SPE 153758

Calculation of Minimum Miscibility Pressure Using Fast Slimtube Simulation
Wei Yan, SPE, Michael L. Michlesen, Erling H. Stenby, SPE, Technical University of Denmark

Copyright 2012, Society of Petroleum Engineers
This paper was prepared for presentation at the Eighteenth SPE Improved Oil Recovery Symposium held in Tulsa, Oklahoma, USA, 14–18 April 2012.
This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Minimum misciblility pressure (MMP) is a critical parameter in designing a miscible gas injection process. It is expected that
100% displacement efficiency on the microscopic scale can be achieved provided the injection pressure is above MMP. Two
approaches are usually employed for equation of state (EoS) based MMP calculation. The slimtube simulation approach is a
numerical simulation of the physical slimtube experiment, which is commonly accepted as the most reliable experimental
method for MMP determination. This approach carries out slimtube simulation runs at a series of pressures and determines the
MMP from the recovery-pressure curve, just as in the experiment. The global approach, which is based on the method of
characteristics analysis of 1D gas injection, finds the MMP by locating the pressure where a key tie-line becomes critical.
Although the global approach is faster, the slimtube approach is still a necessary complementary method especially when the
injection process involves complex phase behavior caused by CO2 or heavy oils. This study addresses how to improve the
computational efficiency of slimtube simulation for MMP calculation. Firstly, a robust and efficient algorithm for rigorous
flash forms the basis of the whole strategy. Secondly, a tie-line distance based approximation (TDBA) method has been
introduced on top of the rigorous algorithm. In the TDBA method, if a new feed composition in a grid block is close enough to
a tie-line previously calculated in the same block, the previous tie-line results can be used with slight adjustment. The
approximation affects the final recovery very little but greatly increases the speed. Thirdly, a higher order method can be
employed to use a fewer number of grid blocks to get the same accuracy in recovery. Finally, a MMP search strategy is
poroposed to reduce the number of slimtube simulations needed. In addition, it is also discussed how to parallelize slimtube
simulations for modern computers with multiple CPU cores to further chop the computation time.
1. Introduction
Gas injection, a widely used enhanced oil recovey method, will probably play a more important role in the future due to
several reasons: (1) gas injection is effective to recover residual oils and its application is likely to be promoted by higher oil
prices; (2) for deeper reservoirs and particularly offshore deeper reservoirs, gas injection as a means to achieve a higher oil
recovery can be crucial to viability of a project; (3) when combined with CO2 capture, gas injection can be used as a way to
mitigate CO2 emission. During gas injection, the injected gas will swell the oil, reduce oil viscosity, and most importantly,
achieve miscibility by exchanging components with the oil. The extent of miscibility increases with pressure, and the threshold
pressure above which the injected gas becomes completely miscible with the oil is known as the minimum miscibility pressure
(MMP). Above the MMP, 100% displacement efficiency can be expected on the microscopic scale.
Development of miscibility between gas and oil is a dynamic process. In a 1-D conceptual picture, the injected gas keeps
exchanging components with the reservoir fluid as it flows through the oil. The miscibility is a combined effect of local phase
equilibrium and flow of the equilibrium phases—although recent studies show that the MMP is not influenced by the mobility
ratio between gas and oil. Early explanations of miscibility always use pseudoternary phase diagrams where the fluid system is
classified into three pseudo components: a gas component, an intermediate component, and an oil component. This conceptual
model is useful to illustrate vaporizing gas drive (VGD) and condensing gas drive (CGD). In VGD, the injection gas becomes
gradually enriched by extracting intermediate components from the reservoir oil and the miscibility is achieved at the
displacement front between the enriched injection gas and the original reservoir oil. In CGD, the injection gas continuously
enriches the reservoir oil by transferring the intermediate components to the oil and the miscibility is achieved near the
injection well between the enriched reservoir fluid and the injection gas. However, the actual miscibility development can be
more complex than pure vaporizing or condensing. This was revealed by Zick (1986) who showed that many gas injection
systems involve both vaporizing and condensing processes. The combined condensing/vaporizing (C/V) mechanism can no
longer be represented by a pseudo ternary phase diagram. The MMP due to a C/V mechanism can be significantly lower than

5. where N is the number of grid blocks. including the initial tie-line. The intersecting key tie-lines are also a good approximation for practical MMP calculation (Wang and Orr.. (1973) extended the model of Cook et al. there exist nc-1 key tie-lines if the solution consists only of shocks. The second type is to simulate the slimtube experiment using either a multicell simulator or a 1-D numerical simulator. he suggested extrapolating the recovery to infinite small grid block size using a plot of the recovery versus 1/N0. and determinating MMP as the pressure corresponding to a recovery of 97% on the exponential curve fitted to the recoveries at different pressures. (1986). Different methods have been proposed for MMP calculations with an equation of state (EoS). Høier and Whitson evaluated that the code was 20 to 100 times faster than multiple slimtube simulations. there exists a third key tie-line called the cross-over tie-line for a quaternary system (Monroe et al. He noticed that the mixing caused by numerical dispersion can be larger than the mixing caused by physical dispersion. (1973). Jaubert et al. The key tie-lines will intersect pairwise for solutions consisting only of shocks. and nc-3 cross over tie-lines. which describes the distance from the critical point in each cell. namely moving excess oil and moving oil and gas according to their phase mobilities. to study miscibility development rather than gas cycling. In addition to the “stagnant oil” option used by Cook et al. gave a more sophisticated description on how to move gas and oil from one cell to the next. (2006) to illustrate that neither the ratio between injected gas and the oil at each batch nor the fluid mobility affects the MMP. Their approach is essentially the same as that of Metcalfe et al. we put both the multicell approach and the 1-D FD approach under the same “slimtube simulation” category since they essentially simulate slimtube experiments. The minimum threshold pressure where multiple contacts just lead to a single phase result is interpreted as the MMP. and continues to any number of cells that satisfy a desired accuracy. The minimum of this physical parameter in each cell is extrapolated to the infinite grid number to get its dispersion free value. MMP can be determined in the laboratory by a slimtube experiment. 2000). 1996). they also introduced two other options. The MMP is usually determined as the break point on the recovery-pressure plot.. Zick tracks the development of miscibility with a physical parameter. The intersecting key tie-lines was used to study MMP for a four-component system (Dindoruk. However. Recently. 1990). To remove the artificial dispersion. Ahmadi and Johns (2008) proposed a different multiple mixing-cell method for MMP calculations. The experiment is time consuming but there are no other easy and reliable experimental methods to substitute it. The earliest attempts in this direction can be traced back to Cook et al. The method does not account for the C/V mechanism and is therefore of limited use. 1986). forms the basis for later studies on miscibility using the multicell approach. The first type is the singlecell approach where the VGD or CGD mechanism is simulated by forward or backward contact in a single cell (Jensen and Michelsen. but with a few improvements for faster MMP calculation. Metcalfe et al. Wang and Orr (1997) proposed the first general algorithm for multi-component systems which also allows injection gas with more than one component. 1998b) proposed a multicell algorithm to compute the MMP for any types of miscibility mechanisms. for a nc component gas injection systems. The work of Metcalfe et al. According to Høier and Whitson (2001). Details on the 1-D analytical theory of gas injection and their applications can be found in Orr’s book (Orr. where extrapolation using 1/N was tried. such as extrapolating the recovery at infinite cells with a single simulation run. Orr et al. Stalkup proposed to study the 1-D displacement behavior with a compositional simulator. 1998). (1998) substaintially increased the speed of MMP calculation for gas injection systems with an arbitrary number of components in either oil or gas. Determination of the MMP using the single-cell approach is equivalent to searching the pressure where the initial tie-line or the injection tie-line becomes critical. In general.2 PVI (pore volume injected) for this 1-D gas injection process. Analysis of 1-D two-phase gas injection problems using the method of characteristics (MOC) shows that in addition to the initial tie-line and the injection tie-line. 1992) and later extended to fluid with 11 components (Johns and Orr. such as the work of Pedersen et al. (1993) showed that the multicomponent miscibility happens when a key tie-line becomes critical.2 SPE 153758 that due to a VGD or CGD mechanism (Zick. A computer program was developed to simulate oil vaporization during gas cycling in a linear laboratory-model oil reservoir. 1990). Metcalfe et al. the multicell approach (also known as the multiple mixing cell simulation or the cell-to-cell simulation) and the 1-D FD simulation approach are actually based on the same formulation for 1-D gas injection. Metcalfe’s multicell approach was also used by Zhao et al. Zick also pointed out that for multicomponent reservoir fluid systems it is not practical to achieve a CGD type miscibility. Zick’s algorithm is similar to the multicell approach proposed by Cook et al. The former provides a physical interpretation of the 1-D gas injection process while the latter provides a mathematical solution to the problem (Jessen. In parallel to the multicell approach is the 1-D compositional simulation approach used by Stalkup (1987). In their method. setting the pressure search region by the saturation pressure and the first contact miscibility pressure. The third type of MMP calculation methods is based on the 1-D analytical theory of gas injection. The equivalence between a cell-to-cell simulation and an implicit single point upstream FD scheme was also implicated in the study of Michelsen (1998). (1998a. (1969) and Metcalfe et al. Jessen et al. The MMP is searched by an automated code to locate where the distance parameter becomes zero. Høier and Whitson (2001) made a comphrensive study of different slimtube simulation algorithms. . The experiment consists of gas injection at a series of pressures into the reservoir oil in a long slimtube packed with sand or glass beads. Cook et al. Although appearing to be two different approaches. Each run at a pressure provides a correponding recovery at 1. Here. such as the difference of phase densities. 2007). the mixing starts with only one gas cell and one oil cell. The transfer of gas and oil between cells has been largely simplified in this method. (1969) suggested using 20 cells as a simplified 1-D reservoir model. including Zick’s multicell algorithm. and Metcalfe et al. the injection tie-line. Stalkup (1990) and Stalkup and Dean (1990) further investigated the effect of numerical dispersion and the sensitivity of gridding in finite difference (FD) simulations of the gas injection process.

Governing equations for 1-D gas injection For 1-D horizontal gas injection under the assumptions of non-deformable porous media. are not essential since one can easily simplify the 1-D FD simulation or complicate the multicell simulation. In addition. And the similarity between Ahmadi and John’s multicell approach and an ESPU scheme is discussed in Section 2. nc (1) i  1. First. capillary effects. Mathematical formulation and physical interpretation of slimtube simulation In the literature. The two types of simulation are based on the same mathematic formulation as outlined in Section 2.1. Phase equilibrium constraints and Eqs. and a simple fractional flow function is adopted. The multicell simulation stresses more on the physical interpretation of the process while the 1-D FD simulation seeks to solve the problem from a mathematical point of view. such as pressure solution. Ci is the overall molar density of component i and Fi is the overall molar flux of component i: np Ci  Ci (z )   xij  j s j (5) j 1 np Fi  Fi (z. For .(2) can be solved using different numerical schemes. 2. 1-D FD simulation is often distinguished from multicell simulation. We then present a comprensive strategy to improve the simulation speed.3. and realistic relative permeability models. appropriate approximation in flash calculations to cut simulation time. dispersion effects. and a proper MMP searching stategy to reduce the number of required slimtube simulations. no capillary effects. and Eqs. 2.1. we review the mathematic formulation of the 1-D gas injection problem and present different numerical schemes for the problem.(2) thus constitute the underlying equations for 1-D gas injection. application of a higher order method to reduce numerical dispersion. In this study. we use slimtube simulation as a general name for both multicell simulation and 1-D FD simulation.SPE 153758 3 The purpose of this study is to speed up slimtube simulations for MMP determination.(2). There are some arguments supporting this distinction. multicell simulation usually uses a tailor-made simulator where pressure is assumed constant. . 1-D FD simulation can employ an existing compositional simulator and can include more details.2. These differences. and negligible dispersion and diffusion effects. nc (2) or Ci Fi  0   with dimensionless time τ and dimensionless distance ξ defined as   vinj t L  PVI (3) x L (4) In Eqs. The proposed algorithm was tested with 18 gas injection systems and the results are compared with those from the MOC approach. capillary and dispersion effects are neglected. the mass conservation equations are given by  n    np    p    v  xij  j f j   0 x  s  ij j j  x  j 1  t  j 1    i  1. In contrast. however. The equivalence between a cellto-cell simulation and an implicit SPU scheme is discussed in Section 2. vD )  vD  xij  j f j (6) j 1 The dimensionless velocity vD is defined as vD  v vinj (7) It is usually assumed that the gas injection is isothermal and local mass transfer is quick enough to reach instantaneous phase equilibrium. . Fast MMP determination using slimtube simulation depends on several factors including an efficient thermodynamic algorithm. The phase equilibrium part can be solved by isothermal flash. the advance in computer hardware can be utilized to parallelize slimtube simulations.

(2) is given by (Cin. *n 1 to the next cell. and those fractions in  vD .k Fk*n 1 flux. k 1Fk 1 .3.kn 1  vD . (9) are both at grid block k and time level n+1.k Fk*n 1 on the LHS of Eq. Flux from cell k at tn+1 Flux from cell k-1 at tn+1  vD .k Fk*n 1 for the fluid leaving the grid block. giving an intermediate stage with overall mass of Ck   vD .k 1 Fi*. The removed extra volume is equivalent to the downstream If we assume the cell volume is unity.2. For cell k in Figure 1. 2. Equivalence between an Implicit SPU scheme and a cell-to-cell simulation: the blocks with the same color have the same equilibrium phase properties (phase densities and phase compositions). its overall mass after batch n is Cnk . The flux from its upstream  vD . can be used for Cnk 1 and  vD . we transfer the extra volume  vD . After flashing the mixture at this intermediate n stage.k Fk*n 1 .k Fk*n 1 + Cnk 1 Material in cell k at tn+1 Figure 1.k 1Fk*n11 (9) or in a vector form n 1 It is worth noting that the two terms Ck and  vD . 2. an expliit SPU method. k )   (vD . The phase densities and phase mole fractions in those two terms should be exactly the same.k Fk*n 1  Cnk   vD .k 1Fk*n11 is added to this cell. including phase densities and phase compositions. However.k    vD . (9) by using its RHS as feed to a flash. the volume of the upstream flux will and the volume of the downstream flux will be  vD . kn1  where Fi * is the velocity scaled overall molar flux of component i : (10) .k1  Cin. The equilibrium phase properties from the flash.k 1Fk*n11 Flash for the mixture at the intermediate stage + Cnk Cnk   vD . The mobility ratio between oil and gas determines how to split the equilibrium phases from n 1 the flash into Ck and  vD .k Fi *. k 1 Fi*. Figure 1 dipicts the solution procedure and reveals its equivalence to a cell-to-cell simulation. The vapor/liquid volume n 1 fractions in Ck are actually the final saturations in the block. k . Implicit SPU scheme The implicit SPU (ISPU) discretization of Eqs.4 SPE 153758 finite difference methods.k Fi *. the phase volume fractions in Cnk 1 and  vD . and the mass remaining in the cell is be  vD . or a higher order explicit method with Total Variation Diminishing (TVD) limitation. We can solve Eq.k 1Fk*n11 Material in cell k at tn  vD . Explicit SPU scheme The explicit SPU (ESPU) discretization of Eqs.k1  Cin. one can employ an implicit single point upstream (SPU) scheme.k 1 C n 1 k . Cnk 1 stands for the fluid remaining in the grid block and  vD .k Fk*n 1 are different. kn11 )  0 (8) Cnk 1   vD .kn  vD .(2) is given by Cin.k Fk*n 1 .k Fk*n 1 are the fractional flow functions for the downstream flux.

(1997) applied a TVD scheme with the van Leer limiter to the 1-D compositional simulations along the streamlines in their 3-D compositional streamline simulations. k and nc mole numbers Cin. Using total variation diminishing (TVD) schemes is one of the most prevailing flux limiting methods used for higher order finite difference simulations.(10) ( vD . For the equilibrium gas and liquid phases from a cell in the middle. In the second contact. and nc  1 unknowns in Eqs. There are several flux limiters that can be chosen (Sweby. Thiele et al. The maximum grid number.4. however. Higer order explicit method with TVD limitation 1-D gas injection simulation can suffer from severe numerical dispersion for certain systems.kn   in. k    vD . We can use a a larger number of grid blocks to achieve a desired accuracy. Peddibhotla et al. can be compared to an explicit SPU scheme. k1  Cin. (1997). the equilibrium liquid and gas phases from the first (negative) flash are brought into contact with a newly added cell of injection gas in the upstream and a newly added cell of initial oil in the downstream. illustrated in Figure 1 of Ahmadi and Johns (2008). We can interpret the newly added cell filled with injection gas as the flux from the injector at tn. the equilibrium gas from the upstream cell with the equilibrium oil in the downstream cell. only nc variables are vD . The procedure repeats itself by adding injection gas in the upstream and initial oil in the downstream. and they suggested using the essentially nonoscillatory (ENO) schemes for such simulations. and Peddibholta et al. (1997). k is dependent on Cnk 1 . This complicates the solution procedure slightly since there are unknowns on both sides of the equations.kn1  Fi*. It is interesting to notice that their procedure shows some similarity to an explicit SPU FD scheme. and the number of flashes increases by one after each contact. The third contact consists of three mixings: newly added gas with the equilibrium oil in the upstream cell. we can interpret the former as the flux to the next adjacent cell at tn. 2. Ahamdi and Johns (2008) proposed a multicell algorithm different from the other multicell ones in the literature. and the latter as the content left in the cell at tn.kn1/2  (12) *n where Fi .kn1/ 2  vD .k Fi *.kn1/2  Fi *. is limited by simulation time since the overall time increases with the square of the grid number. The solution procedure starts with just one upstream cell filled with injection gas and one downstream cell filled with initial oil. A different way to construct the flux was proposed by Thiele and Edwards (2001) and the resulting scheme has the nature of a Godunov formulation. k 1/ 2  ri n. Rubin and Edwards (1993). Higher order methods can lead to spurious oscillation when applied to hyperbolic type gas injection equations. Mallison et al. 1997) and we select the van Leer limiter here:  n i . Dindoruk (1992) presents a solution procedure for the ESPU scheme. kn ) (13) with Φ being a TVD limiter. (1991). k . k 1/ 2  ri n. and the equilibrium gas in the downstream cell with newly added oil. (2005) made an in-depth investigation on the higher-order upwind schemes for two-phase. A second order scheme with TVD limitation can be formulated as follows: Cin. Another way to achieve a higher accuracy is to use higher order methods. and the overall mixture is flashed. This procedure. Here. k 1 Fi *.k 1/2 2 ( Fi *. Appendix A reviews the procedure in brief. 1984. and the newly added cell filled with initial oil as the downstream cell at tn which is to be updated. multicomponent flow. which has the innermost loop for vD . The solution procedure resembles an explicit SPU FD scheme except that the way to determine flux is different in their procedure. Datta Gupta et al..k 1/2 and r is a function of the ratio of adjacent flux difference (14) . Applications of the TVD schemes in black oil type simulations can be found in the works by Rubin and Blunt (1991).SPE 153758 5 np Fi   xij  j f j * (11) j 1 It is worth noting that among independent.k 1/2 1  ri n. The two cells are mixed in a certain ratio in the first contact. One approach to remove this instability in higher resolution methods is to introduce flux limiters. k 1/2 is calculated as Fi *. k1 ). respectively. The two resulting mixtures are flashed and give two sets of equilibrium phases. we use the formulation similar to those of Thiele et al.

2. the previous results will be used either directly or with some adjustment. The first one is to solve the Rachford-Rice equation with the old K-values. The same optimization approach has been employed here in the slimtube simulation code for MMP determination. Efficient flash algorithm for slimtube simulation Two-phase isothermal flash is the work-horse in slimtube simulation. Here. a recently proposed approximation method for flash calculations (Belkadi et al. The adjustment can be made in two ways.kn  Fi*. it can be calculated directly. The TVD scheme has almost the same calculation procedure as the explicit SPU scheme.kn If Φ is set to zero. The distance between the new feed composition and the tie-line from the previous flash. The original TDBA uses two levels of tolerances. The computation expenses for two schemes are therefore very similar. The old flash results will be accepted without any change if the error is less than the smaller tolerance. If the new feed composition is close enough to the previous flash in terms of the tie-line distance.  if e ≤ ε. Optimization of two-phase isothermal flash for slimtube simulation was discussed by Michelsen (1998). k 1/2 r  Fi *. Therefore. and adjustment of the old flash results is needed if the error falls in between the two tolerances.k 1  Fi *. The vapor fraction β corresponding to the minimum distance is given by (z  x)T (y  x)  (y  x)T (y  x) (17) Although β is a result of minimization. The only difference is that Φ must be calculated for each grid. the TVD scheme reduces to the explicit SPU scheme. For slimtube simulations.1. kn1 (15) Fi *. a direct Netwon iteration can always be employed for the new flashes and most of them will converge without a problem. 2006).6 SPE 153758 n i . is employed here to further reduce the time used in flash calculations.. is used as a measure to judge whether to make the approximation or not. The unique solution structure of 1-D gas injection also indicates that most single-phase cells will stay in single-phase in the next time step. flash calculations in the single-phase regions are only needed in the “safety zones” around the borders between the two-phase region and the single-phase regions. The solution to 1-D gas injection is characterized by a leading single phase oil region. Fast flash calculations also require efficient thermodynamic subroutines. we do new flash. For the flashes that fail to converge or converge to trivial solutions with the direct Newton approach. 2011). a further cut in calculation time can be made by combining the evaluation of thermodynamic properties for the two phases into a single evaluation of the quantities of interest. named the tie-line distance. Tie-line distance based approximation (TDBA) The tie-line distance based approximation (TDBA) method. Determination of MMP by fast slimtube simulations It calls for a comphrensive strategy to improve the computational efficiency for MMP determination by slimtube simulations. our experiences show that the “safety zone” approach is robust enough. The tie-line distance d from a feed z to a previous tie-line characterized by equilibrium compositions x and y is given by  d 2  e  min  zi    yi  1    xi  i  2 (16) where the error e is equal to d2. a trailing single phase gas region. The essence of the approach is to use Newton’s method whenever possible. In the two-phase region where the previous two-phase results are available.. 3. The flash is proceeded in two different ways depending on the magnitude of e:  if e > ε. For slimtube simulations. The following issues in diverse aspects should be considered: 3. and a two-phase gas-oil region in between. the second one is to adjust the material balance using the vapor split factors defined by      yi  1    xi  old i    yi (18) . The method is based on the fact that the results from a previous rigorous flash in a grid block could be a good approximation to the new flash in the same block. we use old results but adjustment is needed. a new flash is performed if the error is greater than the larger tolerance. the conventional flash with stability analysis can be used to safeguard convergence. we keep just one level tolerance by removing the smaller one. More details on implementing an efficient two-phase isothermal flash in slimtube simulations can be found in Michelsen (1998). A more rigorous approach is to skip stability analysis in the single phase regions is to use the shadow region method (Rasmussen et al. 3.

To obtain such a RF∞ from a direct simulation run is obviously impossible. Selection of numerical schemes Three different numerical schemes are investigated here. The curvature can be quite large for systems with large numerical dispersion. newi (19) The two options are named TDBA1 and TDBA2. ISPU. 3. ISPU is usually a little faster than ESPU or TVD since both ESPU and TVD need to include isothermal flash in their innermost loops. The choice has to be made between a simulation using ISPU with more grid blocks and larger time steps and a simulation using TVD with the same grid number but smaller time steps.5=0. Therefore. which help to reduce the time for flash calculations since more cells tend to have similar results to the previous ones. some gas injection systems can suffer from severe numerical dispersion which can not be completely removed by recovery extrapolation. ESPU and TVD also lead to less dispersed results with shaper fronts and longer constant K-values regions. The stable time step is estimated by   a  CFL  vmax (21) where a is an empirical parameter to address the difference between vmax and the actual fastest propagation speed which needs to account for the maximum eigenvalue for all the compositions. One can make a rough estimate of (vD·λi)max and use a constant Δτ before the simulation. which was also observed by Høier and Whitson (2001). If the same Δτ/Δξ is used. and TVD. For example. (1998a) also used 1/N0. We use the C7+ recovery in this study as the oil recovery. Both TDBA1 and TDBA2 will preserve material balances in their adjustment. 400. as an implicit scheme. For 1-D gas injection simulation with a uniform initial condiiton. Here. Even with the adaptive time stepping.5 provides a buffer large enough for stable simulations. Another issue worth noting is that the recovery versus 1/N0. Extrapolation to RF∞ with a single simulation run Determination of MMP using the recovery-pressure plot needs the oil recovery at 1. respectively. 200. we can either perform simulation with more grid blocks or resort to a higher order FD scheme without increasing the grid number. we can obviously get the fastest results from ISPU. A new regression is made after each removal until such extremum disappears. if the grid number is the same for three schemes. the recovery points will be removed one by one from the smallest N. In many situations. However. On the other hand. at N=100. we can get a series of recoveries.2M/N PVI. To overcome the problem. and 500.5=0. There is . ISPU can use much larger Δτ/Δξ than ESPU or TVD. we only need to calculate the recovery for the first M grid blocks at 1. Our numerical tests show that a =0.3. an extremum may form between the smallest 1/N0. three schemes will give similar simulation time. 300. This is actually not necessary since a simulation run with N grid blocks provides essentially the simulation results for any M grid blocks if M is smaller than N.g. To calculate the largest eigenvalue for all the compositions along the slimtube is not practical. to get the recovery at 1. Stalkup found that extrapolation using the recovery versus 1/N0. namely. a second degree polynomial has been used here in the recovery fitting. Høier and Whitson (2001) and Jaubert et al.4. In the extreme situation of infinite number of grid blocks. the maximum stable time step can be calculated by   CFL  (vD  i ) max (20) where CFL is the Courant-Friedrichs-Lewy number (1 for ESPU and 0. With the NVC assumption. It is well known that FD simulation with a finite number of grid blocks will have numerical dispersion and the dispersion will decrease with increasing number of grid blocks. e. 2005). Therefore.. vmax.2 PVI. the recovery RF∞ is supposed to be dispersion free.5 gave nearly straight lines. Without the NVC assumption. for extrapolation through a single simulation run. In that case. The recovery points at lower N values may affect the fitting for the recovery points at higher N values.5 for TVD) and (vD·λi)max is the maximum propagation speed (Mallison et al. A practical way to estimate RF∞ is to extrapolate the recoveries at different numbers of grid blocks (N) to N=∞. Our experiences show that ESPU and TVD are around 10-20% slower than ISPU at the same Δτ/Δξ for simulations without the NVC assumption. 3.. Extrapolation seems to require the same number of simulation runs as those recovery points used in the extrapolation. ESPU or TVD will have a maximum time step much smaller than ISPU. For ESPU and TVD. there is no need to iterate over the dimensionless velocity vD in ESPU and TVD. the leading front can be readily located and its propagation speed.5 is not always a straight line. can be updated dynamically. and the simulation times for three schemes are very similar.new 1  i  vi  zi .SPE 153758 7 The new vapor and liquid flows are calculated as li  zi .5 and 1/N0. To account for the curvature.2 PVI for M grid blocks.5 in their extrapolation. especially when it approaches the MMP. However. The fitting procedure is made adaptive since not all the recovery points are equally important to the final trend to 1/N0. we use a different approach by calculating Δτ adaptively during the simulation. ESPU.

Peform slimtube simulation at P3 and the recovery is denoted by RF3. which are called repeatedly in slimtube simulation. the dimensionless velocity vD is always treated implicitly (i. we can parallelize the calculations for all the cells. the calculation at tn+1 always starts at the injector and propagates to the producer. This parallelization is easy to implement provided that one knows what pressures to calculate in advance. with all the cells already calculated at tn. and one can easily realize parallel computation of slimtube simulation tasks on a personal computer. It demands parallel computers which used to be hardly accessible or afforadable for simple simulation tasks. 3. the calculation at time level tn+1 depends only on the results at time level tn. Otherwise. set P3=(P1+P2)/2. RF∞ can be higher than 100%). OpenMP is an Application Program Interface (API) that supports multi-platform shared-memory parallel programming in C/C++ and Fortran. this parallelization approach is not appropriate. The extrapolated recovery at infinite cells is denoted by RF2. Since the calculation for a cell is completely independent from the properties of the other cells at the same time level. (3) If RF2 is higher than 100%. Perform slimtube simulation at this pressure. and RF3 using an exponential function RF  exp(a  bP  cP 2 ) (22) (1) Extrapolate to 100% recovery to get the first MMP PMMP .8 SPE 153758 no obvious answer to the question. 3. but to N2. the overall simulation time is proportional not to N. A similar approach can be applied to the ESPU scheme or the TVD scheme without NVC assumptions. we can select a series of pressures to run slimtube simulations. Perform slimtube simulation at P1. while parallelization of the TVD scheme without NVC requires a slight modification since calculation . The extrapolated recovery at infinite cells is denoted by RF1. we can arbitrarily select a cell at tn+1. The simulation runs above the MMP always give RF∞ equal to or higher than 100% (due to extrapolation. It is supported by different compilers. (4) Select P3=P2+ΔP and calculate the expected recovery at P3 by linear extrapolation based on RF1 and RF2. where ΔP can be set to 50 atm. (8) Compare the new MMP with the previous one.. and even perform the calculations in any arbitrary order. Figure 1(a) shows a typical ISPU scheme. A MMP search strategy is suggested as follows: (1) The first pressure P1 is selected as a pressure just above the saturation pressure (e. parallelization of these “explicit” schemes requires a strategy similar to that for the implicit scheme. It seems that TVD is more suitable to systems with severe numerical dispersion. Under such a situation. (2) Select the second pressure P2=P1+ΔP.6. In principle. The most straightforward way is to parallelize the simulations at different pressures since the simulations at different pressures are completely independent. It should be noted that L can be larger than the maximum number of threads available for the machine.e. Actually. For a sequential calculation. To parallelize such a calculation. How to implement parallelization depends on what numerical scheme is used and whether the NVC assumption is employed.5. reduce ΔP by half and go to step 2. The calculation has such a structure that the the cell to be calculated at tn+1 is always one cell ahead of the cell to be calculated at tn+2 so that the cell to be calculated at a higher time level can always get the required information from the lower adjacent time level. (5) Fit the recoveries RF1. advance in manufacturing technology has made multi-core processors readily available. The parallelization is implemented for all the red dots in the L batches. rounded to 0 or 5 atm). Strategy for searching MMP To determine MMP from the break point in the recovery-pressure plot. We have used OpenMP (Open Multi-Processing) to parallelize our Fortan code for slimtube simulation. RF2.. Parallel computation of slimtube simulations can be realized in different ways. and perform a flash. The simplest situation is that both an explicit scheme (including the explicit scheme with TVD) and the NVC assumption are used. Perform slimtube simulation at P2. If the expected recovery is higher than 100%. (22) and calculate the new MMP. where L batches or simulation at L time levels can be performed simultaneously (L=5 in Figure 1(b)). where the cell properties at xk and tn+1 is determined by the cell properties at xk and tn+1 and the flux from the cell xk-1 and tn+1. (6) Select the new pressure as the mid point of the current MMP and the highest pressure that gives a recovery lower than 100%. Therfore. Parallel computation We can apply parallel computation to further speed up slimtube simulation. calculate its overall composition from the fluxes at tn. including the Intel Fortran Compiler used here. Provided that the results are already available at tn. In the search for MMP where simulation is made essentially at one pressure after another. from the current time level). we can use the approach illustrated in Figure 1(b). (7) Fit all the recoveries smaller than 100% with Eq. Parallelization of the ESPU scheme without NVC can take exactly the same propagration style as depicted in Figure 1(b). In the case that there is no NVC assumption. Parallel computation is commonly used for large scale and complex simulations.g. Nowadays. although the scaled flux F* is taken from the previous time level. stop the iteration and output the MMP. If the deviation is larger than a tolerance (1 atm here). The situation is a bit more complicated if an implicit scheme or the NVC assumption is not used. it is usually helpful to use a larger L to counteract the adverse effect of different workload in the flashes in the L batches. Those runs are not helpful to MMP determination and should be avoided if possible. A more general way to realize parallelization is to parallelize the most time consuming flash calculation. continue to step 6.

In this study. Those recoveries are much lower than the extrapolated recoveries.0 secs for TDBA1 (ε=10-8). the higher order TVD scheme gives much closer recovery predictions to the reference ones.0. www. The calculated MMPs are very close to the MMPs predicted by the MOC method using the phase behavior and gas injection software MI-PVT 2. (b) parallelization of ISPU slimtube simulation. which leads to an overshooting of the MMP. even the time-consuming simulation with 20000 grid blocks can not give a correct MMP prediction for System 2. It is obvious that using an ISPU scheme with a large time step is the fastest way to determine the MMP for this system.8 secs for TDBA1 (ε=10-6). we always use a simple relative permeability model krj=sj and a 5:1 viscosity ratio between oil and gas.1.k 1 Fi . Without recovery extrapolation or a higher order scheme. extrapolated recoveries for ISPU and ESPU at the pressures just below the MMP are far below the extrapolated recoveries from reference simulations with ESPU using 20000 grid blocks.4 secs for TDBA2 (ε=10-6). k Fi *. different schemes give almost the same RF∞ values. For cases with large dispersion. The TVD scheme is used and Δτ/Δξ is fixed at 0. Figure 3(b) shows the simulation results with 500 grid blocks and the VC assumption for different numerical schemes. In contrast. However. showing the severeness of the numerical dispersion in this system. For the most dispersive ISPU scheme with Δτ/Δξ as large as 1. k t  vD .5 (see Figure 4(a)) and the last few recoveries corresponding to large N values are crucial to the extrapolated recovery RF∞. The detailed fluid description for System 2 is provided in Table 2. Results and discussion Table 1 presents the major characteristics of the 18 gas injection systems studied here. The modification only requires that the cell to be calculated at tn+1 is two cells ahead of the cell at tn. The simulations are on 1000 grid blocks and the NVC assumption is not used. The calculated MMPs from slimtube simulation . 17. Simulation of these systems shows numerical dispersion to different extent. All our slimtube simulations for MMP determination use 500 grid blocks. The simulation without approximation takes 42. and thus a MMP closer to the true value. Figure 5 compares the results between TDBA1 and the simulation using the “safety zone” method.kn 1  vD . Extrapolation of the recovery to infinite grid blocks using the recovery-1/N0. Implicit Single Point Upstream Gin. 4.nk11 x 0 t x (a) (b) Figure 2. In the following calculation.4 secs. and 21.SPE 153758 9 of the current cell depends on the information from the downstream cell at the previous time level as well. As shown in Figure 4(b). this conclusion for System 1 does not hold for systems with large numerical dispersion like System 2.06 (Tie-Line Technology.com). the TDBA1 option is used in the runs with TDBA. Figures 6 and 7 present the calculation results using the TVD scheme with the NVC assumpation and the algorithm described in Section 3. The yellow dots represent calculated or known cells while the red ones represent the cells to be calculated. Illustration of parallization of the ISPU slimtube simulation: (a) the ISPU scheme.98GB RAM).2 secs for TDBA2 (ε=10-8). in comparison with 17. 21.5 curve is almost a straight line for cases with small dispersion (see Figure 3(a)). For this system with low numerical dispersion. The simulation time can be cut significantly by using TDBA. with an average difference less than 2 atm.k1  Gin. which does not make any approximation in flash. The non-extrapolated recoveries for the simulation with 20000 grid blocks are also plotted in the figure. Therefoere. a reduction in simulation time at least by half is acheived without noticable adverse effects on recoveries and saturation profiles. The calculation was performed on a Lenovo laptop with an Intel® Core™2 Duo CPU (2. TDBA2 gives similar results which are not shown here. the curve is usually curved up at large N values or small 1/N0.5 plot is needed to reduce numerical dispersion.tie-tech. The recovery-1/N0. the recoveries are also similar to the others.40GHz and 1. It can be seen that TDBA1 gives almost the same recoveries and saturation profiles.

However. we illustrate the potential of parallelization using three slimtube simulation tests for System 2. indicating that TDBA can cut the MMP calculation time by half. In the parallel computation.10 SPE 153758 without approximation and with TDBA using two different tolerances (10-6 and 10-8) are essentially the same. and 8 secs. together with the speedup factors. For simulation at six pressures (300. the predicted MMPs are mostly higher than those from the MOC method. On average. For 3000 grid blocks and 6000 time steps. 340. Extrapolation to recovery at infinite cells is essential to reduce numerical dispersion and can be performed with the information from a single simulation run. Using ESPU increases dispersion and does not relax the time step very much. are also beneficial to understanding the real gas injection process. such as the ESPU scheme with TVD. Conclusions Numerical simulation of the slimtube experiment provides a reliable way to calculate MMP since the calculation procedure resembles the experimental MMP determination procedure. At the tolerance of 10-8. The first test is on parallelization of simulation runs at different pressures.8 secs with parallel computation. All the tests were made on a personal computer with an AMD Phenom II X6 model 1035T processor (2. and 3. however. We have also tried MMP determination using ESPU and ISPU. it takes 80 secs without parallel computation and 16. the former requires more memory and is less flexible in a sequential MMP search algorithm. both with TDBA. A fast flash subroutine with an efficient thermodynamic package forms the basis of the algorithm. where all the flash calculations at the same time level can be computed in a parallel mode. the number of grid blocks. For the longer simulation with ε=10-10. The average MMP calculation time for the TVD scheme with TDBA is around 15 secs for those 12~15 components systems on a common laptop (without parallelization). and 13 secs for TDBA with ε=10-6. TDBA can be applied to get essentially the same recoveries at 50% or even shorter simulation time. There are several cases that ESPU and ISPU give results close to the MOC ones but for System 2. ISPU (ε=10-8 and Δτ/Δξ = 0. The average simulation times for ESPU (ε=10-8). and 400 atm). For a simulation run at 300 atm with 1000 grid blocks and the adaptive time stepping. Their calculation results are compared with the TVD results in Figure 8. could be readily utilized for parallel computation in slimtube simulation. One reason for this is that the overhead other than flash calculation is more significant when the simulation time is short. By comparing the three tests.5). The unique structure of the 1-D gas injection solution can be used to skip most stability analysis in the single phase regions and to apply Newton’s method directly in flash calculations in the two-phase region.1 GHz in turbo mode with maximum three cores being active). the MMP search needs simulation runs at 5 pressures. are based on the same mathematical formulation of 1-D gas injection. and the decrease in simulation time will be affected if the number of threads is larger than three. Finally. ISPU allows much larger time steps and we have tried Δτ/Δξ = 0. the speedup performance is better. Development of such a fast algorithm needs a comphrensive effort in many aspects. The overall simulation is not reduced proportionally with the time step. 12. Another reason is that the turbo mode is only available for three cores. and ISPU (ε=10-8 and Δτ/Δξ = 1. and details in the searching strategy. The additional information provided by MMP determination using slimtube simulation. 320. can be greatly reduced by using TDBA (Figure 7). the algorithm has provided an efficient MMP calculation approach complementary to the fast MOC method. The scalability is satisfactory if we consider that the turbo mode is only available for maximum three cores. for instance. It reduces to 18 secs for TDBA with ε=10-8.6 Ghz. which should be based on a large scale MMP calculation tests. the saturation and composition profiles during the simulation show fluctuations for some systems. The third test is on parallelization of an ESPU slimtube simulation with the NVC assumption. An adaptive time stepping is recommended if an explicit scheme is used. The simulation times for ISPU with larger time steps are shorter than that for TVD. The simulation time decreases with the number of threads used but not in a simple inversely proportional relation. If we consider that a single core always runs faster. We have investigated how to speed up MMP determination based on slimtube simulations. Finally. We also point out that the advance in computer hardware.0) are 15. the times needed for a simulation run at 300 atm with TDBA (ε=10-6 and 10-10) are shown in Figure 9. the fluctuations in the saturation and composition profiles are nearly invisible. it seems that direct parallelization of simulation runs at different pressures is the most efficient since the overhead in running OpenMP is larger for parallelization of flash calculations. 380. the number of loops L is set to 60. a significant part of time will be used in the overhead other than flash calculations. ESPU and ISPU give unacceptable large over-predictions for MMP. Due to higher dispersions in these schemes. can be employed to reduce the number of grid blocks needed. but the recoveries are not affected.0. Different variations of the slimtube simulation. the scaling is almost perfect. Our tests using 18 multicomponent gas injection systems have shown that our code developed using the aforementioned principles give the MMP results similar to those from the MOC method. the results using flash with full stability analysis aree shown in Figure 10. including multicell simulations and 1-D FD simulations. It seems more adequate to use a low dispersed TVD scheme than a little faster but high dispersed ISPU scheme for the systems studied here. . 5. The simulation time. At the tolerance of 10-6. The algorithm presented here can be further refined in terms of.5 and 1. 360. Higher order methods. namely multicore CPU. However. Nevertheless. The average simulation time is 33 secs if there is no approximation in flash.6. especially for the run with ε=10-6. such as saturation/composition profiles and oil recoveries. The second test is on parallelization of an ISPU slimtube simulation using the approach described in Section 3. selection of numerical schemes. a MMP search strategy is crucial to reduce the number of simulation runs at different pressures.

Avaullee. A Very Simple Multiple Mixing Cell Calculation to Compute the Minimum Miscibility Pressure Whatever the Displacement Mechanism. Texas. Dindoruk. nc 6. . A.. Neau. Properly Defining the Classical Vaporizing and Condensing Mechanisms When a Gas Is Injected into a Crude Oil. Industrial & Engineering Chemistry Research 37 (12): 4854-4859.. USA.B. E. USA. k . kn ) (25) n 1 4. M. Paper SPE 11823 presented at the 2008 SPE Annual Technical Conference and Exhibition held in Denver.H.(10) using the newly calculated vD. (10) over all the components. A. Jaubert. Maersk Oil.k  ( Cin. G. 1998b. Stenby. 2008. B.kn1 ) / (  Fi*. Multiple Mixing-Cell Method for MMP Calculations. Michelsen. 1991. G. L. USA.SPE 153758 11 Acknowledgements This study has been carried out under the project “ADORE—Advanced Oil Recovery Methods” funded by the Danish Council for Technology and Production Sciences. k 1  Fi . the procedure can be simplified since vD .L. Lake. and the innermost one for velocity vD . Cook. it should be equal to the overall density calculated from the phase equilibrium:   1       l   v 1 (24) Dindoruk suggests using the overall density ρ instead of vd as iterative variable since it makes the procedure more robust. (24) and compare it with C i 1 n 1 i. C. and DONG Energy..k represents the overall density at tn+1.k 1  Fi *.k     vD . Jaubert. W. L.N.. R. Belkadi. 2008. References Ahmadi...k D . Yan.N.A..W. k is always equal to 1 and no innermost loop is needed.. k (normalization needed). September 21-24. 1969. The solution of vD . (23): nc nc nc i 1 i 1 i 1 vD . Calculate  Cin. Pope. In our code. K. 1992. E..kn . California.k  i . Appendix A. stop iteration. The final procedure for a grid block k is as follows: 1. J. k by Eqs. the same procedure can be used. we use a secant method for ρ instead of direct substitution. Wolff. Comparison of Two Methods for Speeing up Flash Calculations in Compositional Simulations. go to step 3. Johns. High-Resolution Monotonic Schemes For Reservoir Fluid Flow Simulation.k i . Walker. i 1 nc  Fi*. 3. Ph.D. L. J. Paper SPE 142132 presented at the SPE Reservoir Simulation Symposium held in The Woodlands.. Flash the mixture in the block with composition Ci . In-Situ 15(3): 289-317.. Neau. 2011. Journal of Petroleum Technology 21(7): 901-915.k . If the relative error smaller than the tolerance (10-3~10-4).k. For explicit scheme with TVD. otherwise. Spencer. the middle one for space. A.... Calculate vD.B. Department of Petroleum Engineering. Datta Gupta.k i . 21-23 February 201. University of Stanford. Analytical Theory of Multiphase Multicomponent Displacement in Porous Media... Colorado. E. k 1  i 1 i 1 i 1 i 1   nc (23) nc C which is obtained by summing Eqs. Since i 1 n 1 i. nc 2. 1998a. Realistic K Values of C7+ Hydrocarbon for Calculating Oil Vaporization During Gas Cycle at High Pressures... K. and i 1 nc F i 1 *n i .k . the outermost one for time.k by rearranging Eq. Thesis. n 1 5. Use the overall density from the previous time step as initial estimate for ρ. Calculate the new density ρ by Eq.. L.J. L. Avaullee. k 1 at time level n. Industrial & Engineering Chemistry Research 37 (12): 4860-4869.. If it is assumed no volume change on mixing (NVC). Sepehrnoori. Arras. Solution procedure for the ESPU scheme The solution procedure consists of three-level loops. Calculate Ci . k must satisfy the overall mass balance equation: nc nc nc  n 1 n *n *n  C C  v F v       D .

.H. Higher-Order Implicit Flux Limiting Schemes for Black Oil Simulation. F. Orr.. Batycky. Effective Algorithms for the Study of Misccible Gas Injections Processes.K.. Paper SPE 15493 presented at the 61st Annual Technical Conference and Exhibition of Society of Petroleum Engineers held in New Orleans. SPE Reservoir Evaluation & Engineering February: 36-43.. M. November. Lo. Paper SPE 38889 presented at the Annual Technical Conference and Exhibition held in San Antonio. 2006. Adidharma.S. 5(3).R. Global Approach for Calculation of Minimum Miscibility Pressure. F. Eng. R. 1991. E. ... E. L. Xue. Department of Chemical Engineering. Fluid Phase Equilibria 153 (2): 251-263. LA..L.. Thiele. 2001.. Copenhagen.M. 1986. 08-11 June 1997. 1997. C. Orr. Jr. M. October 5-8.. Michelsen. 1986. Dispacement Behavior of the Condensing/Vaporizing Gas Drive Process. Y. 1987. SPE Journal September 297-311. 2007.. Zick.G.H. 647-655. Larsen. K. Increasing the Computational Speed of Flash Calculations with Applications for Compositional.. M. Michelsen.. Towler. Analytical Calculation of Minimum Miscibility Pressure. Jr. Krejberg... Metcalfe. L. H.H. B. L.L. C.L. L..A. Paper SPE 21222 presented at the 11th SPE Symposium on Reservoir Simulation held in Anaheim. Orr. K. Chem.. K. 1990.B. October 5-8.. TX. Jessen.... 1997. A Combined Condensing/Vaporizing Mechanism in the Displacement of Oil by Enriched Gases. M. Pedersen.L. 1987. M.D. Fluid Phase Equilibria 143: 1-12. SPE Reservoir Engineering. Paper SPE 16715 presented at the 62nd Annual Technical Conference and Exhibition of the Society of Petroleum Engineers held in Dallas.. Stenby. Monroe. Fjellerup. 1986. Miscible Gas Displacement of Multicomponent Oils..K. SPE Journal 1(1): 39-50. Datta-Gupta. Physically Based Higher Order Godunov Schemes for Compositional Simulation. Zhao. Texas. M.. F. R. R. Sweby. 2000. SPE Reservoir Eng. 2001. 2001. M. F. Peddibhotla. SPE Journal June: 147-155. 1996. M. October 5-8. Whitson. Stalkup. Høier. G. Paper SPE 20178 presented at the SPE/DOE seventh Symposium on Enhanced Oil Recovery held in Tulsa. Bjurstrom. 45: 7913-7923. PhD Thesis... Paper SPE 66403 presented at the SPE Reservoir Simulation Symposium held in Houston. Denmark. Silva. Radosz. Shelton.M. Thomassen.P. F..I.. Composition Paths in Four-Component Systems: Effect of Dissolved Methane on 1D CO2 Flood Performance.. Multicomponent Flow. J. Tie-Line Publications.M. B. B. Sensitivity to Gridding of Miscible flood Predictions Made With Upstream Differenced Simulations. Using a Multiple-Mixing-Cell Model to Study Minimum Miscibility Pressure Controlled by Thermodynamic Equilibrium Tie Lines. 2005.. M. P.. Gerritsen. K. February 17-20.T.T. 1991...12 SPE 153758 Jessen. M. Fussell. LA. 423-432.. S.. A.. R... Orr.S. Wang. 1990. Fluid Phase Equilibria 137: 101-124. 1997. Edwards. Paper SPE 15599 presented at the 61st Annual Technical Conference and Exhibition of the Society of Petroleum Engineering held in New Orleans.J.L.. California.W. Blunt.. A Streamline-Based 3D Field-Scale Compositional Reservoir Simulator.. Blunt. P.J. W. Technical University of Denmark. P. 1998. M. Stalkup. with Applications for Calculation of Miscibile Displacement. F. September 27-30. J. 1986..M. A Multicell Equilibrium Separation Model for the Study of Multiple Contact Miscibility in Rich-Gas Drives. G. A. Paper SPE 38003 presented at the 1997 SPE Reservoir Simulation Symposium held in Dalia. 1990.. Stalkup. Res.G.. Multiphase Streamline Modeling in Three Dimensions: Further Generalizations and A Field Application.I.. Rasmussen. Jessen.. 1973. Dean. Jr. SIAM Journal on Numerical Analysis 21(5) 995-1011... Effect of Gas Enrichment and Numerical Dispersion on Enriched-Gas-Drive Predictions. K. Holte. D.. Rubin. Transient Simulations. Texas.. High-Order Upwind Schemes for Two-Phase. February 11-14. Oklahoma. Studies of Gas Injection Into Oil Reservoirs by a Cell-to-Cell Simulation Model. Mallison. M. F. High Resolution Schemes Using Flux Limiters for Hyperbolic Conservation Laws. Jr. SPE Res Eval & Eng 9: 32-38.. 2006. Speeding up the Two-Phase PT-flash. Ind. Thiele. 1998.. 1984. April 22-25. TX. 1990. Theory of Gas Injection Processes. Miscibility Variation in Compositionally Grading Reservoirs.R. 1997. K. Michelsen. F. Johns. Denmark. Orr.

227 0.93 315.3 30.j 0.15 393.4 469.69 245.251 0.031507 0.3 72.99 248.01154 0.8 408.003712 0.85 277.12 0.4 27.14 168.12 0.074353 0.85 375.008 0.78 148.040839 0.04 30.3 29.12 0.26 269.4 15.12 0.15 86.33 260.14 97.022477 0.15 394.1 0.01 16.05 387.193 0.437387 0.296 0.10 .062312 0.3 32.229078 0.176 0.32 245.4 529.35 388.04 286.4591 0.15 393.5 547.64 215.SPE 153758 13 Table 1.1 810.25 310. Fluid description for gas injection system 2 CO2 C1 C2 C3 i-C4 C4 i-C5 C5 C6 C7 C8 C9 C22 zoil zgas Tc (K) Pc (atm) ω MW (g/mol) kCO2.10 0.41 168.5 153.6 507.15 394.2 191.23 128.10 0.025386 0.26 149.45 377.16 316.05 368.12 58.12 72.56 304.55 98.1 425.12 353.016557 0.12 0.5 33.07 44.12 0.02 519.64 Table 2.00 358.6 305.12 0.17 369.17 100.14 154.75 376.2 41.9 250.0 37.0315 44.23 170.62 252.5228 1.010436 0.2 460.9 36.25 373.7 204.00 0.46 274.225 0.4 33.1 568.45 237. Overview of the four gas injection systems System EoS N2/CO2/C1 C2-C6 C7+ nc Tres (K) Pbub (atm) MMP (atm) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 SRK SRK PR PR PR PR SRK SRK SRK SRK SRK SRK SRK SRK SRK SRK SRK SRK 24 44 25 42 47 47 25 48 58 29 37 37 33 34 51 48 50 40 32 23 18 16 18 18 29 23 21 28 29 29 30 30 22 23 24 20 44 33 57 42 35 35 46 29 21 43 34 34 37 36 27 29 26 41 15 13 12 12 15 15 15 15 15 15 15 15 15 15 15 15 15 15 374.15 368.034417 0.8 45.20 114.15 72.4854 0.15 372.2 190.15 368.54 365.09 58.2 304.11 299.60 0.15 358.152 0.14 237.8 0.10 0.4 48.098 0.35 249.25 383.65 295.15 383.4 369.45 117.12 0.

110 110 105 ESPU 250atm ESPU 275atm ESPU 300atm ESPU 325atm ESPU 350atm ESPU 400atm ESPU 450atm ESPU 500atm 105 100 C7+ recovery (%) C7+ recovery (%) 100 95 90 85 95 ISPU dt/dx=0.16 ESPU dt/dx=0.1 0.4 0.12 0. (b) RF from different numerical schemes (except for the last one marked with “no extrapolation”).5 TDBA1 Tol=1E‐6 0.1  N=500 80 ESPU dt/dx=0.04 0.1 N=20000 (extrapolation) 80 75 75 70 0 0.1  N=500 75 75 70 0 0.1 N=500 85 TVD dt/dx=0.14 0.06 0.1 N=20000 (no extrapolation) 200 250 300 350 400 450 500 550 Pressure (atm) 1 100 0.6 PVI.7 70 C7+ recovery (%) Gas saturation (a) (b) ∞ Figure 4.3 60 50 No approximation 40 TDBA1 Tol=1E‐6 30 TDBA1 Tol=1E‐8 0.2 1.8 PVI (a) (b) Figure 5.06 0.14 SPE 153758 110 110 105 ESPU 120atm ESPU 170atm ESPU 220atm ESPU 250atm ESPU 300atm ESPU 350atm ESPU 400atm ESPU 500atm 105 100 C7+ recovery (%) C7+ recovery (%) 100 95 90 95 ISPU dt/dx=0.1 10 0 TDBA1 Tol=1E‐8 0 0 100 200 300 400 500 600 Grid block number 700 800 900 1000 0 0.4 0.1 N=500 85 80 80 TVD dt/dx=0. 0.02 0.02 0.2 20 0.8 80 0.1 N=500 90 ISPU dt/dx=1.9 90 0.0 N=500 85 ESPU dt/dx=0.04 0. (b) RF from different numerical schemes.12 0.16 100 150 200 250 300 350 400 450 500 550 Pressure (atm) (a) (b) ∞ Figure 3. 1 1.4 . C7+ recoveries at different pressures for System 2: (a) recoveries from ESPU for N up to 2000.1 0.1 N=500 90 ESPU dt/dx=0.08 1/SQRT(N) 0.2 0.08 1/SQRT(N) 0.6 0.14 0.6 No approximation 0. (b) C7+ recovery. Slimtube simulation (TVD and VC) results for System 2: (a) saturation profile at 0. C7+ recoveries at different pressures for System 1: (a) recoveries from ESPU for N up to 2000.

5 ISPU TDBA Tol=1E‐8 dt/dx=1. 90 80 TVD TDBA  Tol=1E‐8 ESPU TDBA Tol=1E‐8 Deviation in MMP (atm) 70 60 ISPU TDBA Tol=1E‐8 dt/dx=0. 70 Reference 60 TDBA Tol=1E‐8 TDBA Tol=1E‐6 Time (sec) 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 System Figure 7.0 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 ‐10 ‐20 System Figure 8.SPE 153758 15 15 600 MOC No approximation TDBA Tol=1E‐8 TDBA Tol=1E‐6 TVD TDBA  Tol=1E‐8 400 MMP (atm) TVD no approximation 10 Deviation in MMP (atm) 500 300 200 TVD TDBA  Tol=1E‐6 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 ‐5 100 ‐10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 ‐15 System (a) Figure 6. Deviations in MMP between the slimtube results and the MOC results System (b) . MMP calculation results using the TVD scheme: (a) MMP. (b) deviation in MMP. Simulation time for MMP calculation by slimtube simulation using TVD.

. (b) speedup factor.5 2 10 1.16 SPE 153758 35 4 TOL=1E‐6 30 3. Parallel computation of an ISPU slimtube simulation run with 3000 grid blocks for System 2: (a) simulation time.5 Speedup factor Time (sec) 100 80 60 3 2. 140 4.5 TOL=1E‐6 TOL=1E‐10 25 3 Speedup factor Time (sec) TOL=1E‐10 20 15 2.5 0 1 0 1 2 3 4 Number of threads 5 6 7 1 2 3 4 5 6 7 Number of threads (a) (b) Figure 10.5 5 0 1 0 1 2 3 4 5 6 7 1 2 3 Number of threads 4 5 6 7 Number of threads (a) (b) Figure 9.5 120 4 ESPU ESPU 3. (b) speedup factor.5 40 2 20 1. Parallel computing of an ESPU slimtube simulation run with 1000 grid blocks for System 2: (a) simulation time.