Professional Documents
Culture Documents
Hand-out note 11
page 1 of 4
) (
) (
)] + T [( P - P ) + ( P
[(
) (
) (
xgi-1 2
) (
) (
oi-1
oi
) (
cog i-1
)]
- Pcog i - qgi
)i
[(
)]
) (
[(
i = 1, N
i = 1, N
)]
) (
Txw i+1 2 Poi+1 - Poi - Pcowi+1 - Pcowi + Txoi-1 2 Poi-1 - Poi - P cow i-1 - P cow i - qwi
Poit
= Cp ow i Poi -
) + C (S
sgw i
gi
Sgti
) + C (S
swwi
wi
Swit
),
i = 1, N
All coefficients and caillary pressures in the equations above are functions of the solution, ie. the new saturations
and pressures.
IMPES method
We have so far only discussed the IMPES method for solution of the non-linear equations arising in reservoir
simulation. In this method we made the assumption of all coefficients and capillary pressures are evaluated at
old time step level, and then eliminated unknown saturations by combining the three equations and obtained a
pressure equation. Having thus solved for pressures, saturations could be solved for explicitly from one of the
above equations.
Simultaneous solution method
Obviously, another alternative for this type of simplified solution, is to keep all three phase pressures as unknowns
in the equations, and eliminate saturations by the following operations:
-1
Sw i - Sw it
dP
= cow (P oi - Poti ) - (Pwi - Pwti )
dSw i
-1
dPcog
t
t
Sg i - Sg ti =
(Pgi - Pgi ) - (Poi - Poi ) .
dS
g i
Except for the capillary pressures, we now make identical assumptions as for the IMPES method. Then, we may
solve for the three phase pressures. The coefficients would then have the form shown on the next page.
By this method, the simultaneous solution method, we have achieved to keep capillary pressures implicit, while
the other coefficients are explicit. The set of linear equations has tripled in size, however.
page 2 of 4
c
a
cg
Po
cw
cg
cw
cg
cw
cg
cw
cg
cw
cg
cw
cg
cw
cg
co
co
cw
cg
co
co
cw
co
cw
co
cw
co
cw
cw
co
cw
cg
cg
cg
cg
cg
Pg
cw
co
cg
cw
cw
co
cg
cw
cg
cg
cg
cg
cg
cg
cg
cg
cg
cg
cg
cg
cg
cg
cw
cw
Pw
Having thus obtained the three phase pressures, saturations may be interpolated from the capillary pressuresaturation relationships.
Iterative methods
Today, the IMPES method is still being used for field scale simulation, provided changes in pressures and
saturation are reasonably slow. However, generally we need to solve the flow equations implicitly, and in order to
do so, we apply iterative techniques.
For the purpose of illustration of Newtonian iteration, recall the Newton-Raphson method for solving a non-linear
equation with one unknown. Having a nonlinear equation of the form:
F(x) = 0.,
for instance:
x 5 - 3x 3 + 2x - 4 = 0,
we can, by using Taylor expansion, derive an iterative formula:
x k +1 = x k -
F( xk )
.
F( x k )
By iteration, provided that the first guess of the solution is sufficiently close to the correct solution, rapidly
converge to the solution.
If we have two simultaneous non-linear equations:
F(x, y) = 0
G(x, y) = 0,
the iterative formula becomes:
Norwegian University of Science and Technology
Department of Petroleum Engineering and Applied Geophysics
F(x k , y k )
x k +1 = x k -
yk +1 = yk -
F(x k , yk )
page 3 of 4
G
F
( x , y ) - G(x k , yk )
(x , y )
y k k
y k k
J(x k , y k )
G
F
(x k , y k ) - G(x k , y k ) (x k , yk )
x
x
J(x k , yk )
F
G
F
G
(x k , y k )
(x k , yk ) (x k , yk )
(x , y ).
x
y
y
x k k
Our three phase flow equations may be written on a similar form as the above equations. Generally, in one
dimensional flow, we have 3N unknowns. However, due to the structure of the finite difference method used, each
node equation will have only 9 unknowns, so that our system of equations becomes:
Foi (Poi-1, Poi, Poi+1, Sgi-1, Sgi, Sgi+1, Swi-1, Swi, Swi+1) = 0
Fgi (Poi-1, Poi, Poi+1, Sgi-1, Sgi, Sgi+1, Swi-1, Swi, Swi+1) = 0
Fwi (Poi-1, Poi, Poi+1, Sgi-1, Sgi, Sgi+1, Swi-1, Swi, Swi+1) = 0,
i = 1, N
Foik +1 = Foki +
F oi
Foi
Foi
+1
+1
(Poki-1
- Poik-1 ) +
(Poik +1 - Poki ) +
(Poik+1
- Poki+1 )
Poi -1
Poi
Poi +1
F oi
F oi
F oi
+1
(Sg ki-1
- Sg ik-1 ) +
(Sg ki +1 - Sg ik ) +
(Sg k +1 - Sg ik+1 )
Sgi -1
Sg i
Sgi +1 i +1
Foi
F oi
F oi
+1
+1
(Sw ik-1
- Sw ki-1 ) +
(Sw ik +1 - Swik ) +
(Sw ki+1
- Sw ik+1 )
Sw i-1
Sw i
Sw i +1
Fgik +1 = Fgik +
F gi
Fgi
Fgi
+1
+1
(Poki-1
- Poik-1 ) +
(Poik +1 - Poik ) +
(P oik+1
- Poki+1 )
P oi -1
Poi
Poi +1
F gi
F gi
F gi
+1
(Sg ki-1
- Sg ik-1 ) +
(Sg ki +1 - Sg ik ) +
(Sg k +1 - Sgik+1 )
Sgi -1
Sg i
Sgi +1 i +1
Fgi
F gi
F gi
+1
+1
(Sw ik-1
- Sw ki-1 ) +
(Sw ik +1 - Swik ) +
(Sw ki+1
- Sw ik+1 )
Sw i-1
Sw i
Sw i +1
Fwki +1 = Fwik +
F wi
F wi
F wi
+1
+1
(Poik-1
- Poki -1 ) +
(Poik +1 - Poki ) +
(Poki +1
- Poik+1 )
Poi-1
P oi
Poi +1
F wi
F wi
F wi
(Sg k +1 - Sg ik-1 ) +
(Sg ik +1 - Sgik ) +
(Sg k +1 - Sgik+1 )
Sgi -1 i-1
Sgi
Sg i+1 i +1
F wi
F wi
F wi
+1
+1
(Sw ik-1
- Sw ki-1 ) +
(Sw ik +1 - Sw ki ) +
(Sw ki +1
- Swik+1 )
Sw i-1
Sw i
Sw i +1
i = 1, N
page 4 of 4
Thus, for a one-dimensional system we have 3N equations and 3N unknowns, and we can easily solve for
estimates of oil pressures and gas and water saturations. By applying Newtonian iteration until we converge on a
solution within some tolerance, we may obtain a solution to the equations. Our linear equations for iteration
step k+1 would then take the form:
+1
+1
+1
+1
ap o oi Poki-1
+ b pooi Poik +1 + c pooi Poik+1
+ asgoi Sgik-1
+ b sgoiSg ki +1 + c sgoiSg ik+1
+1
+1
+ aswoi Sw ki-1
+ b swoi Sw ki +1 + c swoiSw ik+1
= doi
+1
+1
+1
+1
ap o gi Poki-1
+ b pogi Poik +1 + c pogi Poki+1
+ asggi Sgik-1
+ b sggiSg ik +1 + csggi Sgik+1
+1
+1
+ aswgi Sw ki-1
+ b swgi Sw ki +1 + c swgiSw ik+1
= dg i
k +1
k +1
k +1
apo wiPoi-1
+ bpo wiPoki +1 + c po wiPoki+1+1 + a sgwiSgi-1
+ b sgw iSgik +1 + csgw iSgi+1
k +1
k+1
+ asww iSwi-1
+ bswwiSw ki +1 + c sww iSwi+1
= dw i
i = 1, N
or, on a compact form
r k +1 r k +1
r
v
ai Xi-1
+ bi Xi + ci Xik+1+1 = d i ,
i = 1, N
where
ap o oi
ai = ap o gi
apowi
asgoi
asggi
asgwi
Poik +1
r k +1
Xi = Sg ki +1
Swik +1
aswoi
aswgi
aswwi
b pooi
bi = b pogi
b powi
b sgoi
b sggi
b sgwi
b swoi
b swgi
bswwi
c p o oi
ci = c p o gi
cpowi
csgoi
csggi
c sgwi
cswoi
cswgi
c swwi
do i
r
di = dg i
dw i
A simplified, non-iterative method that was extensively used in the past, uses estimated chord-slopes instead of the
derivatives above, and accepts the solution obtained after the first iteration. This method, called semi-implicit
method, is hardly used today.