Professional Documents
Culture Documents
A R T I C L E I N F O A B S T R A C T
Keywords: The efficient simulation of gas flow in porous media is highly required in petroleum engineering, CO2 seques
Hybrid reduced order model tration, etc. However, it still remains a great challenge attributing to the gas compressibility compared with
POD-DEIM incompressible fluid flows. The commonly-used equation of states (EOS) of the gas are cubic equation and need
Compressible gas flow
to be updated in each iteration, and it also leads to the nonlinearity in flow equations, thus the computational
Porous media
cost mainly stems from the solution of gas EOS. In this paper, a hybrid reduced order model (ROM) coupling the
Peng-Robinson equation of state
proper orthogonal decomposition (POD) and the discrete empirical interpolation method (DEIM) is presented to
accelerate the calculation of compressible single-phase gas reservoir flow, in which the Peng-Robinson EOS (P-R
EOS) is considered to describe the gas states. To this end, in the hybrid ROM framework POD is applied to solve
the flow equation and DEIM is used to solve the P-R EOS, respectively. The selection of POD modes and DEIM
interpolation points, which plays a crucial role in the hybrid ROM, is discussed and carried out carefully. Per
formances of the proposed POD-DEIM-ROM are evaluated and demonstrated by two numerical cases. Simulation
results illustrate that the proposed hybrid ROM displays a satisfactory computational speed-up (two orders of
magnitudes faster) without sacrificing numerical accuracy significantly compared with the standard finite dif
ference method. In addition, DEIM shows excellent acceleration and it is a perfect choice for solving the cubic gas
EOS.
* Corresponding author.
** Corresponding author.
E-mail addresses: yubobox@vip.163.com (B. Yu), shuyu.sun@kaust.edu.sa (S. Sun).
https://doi.org/10.1016/j.jngse.2020.103367
Received 15 November 2019; Received in revised form 2 February 2020; Accepted 9 May 2020
Available online 16 May 2020
1875-5100/© 2020 Elsevier B.V. All rights reserved.
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
2
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
basis functions that were applied to extract the coarse-scale effective control volume, kg=ðm3 ⋅s); p is the pressure, Pa; W is the molecular
transmissibilities by using the same method for that of incompressible weight of gas, kg/mol; Z is the compressibility factor; R is the universal
flows, and the effects of compressibility were considered in the solution gas constant, R ¼ 8:314J=ðmol:KÞ; T is the temperature, K; Ω represents
of the coarse-scale pressure equation or in the reconstruction of the computational domain.
fine-scale fluxes. Numerical examples demonstrated the accuracy of the The following boundary conditions and initial conditions are adop
proposed methods was affected by the modeling of compressibility. ted to close above governing equations,
Krogstad et al. (Krogstadet al., 2011) presented a coarse grid localized
u ⋅ n ¼ uB on ∂ΩN (4)
model order reduction approach to speed up the simulation of
compressible multiphase flow in porous media. In this approach the
p ¼ pB on ∂ΩD (5)
approximation of both flux and pressure fields was based on the POD
coupled multiscale mixed FEM (MsMFEM) on the coarse grid. Numerical p ¼ pinit or Z ¼ Zinit at t ¼ t0 (6)
tests illustrated that the sparse basis POD is more process independent
but it may result in more basis functions compared to standard POD where n represents the outward normal direction; the boundary velocity
models. Although the above works made a step towards the compress uB , boundary pressure pB and the initial pressure pinit and initial
ible porous flow, they all coupled the local model reduction and global compressibility factor Zinit are all given parameters; The subscripts D and
model reduction techniques involving the calculation both on coarse N denote the Dirichlet boundary and the Neumann boundary,
grid and fine grid, and the multiple computation of different snapshots respectively.
for local and global ROMs are also tedious, which complicate the solu
tion process and program coding. To further explore the potential
capability of reduced order modeling for fast simulation of compressible 2.2. Peng-Robinson equation of state
gas flow in porous media, in this paper a hybrid POD-DEIM global model
is proposed considering the simplicity of the global model order To get the compressibility factor Z and solve Equation (3), the
reduction (Ghasemi et al., 2015). By applying this model, the compu compressibility of the gas flow should be considered. Thus, the flow
tational burden associated with the solution of gas flow equation and equation should be coupled with the equation of state (EOS) of gas.
equation of states of gas is expected to be mitigated substantially. The There are several choices for the EOS, such as Var der Waals EOS,
main contribution of this study is to couple the global model order Redlich-Kwong EOS, Soave-Redlich-Kwong EOS, Peng-Robinson EOS,
reduction methods POD and DEIM in a hybrid ROM in the gas flow and so on. In this work, the most popular equation of state describing
considering the Peng-Robinson EOS (P-R EOS), in which the POD is used natural gas flow in petroleum industry, Peng-Robinson EOS (P-R EOS), is
to solve the flow equation and the DEIM is applied to solve the P-R EOS applied to characterize the gas states in porous media. The typical form
associated with the projected non-linear terms in the POD reduced of P-R EOS reads,
system. Though the current work is still at initial stage and more efforts
need to be paid, to the best of authors’ knowledge, there is rare report on p¼
RT aðTÞ
(7)
the solution of EOS based on applying the DEIM. v b v2 þ 2bv b2
The structure of this paper is organized as follows. In Section 2, the For the convenience of computation, the P-R EOS can be reformu
model problem is introduced in detail, including the equation of state of lated as a cubic equation in terms of Z,
gas and the derivation of pressure equation. In Section 3, the conven � �
tional methodology for solving the model problem is shown. In Section Z 3 ð1 BÞZ 2 þ A 2B 3B2 Z AB B2 B3 ¼ 0 (8)
4, the establishment of proposed POD-DEIM-ROM is presented, and the
In above equation, the parameters A and B are defined as follows,
selection of POD modes and DEIM interpolation points is discussed. In
Section 5, the proposed POD-DEIM-ROM is validated by two numerical aðTÞp bp
A¼ ; B¼ (9)
cases and its reconstruction/prediction accuracy and computational R2 T 2 RT
speed-up are demonstrated. Concluding remarks of this work are sum
2 0:45724R2 Tc2
marized in Section 6. where aðTÞ ¼ aðTc Þð1 þ mð1 Tr0:5 ÞÞ , in which aðTc Þ ¼ pc , Tr is
0:0778RTc
the reduced temperature defined as Tr ¼ TTc ; b ¼ bðTc Þ ¼ pc , Tc and
2. Model problem
pc are cirtical temperature and critial pressure, respectively; parameter
m is computed by,
2.1. Governing equations
�
0:37464 þ 1:54226ω 0:2699ω2 ; 0 < ω < 0:5
m¼ (10)
In this study, the compressible single-phase gas flow in porous media 0:3796 þ 1:485ω 0:1644ω2 þ 0:01667ω3 ; 0:1 < ω < 2:0
is considered as the model problem to introduce the proposed POD-
The acentric factor ω (a measure of the non-sphericity of molecules)
DEIM-ROM. The governing equations of this model problem are
of gas species in Equation (10) is defined as,
described by the mass conservation equation, Darcy’s law (by ignoring
� � ��� ��
gravity’s contribution) and equation of state, which are respectively 3 pc Tc
ω ¼ log10 1 1 (11)
given below, 7 patm Tb
∂ρ
φ þ r ⋅ ðρuÞ ¼ q in Ω (1) where patm is the atmosphere pressure; Tb is the boiling temperature.
∂t
k
u¼ rp in Ω (2) 2.3. Pressure equation
μ
pW ¼ ρZRT (3) For the convenience of numerical solving, the original mass conser
vation equation (1) with primary unknow ρ is reformulated into a
where φ is the porosity of the porous media; ρ is the gas density, kg= m3 ; possion-type equation regarding the pressure. To transform the mass
u is the Darcy velocity, m/s; μ is the dynamic viscosity of gas, Pa.s; k is conservation equation, the compressibility coefficient needs to be
the diagonal permeability, m2 ; q is the mass accumulation rate in a calculated first. The fluid compressibility coefficient at isothermal con
dition is expressed as,
3
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
� �
1 ∂ρ Step 1: Give the initial calculation parameters, physical parameters
cf ¼ (12)
ρ ∂p T
and set the boundary conditions.
Step 2: Solve the P-R EOS (8) implicitly to get the compressibility
When P-R EOS is applied to describe the fluid state, the compress factor Z. Equation (8) is a cubic polynomial, the roots of Equation (8)
ibility coefficient can be derived as, can be found by using algebraic methods if the coefficients are real
� �
1 1 1 BZ 2 þ A 2B 6B2 Z 2AB 2B2 3B3 numbers.
cf ðZÞ ¼ þ � (13) Step 3: Calculate the compressibility coefficient cf explicitly ac
p pZ 3Z 2 2ð1 BÞZ þ A 2B 3B2
cording to Equation (13).
Substituting Equation (12) into the mass conservation equation (1) Step 4: Solve the pressure equation (14) implicitly to get the pressure
yields the following pressure equation, p with prescribed tolerance. In this work, the finite difference
� � method (FDM) is applied to discrete the pressure equation (14),
∂p k ZRT
φcf ðZÞp ¼ r ⋅ p rp þ q in Ω (14) among which the second-order central difference scheme is used to
∂t μ W
discretize the second-order nonlinear term and the first-order dif
ference scheme is adopted to discretize the unsteady term. The
3. Conventional methodology generated algebraic equations are solved by the Gauss-Seidel itera
tion and the numerical solution of pressure p can be obtained with
Here the conventional numerical methods for solving the governing prescribed tolerance.
equations ((2), (8), (13) and (14) of compressible single-phase gas flow Step 5: Calculate the Darcy velocity u explicitly according to Equa
in porous media are first briefly introduced. To solve this model prob tion (2);
lem, the semi-implicit and fully implicit algorithms can be applied. In Step 6: Update the related variables, parameters and boundary
this study, the semi-implicit algorithm is adopted due to its simple and values in the current time step;
easy implementation. Following semi-implicit algorithm, the governing Step 7: Go to the next time step and repeat steps 2e6 until the pre
equations (14) and (8) are solved implicitly and equations (2) and (13) scribed simulation time is reached.
are solved explicitly. The numerical solving procedures are summarized
as follows and the flow chart depicting the entire solution procedure is From above numerical procedures shown in Fig. 2, it can be clearly
also presented in Fig. 2.
4
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
seen that in step 2 and step 4 the P-R EOS (8) and pressure equation (14) problem in a high-dimensional space. To clearly show the difference of
need to be solved implicitly. Thus, the computational burden of this the improved semi-implicit method based on POD-DEIM between the
model problem mainly stems from the implicit solution of these two conventional semi-implicit method based on FDM, the entire solution
equations, namely, the iterative solution of pressure equation by using procedure is briefly summarized in Fig. 3.
Gauss-Seidel method and the algebraic solution of P-R EOS by finding The cornerstone of the improved semi-implicit algorithm is to
the roots of a cubic polynomial. Actually, the solution of P-R EOS takes construct the POD-ROM-ROM. Therefore in the following content, the
most of the CPU time, this point will be demonstrated through the nu establishment of POD-ROM for pressure equation and the establishment
merical examples in Section 5. of DEIM-ROM for P-R EOS are presented in detail, respectively.
5
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
coefficient cðtÞ and basis functions ΦðxÞ: p ¼ cðtÞΦðxÞ, then Equation empirical interpolation method (DEIM) (Chaturantabut and Sorensen,
(15) is written as, 2010) is applied innovatively, which is a newly developed model order
reduction technique in 2010, to solve the P-R EOS in a reduced
AM�M ΦM�N cN ¼ bM (16)
dimensional space that only depends on the selected interpolation points
In order to establish the POD-ROM for pressure equation, the discrete in the computational domain. The cornerstone of DEIM is to approxi
pressure equation (16) will be projected into a low-dimensional space mate the nonlinear function by projecting it onto a subspace that ap
spanned by the selected N basis functions (or called POD modes) of proximates the space generated by the nonlinear function and that is
pressure ΦM�N (M≫N) using the projection. The basis functions of spanned by a basis of dimension N≪M.
pressure can be obtained offline by the eigenvalue decomposition or Suppose the first term and last term of pressure equation (14) can be
singular value decomposition (SVD) of snapshots consisting of pressure expressed as a function of Z namely fðZÞ, the approximation of fðZÞ can
samples at different evolution time. For those who are interested in more be formulated using DEIM as follows,
details of constructing the basis functions from snapshots, it is recom � 1
mended to study the previous related kinds of literature (Li et al., 2019). f ðZÞ � ~f ðZÞ ¼ Uc ¼ U PT U PT bf ðZÞ (20)
Projecting Equation (16) into the low-dimensional space ΦM�N yields,
where bf ðZÞ is calculated on selected interpolation points with DOF of N;
ΦTM�N AM�M ΦM�N cN ¼ ΦTM�N bM (17) ~fðZÞ is the approximation of fðZÞ with DOF of MðM ≫NÞ; U is the basis
The above equation can be further rearranged as, functions obtained by eigenvalue decomposition or singular value
decomposition (SVD) of the snapshots of fðZÞ; P is the interpolation
~ N�N cN ¼ b
A ~N (18) point matrix.
It can be seen from Equation (20) that the P-R EOS can be solved only
where A ~ N�N ¼ ΦT AM�M ΦM�N is the reduced order matrix of discrete
M�N on selected interpolation points instead of in the whole computational
~N ¼ ΦT bM is the reduced order
coefficient with the subscript N≪M; b domain, which greatly reduce the computational cost. A notable prob
M�N
matrix of discrete source term; cN is the reduced order matrix of un lem associated with DEIM-ROM of P-R EOS is that the computational
known spectral coefficient. efficiency and numerical accuracy strongly depend on the basis func
Equation (18) is the final POD-ROM for pressure equation (14), the tions U and interpolation point matrix P, which are both determined
spectral coefficients can be computed efficiently due to the low degree of offline. When the basis functions U are determined, the selection of
freedom (DOF) of POD-ROM. In Table 1, the construction of POD-ROM interpolation points is the key point and needs to be paid special
for pressure equation is presented and summarized step by step. It attention. In this work, the DEIM algorithm originally proposed by
should be noted that POD modes should be selected carefully because it Chaturantabut et al. (Chaturantabut and Sorensen, 2010) is used to
can exert considerable effects on the prediction accuracy and speed-up select the representative interpolation points, as shown below in
of POD-ROM. In this study, the following accumulative energy contri Table 2. From Table 2, actually it can be found that the interpolation
bution of different POD modes (Sirovich, 1987) is defined to determine points are selected where the interpolation error is the largest, interested
the optimal POD modes number, readers can refer to (Chaturantabut and Sorensen, 2010) for more
technical details.
X
n X
N
En ¼ λj n λi � 100% (19)
j¼1 i¼1 5. Results and discussion
where λ represents the eigenvalue obtained by the eigenvalue decom In this section, numerical simulations of two-dimensional single-
position or SVD and is sorted in descending order; En is the accumulative phase natural gas flow (the main component of natural gas, methane, is
energy contribution of the first n POD modes to the total energy of N considered) in porous media are performed to validate the computing
POD modes. performances of POD-DEIM-ROM through two cases. In numerical case
1, the reconstruction accuracy and acceleration effect of proposed POD-
DEIM-ROM are displayed and evaluated; in case 2, the predictability of
4.2. DEIM-ROM for P-R EOS POD-DEIM-ROM is further verified and proofed. It should be mentioned
that all simulations are carried out on an Intel Xeno E5-2640, CPU2.60
From the original pressure equation (14), it is obvious to see that the GHz PC with 64.00 GB of RAM in this work.
first nonlinear term φcf ðZÞp and the last nonlinear term ZRTq
W are related
to the compressibility factor Z. Thus, it will bring a problem that the
5.1. Numerical case 1
POD-ROM (18) is still related to the original full-order model problem,
which deteriorates the speed-up of POD-ROM. In order to settle this
In case 1, the permeability distribution of porous media is shown in
demerit of POD method with nonlinear terms, in this work the discrete
Table 2
Table 1
DEIM algorithm.
Construction of POD-ROM for pressure equation.
Algorithm 2: DEIM-ROM
Algorithm 1: POD-ROM
INPUT: fsl gm n
l¼1 ⊂R linerly independent
Offline 1: Solve the original pressure equation under different calculation
OUTPUT: ℘ ¼ ½℘1 ; ⋯; ℘m �T 2 Rm
!
stage conditions to obtain the representative snapshot ensembles of
pressure 1: [jθj; ℘1 ] ¼ maxfjs1 jg
2: Compute the snapshot kernel and solve the eigenvalue problem !
2: U ¼ ½s1 �; P ¼ ½e℘1 �; ℘ ¼ ½℘1 �
using SVD or eigenvalue decomposition
3: for l ¼ 2 to m do
3: Compute the basis functions
Online 4: Project the discrete pressure equation into a low-dimensional 4: SolveðPT UÞc ¼ PT sl for c
stage space spanned by the selected basis functions and establish the 5: r ¼ sl Uc
POD-ROM
6: [jθj; ℘l ] ¼ maxfjrjg
5: Compute the spectral coefficient by solving the POD-ROM
! !
6: Reconstruct the pressure using the spectral coefficient and 7: U←½U sl �; P←½P e℘l �; ℘ ←½℘ ℘l �
selected basis functions 8: end for
6
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
Table 3
Calculation parameters in numerical case 1.
Parameter Value Parameter Value
φ 0.2 W 1:6 � 10 2
kg=mol
p0 1:01325 � 10 Pa 5 μ 1:1067 � 10 5
Pa⋅s
p1 1:01325 � 10 Pa 6 T 298 K
p2 1:01325 � 105 Pa lx � ly 100 m � 100 m
k1 1:0 � 10 1
D nx � ny 100 � 100
k2 1:0 � 10 3
D Tc 190.58 K
R 8.314 J=ðmol:kÞ Tb 111.63 K
q 0 pc 4:604 � 106 Pa
Δt 1:5 � 10 4
day simulation time 30 days Fig. 5. Variation of accumulative energy contribution with different
POD modes.
7
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
thus the computational speed-up for solving the pressure equation is not
obvious. Overall, the efficient solving of P-R EOS contributes most of the
computational saving in this case, the computational speed up of POD-
DEIM-ROM is very attractive, which is two orders of magnitude faster
than the traditional FDM.
As mentioned before, the selection of POD modes and DEIM inter
polation points can exert significant impacts on numerical accuracy and
speed-up of POD-DIM-ROM. To investigate those influences induced by
POD modes and DEIM interpolation points selection, here the average
relative errors of pressure, Darcy velocity and compressibility factor
with different POD mode numbers and DEIM interpolation point
numbers are presented in Fig. 9, the profile of speed-up ratio with
different POD mode numbers and DEIM interpolation point numbers is
shown too in Fig. 9. From Fig. 9a ~9c, it can be seen that POD modes
number has critical effects on the computational accuracy of POD-DEIM-
ROM. With the increase of POD modes number, the average relative
errors of pressure, Darcy velocity and compressibility factor all decrease,
but this trend would be reduced slowly when enough POD modes are
adopted. Compared with the effect of POD modes number, the selection
of DEIM interpolation points demonstrates slight influences on accuracy
of pressure and Darcy velocity, it is hard to find any distinctions in the
average relative errors of pressure and Darcy velocity when different
Fig. 6. Selected interpolation points in DEIM-ROM, the indexes of these points
DEIM interpolation points are selected. However, Fig. 9c indicates that
are: (100, 50), (38, 53), (29, 71), (25, 50), (35, 67), (46, 50), (12, 50), (27, 27),
the selection of DEIM interpolation points affects the accuracy of
(32, 31), (42, 58).
compressibility factor substantially, generally the increasing interpola
tion points lead to decreasing average relative errors. It also implies that
where εM denotes the average relative error; M denotes p, u, v, Z, in this numerical case 10 interpolation points are enough to capture the
respectively; M1 denotes the results computed by POD-DEIM; M2 de compressibility factor, more extra points have no contribution to the
notes the results computed by FDM; nx and ny denote the grid number calculation accuracy even sometimes exert negative effects. From
along x and y directions. Fig. 9d, it is obvious to find that with the rise of POD modes number and
From Table 4, it can be found that the pressure and compressibilty DEIM interpolation points number, the computational speed-up would
factor have an excellent reconstruction accuracy, especially the average decline almost linearly. The effect of POD modes number on the accel
relative error of compressibility factor is only 0:0026%. It is obvious to eration overweighs that of DEIM interpolation points. It is also inter
observe the average relative error of Darcy velocity is larger than those esting to see that when POD modes number and DEIM interpolation
of pressure and compressibility factor, the main reason can be clarified points are lager enough, the influence of interpolation points number on
by Equation (2) that the relative error of Darcy velocity is influenced by speed-up will be submerged. Fig. 9 also proofs that our selection of POD
both the permeability of porous media and fluid viscosity. Overall, the modes and DEIM interpolation points in subsection 5.1.1 is reasonable.
average relative errors can verify that the proposed POD-DEIM-ROM In conclusion, the selection of POD modes and DEIM interpolation
possesses excellent computational performance from the perspective points should be implemented carefully, trade-offs between accuracy
of reconstruction accuracy is this case. and speed-up should be considered.
Except for the reconstruction accuracy, the computational speed-up
of POD-DEIM-ROM is much more concerned, which is the main aim of 5.2. Numerical case 2
this work. Therefore, the CPU time of POD-DEIM-ROM and FDM
consumed in the simulation is recorded and analyzed in detail, as shown In this case, the predictability of POD-DEIM-ROM is validated and
in Fig. 8. In this figure, the left part represents the CPU time of FDM and demonstrated with different boundary values at offline stage and online
the right part represents the CPU time of POD-DEIM-ROM consumed for stage. The permeability distribution of computational domain is shown
simulating 13.5 days. The black bar is the total CPU time consumed in in Fig. 10 where the two blue bars denote low permeable regions, and
this case, the red bar represents the CPU time used to solve the pressure the boundary type is same as that in numerical case 1. Detailed calcu
equation, and the blue bar represents the CPU time consumed to solve lation parameters are presented in Table 5.
the P-R EOS. It can be seen from the comparison that in FDM 97:4% of At the offline stage, the representative pressure samples and fðZÞ
the CPU time was consumed to solve the P-R EOS, however, in POD- samples are calculated with four groups of pressure setting where p1 is
DEIM-ROM the corresponding CPU time was deceased substantially to set as 3:03975 � 105 Pa and 4:053 � 105 Pa, p2 is set as 1:519875 � 105
a very small value, 15:3%. Overall, the absolute CPU time of POD-DEIM- Pa and 2:533125 � 105 Pa, respectively. 500 samples in each group are
ROM was largely decreased compared with that of FDM, and the speed- considered thus totally 2000 samples are obtained respectively for
up ratio of POD-DEIM-ROM to FDM is up to 58.4. It indicates that the pressure and fðZÞ. Readers of interest can refer to (Li et al., 2019) for
proposed POD-DEIM-ROM is an efficient acceleration approach that can more details on how to get representative samples.
largely improve the numerical efficiency of the traditional FDM. More
over, Fig. 8 also proofs that DEIM-ROM is a perfect choice to solve the P- 5.2.1. Determination of POD modes and DEIM interpolation points
R EOS, the speed-up ratio of DEIM-ROM for solving the P-R EOS is up to Just as numerical case 1, first the number of POD modes and DEIM
371 compared with that of FDM. To the best of authors’ knowledge, interpolation points should be determined. Using the same approach
there is no report on the solution of P-R EOS using DEIM approach. For presented in subsection 5.1.1, in this case 10 POD modes and 10 DEIM
the pressure equation, the calculation time is also reduced when using interpolation points are selected too. Fig. 11 illustrates locations of the
the POD-ROM, but the acceleration is not obvious compared with the selected 10 interpolation points.
DEIM-ROM. The main reason can be explained from the aspect that in
this case the pressure equation part can be easier solved, and the
convergence criterion for iterative solution of pressure is not so strict,
8
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
Fig. 7. Comparison of reconstruction accuracy of POD-DEIM-ROM (red solid line) and FDM (blue dashed line) in numerical case 1.
9
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
Fig. 9. Influence of POD modes and DEIM interpolation points numbers on accuracy and speed-up in numerical case 1.
10
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
Table 5
Calculation parameters in numerical case 2.
Parameter Value Parameter Value
φ 0.2 W 1:6 � 10 2
kg=mol
p0 1:01325 � 10 Pa 5 μ 1:5 � 10 4
Pa⋅s
p1 4:053 � 105 Pa lx � ly 100 m � 50 m
p2 1:01325 � 105 Pa nx � ny 100 � 50
k1 1:5 � 10 1
D Tb 111.63 K
k2 1:0 � 10 2
D Tc 190.58 K
R 8.314 J=ðmol:kÞ T 285.87 K
q 0 pc 4:604 � 106 Pa
Δt 3:0 � 10 4
day simulation time 30 days
Fig. 11. Selected interpolation points in DEIM-ROM, the indexes of these Fig. 13. Comparison of Darcy velocity predicted by POD-DEIM-ROM (red solid
points are: (100, 50), (8, 1), (28, 32), (1, 50), (43, 1), (22, 1), (25, 41), (72, 14), line) and FDM (blue dashed line) in numerical case 2.
(15, 37), (33, 50).
DEIM approaches is presented to accelerate the simulations of single-
total CPU time, and the computational efficiency of DEIM solving P-R phase compressible gas flow in porous media, which is highly deman
EOS is as high as 359.2 times of FDM. The absolute CPU time of POD- ded in petroleum engineering. The establishment of POD-DEIM-ROM is
DEIM-ROM was largely decreased compared with FDM (from 6076.8s introduced in detail, and the key point of POD-DEIM-ROM that how to
to 84.1s), and the speed-up ratio of POD-DEIM-ROM to FDM is up to 72. select POD modes and DEIM interpolation points is demonstrated. The
It indicates that the DEIM-ROM is an efficient approach and excellent numerical performances including the reconstruction and prediction
choice to solve the P-R EOS. Again, the attractive advantage of POD- accuracy and computational speed-up of proposed POD-DEIM-ROM are
DEIM-ROM is well illustrated from the perspective of computational validated through two numerical cases. The main concluding remarks of
efficiency, which is vital for engineering applications in the develop this work can be summarized as follows.
ment of natural gas reservoirs.
(1) A POD-DEIM-ROM is established in this study for fast calculation
6. Conclusions of compressible natural gas flow in porous media considering the
Peng-Robinson EOS, in which the POD-ROM is used to solve the
In this study, a hybrid reduced order model coupling the POD and pressure equation and the DEIM-ROM is applied to solve the
Peng-Robinson EOS, respectively. By using these two reduced
Fig. 12. Comparison of pressures predicted by POD-DEIM-ROM (red solid line) and FDM (blue dashed line) in numerical case 2.
11
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
References
Aarnes, J.E., Efendiev, Y., 2008. Mixed multiscale finite element methods for stochastic
porous media flows. SIAM J. Sci. Comput. 30 (5), 2319–2339.
Aarnes, J.E., Krogstad, S., Lie, K.-A., 2006. A hierarchical multiscale method for two-
phase flow based upon mixed finite elements and nonuniform coarse grids.
Multiscale Model. Simul. 5 (2), 337–363.
Alotaibi, M., Calo, V.M., Efendiev, Y., Galvis, J., Ghommem, M., 2015. Global–local
nonlinear model reduction for flows in heterogeneous porous media. Comput.
Methods Appl. Mech. Eng. 292, 122–137.
Bastian, P., Helmig, R., 1999. Efficient fully-coupled solution techniques for two-phase
flow in porous media: parallel multigrid solution and large scale computations. Adv.
Water Resour. 23 (3), 199–216.
Chaturantabut, S., Sorensen, D.C., 2010. Nonlinear model reduction via discrete
empirical interpolation. SIAM J. Sci. Comput. 32 (5), 2737–2764.
Chaturantabut, S., Sorensen, D.C., 2011. Application of pod and deim on dimension
reduction of non-linear miscible viscous fingering in porous media. Math. Comput.
Model. Dyn. Syst. 17 (4), 337–353.
Ghasemi, M., Yang, Y., Gildin, E., Efendiev, Y., Calo, V., et al., 2015. Fast multiscale
reservoir simulations using pod-deim model reduction. SPE Reservoir Simulation
Symposium. Society of Petroleum Engineers.
Ghommem, M., Presho, M., Calo, V.M., Efendiev, Y., 2013. Mode decomposition methods
for flows in high-contrast porous media. global–local approach. J. Comput. Phys.
253, 226–238.
Gildin, E., Ghasemi, M., Romanovskay, A., Efendiev, Y., et al., 2013. Nonlinear
complexity reduction for fast simulation of flow in heterogeneous porous media. SPE
Fig. 15. Comparison of CPU time of POD-DEIM-ROM and FDM in numerical Reservoir Simulation Symposium. Society of Petroleum Engineers.
case 2. Jenny, P., Lee, S., Tchelepi, H.A., 2003. Multi-scale finite-volume method for elliptic
problems in subsurface flow simulation. J. Comput. Phys. 187 (1), 47–67.
Kani, J.N., Elsheikh, A.H., 2019. Reduced-order modeling of subsurface multi-phase flow
order models, the original full order problem can be solved effi models using deep residual recurrent neural networks. Transport Porous Media 126
ciently in a low dimensional space with small degrees of freedom. (3), 713–741.
Krogstad, S., et al., 2011. A sparse basis pod for model reduction of multiphase
(2) The determination of POD modes and DEIM interpolation points
compressible flow. SPE Reservoir Simulation Symposium. Society of Petroleum
has appreciable effects on the prediction accuracy and compu Engineers.
tational speed-up of POD-DEIM-ROM, overall the influence of Li, M.. World natural gas 2018-2050: World energy annual report (part 3). URL. http
POD modes number overweighs that of DEIM interpolation ://peakoilbarrel.com/world-natural-gas-2018-2050-world-energy-annual-report-par
t-3/.
points. Therefore, trade-offs between numerical accuracy and Li, J., Zhang, T., Sun, S., Yu, B., 2019. Numerical investigation of the pod reduced-order
computational acceleration should be considered carefully when model for fast predictions of two-phase flows in porous media. Int. J. Numer.
selecting the POD modes and DEIM interpolation points. Methods Heat Fluid Flow 29 (11), 4167–4204.
Lunati, I., Jenny, P., 2006. Multiscale finite-volume method for compressible multiphase
(3) The reconstructed and predicted results of proposed POD-DEIM- flow in porous media. J. Comput. Phys. 216 (2), 616–636.
ROM match well with those of standard finite difference method, Moridis, G.J., Collett, T.S., Dallimore, S.R., Satoh, T., Hancock, S., Weatherill, B., 2004.
especially the pressure and compressibility factor. It indicates Numerical studies of gas production from several ch4 hydrate zones at the mallik
site, mackenzie delta, Canada. J. Petrol. Sci. Eng. 43 (3–4), 219–238.
that the POD-DEIM-ROM possesses satisfactory prediction accu Sirovich, L., 1987. Turbulence and the dynamics of coherent structures. i. coherent
racy. Compared with FDM, the computational speed-up of POD- structures. Q. Appl. Math. 45 (3), 561–571.
DEIM-ROM is attractive, generally the simulation using POD- Vermeulen, P., Te Stroet, C.B., Heemink, A.W., 2006. Model inversion of transient
nonlinear groundwater flow models using model reduction. Water Resour. Res. 42
(9), W09417.
12
J. Li et al. Journal of Natural Gas Science and Engineering 79 (2020) 103367
Wang, J., Zhang, X., Bengough, A.G., Crawford, J.W., 2005. Domain-decomposition Wu, Y.-S., Zhang, K., Ding, C., Pruess, K., Elmroth, E., Bodvarsson, G., 2002. An efficient
method for parallel lattice Boltzmann simulation of incompressible flow in porous parallel-computing method for modeling nonisothermal multiphase flow and
media. Phys. Rev. 72 (1), 016706. multicomponent transport in porous and fractured media. Adv. Water Resour. 25
Wang, Y., Yu, B., Sun, S., 2016. Pod-galerkin model for incompressible single-phase flow (3), 243–261.
in porous media. Open Phys. 14 (1), 588–601.
13