3):189–195, June 2005.
[52] M.K. Hwang. An in-situ combustion process simulator with a moving-front
representation. Society of Petroleum Engineers Journal, 22(2):271–279, April
1982.
[53] Y. Ito and A.K. Chow. A field scale in-situ combustion simulator with chan-
neling considerations. SPE Reservoir Engineeringg, 3(2):419–430, May 1988.
[54] S. Javad, P. Oskouei, R.G. Moore, B. Maini, and S.A. Mehta. Front self-
correction for in-situ combustion. Journal of Canadian Petroleum Technology,
50(3):43–56, March 2011.
[55] G.R. Jenkins and J.W. Kirkpatrick. Twenty years’ operation of an in-situ
combustion project. Journal of Canadian Petroleum Technology, 18(1), Jan-
Mar 1975.
BIBLIOGRAPHY 209
[56] K. Jessen, M.G. Gerritsen, and B.T. Mallison. High-resolution prediction of
enhanced condensate recovery processes. SPE Journal, 13(2):257–266, June
2008.
[57] K. Jessen and F.M. Orr Jr. Gravity segregation and compositional streamline
simulation. In SPE89448 presented at the SPE/DOE Symposium on Improved
Oil Recovery. Society of Petroleum Engineers, 2004.
[58] E.A. Jimenez, A. Datta-Gupta, and M.J. King. Full-field streamline tracing in
complex faulted systems with non-neighbor connections. SPE Journal, 15(1):7–
17, March 2010.
[59] C.S. Kabir, N.I. Al-Khayat, and M.K. Choudhary. Lessons learned from en-
ergy models: Iraq’s south rumaila case study. SPE Reservoir Evaluation &
Engineering, 11(4):759–767, August 2008.
[60] Jeremy E. Kozdon. Numerical Methods with Reduced Grid Dependency for
Enhanced Oil Recovery. PhD thesis, Stanford University, 2009.
[61] Morten Rode Kristensen. Development of Models and Algorithms for the Study
of Reactive Porous Media Processes. PhD thesis, Technical University of Den-
mark, 2008.
[62] Archana Kumar. Hybrid simulation of the eulerian method and the streamline
method: Investigating the role of cross-product in cross-over time between the
two methods. Master’s thesis, Stanford University, 2010.
[63] M. Kumar. A cross-sectional simulation of west heidelberg in-situ combustion
project. SPE Reservoir Engineering, 6(1):46–54, February 1991.
[64] M. Kumar and A.M. Garon. A method for predicting oxygen and air fireflooding
performance. SPE Reservoir Evaluation & Engineering, 6(2):245–252, May
1991.
BIBLIOGRAPHY 210
[65] M.A. Langley. In-situ combustion experiences, husky oil operation. In Technical
Meeting/Petroleum Conference Of The South Saskatchewan Section. Petroleum
Society of Canada, 1985.
[66] A. Lapene, L.M. Castanier abd G. Debenest, M. Quintard, A.M. Kamp, and
B. Corre. Effects of water on kinetics of wet in-situ combustion. In SPE93243
presented at the SPE Western Regional Meeting. Society of Petroleum Engi-
neers, 2009.
[67] A. Lapene, D.V. Nichitae, G. Debenesta, and M. Quintard. Three-phase free-
water flash calculations using a new modified rachfordcrice equation. Fluid
Phase Equilibria, 297(1):121–128, October 2010.
[68] Alexandre Lapene. Experimental and numerical study of in-situ combustion of
heavy oil. PhD thesis, Institut de Mecanique des Fluides de Toulouse, 2010.
[69] E.C. Lau and K.E. Kisman. Attractive control features of the combustion
override split - production horizontal well process in a heavy oil reservoir. In
Annual Technical Meeting, Calgary, Alberta. Petroleum Society of Canada, Jun
1994.
[70] H. Lof, M. Gerritsen, and Marco Thiele. Parallel streamline simulation. In
SPE113543 presented at the Europec/EAGE Conference and Exhibition. Society
of Petroleum Engineers, 2008.
[71] Bradley Thomas Mallison. Streamline-Based Simulation of Two-phase, Multi-
component Flow in Porous Media. PhD thesis, Stanford University, 2004.
[72] B.T. Mallison, M.G. Gerritsen, and S.F. Matringe. Improved mappings for
streamline-based simulation. SPE Journal, 11(3):294–302, September 2006.
[73] E.J. Manrique, V.E. Muci, and M.E. Gurfinkel. Eor field experiences in carbon-
ate reservoirs in the united states. SPE Reservoir Evaluation & Engineering,
10(6):667–686, December 2007.
BIBLIOGRAPHY 211
[74] D.M. Marjerrison and M.R. Fassihi. A procedure for scaling heavy-oil combus-
tion tube results to a field model. In SPE24175 presented at the SPE/DOE
Enhanced Oil Recovery Symposium. Society of Petroleum Engineers, 1992.
[75] D.M. Marjerrison and M.R. Fassihi. Performance of morgan pressure cycling
in-situ combustion project. In SPE27793 presented at the SPE/DOE Improved
Oil Recovery Symposium. Society of Petroleum Engineers, 1994.
[76] J.C. Martin and R.E. Wegner. Numerical solution of multiphase, two-
dimensional incompressible flow using streamtube relationships. Society of
Petroleum Engineers Journal, 19(5):313–323, October 1979.
[77] M.L. Michelsen and J.M. Mollerup. Thermodynamic Models: Fundamentals
and Computational Aspects. Tie-Line Publications, Holte, Denmark, 2004.
[78] C.A. Miller. Stability of moving surfaces in fluid systems with heat and mass
transport iii. stability of displacement fronts in porous media. AIChE Journal,
21(3):474–479, May 1975.
[79] A.R. Montes, D. Gutierrez, R.G. Moore, S.A. Mehta, and M.G. Ursenbach. Is
high-pressure air injection (hpai) simply a flue-gas flood? Journal of Canadian
Petroleum Technology, 49(2):56–63, February 2010.
[80] R.G. Moore, C.J. Laureshen, M.G. Ursenbach, S.A. Mehta, and J.D.M. Bel-
grave. A canadian perspective on in situ combustion. Journal of Canadian
Petroleum Technology, 38(13), 1999.
[81] R.J. Morgan. Can horizontal wells inject new life into an old combustion pilot?
eyehill thermal project update. In Technical Meeting/Petroleum Conference Of
The South Saskatchewan Section. Petroleum Society of Canada, 1983.
[82] T.N. Nasr, G. Beaulieu, H. Golbeck, and G. Heck. Novel expanding solvent-sagd
process es-sagd. Journal of Canadian Petroleum Technology, 42(1), January
2003.
BIBLIOGRAPHY 212
[83] T.W. Nelson and J.S. McNeil. How to engineer an in-situ combustion project.
The Oil and Gas Journal, page 58, June 1961.
[84] J. Nilsson, M. Gerritsen, and R. Younis. A novel adaptive anisotropic grid
framework for efficient reservoir simulation. In SPE93243 presented at the SPE
Reservoir Simulation Symposium. Society of Petroleum Engineers, 2005.
[85] J. Nodwell, R.G. Moore, M.G. Ursenbach, C.J. Laureshen, and S.A. Mehta.
Economic considerations for the design of in-situ combustion projects. Journal
of Canadian Petroleum Technology, 39(8), August 2000.
[86] Olufolake Ogunbanwo. Uncertainty assessment on in-situ combustion simula-
tions using experimental design. Master’s thesis, Stanford University, 2011.
[87] I. Osako, M. Kumar, V. Hoang, and G. Balasubramanian. Evaluation of stream-
line simulation application to heavy oil waterflood. In SPE122922 presented at
the Latin American and Caribbean Petroleum Engineering Conference. Society
of Petroleum Engineers, 2009.
[88] Ichiro Osako. A Rigorous Compressible Streamline Formulation for Black Oil
and Compositional Simulation. PhD thesis, Texas A&M University, 2006.
[89] U. Pasarai and N. Arihara. Application of streamline method to hot waterflood-
ing simulation for heavy-oil recovery. In SPE93149 presented at the SPE Asia
Pacific Oil and Gas Conference and Exhibition. Society of Petroleum Engineers,
2005.
[90] R. Penney, R. Moosa, G. Shahin, F. Hadhrami, A. Kok, G. Engen, O. van
Ravesteijn, K. Rawnsley, and B. Kharusi. Steam injection in fractured car-
bonate reservoirs: Starting a new trend in eor. In IPTC10727 presented at
the International Petroleum Technology Conference. Society of Petroleum En-
gineers, 2005.
[91] H.J-M. Petit, P.L. Thiez, and P. Lemonnier. History matching of a heavy-
oil combustion pilot in romania. In SPE20249 presented at the SPE/DOE
Enhanced Oil Recovery Symposium. Society of Petroleum Engineers, 1990.
BIBLIOGRAPHY 213
[92] D.W. Pollock. Semianalytical computation of path lines for finite-diffference
models. Ground Water, 26(6):743–750, 1988.
[93] M. Prats. Thermal Recovery. 1986.
[94] S. Roychaudhury, N.S. Rao, S.K. Sinha, S. Sur, K.K. Gupta, A.V. Sapkal, A.K.
Jain, and J.S. Saluja. Extension of in-situ combustion process from pilot to semi-
commercial stage in heavy oil field of balol. In SPE37547 presented at Inter-
national Thermal Operations and Heavy Oil Symposium. Society of Petroleum
Engineers, 1997.
[95] P.H. Sammon. Dynamic grid refinement and amalgamation for compositional
simulation. In SPE79683 presented at the SPE Reservoir Simulation Sympo-
sium. Society of Petroleum Engineers, 2003.
[96] P.S. Sarathi. In-Situ Combustion Handbook C Principles and Practices. Na-
tional Petroleum Technology Office, U.S.D.O.E., Tulsa, OK, 1998.
[97] W.M. Schulte and A.S. de Vries. In-situ combustion in naturally fractured
heavy oil reservoirs. SPE Journal, 25(1):67–77, February 1985.
[98] J.P. Seidle and L.E. Arri. Use of conventional reservoir models for coalbed
methane simulation. In SPE21599 presented at the CIM/SPE International
Technical Meeting. Society of Petroleum Engineers, 1990.
[99] C.J. Seto, K. Jessen, and F.M. Orr Jr. Compositional streamline simulation
of field scale condensate vaporization by gas injection. In SPE79690 presented
at the SPE Reservoir Simulation Symposium. Society of Petroleum Engineers,
2003.
[100] C.J. Seto, K. Jessen, and F.M. Orr Jr. Using analytical solutions in composi-
tional streamline simulation of a field-scale co2-injection project in a conden-
sate reservoir. SPE Reservoir Evaluation & Engineering, 10(4):393–405, August
2007.
BIBLIOGRAPHY 214
[101] STARS. STARS Version 2009 User’s Guide. Computer Modeling Group Ltd.,
Calgary, Canada, 2009.
[102] S. Stokka, A. Oesthus, and J. Frangeul. Evaluation of air injection as an ior
method for the giant ekofisk chalk field. In SPE97481 presented at the SPE
International Improved Oil Recovery Conference in Asia Pacific. Society of
Petroleum Engineers, 2005.
[103] T.W. Stone and J.S. Nolen. Practical and robust isenthalpic/isothermal flashes
for thermal fluids. In SPE118893 presented at the SPE Reservoir Simulation
Symposium. Society of Petroleum Engineers, 2009.
[104] M. Thiele, R.P. Batycky, S. Pollitzer, and T. Clemens. Polymer-flood modeling
using streamlines. SPE Reservoir Evaluation & Engineering, 13(2):313–322,
April 2010.
[105] M.R. Thiele. Streamline simulation. In 8th International Forum on Reservoir
Simulation, 2005.
[106] M.R. Thiele and R.P. Batycky. Using streamline-derived injection efficiencies for
improved waterflood management. SPE Reservoir Evaluation & Engineering,
9(2):187–196, April 2006.
[107] M.R. Thiele, R.P. Batycky, and M.J. Blunt. A streamline-based 3d field-scale
compositional reservoir simulator. In SPE38889 presented at the SPE Annual
Technical Conference and Exhibition. Society of Petroleum Engineers, 1997.
[108] M.R. Todd and W.J Longstaff. The development, testing and application of a
numerical simulator for predicting miscible flood performance. Trans. AIME,
pages 874–882, 1972.
[109] S. Turek, L. Rivkind, J. Hron, and R. Glowinski. Numerical analysis of a new
time-stepping θ-scheme for incompressible flow simulations. 2005.
BIBLIOGRAPHY 215
[110] M.G. Ursenbach, R.G. Moore, and S.A. Mehta. Air injection in heavy oil reser-
voirs - a process whose time has come (again). Journal of Canadian Petroleum
Technology, 49(1):48–54, January 2010.
[111] F.K. Veedu, M. Delshad, and G.A. Pope. Scaleup methodology for chemical
flooding. In SPE135543 presented at the SPE Annual Technical Conference and
Exhibition. Society of Petroleum Engineers, 2010.
[112] G. K. Youngren. Development and application of an in-situ combustion reservoir
simulator. SPE Journal, 20(1), February 1980.