Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

Annu. Rev. Fluid Mech. 2005. 37:21138 doi: 10.1146/annurev.uid.37.061903.175748 Copyright c 2005 by Annual Reviews.

All rights reserved

MODELING FLUID FLOW IN OIL RESERVOIRS


Margot G. Gerritsen1 and Louis J. Durlofsky1,2
1

Department of Petroleum Engineering, Stanford University, Stanford, California; email: margot.gerritsen@stanford.edu 2 ChevronTexaco Energy Technology Company, San Ramon, California
Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Key Words reservoir simulation, heterogeneity, multiscale methods, upscaling, compositional modeling Abstract Efciently and accurately solving the equations governing uid ow in oil reservoirs is very challenging because of the complex geological environment and the intricate properties of crude oil and gas at high pressure. We present these challenges and review successful and promising solution approaches. We discuss in detail the modeling of uid ow in reservoirs with strongly varying rock properties. This requires subgrid-scale models that accurately represent the ow physics due to ne-scale uctuations. A second focus is on the complex multiphase, multicomponent systems that describe miscible gas injection processes for enhanced oil recovery and CO2 sequestration.

1. INTRODUCTION
The eld of uid ow simulation in petroleum reservoirs has seen major advances in the last few decades. These advances partly reect the increasing computer power that has facilitated the application of more sophisticated and detailed models; but mostly the advances are the result of intense research in reservoir uid ow modeling driven by the petroleum industry. The decline of the easiest-toproduce petroleum accumulations motivates the industry to optimize production from existing reservoirs using enhanced oil recovery (EOR) processes, and to move to high-risk developments of reservoirs in more challenging physical environments. In EOR processes uids that may not naturally occur in the reservoir are injected. The injected uids alter the ow properties of the reservoir and promote oil production. Examples of high-risk developments are offshore production in deep water and oil production in the Arctic. To maximize the production and reduce the risks of failure of these projects, the oil industry increasingly relies on computational performance prediction and optimization tools, of which reservoir uid ow simulation is an integral part. Efciently and accurately solving the reservoir uid ow equations is very challenging because of the complex geological environment and the intricate properties of crude oil and gas at high pressure. The aim of this review is to provide
0066-4189/05/0115-0211$14.00

211

212

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

an understanding of the physics of reservoir uid ows, present the challenges in simulating them, and review successful and promising solution approaches. Because of the breadth of the eld, and the limited space in this review, we necessarily limit detailed discussion to two main topics. The rst topic is the efcient modeling of ow in reservoirs with strongly varying heterogeneity, which requires subgrid-scale models that accurately represent the ow physics due to ne-scale uctuations. This topic presents interesting technical challenges and has a signicant impact on reservoir performance prediction. Second, we discuss the complex and strongly nonlinear multiphase, multicomponent systems that describe production of light oil reservoirs and miscible gas injection processes for EOR and CO2 sequestration. We emphasize the challenges in modeling transport of components in miscible gas injection processes that have great economic and environmental relevance. We note that with these two choices, we touch on key challenges in the two main parts of a reservoir simulation: ow and transport.

1.1. Fluid Flow in Oil Reservoirs


Typically, a petroleum reservoir is a body of deep underground sedimentary rock that contains a mixture of uids in the interstitial spaces (about 10100 m across) between grains (see Figure 1a). The volume fraction of void space, referred to as porosity, generally ranges from 0.10.3. Generally, three uid phases exist; we designate them aqueous (the phase containing predominantly water), liquid (the phase containing liquid hydrocarbons, which we also refer to as oil), and the vapor or gas phase, which contains gaseous hydrocarbons. Each phase may consist of many chemical components. For example, oil is a very complex mixture of hundreds of hydrocarbons with different chemical properties. Due to differences in density, the phases are frequently found segregated in the reservoir with higher gas saturations in the upper reservoir rock, followed by oil, with water primarily present in the lower layers. At the pore scale, surface tension controls the uid conguration. Reservoir pressure and temperature are generally very high (of the order of thousands of psia and hundreds of degrees centigrade, respectively). Both microscopic and macroscopic effects control the movement of uids in the reservoir. The microscopic effects include viscosity and interfacial tension (IFT). For light crudes, the viscosity is comparable to or less than that of water, whereas heavy oils may have viscosities comparable to that of tar. Interfacial tension between reservoir uids may cause uids with lower saturation to become disconnected in the pores and unable to establish continuous ow paths. Also, the saturation of the immobile uid in the crooks and crevices of the rock can be substantial. The presence of one uid can inhibit the ow of another because of the resistance to change in interfacial geometry. Macroscopically, uid ow is controlled by reservoir heterogeneity and mobility differences between the uids. Reservoir rocks are usually formed from nonuniform deposits that vary in time and space (Figure 1b). Erosion, faulting, and fracturing further complicate the heterogeneity. Consequently, the rock

FLUID FLOW IN OIL RESERVOIRS

213

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Figure 1 (a) Outcrop showing rock heterogeneity with a superimposed typical grid cell used in reservoir simulations. (b) Slice of berea sandstone reservoir rock. Actual dimensions of section shown 400 2802 .

214

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

permeability may vary by orders of magnitude [typically of the order of a millidarcy to a darcy with 1 darcy (d) = 1012 m2 = the permeability that gives a ow of 1 cm s1 to a uid with a viscosity of 1 centipoise under a pressure gradient of 1 atm cm1 ]. Mobility differences may severely impact the efciency of gas or water injection. The more mobile displacing uid seeks pathways formed by connected high-permeability zones, faults, and fractures. Because these pathways may extend over large distances, large areas of oil may be bypassed. In gas injection processes gravity segregation can also greatly affect the global sweep. Viscosities, IFT, and mobility differences vary throughout the reservoir and depend on phase saturations, phase interactions at the prevailing pressure and temperature, and molecular composition of the phases. Chemical components may transfer between contacting phases, altering the uid properties of both. Interactions between the uids and the reservoir rock may also impact performance. For example, pressure reduction during depletion can cause (local) rock compaction that affects porosity as well as permeability. Thermal effects are generally very small due to the large heat capacity of the rock. However, in thermal recovery methods, such as steam ooding and in situ combustion (described below), temperature effects must be included. In situ combustion is further complicated by the presence of fast chemical reactions.

1.2. Oil Production


There are several mechanisms for primary production in which the natural drive mechanisms present in the reservoir are exploited. In dissolved gas drives, oil is displaced toward production wells by gas that evolves from the oil as production lowers the reservoir pressure. Gas cap drives exist if the reservoir initially contains free gas overlying the oil. Expansion of this gas cap pushes the oil toward the wells. In natural water drives, it is expanding brine that underlies the reservoir that displaces the oil. It is very common to combine primary production with water ooding, also referred to as secondary production, where water (possibly produced along with the oil) is injected back into the reservoir. Water ooding is relatively inexpensive and helps maintain reservoir pressure, but capillary forces between the water and oil phases generally act to retain oil in the rock, causing a substantial fraction of the oil to remain undisplaced. Despite the very high reservoir pressures and temperatures, the percentage of oil originally in place that is recovered by primary and secondary production is usually only 20%40%. In the United States alone this means 300 billion barrels of the estimated 400 billion barrels of original oil in place will not be recovered using these production techniques. A substantial fraction of the residual oil may be recovered using EOR processes. In thermal recovery, heating reduces the viscosity of the oil. Steam ooding has had long-standing commercial success, but can not be used effectively in deep reservoirs. In situ combustion, in which heat is generated by burning part of the oil in the reservoir, is an economically attractive alternative in recovering heavy oil from shallow reservoirs and lighter oil from deep reservoirs

FLUID FLOW IN OIL RESERVOIRS

215

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

(Gerritsen et al. 2004, Sarathi 1999). It is also an ideal process for producing oil from thin formations, being most effective in 1050-ft thick sandbodies. An added advantage is that elevated temperatures ahead of the combustion zone cause the oil to crack into lighter components, increasing its value. It is now well established that injecting gases such as CO2 , methane, enriched hydrocarbon gases, or nitrogen into an oil reservoir can lead to efcient local displacements and small residual oil saturations if the displacement pressure is sufciently high such that miscibility develops and IFT is (near) zero. CO2 injection has proven to be one of the most efcient EOR methods since it was rst tried in Texas in 1972. EOR techniques have been hampered by their relatively high cost and, in some cases, by the unpredictability of their effectiveness. Today, EOR produces less than 15% of petroleum in the United States, with gas injection accounting for about 40% of this EOR production. Worldwide, only 1% (or approximately 1 million barrels of oil per day) of world production comes from gas injection projects. But the potential for EOR techniques, in particular CO2 injection and in situ combustion projects, is high. For example, in the United States alone there are roughly 125 billion barrels of heavy oil (corresponding to about 18 years of U.S. petroleum consumption) that could be targeted by in situ combustion. CO2 injection can potentially be applied to a large number of reservoirs. Also, CO2 injection is a potential means to sequester carbon dioxide from power plants and other energy facilities.

1.3. Challenges in Reservoir Fluid Flow Simulation


Clearly, accurate and efcient simulation of reservoir uid ows presents many challenges. Feasible grid-block sizes are generally large compared to at least some of the length scales associated with the geological characterization or uid ow processes, and multiscale approaches are necessary (see section 3). Geological models may include O(107 108 ) cells for a typical reservoir, whereas practical industrial models typically contain O(105 106 ) grid blocks (depending on the type of model), with the model size often determined such that the simulation can be run in a reasonable time frame (i.e., overnight) on the available hardware. Also, rock properties are uncertain. Geostatisticians create a statistical ensemble of possible permeability distributions that reect measurements from a range of scales and resolutions (such as seismic studies, drill core samples, production data, and models of the geological structure). Monte-Carlo or stochastic approaches are necessary to assess the effects of this uncertainty on performance, dramatically increasing the computational costs of a reservoir simulation study. In compositional simulation, component transfer between phases introduces very strong nonlinearities in the equations describing transport and increases the computational requirements signicantly, as discussed in section 4. But, despite the modeling and computational challenges and inherent uncertainties, reservoir uid ow simulation has been widely adopted by the petroleum industry as a reservoir management tool.

216

GERRITSEN

DURLOFSKY

2. MODELING RESERVOIR FLUID FLOW


Crude oil generally contains some amount of dissolved gas and invariably occurs in conjunction with water. In many cases it is acceptable to assume that the oil and gas compositions are xed, and that the solubility of the gas in the oil depends on pressure only. Then it is possible to consider a single oil pseudo-component and a single gas pseudo-component. If oil and gas equilibrium compositions vary strongly as a function of space and time, a compositional formulation is needed that includes a larger number of (pseudo-) components and appropriate equations of state. We rst consider the general case of multiple components and then discuss some simplied representations. We label the n p phases present in the reservoir with the subscript j , and may refer to specic phases using the subscripts a , l , or v for aqueous, liquid, and vapor. The local volume fraction of phase j is referred to as the saturation S j . Components are designated by the subscript i . A key quantity is the mass fraction of component i in phase j , which we designate yi, j .

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

2.1. Compositional and Black-Oil Models


Fully compositional models must be used when the uid ow depends strongly on component transfer between phases. In fact, many EOR processes, including miscible gas injection, are specically designed to take advantage of the phase behavior of multicomponent uid systems. Compositional modeling is also required in modeling depletion and/or cycling of retrograde reservoirs and reservoirs with highly volatile oils. In these cases, the phase compositions are away from the critical point, which simplies the behavior of the uid system. If we consider transport to occur by both convection and dispersion, the mass balance for component i in an isothermal system gives (Acs et al. 1985) t j yi, j S j +
j j

c = 0, i = 1, . . . , n c , ( j yi, j u j + j S j Di, j yi, j ) + m (1)

where is porosity (volume fraction of void space); j is the phase density; u j is the phase Darcy velocity, dened as the supercial velocity (volumetric ow rate divided by area) of phase j ; Di, j is the dispersivity of component i in phase j ; c is the mass source term (where the tilde indicates per unit volume); and n c is m the number of components. The equations in Equation 1 are also referred to as the transport equations. The Darcy velocity is related to the phase pressure gradient via the semiempirical representation uj = kr j k p j , j j = 1, . . . , n p . (2)

In Equation 2, kr j is the relative permeability to phase j , j is the viscosity of phase j , p j is the pressure of phase j , and n p is the number of phases. The

FLUID FLOW IN OIL RESERVOIRS

217

absolute permeability tensor k appears in Equation 2 describing ow, and not in the transport equations. The phase pressures can differ as a result of capillary pressure effects. Equation 2 can be generalized to include the effects of gravity through the introduction of the phase potential j in place of p j . For single-phase ow, kr j = 1 and Equation 2 defaults to the single-phase Darcys law. For multiphase ows, kr j < 1, due to the interference between phases and the resulting reduction in total ow for a given p . Relative permeability is generally a strong function of one or more of the phase saturations but does not depend directly on uid ow properties (Lake 1989). However, when miscibility develops, relative permeabilities vary with IFT. Extending the single-phase Darcys law to multiphase ow using relative permeability functions is generally accepted by the reservoir simulation community, although, as pointed out in a previous Annual Reviews article (Adler et al. 1988), it is an incomplete model that is valid only at small capillary numbers. Alternative models have been proposed in recent years (Gray & Hassanizadeh 1998), but these have yet to be explored for reservoir simulation applications. The design of three-phase relative permeability models is an area of active research. Blunt (2000) presents a good overview of the most commonly used empirical models for three-phase relative permeability. Most simulators still use empirical correlations for three-phase relative permeability based on two-phase relative permeability data. Besides the IFT effect, three-phase relative permeabilities may also depend strongly on hysteresis between imbibition and drainage (Lenhard & Parker 1987). A unied theory for handling the effects of IFT and hysteresis on three-phase relative permeabilities does not currently exist. Phase viscosities are generally given by empirical correlations that consider the viscosity a function of the mole fractions and molar density for the phase. The polynomial Lohrenz-Bray-Clark correlation (Lohrenz et al. 1964) is popular because of its simplicity and ease of implementation. It was developed primarily for low-density gases, but alternative (and generally more complex) models are available (Aziz & Settari 1979). If we consider a system of n h hydrocarbon components plus a water component (total number of components n c = n h + 1) existing in n p phases (typically n p = 3), there are a total of n p (n c + 2) equations and unknowns. The unknowns in each grid block include a total of n c n p values for yi, j and n p values each of S j and p j . The governing equations include n c conservation equations of the form of Equation 1, n p 1 capillary pressure relationships (which relate the pressure in one phase to that in a reference phase), n p phase constraints ( i yi, j = 1 for each phase j ), one saturation constraint ( j S j = 1), and n c (n p 1) equilibrium relationships describing how each component partitions between phases. Of the n p (n c + 2) unknowns, only n c are primary unknowns. The remaining unknowns can be computed algebraically once the primary unknowns are determined. There are many possible sets of primary variables that can be used for compositional problems. Cao (2002) reviews several of these formulations. To calculate the mole fractions of the components in each phase, we assume that chemical equilibrium is achieved instantaneously. An adequate thermodynamic

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

218

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

description requires a cubic equation of state (EOS) such as the Soave-RedlichKwong equation (Soave 1972) or the Peng-Robinson equation (Peng & Robinson 1976). Typically, solving these EOSs comprises the bulk of the computational work in a compositional simulation. Extensive research in the last decade resulted in a much-increased efciency of the algorithms. For a thorough discussion of thermodynamic models and their computational complexity, see Michelsen & Mollerup (2004). When the oil is not overly volatile, the phase behavior representation can be simplied and the full compositional model is not required. The system can then be represented in terms of two hydrocarbon pseudo-components, designated oil and gas. These are dened as the phases at stock tank (standard) conditions. This formulation is called a black-oil model. In its most general form this model can include the effects of mass transfer of all components between phases (e.g., oil can vaporize into the vapor phase), although in many cases the only interaction considered is gas dissolved in the oil phase. Refer to Aziz & Settari (1979) for full details of this formulation.

2.2. Oil-Water Model


The compositional and black-oil models described above can be further simplied for two-phase immiscible displacements (no mass transfer between phases). The resulting formulation is useful both for describing some real cases (dead-oil systems) and for developing and studying numerical solution procedures. Additional simplication arises if we neglect the effects of uid and rock compressibility (generally acceptable for liquid systems) as well as capillary pressure. The latter assumption is often appropriate in models describing displacements at the length scales associated with reservoir simulation grid blocks. By summing the conservation equations for oil and water, we form the pressure equation as t , (kt ( Sw ) p ) = q where we have introduced the total mobility t , dened via t ( Sw ) = kr w ( Sw ) kr o ( Sw ) + . w o (4) (3)

j = m j / j and the total volumetric source term is q t = q w + In Equation 3, q o . The pressure equation as written is elliptic in the absence of compressibility. q Because the total mobility depends on saturation, the pressure eld changes as the displacement evolves. The total velocity ut is dened as the sum of the water and oil Darcy velocities (ut = uw + uo ) and can be computed from the pressure via ut = t k p . The water Darcy velocity can be expressed as uw = f ( Sw )ut , where f ( Sw ) is the Buckley-Leverett fractional ow function given by f ( Sw ) = w /t , with w = kr w /w . This yields the water conservation equation

FLUID FLOW IN OIL RESERVOIRS

219

Sw w , + [ut f ( Sw )] = q t

(5)

also called the water saturation equation or the water transport equation. It describes the movement of saturation fronts in the reservoir and is purely hyperbolic as written. Including capillary pressure effects renders this equation parabolic. Representing more general compositional or black-oil systems in this way is also possible, as we discuss below.

2.3. Numerical Formulations


Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Practical models contain of the order of 106 (black-oil model) or 105 (compositional simulation) grid blocks, with the model size often determined such that the simulation can be run in a reasonable time frame (i.e., overnight). In compositional simulation, the system of appropriate primary variables can be solved using varying degrees of implicitness. We now briey discuss various numerical formulations (our intent is not to provide a detailed review of numerical procedures for the general system). The most stable approach is a fully implicit solution technique but this leads to large matrices (containing n c primary unknowns per grid block), which are often ill-conditioned. Consequently, fully implicit methods are limited foremost by the number of components. To reduce the level of implicitness, and thus solve smaller linear systems, the system can be reformulated into a pressure equation (which can be viewed as an overall volume balance) plus a sequence of n c 1 component conservation equations (Coats 2000, Trangenstein & Bell 1989, Wong et al. 1990). The conservation equations have a strongly hyperbolic character, and are therefore amenable to explicit solution. The pressure equation is always solved implicitly. The overall technique is called an IMPEC procedure (implicit in pressure, explicit in composition). The IMPEC methods are limited by, often severe, stability restrictions on the time step size, but the solutions do not suffer less smoothing than fully implicit methods, which can strongly affect performance prediction of compositional problems (see section 4). Intermediate degrees of implicitness are also possible and often provide enhanced efciency. One such technique is an implicit pressure and saturation (IMPSAT) procedure, in which pressure and n p 1 saturations (but not compositions) are determined implicitly (Cao & Aziz 2002). Adaptive implicit (AIM) approaches enable varying degrees of implicitness, with some grid blocks treated more implicitly than others (Thomas & Thurnau 1983). For example, blocks near fronts might be treated fully implicitly (or via IMPSAT), whereas the bulk of the domain is treated using IMPEC. Computational efciency can also be improved through parallelization, or reducing the number of (pseudo-) components in the simulation. However, the latter may lead to inaccurate representation of the physics. Recently, efforts were made to include adaptive mesh renement (AMR) in compositional simulation (e.g., Sammon 2003). AMR focuses computational effort in regions near displacement fronts to accurately capture the local displacement efciency.

220

GERRITSEN

DURLOFSKY

Although there has been much progress in recent years, conventional simulators based on nite difference or nite element discretization of the three-dimensional governing system of equations may not provide solutions for realistic compositional problems in reasonable time frames, even after code parallelization and adding AMR capability. We discuss recently developed solution methods for compositional simulation of gas injection processes based on Euler-Lagrangian-type approaches in section 4. A basic issue in any simulator is the numerical representation of the ow terms in the governing equations. The block-to-block ow in simulators is determined by the transmissibility, which relates interblock ow to the pressure difference between neighboring blocks. Standard nite difference (or nite volume) techniques are based on diagonal-tensor permeabilities and orthogonal coordinate systems. Under these assumptions, the calculation of interblock transmissibilities, which account for varying permeability, is straightforward. For example, transmissibility in the 2(k )i +1/2 yi z i x -direction between two blocks i and i + 1 is given as (Tx )i +1/2 = x x , i +1 + x i where xi , yi , and z i represent the grid-block dimensions and (k x )i +1/2 is the weighted harmonic average of (k x )i and (k x )i +1 . Extensions to handle more general systems, such as those with discontinuous full-tensor permeability elds, were presented, within a nite volume context, by Aavatsmark et al. (1996), Edwards & Rogers (1998) and Lee et al. (2002). Numerous nite element procedures have also been developed for solving the reservoir ow equations. Mixed nite element methods are well suited for reservoir simulation because they provide local conservation over the primary grid. Many of the original applications of these and other nite element methods are described in the extensive review by Russell & Wheeler (1983). Some recent work has focused on the use of multiblock models (e.g., Wheeler et al. 2002) and multiscale methods (discussed below in section 3.2). Alternate procedures, based on the solution of stochastic equations (in which porosity and permeability are treated as random variables), are widely studied within the context of groundwater hydrology and are also under study for reservoir simulation (Jarman & Russell 2003, Zhang et al. 2000). These approaches are, however, not yet sufciently developed for application in practical reservoir simulation studies. Outstanding issues include difculties encountered when applying these approaches to general permeability elds (e.g., elds characterized by multipoint geostatistics) and the inherent complexity associated with the incorporation of additional physics into the stochastic formulation.

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

3. COARSE-SCALE MODELS OF HETEROGENEOUS RESERVOIRS


The equations presented above are strictly valid when ne-scale effects are fully resolved. However, we are interested in solutions over grid blocks that are large compared to at least some of the length scales associated with the geological characterization (as is evident from the ne-scale detail illustrated in Figure 1b).

FLUID FLOW IN OIL RESERVOIRS

221

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

The coarse-scale modeling procedures considered here fall within two broad (but related) classes: upscaling techniques and multiscale methods. We view upscaling techniques as methods in which coarse-scale equations of a prescribed analytical form are solved. This form may differ from the underlying ne-scale equations. It is either derived via homogenization or volume-averaging procedures, developed based on physical arguments, or, as is often the case in practice, simply assumed to be the same (or nearly so) as the ne-scale equations. The coefcients in these upscaled equations appear explicitly and are typically computed in a preprocessing step from appropriate solutions over regions of the ne-scale model. By contrast, in multiscale methods, the coarse-scale equations are not expressed analytically, but rather are formed and solved numerically. Fine-scale information may be carried throughout the simulation and used at various stages, and different grids are generally used for ow and transport computations.

3.1. Computation of Upscaled Parameters


Upscaling within the context of reservoir simulation has been the subject of intensive research. Several reviews of numerical upscaling procedures have appeared in recent years (Christie 1996, Farmer 2002, Renard & de Marsily 1997, Wen & G omez-Hern andez 1996). These reviews describe a wide range of procedures and should be consulted for an exhaustive coverage of recent work. Many related issues are also discussed in the hydrology and applied physics literatures. Upscaling methods can be classied in terms of the types of parameters that are upscaled (single- or two-phase ow parameters) and in terms of the way in which the upscaled parameters are computed (e.g., using local or global calculations). It is often possible to develop coarse models of reasonable accuracy for multiphase ow with only the single-phase ow parameters (e.g., absolute permeability) upscaled. In other cases, two-phase ow parameters [e.g., kr j and pc (capillary pressure)] are also effectivized. In any upscaling method, the two main issues that must be addressed are the analytical form of the coarse-scale equation and the way in which the parameters in the coarse-scale model are computed.
3.1.1. COARSE-SCALE MODEL FOR SINGLE-PHASE FLOW

Here we consider steady, single-phase incompressible ow with no source terms. The system is characterized via an idealized two-scale model for permeability k(x, y) that varies on two distinct scales: the slow scale x and the fast scale y. Homogenization procedures, applied within the context of reservoir simulation by, for example, Bourgeat (1984) allow this fully resolved k(x, y) to be replaced by an effective permeability tensor k (x) in certain cases. This means that effects on the y-scale need not be resolved in solving the global pressure equation, which leads to signicant computational savings. The upscaling of k(x, y) to k (x) is mathematically valid only when the region over which k is computed is large relative to the fast (y) scale of variation (e.g., the correlation length of the heterogeneity eld). In addition, special treatment is required near boundaries or sources. In cases where these restrictions are not

222

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

satised, upscaled permeability can still be computed, but it is more properly referred to as the equivalent grid-block permeability tensor rather than an effective permeability [see Durlofsky (1991) for more discussion of this point]. In purely local upscaling methods, the problem solved to determine k (x) involves only the ne cells that reside in the coarse block of interest. In general, a local problem must be solved for each coarse block in the domain. Common choices for local boundary conditions are xed pressureno ow conditions or periodicity. The coarse-scale permeability k can then be computed through u = k p , with the averaged velocity u and averaged pressure gradient p . The averages can be computed as either outlet averages or volume averages. If outlet averages are used, the cross terms of k are undetermined if xed pressureno ow conditions are applied. Using volume averages enables the determination of the off-diagonal terms of k for any boundary condition, although the symmetry of k is not generally guaranteed. Symmetry can be approximately recovered by using a least square procedure (Wen et al. 2003a). Note that k may also be computed based on averages of other quantities (e.g., Zijl & Trykozko 2001). Periodic boundary conditions have some desirable features in that the resulting k is guaranteed to be symmetric and positive denite (Boe 1994, Durlofsky 1991, Pickup et al. 1994). In addition, the computed k is the same with either outlet or volume averaging. Any local method can alter large-scale permeability connectivity, which can sometimes lead to error (as is illustrated below). Upscaled transmissibility (T ) can be computed directly during the upscaling step (as opposed to being determined from the k in a subsequent calculation). In this case, the purely local problem may be shifted such that it is centered on the target interface rather than the target block. The upscaled transmissibility can then be calculated as the ratio of the total ux through the interface to the imposed pressure drop.
3.1.2. NONLOCAL CALCULATION OF COARSE-SCALE SINGLE-PHASE FLOW PARAMETERS

The purely local upscaling approach discussed above can be improved on by using nonlocal information when calculating k or T . Extended local procedures introduce the effects of neighboring regions in the upscaling computations. Boundary conditions are imposed on the edges of the extended region (target block plus neighboring cells), although averaged quantities and the resulting upscaled parameters are usually computed only over the target coarse block or interface. Extended local upscaling procedures have been applied by numerous researchers (e.g., G omez-Hern andez & Journel 1994, Holden & Lia, 1992, Wen et al. 2003a, Wu et al. 2002). Methods of this type have the advantage of reducing the effect of the assumed boundary conditions on the upscaled quantities and provide improved coarse-scale descriptions over purely local methods in many cases. However, for some types of geological models, such as channelized systems with highly discontinuous permeability descriptions, these methods still may not provide results of great accuracy. In such cases it may be necessary to introduce global ow information into the upscaling procedure.

FLUID FLOW IN OIL RESERVOIRS

223

There are various global single-phase parameter upscaling methods (Holden & Nielsen 2000, Moulton et al. 1998, Pickup et al. 1994, White & Horne 1987). These methods have the apparent limitation of requiring a global ne-scale solution for the particular ow scenario. However, this may not pose a major problem in cases for which the ultimate goal is a two-phase or multiphase ow simulation because the global computation we are considering here is a single-phase steady state solution. A limitation of global procedures is that the coarse-scale parameters found are not always positive. This problem can be circumvented by applying optimization procedures with the coarse-scale parameters constrained to be positive (Holden & Nielsen 2000). An approach that avoids some of the difculties associated with global methods, while still effectively using global ow information, is the recently developed local-global method of Chen et al. (2003). This method uses global coarse-scale pressure information to provide boundary conditions for the extended local calculation of k or T . The method is iterated until self-consistency between the upscaled properties and the global ow eld is achieved, which typically requires only a few iterations. In reservoir simulation, global ow is almost always driven by wells rather than xed-pressure or xed-rate boundary conditions. Therefore, it is important that near-well effects be captured in coarse-scale models. Near-well upscaling is complicated by the fact that the pressure gradient, which is assumed constant in local upscaling procedures, is not constant in the well region. Several techniques have been presented to address near-well upscaling (Ding 1995, Mascarenhas & Durlofsky 2000, Muggeridge et al. 2002) and they are effective in practice. The local-global procedure described above has also been extended to capture the effects of wells on the coarse scale. An example of the performance of local-global upscaling on a complex permeability eld is shown in Figure 2. The permeability eld (Figure 2a) derives from a channelized geological model (see Christie & Blunt 2001 for details). This is a particularly difcult example because the complex connectivity in the high permeability channels is easily lost with simpler upscaling techniques. The ne-scale model in this case is 220 60 grid blocks and the upscaled model is 22 6. We specify xed pressures over portions of the left and right edges of the model (this boundary specication is representative of partially penetrating wells) and compute the total ow rate Q through the system. The initial estimate for Q, computed on a coarse model generated using an extended local transmissibility upscaling method, is in error by 58%. After two iterations of the local-global procedure, this error is reduced to 2.3%. This illustrates the upscaling errors that can result from the application of local procedures in highly heterogeneous systems and the improvement that can be obtained by introducing global coupling into the upscaling calculation. Using ow-based and other gridding procedures can signicantly improve the quality of the coarse model. A number of procedures based on single-phase ow calculations (e.g., Portella & Hewett 2000, Wen et al. 2003b and references therein) enhance solution accuracy. Alternate procedures based on the solution of elliptic

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

224

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Figure 2 (a) Channelized permeability eld and inow regions (red indicates high permeability, blue indicates low permeability). (b) Total ow rate through the system for coarse models generated using extended local (iteration 0) and local-global upscaling procedures. The ne-scale result is matched closely after only 1 or 2 iterations.

grid generation equations have not been extensively investigated within the context of reservoir simulation, although such techniques may be useful.
3.1.3. COARSE-SCALE MODELS FOR TWO-PHASE FLOW

A key requirement for accurate coarse-scale models of two-phase or multiphase systems is that they be built on top of accurate single-phase upscaling procedures. This is because absolute permeability (or transmissibility) appears in two-phase ow problems in much the same way as in single-phase systems. The simplest two-phase parameter upscaling entails the calculation of so-called pseudo-relative permeability functions. Procedures based on pseudo-relative permeability involve the use of coarse-scale equations of the same form as the ne-scale equations. The upscaled relative c c permeability and capillary pressure functions, designated kr j ( S ) and pc ( S ), are c generally functions of S (saturation on the coarse scale) only. The kr j may vary in each coarse grid block and may also be direction dependent. Because relative permeability (and pseudo-relative permeability) functions are typically stored as tables of data, the additional information that must be carried when using pseudofunctions can be substantial. A number of different local and nonlocal procedures for computing upscaled two-phase ow parameters have been developed. Barker & Dupouy (1999) and Darman et al. (2002) presented descriptions and comparisons of various procedures. The upscaling of two-phase parameters may be especially important in homogenizing small-scale effects (e.g., cm scale) up to geocellular-scale descriptions (e.g., 10 m). At very small scales capillary forces are important, and the variation of capillary pressure with rock type (facies) can have a dominant effect on the upscaled quantities. Numerous investigators have addressed this issue; see, e.g., Pickup & Stephen (2000) and Virnovsky et al. (2004). c Complications arise in two-phase parameter upscaling because the kr j ( S ) are often sensitive to the local boundary conditions used in the upscaling computation.

FLUID FLOW IN OIL RESERVOIRS

225

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

This lack of robustness has been addressed by using extended local and global methods as well as effective ux boundary conditions, which attempt to approximate the local velocity eld using analytical approximations (Wallstrom et al. 2002). Portella & Hewett (2000) and Aarnes & Espedal (2002) applied global single-phase ow solutions to determine appropriate boundary conditions for calculating upscaled relative permeabilities. These and other methods have been successful in some cases, but the potentially strong boundary condition dependence of upscaled relative permeabilities represents an important practical limitation. The lack of robustness may also be due in part to the limited functionality of the coarsescale equation (i.e., the form of the coarse model is incomplete). Recently, some investigators attempted to address this issue. Within stochastic frameworks, Langlo & Espedal (1994) and Lenormand & Fenwick (1998) introduced models that contain dispersive terms in addition to a modied convective ux function f ( S c ). Efendiev & Durlofsky (2003) recently extended these ideas to coarse-scale deterministic models. The dispersive terms represent small-scale subgrid effects and appear even in systems with no diffusion on the ne scale. This model was more robust than standard procedures for the coarse grid modeling of transport, in some cases.

3.2. Multiscale Methods


Multiscale nite-element and nite-volume approaches make up an important class of techniques designed to account for ne-scale information in coarse-scale solutions. Most methods discussed below provide a specialized treatment of the ow (pressure) equation only. In most cases the transport equations are solved on the ne grid after reconstruction of the ne grid velocity eld based on the solution of the ow equation. Because the computational costs of the ow equation are generally (much) higher than those of the transport equations in noncompositional models, owing to implicitness and ill-conditioning, ne-grid transport solves can generally be afforded in such cases. Thus, these methods require that both a coarse and ne grid be used, and that ne-scale data be retained during the computations. Within the context of reservoir simulation, most of the multiscale methods to date have been based on nite element procedures. Hou & Wu (1997), in a key paper, presented a technique that entails the construction of multiscale basis functions from local solutions of the pressure equation. They applied an oversampling procedure in which the local basis functions are computed using a small amount of ne-scale information from neighboring elements. This approach is somewhat akin to using border regions when calculating upscaled permeability. Efendiev (1999) coupled the multiscale nite element solution with a coarse-scale model for unit mobility ratio transport (i.e., transport of a passive scalar) to give a model that did not require the global ne-scale solution of the saturation equation. The transport model was based on a moment expansion (Efendiev et al. 2000) and included a dispersivity term that is driven by subgrid velocity uctuations (determined from the multiscale pressure solution). In subsequent work, Chen & Hou (2003) presented a mixed multiscale method. Unlike the method of Hou & Wu (1997), this method retains local mass

226

GERRITSEN

DURLOFSKY

conservation and is therefore better suited for transport calculations. Chen & Hou simulated unit mobility ratio transport on the ne scale using the reconstructed velocity eld as well as the coarse grid dispersivity model of Efendiev et al. (2000). Chen & Yue (2003) extended the multiscale method to account for well-driven ow. Aarnes (2004) reported further developments involving a mixed multiscale method and well effects. Arbogast (2002) and Arbogast & Bryant (2002) developed variational multiscale nite element methods. Their approach differs from the methods discussed above in that they represent the discrete solution as the sum of a coarse grid component and a subgrid component. Different mixed nite element basis functions represent these two components. Arbogast & Bryant (2002) demonstrated numerical results of good accuracy for some highly heterogeneous permeability descriptions. Another nite element technique with a multiscale component is the mortar upscaling method of Peszynska et al. (2002). This approach was formulated within a multiblock context and can handle nonmatching subdomains as well as different physical models in different domains. Jenny et al. (2003) introduced a multiscale nite volume procedure that represents a generalization of previous ux-continuous techniques (e.g., Lee et al. 2002) to account for subgrid effects. Flux-continuous nite volume techniques require local solutions, which enforce pressure and ux continuity between cells, to determine the nite volume stencil. Jenny et al. included ne-scale information in these local solutions to provide coarse-scale transmissibilities that incorporate the effects of the underlying nescale permeability. Multiscale methods that incorporate velocity reconstruction and the subsequent solution of the transport equation on the ne grid are in some ways analogous to previously developed dual-grid procedures (e.g., Gautier et al. 1999, Gu erillot & Verdi` ere 1995). These techniques entail an upscaling step, the coarse grid solution of the pressure equation, velocity reconstruction, and the solution of the transport equation on the ne grid. In most dual-grid procedures, standard (local) upscaling techniques are applied. Using a more sophisticated upscaling procedure leads to more accurate reconstructed velocity elds and hence more accurate ne-scale saturation solutions for highly heterogeneous systems (Chen et al. 2003). In comparing the accuracy of upscaling and multiscale procedures, it is necessary to distinguish between the impact of the treatment of the ow (pressure) equation, whether multiscale or upscaled, and the effect of the scale on which the transport equations are solved. In many cases it is the solution of the transport equations on the ne-scale rather than the multiscale treatment of the pressure equation that is largely responsible for the signicant improvement in accuracy over procedures that use coarse grids for both ow and transport.

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

3.3. Key Challenges in Coarse-Scale Reservoir Modeling


Although there has been extensive progress in the efcient modeling of ow and transport in heterogeneous reservoirs, many outstanding issues remain.

FLUID FLOW IN OIL RESERVOIRS

227

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Specically, robust and practical techniques for the coarse-scale representation of transport (i.e., the saturation equation) have yet to be presented. Much less work has been directed toward the upscaling of three-phase ow or compositional problems, although there is a clear need for such capabilities. Applying upscaling and multiscale procedures in conjunction with complex unstructured grids, as might be required, for example, to resolve complex geology and multilateral wells, is still in its early stages. Grids of this type are now used in many reservoir simulation applications, so it is important that compatible upscaling techniques be developed. Using dynamic coarse grids (either fully adaptive or occasionally updated) in conjunction with upscaling procedures is another area that might be usefully pursued. It will also be of interest to effectively combine multiscale (reconstruction) techniques with fully coarse-scale models. Then upscaled representations could be applied in some portions of the domain, whereas a multiscale representation could be used in key active regions. This sort of general framework was previously suggested (e.g., E et al. 2003), although the development of such methods within the context of reservoir simulation has not yet been attempted.

4. MODELING MISCIBLE GAS INJECTION PROCESSES


In this section we focus on the challenging physics and simulation of miscible gas injection processes. Because of the relatively high costs of such processes, as discussed in section 4.3, reliable performance prediction tools are of great interest to the petroleum industry.

4.1. Miscible Gas Injection


When gas injected at sufciently high pressure contacts oil, components in the oil transfer into the gas (or vapor) phase, and components in the gas dissolve in the oil to establish chemical equilibrium. The resulting liquid and vapor phases move through the porous medium at different ow velocities that depend on the local pressure gradient and the viscosity and saturation of each phase. The vapor phase encounters oil with a composition different from the equilibrium liquid composition created at rst contact, and additional component transfers occur. In this way a series of composition changes may be induced, with the most volatile components being displaced faster. In the transition zone the IFT and capillary pressure between the vapor and liquid phases are reduced, thus improving the mobility of the oil. The residual oil saturation is reduced signicantly as compared to water ooding because the injected gas can vaporize and transport components trapped in the immobile oil that sits in the crooks and crevices. When the pressure is at or above the Minimal Miscibility Pressure (MMP), the two phases are fully miscible. Then the IFT and capillary pressure are reduced to zero, and a very high, piston-like local displacement efciency is achieved.

228

GERRITSEN

DURLOFSKY

We illustrate the behavior of miscible gas injection processes in one dimension. Because gas injection processes are generally very strongly advection dominated, we ignore physical diffusion and hydrodynamic dispersion. The conservation equations (Equation 1) describing the transport of the components can be written as Ci (v D Fi ) + = 0, i = 1, . . . , n c .
n

(6)

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

p The total molar density of component i is given by Ci = j = 1 yi , j j S j , with np n p the number of phases, and the total molar ux is v D Fi = v D j = 1 yi , j j f j . Here j , S j , and f j refer to the molar density, saturation and fractional ow of each phase, respectively. The total dimensionless velocity is given by v D = v/vin j , where vin j denotes the constant injection velocity. The velocity is determined from the ow (pressure) equation. The dimensionless time and spatial variables are

vin j t , L

x , L

with length of the medium L and distance from the inlet x . The injection takes place at = 0. The partitioning of the components over the phases is determined by solving phase equilibrium equations in the multiphase regions. The fractional ow functions depend on relative permeabilities and viscosities of the phases. For two-phase ows, fl = 1 f g , and the fractional ow of the gas is given by f = krg ( S ) , krg ( S ) + g krl ( S ) l

where S is the gas saturation. We discussed the relative permeability functions in section 2.1. The commonly used two-phase relative permeability functions retain the hyperbolic character of the two-phase system (Equation 6). However, most three-phase relative permeability models used today introduce regions with elliptic behavior when applied to one-dimensional immiscible three-phase ow (with capillary effects neglected). Juanes & Patzek (2004) give an overview of this interesting and controversial phenomenom. In the last 10 years, much progress was made in formulating (semi-) analytical solutions for the hyperbolic system (Equation 6) with constant initial and boundary conditions (the Riemann problem), computed by a characteristic decomposition of the system. Jessen et al. (2001) give solutions for two-phase, multicomponent systems. Solutions for miscible three-phase ow are currently limited to three components.

4.2. An Illustration: Two-Phase Three-Component System


For illustration purposes we show solutions for two-phase, three-component ows with a constant partitioning of components between phases; that is, with constant

FLUID FLOW IN OIL RESERVOIRS


y

229

equilibrium ratios K i = yii,,g , which are also known as the K-values. For constant l K-values and under ideal mixing, v D is constant and equal to 1 (in the general case, v D varies in time and space due to volume change on mixing as a result of component transfer between phases). We inject a mixture of C1 (methane) and C2 (ethane) into a mixture of C2 and C3 (propane). Figure 3a (left) shows solution proles for C1 and C2. In Figure 3a (right) the composition path of the solution is plotted in a ternary phase diagram. The vertices of the triangle represent pure component mixtures and the opposite edges correspond to the binary mixtures lacking that component. Although phase space does not explicitly show the time dependence of the solution, it lends insight into the behavior of the system that cannot be determined by examining proles alone, and it offers an attractive validation of numerical methods as used in, for example, Mallison et al. (2003). The solutions generally consist of segments of shocks and rarefactions, whose number increases with n c . The eigenvectors of the system give candidate directions in phase space along which the composition path may vary, and the eigenvalues give the propagation speeds. The correct composition path is determined by requiring that the path satisfy certain physical rules, akin to the entropy condition for shocks in the Euler equations. In Figure 3a (right) the dew and bubble point curves of the three-component system are drawn as two key tie lines that extend through the initial composition (a) and the injection composition (f). Transitions to and from the two-phase region occur as shocks (ab and ef) along tie line extensions. In the two-phase region the solution consists of short rarefactions along the key tie lines, a rarefaction connecting the key tie lines and a zone of constant state that appears as the point (d). At the transition point (c) the propagation speeds of both rarefactions are equal (equal eigenvalue point). Multicomponent systems are weakly hyperbolic because the eigenvectors associated with equal eigenvalue points are not independent, which poses extra challenges to the numerics. The number of key tie lines increases with n c . Each additional key tie line introduces an additional nontie line path, a zone of constant state, and possibly a tie line rarefaction. For a given gas/oil displacement, the lengths of the key tie lines dictate the efciency of the displacement (in 1D). When all of the tie line lengths are long, two-phase ow dominates and the displacement is inefcient. If one of the key tie lines is short, then the phases that form will be similar in composition and the overall efciency of the displacement will increase. In the limit that a tie line has length zero, miscibility develops and an optimal, piston-like displacement is achieved.

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

4.3. Key Challenges in Simulating Gas Injection Processes


Although the local displacement efciency can be very high in miscible gas injection processes, the injected gas may contact only a small portion of a reservoir because low-viscosity gas nds the high-permeability ow paths. In addition, gravity can cause low-density gas to override the oil in place. So, the global sweep efciency of a gas ood may not be high. The process performance of gas injection

230

GERRITSEN

DURLOFSKY

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Figure 3 Solution proles (left) and phase diagrams (right) for a two-phase threecomponent illustrative example. (a, top) Exact solutions to the Riemann problem. (b, middle) Numerical solutions for a rst-order upwind method (400 grid points). (c, bottom) Numerical solutions for the third-order ENO scheme (400 grid points).

FLUID FLOW IN OIL RESERVOIRS

231

schemes depends on this balance between local displacement efciency and global sweep efciency, and both need to be captured accurately by a performance prediction tool. This presents three substantial challenges for compositional reservoir simulators. First, because the global sweep depends foremost on the underlying heterogeneity, high-resolution reservoir models are required. Solver efciency is paramount because the computational costs per grid cell are high owing to the potential large number of unknowns and the high expense of thermodynamic equilibrium calculations. Alternatively, subgrid models could be developed for compositional models as indicated in section 3. Second, the component transfer between phases strongly affects the prediction of local displacement efciency. Thus, the number and type of (pseudo-) components can strongly affect the accuracy of the simulation. Although 6 or more pseudo-components often sufce, 10 to 15 components are sometimes needed to match phase behavior. Physical models, such as viscosity and relative permeabilities, can also affect predictions. This is especially true in three-phase ows where predicted recovery can range widely depending on the selection of relative permeability models alone. Third, the equation of state introduces a very strong nonlinear coupling between the uid ow equations, which renders the simulation sensitive to modeling and discretization errors. Figure 3b,c shows the solution proles and the composition path of the example problem computed with a rst-order upwind scheme (commonly used in reservoir simulation) and a third-order essentially nonoscillatory (ENO) scheme, respectively (Mallison et al. 2003). The errors made by the diffusive rst-order scheme in capturing the leading shock speed are mainly due to the error in composition path, and originate at the equal eigenvalue point. The composition of the gas phase at the leading shock is less rich in the second component and hence displaces oil less efciently. Oil predicted by the analytical solution to be displaced is left behind and accumulates in the zone of constant state, as the observed dip in the solution proles in Figure 3b (left). The ENO scheme captures the phase behavior near the equal eigenvalue point more accurately, leading to much improved results. In this simple example the errors are not very large. In realistic cases, the nonlinearities in the system are much stronger and numerical errors are greater (Mallison et al. 2004). Also, the number of grid points (400) used in these 1D computations are much larger than can be used in practice.

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

4.4. Novel Compositional Simulation Methods for Gas Injection


The combination of high grid resolution and accurate representation of phase behavior required for reliable performance predictions leads to very high computational expense. Parallelization and adaptive mesh renement strategies improve efciency but generally not to the required degree. As an alternative to the more traditional Eulerian methods described in section 2.1, we mention two Euler-Lagrangian-type methods that show promise for

232

GERRITSEN

DURLOFSKY

gas injection processes: the Euler-Lagrange Localized Adjoint Method (ELLAM) and the compositional streamline method. For compositional problems, both approaches are still in early stages of development. ELLAM was rst developed by Celia et al. (1990) for solving constant-coefcient advection-diffusion problems, and has subsequently been applied to various linear and nonlinear transport problems. It was specically designed to conserve mass while taking large time steps, and to allow systematic treatment of any type of boundary condition. It also naturally accommodates spatial and temporal renements, which is attractive for problems with propagating fronts such as gas injection processes, but efforts in this direction have so far been limited to one dimension. As Russell & Celia (2002) mention, there are a few obstacles to wider use of ELLAM: the absence of a robust mechanism to control nonphysical oscillations and the need to improve efciency of current implementations. Chen et al. (2000) and Qin et al. (2000) reported a rst extension to compositional simulation. Applying ELLAM to the compositional problem requires linearization, which affects phase behavior. Finding a suitable linearization, especially in near-miscible displacements, is an outstanding research problem. In heterogeneous reservoirs the time scale at which uids ow along streamlines is generally much shorter than the scale at which the streamline locations change signicantly. This observation motivates decoupling of the transport problem into a sum of one-dimensional problems along streamlines that can be solved independently and very efciently between pressure updates, as discussed in the overview paper by King & Datta-Gupta (1998). Streamline methods have been used successfully in predicting the global sweep of water oods in heterogeneous reservoirs, and have found applications as fast screening tools in optimizationand history-matching problems (see, e.g., Batycky et al. 1997). Cross-ow is ignored between pressure updates, but efcient corrections have been developed for capillary and gravity effects (Bradvedt et al. 1996). Very recently, the streamline methods were extended to compositional simulation for gas injection processes that are generally strongly dominated by heterogeneity (Jessen & Orr 2002, Mallison et al. 2003). The phase equilibrium equations are solved in points on the streamlines. Because the streamline solves are independent and thus inherently parallel, the computational efciency of compositional streamline simulation is signicantly higher than that of Eulerian approaches. High-order methods, such as the ENO scheme discussed in section 4.2, can be applied along streamlines, which, together with adaptive mesh renement, further improve efciency. In contrast to the ELLAM methods, the full nonlinear formulation of the mass balance equations is used and phase behavior can be adequately captured, but streamline methods are not mass conservative. So, in some sense, the methods are complementary. Intense research efforts are ongoing in both solution approaches. Figure 4 shows a two-dimensional example of a near-miscible CO2 injection into a heterogeneous reservoir simulated using compositional streamline simulation (Mallison 2004). In Figure 4a the two-dimensional permeability eld is given, which was taken from layer 19 of the Tenth SPE Comparative Solution

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

FLUID FLOW IN OIL RESERVOIRS

233

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Figure 4 (a) Permeability eld (in Darcy). The extent of the domain is 670 m (220 grid cells) in the x-direction and 266 m (60 grid cells) in the y-direction. (b) Sample streamline distribution during CO2 injection. (c) Mole fraction of CO2 after 1300 days of injection.

Project (Christie & Blunt 2001). A sample streamline distribution during simulation is shown in Figure 4b, which illustrates that the highly mobile gas seeks high permeability ow paths. We injected CO2 along the left boundary x = 0 at a total rate of 5m 3 per day. Production takes places at the right boundary x = 670 m, where the pressure was xed at 235 atm. Zero ux boundary conditions were applied at the other boundaries. The third-order ENO scheme was used to solve the transport equations along the streamlines. The mole fraction of CO2 in

234

GERRITSEN

DURLOFSKY

the reservoir after 1300 days of injection (Figure 4c) shows high local displacement efciency but a low global sweep as a result of the strongly heterogeneous medium.

5. ADDITIONAL RESEARCH DIRECTIONS


There are many important practical aspects of reservoir simulation that are not discussed in this review. For example, using so-called smart wells (i.e., wells equipped with downhole sensors and chokes) can substantially increase oil production while simultaneously decreasing the production of unwanted water or gas. Accurate representation of these wells in simulators poses a challenge, as does determination of optimum choke settings. The modeling of naturally fractured reservoirs is another active area of research. This problem is complicated because of the wide range of length scales that appear in the problem and because of the need to accurately model transfer between the fractures and the matrix. In current practice, interacting continua (e.g., dual porosity) procedures are commonly used, although discrete fracture modeling techniques are beginning to nd some application. Furthermore, the simulation of coupled multiphase ow and geomechanics is important for predicting oil recovery as well as for determining the onset of well or facility failure. This problem is demanding because the ow and geomechanical effects, although tightly coupled, require different numerical treatments. We hope that this article has provided an understanding of the physics of uid ow in oil reservoirs and an overview of challenges faced when modeling reservoir uid ows. Finally, it is important to emphasize that research in reservoir simulation is largely driven by practical concerns. The future directions of the technology will therefore be determined, at least in part, by advances in geological modeling and by developing new recovery processes and oil eld technologies.
The Annual Review of Fluid Mechanics is online at http://uid.annualreviews.org

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

LITERATURE CITED
Aarnes J, Espedal MS. 2002. A new approach to upscaling for two-phase ow in heterogeneous porous media. In Fluid Flow Transport Porous Med.: Mathematical and Numerical Treatment, vol. 295, ed. Z Chen, RE Ewing. Providence, RI: AMS Contemp. Math. Aarnes J. 2004. On the use of a mixed multiscale nite element method for greater exibility and increased speed or improved accuracy in reservoir simulation. SIAM Multi. Mod. Sim. 2:42139 Aavatsmark I, Barkve T, Boe O, Mannseth T. 1996. Discretization on non-orthogonal, quadrilateral grids for inhomogeneous, anisotropic media. J. Comp. Phys. 127:2 14 Acs G, Doleschall S, Farkas E. 1985. General purpose compositional model. Soc. Petrol. Eng. J. 25:54352 Adler PM, Brenner H. 1988. Multiphase ow in porous media. Annu. Rev. Fluid. Mech. 20:3559

FLUID FLOW IN OIL RESERVOIRS Arbogast T. 2002. Implementation of a locally conservative numerical subgrid upscaling scheme for two-phase Darcy ow. Computat. Geosci. 6:45381 Arbogast T, Bryant SL. 2002. A two-scale numerical subgrid technique for waterood simulations. Soc. Petrol. Eng. J. 7:44657 Aziz K, Settari A. 1979. Petroleum Reservoir Simulation. Condon: Appl. Sci. Publ. Barker JW, Dupouy P. 1999. An analysis of dynamic pseudo-relative permeability methods for oil-water ows. Petrol. Geosci. 5:385 94 Batycky RP, Blunt M, Thiele MR. 1997. A 3D eld-scale streamline based reservoir simulator. Soc. Petrol. Eng. Res. Eng. 12:246 54 Blunt MJ. 2000. An empirical model for threephase relative permeability. Soc. Petrol. Eng. J. 5:43545 Boe O. 1994. Analysis of an upscaling method based on conservation of dissipation. Transport Porous Med. 17:7786 Bourgeat A. 1984. Homogenized behavior of two-phase ows in naturally fractured reservoirs with uniform fractures distribution. Comput. Method. Appl. Mech. Eng. 47:205 16 Bradtvedt F, Gimse T, Tegnander C. 1996. Streamline computations for porous media ow including gravity. Transport Porous Med. 25(1):6378 Cao H, Aziz K. 2002. Performance of IMPSAT and IMPSAT-AIM models in compositional simulation. Presented at the Soc. Petrol. Eng. Annu. Tech. Conf. Ex., San Antonio, Texas Cao H. 2002. Development of techniques for general purpose simulators. PhD thesis. Stanford Univ. Celia MA, Russell TF, Herrera I, Ewing RE. 1990. An Eulerian-Lagrangian localized adjoint method for the advection-diffusion equation. Adv. Water Resour. 13:187206 Chen Y, Durlofsky LJ, Gerritsen MG, Wen XH. 2003. A coupled local-global upscaling approach for simulating ow in highly heterogeneous formations. Adv. Water Resour. 26:104160

235

Chen Z, Qin G, Ewing RE. 2000. Analysis of a compositional model for uid ow in porous media. SIAM J. Appl. Math. 60:74777 Chen Z, Hou TJ. 2003. A mixed multiscale nite element method for elliptic problems with oscillating coefcients. Math. Comput. 72:54176 Chen Z, Yue X. 2003. Numerical homogenization of well singularities in the ow transport through heterogeneous porous media. SIAM J. Multi. Mod. Sim. 1:260303 Christie MA. 1996. Upscaling for reservoir simulation. J. Petrol. Tech. 48:100410 Christie MA, Blunt MJ. 2001. Tenth SPE comparative solution project: a comparison of upscaling techniques. Soc. Petrol. Eng. Res. Eval. Eng. 4:30817 Coats KH. 2000. A note on IMPES and some IMPES-based simulation models. Soc. Petrol. Eng. J. 5:24551 Darman NH, Pickup GE, Sorbie KS. 2002. A comparison of two-phase dynamic upscaling methods based on uid potentials. Computat. Geosci. 6:527 Ding Y. 1995. Scaling up in the vicinity of wells in heterogeneous eld. Presented at Soc. Petrol. Eng. Res. Sim. Sym., San Antonio, Texas Durlofsky LJ. 1991. Numerical calculation of equivalent grid block permeability tensors for heterogeneous porous media. Water Resour. Res. 27:699708 Edwards MG, Rogers CF. 1998. Finite volume discretization with imposed ux continuity for the general tensor pressure equation. Computat. Geosci. 2:25990 Efendiev YR. 1999. The multiscale nite element method (MsFEM) and its applications. PhD thesis. Calif. Inst. Technol. Efendiev YR, Durlofsky LJ, Lee SH. 2000. Modeling of subgrid effects in coarse scale simulations of transport in heterogeneous porous media. Water Resour. Res. 36:2031 41 Efendiev YR, Durlofsky LJ. 2003. A generalized convection-diffusion model for subgrid transport in porous media. SIAM Multi. Mod. Sim. 1:50426

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

236

GERRITSEN

DURLOFSKY lems in subsurface ow simulation. J. Comp. Phys. 187:4767 Jessen K, Wang Y, Ermakov P, Zhu J, Orr FM Jr. 2001. Fast, approximate solutions for 1D multicomponent gas-injection problems. Soc. Petrol. Eng. J. 6(4):44251 Jessen K, Orr FM Jr. 2002. Compositional streamline simulation. Presented at the SPE Annu. Tech. Conf. Ex., San Antonio, Texas Juanes R, Patzek TW. 2004. Relative permeabilities for strictly hyperbolic models of threephase ow in porous media. Trans. Porous Med. 57:12552 King MF, Datta-Gupta A. 1998. Streamline simulation: a current perspective. In Situ 22(1):91140 Lake LW. 1989. Enhanced Oil Recovery. Englewood Cliffs, NJ: Prentice-Hall Langlo P, Espedal MS. 1994. Macrodispersion for two-phase, immisible ow in porous media. Adv. Water Resour. 17:297316 Lee SH, Tchelepi HA, Jenny P, de Chant LJ. 2002. Implementation of a ux-continuous nite-difference method for stratigraphic, hexahedron grids. Soc. Petrol. Eng. J. 7:267 77 Lenhard RJ, Parker JC. 1987. A model for hysteretic constitutive relations governing multiphase ow, 2. permeability-saturation relations. Water Resour. Res. 23(12):2197 206 Lenormand R, Fenwick D. 1998. MDU: a model for dynamic upscaling. In Proc. Int. Energy Agency Meeting, Carmel, Calif. Lohrenz J, Bray BG, Clark CR. 1964. The viscosity of pure substances in dense liquid and gaseous phases. J Petrol. Technol. 1:1171 76 Mallison BT. 2004. Streamline-based simulation of two-phase, multicomponent ow in porous media. PhD thesis. Stanford Univ. Mallison BT, Gerritsen MG, Jessen K, Orr FM Jr. 2003. High order upwind schemes for twophase multicomponent ows. Presented at the Soc. Petrol. Eng. Res. Sim. Symp., Houston, Texas Mascarenhas O, Durlofsky LJ. 2000. Coarse scale simulation of horizontal wells in

E W, Engquist B, Huang Z. 2003. Heterogeneous multiscale method: a general methodology for multiscale modeling. Phys. Rev. B 67:92101 Farmer CL. 2002. Upscaling: a review. Int. J. Numer. Method. Fluids 40:6378 Gautier Y, Blunt MJ, Christie MA. 1999. Nested gridding and streamline-based simulation for fast reservoir performance prediction. Computat. Geosci. 3:295320 Gerritsen MG, Kovscek A, Castanier L, Nilsson J, Younis R, He B. 2004. Experimental investigation and high resolution simulator of insitu combustion processes; 1. simulator design and improved combustion with metallic additives. Presented at the Soc. Petrol. Eng. Int. Therm. Oper. Heavy Oil Sym., Bakerseld, Calif. G omez-Hern andez JJ, Journel AG. 1994. Stochastic characterization of grid block permeability. Soc. Petrol. Eng. Formation Eval. 9:9399 Gray WG, Hassanizadeh SM. 1998. Macroscale continuum mechanics for multiphase porous media ow including phases, interfaces, common lines and common points. Adv. Water Resour. 21:26181 Gu erillot DR, Verdi` ere S. 1995. Different pressure grids for reservoir simulation in heterogeneous reservoirs. Presented at Soc. Petrol. Eng. Res. Sim. Symp., San Antonio, Texas Holden J, Nielsen BF. 2000. Global upscaling of permeability in heterogeneous reservoirs: the output least squares OLS method. Transport Porous Med. 40:11543 Holden L, Lia O. 1992. A tensor estimator for the homogenization of absolute permeability. Transport Porous Med. 8:3746 Hou TY, Wu XH. 1997. A multiscale nite element method for elliptic problems in composite materials and porous media. J. Comp. Phys. 134:16989 Jarman KD, Russell TF. 2003. Eulerian moment equations for 2-D stochastic immiscible ow. SIAM Multi. Mod. Sim. 1:598608 Jenny P, Lee SH, Tchelepi HA. 2003. Multiscale nite-volume method for elliptic prob-

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

FLUID FLOW IN OIL RESERVOIRS heterogeneous reservoirs. J. Petrol. Sci. Eng. 25:13547 Michelsen ML, Mollerup JM. 2004. Thermodynamic Models: Fundamentals and Computational Aspects. Holte, Denmark: TIE-LINE Publ. Moulton JD, Dendy JE, Hyman JM. 1998. The black box multigrid numerical homogenization algorithm. J. Comput. Phys. 142:80 108 Muggeridge AH, Cuypers M, Bacquet C, Barker JW. 2002. Scale-up of well performance for reservoir ow simulation. Petrol. Geosci. 8:13339 Peng DY, Robinson DB. 1976. A new twoconstant equation of state. Ind. Eng. Chem. Fund. 15:5964 Peszynska M, Wheeler MF, Yotov I. 2002. Mortar upscaling for multiphase ow in porous media. Computat. Geosci. 6:73100 Pickup GE, Ringrose PS, Jensen JL, Sorbie KS. 1994. Permeability tensors for sedimentary structures. Math. Geol. 26:22750 Pickup GE, Stephen K. 2000. An assessment of steady-state scale-up for small-scale geological models. Petrol. Geosci. 6:20310 Portella RCM, Hewett TA. 2000. Upscaling, gridding, and simulation using streamtubes. Soc. Petrol. Eng. J. 5:31523 Qin G, Wang H, Ewing RE, Espedal MS. 1997. Numerical simulation of compositional uid ow in porous media. In Lecture Notes in Mathematics 606. Heidelberg: SpringerVerlag Renard P, de Marsily G. 1997. Calculating effective permeability: a review. Adv. Water Resour. 20:25378 Russell TF, Wheeler MF. 1983. Finite element and nite difference methods for continuous ows in porous media. In The Mathematics of Reservoir Simulation, ed RE Ewing, pp. 35106. Philadelphia: SIAM Russell TF, Celia MA. 2002. An overview of research on Eulerian-Lagrangian localized adjoint methods (ELLAM). Adv. Water Resour. 25:121531 Sammon PH. 2003. Dynamic grid renement and amalgamation for compositional simu-

237

lation. Presented at the Soc. Petrol. Reser. Sim. Symp., Houston Sarathi P. 1999. In-situ combustion handbookprinciples and practices. DOE/PC/ 91008-0374 Rep., OSTI ID: 3174, National Petroleum Technology Ofce, U.S. D.O.E. Soave G. 1972. Equilibrium constants from a modied Redlich-Kwong equation of state. Chem. Eng. Sci. 27:1197203 Thomas GW, Thurnau DH. 1983. Reservoir simulation using an adaptive implicit method. Soc. Petrol. Eng. J. 23:75968 Trangenstein JA, Bell JB. 1989. Mathematical structure of compositional reservoir simulation. SIAM J. Sci. Stat. Comput. 10:817 45 Virnovsky GA, Friis HA, Lohne A. 2004. A steady-state upscaling approach for immiscible two-phase ow. Transport Porous Med. 54:16792 Wallstrom TC, Hou S, Christie MA, Durlofsky LJ, Sharp DH, Zou Q. 2002. Application of effective ux boundary conditions to twophase upscaling in porous media. Transport Porous Med. 46:15578 Wen XH, Durlofsky LJ, Edwards MG. 2003a. Use of border regions for improved permeability upscaling. Math. Geol. 35:521 47 Wen XH, Durlofsky LJ, Edwards MG. 2003b. Upscaling of channel systems in two dimensions using ow-based grids. Transport Porous Med. 51:34366 Wen XH, G omez-Hernandez JJ. 1996. Upscaling hydraulic conductivities in heterogeneous media. J. Hydrol. 183:932 Wheeler JA, Wheeler MF, Yotov I. 2002. Enhanced velocity mixed nite element methods for ow in multiblock domains. Computat. Geosci. 6:31532 White CD, Horne RN. 1987. Computing absolute transmissibility in the presence of ne-scale heterogeneity. Presented at Soc. Petrol. Eng. Res. Simul. Symp., San Antonio, Texas Wong TW, Firoozabadi A, Aziz K. 1990. Relationship of the volume-balance method of compositional simulation to the

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

238

GERRITSEN

DURLOFSKY two-phase ow in heterogeneous reservoirs. Soc. Petrol. Eng. J. 5:6070 Zijl W, Trykozko M. 2001. Numerical homogenization of the absolute permeability using the conformal-nodal and the mixed-hybrid nite element method. Transport Porous Med. 44:3362

Newton-Rhapson Method. Soc. Petrol. Eng. Reserv. Eng. 5:41522 Wu XH, Efendiev YR, Hou TY. 2002. Analysis of upscaling absolute permeability. Discrete Cont. Dyn. Sys. B 2:185204 Zhang DX, Li LY, Tchelepi HA. 2000. Stochastic formulation for uncertainty analysis of

Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

Annual Review of Fluid Mechanics Volume 37, 2005

CONTENTS
Annu. Rev. Fluid Mech. 2005.37:211-238. Downloaded from www.annualreviews.org by Colegio de Postgradudados COLPOS on 10/19/11. For personal use only.

ROBERT T. JONES, ONE OF A KIND, Walter G. Vincenti GEORGE GABRIEL STOKES ON WATER WAVE THEORY,
Alex D.D. Craik

1 23 43

MICROCIRCULATION AND HEMORHEOLOGY, Aleksander S. Popel


and Paul C. Johnson

BLADEROW INTERACTIONS, TRANSITION, AND HIGH-LIFT AEROFOILS IN LOW-PRESSURE TURBINES, Howard P. Hodson
and Robert J. Howell

THE PHYSICS OF TROPICAL CYCLONE MOTION, Johnny C.L. Chan FLUID MECHANICS AND RHEOLOGY OF DENSE SUSPENSIONS, Jonathan
J. Stickel and Robert L. Powell

71 99 129 151 183 211 239 263 295 329 357 393 425 457

FEEDBACK CONTROL OF COMBUSTION OSCILLATIONS, Ann P. Dowling


and Aimee S. Morgans

DISSECTING INSECT FLIGHT, Z. Jane Wang MODELING FLUID FLOW IN OIL RESERVOIRS, Margot G. Gerritsen
and Louis J. Durlofsky

IMMERSED BOUNDARY METHODS, Rajat Mittal


and Gianluca Iaccarino

STRATOSPHERIC DYNAMICS, Peter Haynes THE DYNAMICAL SYSTEMS APPROACH TO LAGRANGIAN TRANSPORT IN OCEANIC FLOWS, Stephen Wiggins TURBULENT MIXING, Paul E. Dimotakis GLOBAL INSTABILITIES IN SPATIALLY DEVELOPING FLOWS: NON-NORMALITY AND NONLINEARITY, Jean-Marc Chomaz GRAVITY-DRIVEN BUBBLY FLOWS, Robert F. Mudde PRINCIPLES OF MICROFLUIDIC ACTUATION BY MODULATION OF SURFACE STRESSES, Anton A. Darhuber and Sandra M. Troian MULTISCALE FLOW SIMULATIONS USING PARTICLES,
Petros Koumoutsakos

vii

You might also like