PAPER 2009-021

Global and Local History Matching

Optimization. The Successful Key of
Reservoir Simulation: High Heterogeneous
Field CaseStudy
CUONG T. Q. DANG, NGOC T. B. NGUYEN, WISUP BAE, TAEMOON CHUNG
Sejong Univerisity

THUOC H. PHUNG
Vietsovpetro Joint Venture

WONGYU LEE
Korea National Oil Cooperation

This  paper  is  accepted  for  the  Proceedings  of  the  Canadian  International  Petroleum  Conference  (CIPC)  2009,  Calgary, 
Alberta,  Canada,  16‐18  June  2009.    This  paper  will  be  considered  for  publication  in  Petroleum  Society  journals. 
Publication rights are reserved. This is a pre‐print and subject to correction. 

This paper describes the integration of mathematical

models (SIMPLEX and Steepest Descent algorithm) with global
Abstract modification (through deeply geological studies) to increase the
efficiency of history matching. Finally, the authors apply the
Reservoir characterization and forecasting require the above method to the Lower Miocene reservoir at White Tiger
construction of reliable and history-matched reservoir models. field, a highly heterogeneous reservoir, and have been
The goal of all history matching for the reservoir model is to producing under naturally edge-water mechanism. A model that
generate accurate predictions. History matching can be carried covers the upper formation of White Tiger field with 27
out either by standard history matching techniques (composed production wells, 8 injection wells is conditioned to 23 years of
of a fixed geological model with global modifications and local production history. Otherwise, there is no an absolute reservoir
adjustments), or by automatic computation of a set of parameter model and an error between history data and simulation results
values so as to minimize the prescribed cost function. often occur. So, the authors supposed an experimental method
to estimate this error. These obtained results can be used to
An important aspect of history matching results is to
optimize the history matching procedures. This research is also
quantify the contribution of uncertainties in underlying
an effective approach for history matching oil-gas reservoir
parameters such as porosity, permeability, structure, fault
models.
system, aquifer support, heterogeneous limit, etc. Thus, a
detailed geological understanding of reservoirs will have a
significant effect on history matching results.

1
Introduction The main objective of this paper is to present the idea to
combine the Simplex method with steepest descent algorithm to
In recent years, optimization methods with application to solve history matching optimization problem. The history
reservoir characterization and uncertainty assessment received matching progress was run by the IMEXTM simulator. Then,
an increasing interest. Reservoir simulation is used in several this approach is applied for the Lower Miocene sequence – a
important tasks in the petroleum industry. Since the computer high heterogeneous reservoir in White Tiger field.
system were limited in the past, it is difficult to solve certain The Simplex Method
problems in complex real fields. However, nowadays computer
sciences, mathematics algorithm and automation tools have The simplex algorithm is regarded as one of the most
developed significantly. Numerical reservoir simulation has important algorithms of the twentieth century. The method,
become an important tool for several researches and practical invented by George Dantzig in 1947, was an iterative
activities in the petroleum industry. Although there are several processing for solving linear programming. Then, the downhill
limitations and errors involved in the process of reservoir simplex optimization was improved by Spendley (1962), Nelder
modeling using available numerical simulators, they are still the and Mead (1965), Press et al (1992) (Figure 2). This solution
most viable and reliable way to perform reservoir behavior and has been applied since 1960s and it is still widely used for many
production forecasting. optimization problems. The theory of the simplex method
solves an unconstrained minimization problem in n dimensions
History matching is absolutely necessary and important by maintaining at each iteration n+ 1 point that defines a
for reservoir simulation. It is an inverse problem to partial simplex. This simplex is updated by applying certain
differential equation in mathematics (Figure 1). Usually the transformations to it so that it "rolls downhill" until an optimal
limited static data on the geological and geophysics background solution is found (if it exists). For two variables, a simplex is a
of the reservoir are available from well tests, seismic surveys, triangle, and the method is a pattern search that compares
logs etc. Applications of reservoir simulations, which intend to function values at the three vertices of a triangle. The worst
reproduce measured well production data in the basis of vertex, where f(x,y) is largest, is rejected and replaced with a
unknown model parameters, define a procedure to solve the new vertex. A new triangle is formed and the search is
inverse problem of reservoir modeling (R.W. Schulze-Riegert et continued. The process generates a sequence of triangles
al, 2001)10. (which might have different shapes), for which the function
Traditional history matching comprises the adjustment of values at the vertices get smaller and smaller. The size of the
reservoir parameters in the model, such as porosity, triangles is reduced, and the coordinates of the minimum point
permeability, net thickness, fluid properties, structure, etc, until are found. A standard (minimization) linear programming
the simulated performance matches the measured data. This is problem consists of
the trial-and-error procedure; the qualities of history matching 1. The objective function is to be minimized.
procedures depend on so much of the knowledge and
experience of the reservoir engineer. For a complex and 2. All the variables involved in the problem are non-
heterogeneous reservoir, it usually takes a long time to match negative.
simulation results with production history data
3. Each constraint may be written so that the expression
To reach success in reservoir simulation and improve with the variables is greater than or equal to a non-negative
history matching results, compared with traditional methods, a constant
number of previous works have concentrated on local gradient
based optimization methods (Carter et al., 1974; Chen et al.,
1974; Anterion et al., 1989; Wu et al., 1999; Vasco et al., 1999;
Landa et al., 2000). These techniques are commonly used to
The iteration of Simplex Algorithm
solve the inverse problem, and it is quite efficient in converging Firstly, we must define an initial trial point x1 = (x11,
to a local minimum in the objective function. However, there is x12,…, x1n) and a first vector of parameters ranges h = (h1, h2,
no guarantee that this will be the global minimum. In general, …, hn).
the objective function is likely to have many minima. Only
those minima that give a sufficiently good match to the data are Then, the next trial points will be defined as
of interest (Susana Gomez et al., 2001)13. Otherwise, many x1, xk+1 = (x11, …, x1k + dk, …., x1n) for k = 1, …,n
authors have already proposed global minimization approaches
in the field. Global optimization methods have been used to At each step of the algorithm, the (n+1) trial points are
overcome some of the pitfalls of gradient based methods. sorted in the order of decreasing objective values,
However, global convergence – even to an approximation of the Q1≤Q2≤…≤Qn+1
solution – usually requires a huge number of iterations, After that, a reflection point xr could be computed by
convergence speed is usually low. As a matter of fact, this price
is often too high for the reservoir history matching problem,  

where the computation of the objective function is most x r  x   ( x  x n 1 )

expensive (S. Mantica et al. 2001)11. Gómez et al. (2001) have
 n
xi
coupled a gradient method to a tunneling method with global
optimization features. In order to accelerate the computation of
where x 
n
i 1
is the centroid of the n best point (all

large numbers of independent simulation, Leitão et al. have vertices except for xn+1) and ρ(0< ρ≤1) is the reflection
used the direct optimization methods in connection with parallel coefficient.
computing. Most recently, Genetic Algorithms have been
applied to reservoir characterization by Romero et al2. The objective function is evaluated at the new trial point
xr. The next step will be depended on this value:

2
  If Q1≤ Qr < Qn, accept the reflected point xr and direction decrease the fastest and would be the negative gradient
terminate the iteration. of F(x) (Figure 3). The main idea of this method is found, where
it uses a zig-zag like path from an arbitrary point x0 and
 If Qr < Q1, calculate the expansion point xe: gradually slide down the gradient until it converges to the actual
 
xe  x   ( xr  x) with  is the expansion point of minimum.
coefficient and compute Qe = Q(xe). If Qe < Qr, accept The convergent speed of the steepest descent adaptive
xe, and terminate the iteration; otherwise (if Qe≥Qr), algorithm is determined by the spread value of the correlation
accept xr and terminate the iteration. matrix. When the correlation matrix is singular, the algorithm
will take long time to converge. The iteration move in the
 If Qn ≤ Qr < Qn+1, perform an outside contraction by direction of steepest descent (opposite the gradient direction)
computed: until the minimum is approximately reached.
 
xc  x   ( x r  x) with  is the contraction
coefficient
and evaluate Qc = Q(xc). If Qc ≤ Qr, accept xc and Workflow of steepest descent
terminate the iteration (0)
Initial guess x

 If Qr ≥ Qn+1, perform inside contraction by calculated: x

(k)
Stop!
is minimum
 f(x
(k)
  Compute )

xcc  x  ( x  x n 1 ) Yes

and evaluate Qcc = Q(xcc). If Qcc < Qn+1, accept xcc and ∥ ▽ f(x ) ∥ ﹤ ε
(k)

terminate the iteration. k=k+1

No

Determine α
(k)
α {α1，…αn}
Polynomial fit : cubic
,
Main features x
(k+1)c (k)
=x - α
(k)
 f(x
(k)
)
Region elimination :

The advances of the simplex method can be described as

follows:
 It is highly efficient in practice and converging
expected polynomial time for certain distribution of
random inputs. The method does not depend on the
Steepest descent algorithm
(correct) initial starting points for convergence.  Step1: Estimate a starting design x(0) and set the iteration
counter k=0. Select a convergence parameter ε > 0
 For computational purposes it is still the most
efficient general tool for solving linear programs. The  Step 2: Calculate the gradient of f(x) at the point x(k) as c(k)
method is fairly efficient at each iteration (typically
1-2 function evaluations) =  f(x(k)). Calculate c  c T c . If c <ε, then stop
the iteration process as x*=x(k) is a minimum point.
 From a complexity point of view it is the most
Otherwise, go to step 3.
promising candidate for a strongly polynomial time
linear programming algorithm  Step 3: Let the search direction at the current point x(k) as
d(k) = -c(k)
 The simplex method does not need matrices to be set
up inverted, and no derivatives need to be estimated.  Step 4: Calculate a step size α(k) to minimize f( x(k) +
The simplex method can be used for solving α(k)d(k)). A one-dimensional search is used to determine
smoothing function (such as Archie equation) as well α(k).
as non-smoothing functions when the derivative-
based methods fail.  Step 5: Update the design as x(k+1) = x(k) + α(k)d(k). Set k =
k+1 and go to step 2.
The simplex method is an easy-to-use tool to optimize
complex functions, even if these function are non differentiable It is not hard to see why the method of steepest descent is
or non-continuous. This is one of the most important features so popular among many mathematicians; it is very simple, easy
because it can be combined with classical optimization methods to use, and each repetition is fast. This method does not need a
such as steepest descent method to reduce the limitation of second derivative, and it is usually much better than conjugate
classical methods. or a quasi-Newton algorithm. However, the biggest advantage
of this method lies in the fact that it is guaranteed to find the
minimum through numerous times of iterations as long as it
exists. However, this method also has some significant flaws. If
Steepest Descent Method it is used on a badly scaled system, it will end up going through
The applications of gradient-based, history-matching an infinite number of iterations before locating the minimum,
techniques are becoming more widely used as appropriate and since each of steps taken during iterations are extremely
software becomes commercially available. The method of small, the convergence speed is pretty slow, and this process
steepest descent is the oldest and simplest (also called Cauchy’s may an extremely long time. Although a larger step size will
method) of the gradient methods for minimization of a real increase the convergent speed, it could also result in estimation
function defined on Rn. The basic theory of this method is with large error.
defined and differentiable within a given boundary, so its

3
 whole history matching process but it can be altered after each
iteration step. The results of our solution will be presented in
Optimization Techniques Application to the next section.
Reservoir History Matching
History matching is defined by finding a set of model
parameters that minimize the difference between a desired Field Application
solution (production history data) and the simulated solution Reservoir Simulation
(the calculated production data). For a general history matching
problem, the objective function is an implicit function of the The simulation model describes a black oil sandstone
optimal vector; it needs to carry out running simulation to gain reservoir located offshore of Vietnam. This reservoir is
an objective function and it takes the uppermost in computing extremely heterogeneous and complex in geological
costs. An additional term is commonly added to the objective characterizations. The complicated fractured system and natural
function to constrain deviations from the initial model or from edge-water mechanism caused some problems in reservoir
underlying relationships among parameters being varied. There simulation processing. The PVT characteristic of crude oil in
are different functional dependencies used in the literature. The Lower Miocene was showed in Table 1. Base on the results of
objective used for calculations presented in this paper was integrated geological and production history, the authors used
defined by IMEXTM in CMG software to construct a reservoir model for
the Lower Miocene reservoir – White Tiger field. Due to high
1 Yi ,obs
j , k  Yi , j , k ( x )
calc

2
complex nature of the geological characteristics, differences and
f ( x)  
2 i j k  i2, j ,k disconnection between Southern dome reservoir and Northern
dome reservoir; so authors created two separated models for
where, Yobs is the measured values of observable quantities Lower Miocene reservoir. The Southern structure model
such as bottomhole pressure, water cut or gas/oil ratio and consisted of 44220 grids (67x110x6) and the Northern structure
Ycalc(x) are their calculated values from a reservoir simulation consisted of 48000 grids (80x120x5) with over 50 wells. Well
with a certain set of values of the parameter x. The difference locations were originally selected to give reasonable pattern
between the observed values yobs and the calculated values ycalc coverage over those regions where the oil accumulation was
defines the quality of the history matching. These parameters considered to be the greatest. The simulation model contains
characterize the reservoir (e.g., porosity or permeability cell approximately 23 years of history. Production started in 1986
values) and have to be found so that the simulator closely and was updated until late 2007. Depletion was improved by
approximates the data. The index k is the type of observable water injection, which started 2.5 years after the earliest oil
data, the index j runs over the number of wells, the index i is the production.
measurement date for each well and σ2i define the variance of
the observed values.
History Matching
The object function can be defined to assess the match of
the dynamic data. We used the least square formulation which CMG is provides a detailed history matching of oil and
quantifies the errors between the simulation results and the water production over the entire 23 years history for each well.
observed data. So, a history problem can be posted as an By using history match analysis, the authors estimate a reservoir
inequality constrained nonlinear optimization problem: description for project areas for which production data were
provided. If history matching jobs are not producing good
Min Q = f(X) X  En results, the reservoir model will not reflect actual production.
s.t gi(X) ≥ 0 i = 1, …, m Then there are big differences between model and field
behavior. For this reason, its calculated results will not be
where the optimal vector X is a multi-dimensional space vector, accepted. If history matching is adequate, we will get a
the objective function f(X) and the inequality constrained confident reservoir simulation model on which give out optimal
function vector G(X) are different for different history matching production methods for all reservoirs. In this section, we further
problem. discussed our history matching procedures and presented a
In this study, the authors combined two methods (steepest method to evaluate errors between the model and history data.
descent and simplex algorithm) to solve the convergent problem In the history matching procedure, we combined Simplex
in history matching (Figure 4). This solution can take full and Steepest Decent algorithm to get the best results including
advantage of the gradient-based method (fast rate of success in matching and time minimization. The method was
convergence); it also minimizes the drawback when two tested being used to modify porosity and permeability of new
algorithms are combined. In our research, the reduced-order reservoir model for the White Tiger field. In this test, the three
technique which was presented by Nguyen The Duc et al., dimensional porosity and permeability distributions were
20087, is also applied to History matching to reduce the amount selected to be the modified parameters. Other parameters were
of matched parameters. In the large oil field, the number of not modified and their values were the same as in the basic
model parameters selected to modify the history matching reservoir model. The porosity and permeability distributions
process may be very large. In the reduced-order technique, obtained by the geological model are used as initial values for
groups of parameter in the same grid plane are combined and the iteration procedure.
related through a single parameter (Nguyen The Duc et al.,
2008)7. By using this constraint, the number of parameters The results of the optimization techniques were shown by
decreases from ni x nj x nk to ni, nj, nk parameters. This reducing the objective function values. The decrease of the
technique can be considered a good application for large oil average well pressure and water cut rate had a large
field. However, this constraint is not reasonable to use in the contribution to decrease the objective function value. Due to a

4
large well number, 6 wells that had the highest production were parameters such as cumulative oil, water cut, injection
shown in Figure 12-17. mass, and others.
2. There are differences of some important parameters
between simulation data and actual reservoir such as
Experimental Evaluation of History STOIIP and its distribution; porosity map and relative
Matching Results permeability curves.

To prove the exact level of the model and evaluate the 3. Some parameters such as reservoir rock compressibility
error between simulation and history data, the authors used the and fluid viscosity were determined in the lab. Then these
“reverse history matching” method. This reservoir was results were applied for the whole reservoir model. The
produced from November 1, 1986 until now, however the accuracy of these results can not be tested. This also
authors did not construct the model through this long time. By contributes to the significant error between the actual field
dividing the procedure into small stages, the authors have not and the model.
only assessed sensitivity and certainty of reservoir parameters 4. The grid size in reservoir simulation is usually in coarse
but also taken out the error between simulation and history data. scale (e.g., 250 x 250 x 50 m3), the space volume of these
However, it was noticed that all of history data of each specific grid are very large. If we apply porosity–permeability
times must be originally kept and not to apply the new data in value for all block, the results in simulation and production
previous times for new model. forecasting will not be exact.
Operation activities of the well vary from time to time. It is
difficult to update all of this new information for the reservoir
The first step
model. This is a disadvantage for history matching.
The reservoir model was constructed since the beginning
of field studies until December 31, 1999. Then, history
matching was also done for model in this period. When the Acknowledgements
result of history matching was good, authors forecasted the
Acknowledgment is extended to R&E Institute-
model in 3 years from January 1, 2000 to December 31, 2002
Vietsovpetro JVC for their permission and kind support to
(Table 2). Finally the authors evaluated the error between model
construct reservoir model for this research.
and real data. The level and timely adjusted characteristics of
model could be checked based on these results.
NOMENCLATURE
The second step Q = objective function
After assessing the error of the 1999 model, the authors X = multi-dimensional space vector
continued to do history matching for model until December 31,
2003. Then we forecasted the three-year reservoir performance G(X) = inequality constrained function vector
from January 1, 2004 to October 1, 2007 and took out error Yobs = measured value
between model and history data in this stage (Table 3). The
results of 2003 model are more exact than 1999 model because Ycalc = simulation value
its error is smaller. x = history-matching parameter
x0 = initial guess
Conclusion xr = reflection point

In this paper, the authors have proposed a coupling of the ρ = reflection coefficient
simplex method and steepest descent for the purpose of history σ = variance of the observed value
matching. The combination between two algorithms not only
reduces limitation of gradient based techniques but also γ = contraction coefficient
improves convergent speed of history matching problem. This
approach has been tested and it would be particularly useful for
history matching in large fields due to its advantages. Subscripts
Finally, the history matching results of two reservoirs i = the index of the measurement times for each well
were showed in Figure 10-11. From our experience in matching
more than 50 wells and based on the error evaluation of the two j = well index
stages, authors draw out some conclusions about error between k = type of observation
real data and simulation:
n = iteration step
1. There may be errors found in history matching due to the
difference between the number of wells (production wells
and injection wells) in model and the actual field because
of some wells which will be added into reservoir without REFERENCES
prediction at present. These errors will affect on certain
1. B. Griess, A. Daib and R. Schulze-Riegert.:
“Application of Global Techniques for Model

5
 Numerical Annalysis And Modelling (2005), Volume
2, 131-137.
13. Susana Gomez, Oliver Gosselin and John W. Barker.:
“Gradient-Based History Matching with Global
2. C.E. Romero, J.N Carter, A.C. Gringarten, R.W. Optimization Method”, SPE Journal, 200-208, June
Zimmerman.:“ A Modified Genetic Algorithm for 2001.
Reservoir Characterisation”, paper SPE 64765
presented at the International Oil and Gas Conference 14. Vietsovpetro Joint Venture EOR Study – annual
and Exhibition in China, 7-10 November 2000, reports, 1998 – 2007.
Beijing China. 15. Vietsovpetro Joint Venture Field Development Plans,
3. D.J. Schiozer.: “Use of Reservoir Simulation, Parallel 1998 & 2003.
Computing and Optimization Techniques to 16. Vietsovpetro Reservation Reports, 2002.
Accelerate History Matching and Reservoir
Management Decisions”, paper SPE 53979 presented 17. Xu Wang.: “Method of Steepest Descent and its
at the SPE Latin American and Caribbean Petroleum Applications”, November 25, 2008.
Engineering Conference, Caracas, Venezuela, 21-23
April 1999.
4. D.S. Chen, J.H. Fang and H.C. Chen.: “Novel
Approaches to the Determination of Archie
Parameters I: Simplex Method”, SPE Advance
Technology Series, Vol.3, No.1.
5. Jeffrey C. Lagarias, James A. Reeds, Margaret H.
Wright and Paul E. Wright.: “Convergence Properties
of the Nelder-Mead Simplex Method in Low
Dimensions”, SIAM J. OPTIM, Vol. 9, 112-147.
6. M.K. Choudhary, S. Yoon and B.E. Ludvigsen.:
“Application of Global Optimization Methods for
History Matching and Probabilistic Forecasting-Case
Studies”, paper SPE 105208 presented at the 15th SPE
Middle East Oil & Gas Show and Conference,
Bahrain, 11-14 March 2007.
7. Nguyen The Duc, Duong Ngoc Hai and Nguyen Minh
Toan.: “Applications Global and Local optimization
Algorithms to History Matching for Reservoir
Models”, Fractured Basement Reservoir Conference,
Vung Tau, Viet Nam, 2008.
8. R. W. Schulze-Riegert, O. Haase, A. Nekrassov.: “
Combined Global and Local optimization Techniques
Applied to History Matching”, paper SPE 79668
presented at the SPE Reservoir Simulation
Symposium, Houston, Texas, U.S.A, 3-5 February
2003.
9. R.W. Schulze-Riegert, O. Haase and D.T. Rian and
Y.-L You.:“ Evolutionary Algorithms Applied to
History Matching of Complex Reservoirs”, SPE
Reservoir Evaluation & Engineering, 163-173, 2002.
10. R.W. Schulze-Riegert, O. Haase and D.T. Rian.:
“Optimization Methods for History Matching of
Complex Reservoirs”, paper SPE 66393 presented at
the SPE Reservoir Simulation Symposium, Houston,
Texas, 11-14 February 2001.
11. S. Mantica, A. Cominelli and G. Mantica.:
“Combining Global and Local Optimization
Techniques for Automatic History Matching
Production and Seismic Data”, SPE Journal, June
2002, 123-130.
12. Shuguang Wang, Guozhong Zhao, Luobin Xu, Dezhi
Guo and Shuyan Sun.: “Optimization for Automatic
History Matching”, International Journal of

6
Table 1: PVT characterization of Lower Miocene Reservoir

Reservoir temperature 1200C

Oil Density 865.3 kg/m3
Gas Density 1.1 kg/m3
Water Compressibility 4.5 x 10-5 1/kg/cm2
Viscosity of water at reference pressure 0.2981 cp
Pressure dependence of water viscosity 2 x 10-5 cp/kg/cm2
Water density 1005.3 kg/m3
Water formation volume factor 0.96228
Reference pressure 248 kg/cm2

Table 2: Comparison between actual data and simulation (period 2000-2002)

2000 - 2002 STAGE

2000 2001 2002 Results of 2000- 2002 Stage
Object

Index Simulation Error Simulation Error Simulation Error Simulation Error

Actual Actual Actual Actual
1999 (%) 1999 (%) 1999 (%) 2004 (%)
Operational Production Well
20.0 16.0 20.0 19.0 16.0 15.8 19.0 16.0 15.8 19.0 16.0 15.8
Amount
Operational Injection Well
Lower Miocene

4.0 4.0 0.0 5.0 4.0 20.0 5.0 4.0 20.0 5.0 4.0 20.0
Amount
Cumulative Production Oil
227.6 156.4 31.3 237.7 146.1 38.5 199.8 128.0 35.9 665.1 430.5 35.3
(Thousands ton)

Water Cut, % 46.8 37.1 20.7 52.1 45.0 13.6 60.5 53.1 12.2 53.1 45.1 15.2
Mass of Injection
(Thousand tons) 222.0 170.0 23.4 269.0 230.0 14.5 293.0 287.0 2.0 784.0 687.0 12.4

Table 3: Comparison between actual data and simulation (period 2005-2007)

2005- 2007 STAGE

2005 2006 2007 Results of 2005- 2007 Stage
Object

Index Simulation Error Simulation Error Simulation Error Simulation Error

Actual Actual Actual Actual
2004 (%) 2004 (%) 2004 (%) 2004 (%)

Operational Production Well 27.0 25.0 7.4 27.0 26.0 3.7 27.0 26.0 3.7 27.0 26.0 3.7
Amount
Operational Injection Well
7.0 6.0
Lower Miocene

7.0 5.0 28.6 7.0 5.0 28.6 7.0 6.0 14.3 14.3
Amount

Cumulative Production Oil

209.7 175.6 16.3 206.5 197.5 4.4 140.8 134.9 4.2 557.0 508.0 8.8
(Thousands ton)

Water Cut, % 60.2 52.9 12.1 69.3 54.3 21.6 76.2 72.5 4.9 68.6 59.9 12.6

Mass of Injection
854.0 723.4 15.3 951.4 854.7 10.2 577.2 535.4 7.2 2382.6 2113.5 11.3
(Thousand tons)

7
 Direct Problem

System
INPUT Equations
Parameters Model ?OUTPUT?
Simulator

History Inverse Problem

Matching

System
?INPUT? Equations
?Parameters? Model
OUTPUT
Simulator

Figure 1 – History matching idea

Figure 2 – Downhill Simplex Method

Data Optimization processing Porosity &

Min objective function Permeability
Value

CMG simulator

Figure 3 - Steepest Descent Method Figure 4 – Optimization Processing

Figure 5 – Reservoir Modeling Results Figure 6 – Permeability Distribution

after Local and Global Optimization

8
Figure 7 – Porosity Distribution Figure 8 – Oil Saturation Distribution
after Local and Global after Local and Global Optimization

Figure 10 - History Matching Results of Northern Dome

Structure

Figure 9 – Comparison of initial pressure model and current pressure model

9
Figure 10 - History Matching Results of Northern Dome Structure

Figure 11 - History Matching Results of Southern Dome Structure

10
 Figure 12 - History Matching Results of Figure 13 - History Matching Results
Well 1A of Well 1B

Figure 14 - History Matching Results Figure 15 - History Matching Results

of Well 1C of Well 1D

Figure 16 - History Matching Results Figure 17 - History Matching Results

of Well 1E of Well 1F

11