Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Journal of Petroleum Science and Engineering 210 (2022) 109894

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering


journal homepage: www.elsevier.com/locate/petrol

Development of a method for adjusting rock compaction parameters and


aquifer size from production data and its application to Nam-Su fractured
basement reservoir of Vietnam
Le Ngoc Son a, Phan Ngoc Trung a, Yoshihiro Masuda b, Sumihiko Murata c, Nguyen The Duc d,
Sunao Takagi e, Ahmad Ghassemi f, *
a
Vietnam Oil and Gas Group (PETROVIETNAM), Viet Nam
b
School of Engineering, The University of Tokyo, Japan
c
Graduate School of Engineering, Kyoto University, Japan
d
Institute of Mechanics, Vietnam Academy of Science and Technology (VAST), Viet Nam
e
Japan Oil, Gas and Metals National Corporation (JOGMEC), Japan
f
The University of Oklahoma, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: The Nam-Su is a major naturally fractured basement reservoir (FBR), offshore Vietnam. To date, simulation
Aquifer size models of Nam-Su FBR have failed to give an adequate history match without invoking the presence of large
Computer-aided engineering aquifers. In a previous study (Son et al., 2007), the authors investigated several possible alternatives to achieve a
Fractured basement reservoir
satisfactory history match. They modeled porosity reduction by compaction along with the change of aquifer size
History matching
Permeability anisotropy
and produced an improved history match that is consistent with the geological nature of the system. The key to
Reservoir simulation such modeling is reservoir rock compressibility. Estimating the values of rock compressibility from cores is a
Rock compaction challenge in FBR’s due to the failure of coring from a naturally fractured interval, and thus reliable values are not
Reservoir compressibility available. We developed a computer-assisted history matching method to resolve these two problems together
and save significant time compared to the manual trial-and-error methods used to adjust these parameters
generally used in Vietnam. The methodology developed has been applied successfully to determine rock
compaction coefficients and adjust aquifers’ sizes of the Nam-Su FBR. Comparisons between our adjusted model
and the existing model show considerable improvement between computed and measured values. Simulators can
always be adjusted to obtain a history match even with geologically unrealistic values. The approach outlined
here is more physically realistic than existing approaches and hence should provide/provides better production
and other forecasts.

1. Introduction psia at − 3700 mSS (subsea), and the Original Oil In Place (OOIP) is
estimated as hundreds mmstb by the Operator. The structure is a large,
In Vietnam, fractured granite and other igneous and metamorphic elongated anticline, approximately 15 km by 2.5 km in size, trending
rocks have been major oil and natural gas reservoirs (Nguyen et al., NE-SW. No clear oil-water contact (OWC) has been found, and a massive
2011; Dang et al., 2011). For simplicity, we term these fractured base­ 1000 m thick oil column was proven during appraisal data of the field.
ment reservoirs (FBRs). The Nam-Su is an FBR in the Cuu Long Basin, Therefore, a structural spill point at − 4000mSS is assumed for the OWC.
offshore Vietnam continental shelf (Hung and Le, 2004). The reservoir Because of the reservoir rock’s very low porosity and permeability,
consists of fractured/weathered granite and granodiorite sealed by natural rock fractures are thought to provide the primary storage and
highly over-pressured shales, which act both as seal and source rock. The flow paths. There is almost no flow contribution from the rock matrix.
Nam-Su FBR was discovered in 2001 and put on production in 2003. The Outcrop analog studies and coring/logging analysis show that the
oil is 36oAPI gravity with a GOR of 210 scf/stb and a bubble point fracture distribution is very discrete. Fracture halos around faults
pressure (Pb) of 1200 psia. The initial reservoir pressure (Pres,i) was 5400 (Halos) have been identified from well’ results (borehole log, borehole

* Corresponding author.
E-mail address: ahmad.ghassemi@ou.edu (A. Ghassemi).

https://doi.org/10.1016/j.petrol.2021.109894
Received 30 December 2020; Received in revised form 9 November 2021; Accepted 21 November 2021
Available online 25 November 2021
0920-4105/© 2021 Elsevier B.V. All rights reserved.
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

image, mud loss, and production log data) and high-resolution 3D


seismic interpretation (Li et al., 2004).
The original simulation model was constructed based on the above
geological interpretations using a rectangular grid and using the
ECLIPSE reservoir simulator (Schlumberger, 2009a, 2009b). However,
the geological modeling of FBRs cannot be processed as conventional
sandstone reservoirs. The fracture system is difficult to distribute in a 3D
model. The Halo model (Darmadi et al., 2013) was adapted to solve
fracture distribution in the basement reservoir to address this issue. The
structural halo model is mainly based on a modeling approach that uses
seismically imaged faults and top surface reservoir for distributing a
porosity network. It is a simple but powerful application to characterize
a discrete fracture network model without complex geostatistical
modeling (Guttormsen et al., 2008). The porosity distribution in the
model followed the “Halo” concept, which assumed that a seismic-scale
master fault had subordinate lower-order fractures connected to it. And
the lower order fracture density decreased with i) increasing vertical
depth from reservoir top and ii) increasing horizontal distance from the
master fault. Then the permeability was estimated by
porosity-permeability relationships (applied for Nam-Su Full Field
Development Plan reports). In general, matching actual field dynamic
data could be acceptably achieved by adjusting both the pore volume
(for both the reservoir porosity and aquifer volume) and connectivity Fig. 2. Geochemical and stable isotope results for Nam-Su produced
(including permeability and flow boundary conditions). The water source.
history-matched model could give a generally acceptable prediction of
field performance, but the global and local modification of porosity and
the same hydrostatic gradient with the reservoir, and this pressure re­
aquifer size without strong geological support indicated the existing
mains constant when the reservoir is produced. Therefore, the aquifer
uncertainty on the volume-related input parameters of the model. After
size was unknown and treated as one of the adjusted parameters for
more than ten years of production, significant questions remain. In
reservoir modeling in this study. The assumed sizes of the aquifers were
particular, the level and source of the reservoir energy are unclear, and
determined by the manner of ‘trial and error’ during the manual history
the true OOIP is unknown.
matching process for the reservoir model performed by the field oper­
When production began, the reservoir pressure dropped quickly, and
ator. Table 1 shows the assumed sizes of the aquifers in the existing
little formation water was produced (Fig. 1), so there was no clear
Nam-Su reservoir simulation model (model OLD).
indication of aquifer support. The Operator then carried out water in­
Besides the energy support from the aquifer, the authors investigated
jection, which reduced the pressure decline but caused direct water
three possible scenarios that could explain the maintenance of reservoir
breakthrough in producers. Later on, water injection was stopped
pressure (Pres) (Son et al., 2007), including i) increasing pore volume; ii)
because of the high water cut in producers and the stability of reservoir
changing rock compressibility (Cr), and iii) porosity reduction (fracture
pressure. The maintenance of reservoir pressure without water injection
closure) due to compaction by changing Cr with effective pressure
(Fig. 1), as well as the produced water analysis results (geochemical and
reduction during production. The study results indicated that by
stable isotope data - Fig. 2), indicated the possible existence and impact
applying scenarios ii) and iii), the model could give very good history
of an aquifer. A series of flow simulations were tried using a single
matching without invoking unrealistically large aquifer volumes. A
porosity approach. However, as expected, each result could not be his­
reservoir modeling study has been carried out (Son et al., 2007) to solve
tory matched satisfactorily without invoking enormous active aquifers
these issues. However, adjustments of rock compaction coefficients and
from the sides of the reservoir. There is no direct information about
aquifer size were carried out by a “trial and error” approach. This
aquifer parameters, including its volume and pressure. However, for
manual history matching method was found to be very time-consuming
modeling, the aquifer pressure is simplified by assuming that it follows
and non-unique. Therefore, to develop an improved model, an alterna­
tive automatic history matching method using Computer-aided Engi­
neering (CAE) for determining and calibrating the rock compaction
coefficients and aquifer size has been proposed and applied in our work
presented here.
Compared with manual history matching, automatic history

Table 1
Water volumes of aquifers before and after adjustment.
No Active direction Water Volume (BSTB)

OLD Model NEW Model

1 Western 2.00 1.08


2 Southern 3.00 1.52
3 Eastern 2.00 1.03
4 Western 2.50 1.35
5 Western 3.00 1.72
6 Eastern 2.00 1.11
7 Eastern 5.50 2.98
8 Eastern 0.90 0.48
Total 20.90 11.27
Fig. 1. Nam-Su’s reservoir pressure, oil production, and water injection rate.

2
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

matching can avoid randomness and blindness caused by researchers 2.1. Value of the objective function
and reduce the workload significantly. One of the first studies on auto­
matic history matching was done by Jacquard and Jain (1965). They The value of the objective function is calculated as the combination
have developed a method to modify the permeability map for mini­ of two overall standard deviations (root-mean-square deviations) be­
mizing the difference between observed and actual pressures at the tween measured and calculated water rate and bottom-hole shut-in
wells. Chavent et al. (1975) studied history matching in single-phase oil pressure in all wells and at all times that have measured data:
reservoirs. Later on, Fasanino et al. (1986) looked at single-phase history ( )1
matching of 2D gas reservoirs using the adjoint method combined with E = α2R .ER2 + α2P .EP2 2 (1)
geostatistical information and the pilot point technique. And, Li et al.
(2003) studied history matching of three-dimensional, three-phase flow where: ER is the overall standard deviation between measurement and
production data. Their study minimizes the mismatch in flowing well­ calculation of water rate calculated by equation (2), EP is the overall
bore pressure, producing gas-oil ratio (GOR) and water-oil ratio (WOR). standard deviation between measurement and calculation of bottom-
Reviews of studies on automatic history matching methods can be found hole pressure calculated by Equation (4); αR and αP are the weighted
in Oliver and Chen (2011) and Rwechungura et al. (2011). factors.
Because porosity and permeability are two of the most important The overall standard deviation between measurement and calcula­
parameters in reservoir simulation models, most studies have developed tion of the water rate is determined from the mismatches between the
a method for modifying permeability and/or porosity distributions. measurement and calculation of all the wells and at all measuring times:
Only a study reported by Mittermeir et al. (2004) developed an auto­ ⎛NW NOj ( )2 ⎞12
∑∑
matic history matching method to modify aquifer parameters to the qw obs
− qwsim
⎜j=1 i=1 j,i j,i ⎟
authors’ knowledge. The modification of aquifer size is also of interest in ⎜
ER = ⎜

⎟ (2)
our work. In our opinion, because both rock compaction coefficients and ∑
⎝ NW ⎠
NOj
aquifer size strongly affect reservoir pressure, they should be adjusted j=1

simultaneously.
In automatic history matching, parameter adjustment is performed where j is the index of the well, NW is the number of the wells, i is the
by minimizing the objective function using an optimization algorithm. index of the measurement times, NOj is the number of measurement
Many optimization algorithms have been used in different automatic times of the jth well, qwobs
j,i is the measured water rate of the well, and
history matching studies. These include both global (non-gradient) and qwsim
j,i is the calculated water rate of the jth well at the ith measurement
local (gradient) optimization algorithms. Non-gradient algorithms ∑
time. The denominator in equation (2), NW j=1 NOj , represents the number
covered include evolutionary strategies, simulated annealing, genetic
algorithm, ensemble Kalman filter, and many other algorithms or hybrid of the water rate measurement data of all wells. The degree of water rate
approaches (see, e.g., Gómez et al., 2001; Ouenes et al., 1994; Mantica match at jth well is evaluated as follows:
et al., 2001; Liu and Oliver, 2005; Wen and Chen, 2007). Gradient al­ ⎛NO ( )2 ⎞12
∑j
gorithms covered include the steepest descent, conjugate gradient, ⎜
obs sim
qwj,i − qwj,i ⎟
Gauss-Newton, Quasi-Newton, Variable-Metric (see, e.g., Yang and ⎜
ER,j = ⎜ i=1

⎟ (3)
⎝ NOj ⎠
Watson, 1988; Makhlouf et al., 1993; Chu et al., 2000; Wu, 2001; Zhang
and Reynolds, 2002). Non-gradient algorithms usually take a long time
to converge. On the other hand, the gradient methods are only conver­ Similarly, the overall standard deviation between measurement and
gent to the local minimum. An improved optimization algorithm named calculation of the well pressure between measurement and calculation is
Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm calculated according to the expression:
has been developed by Spall (1992, 1998). The essential feature of SPSA
⎛NW NOj ( )2 ⎞12
is gradient approximation that requires only two objective function ∑ ∑ obs sim
evaluations regardless of the dimension of the optimization problem. p
⎜j=1 i=1 j,i − pj,i ⎟
⎜ ⎟
These two evaluations are made by simultaneously randomly varying all EP = ⎜ ⎟ (4)
⎝ ∑
NW ⎠
the variables. The SPSA algorithm is especially efficient for NOj
j=1
high-dimensional problems by providing a good solution for a relatively
small number of objective function evaluations. The application of SPSA
where pobs sim
j,i is the measured pressure and pj,i is the calculated pressure at
algorithms to do automatic history matching was carried out by Gao
the ith measurement time of the jth well; other symbols have the same
et al. (2004).
meaning as in equation (2).
Descriptions of our automatic history matching method for cali­
The degree of pressure matching at jth well is evaluated as follows:
brating the rock compaction coefficients and aquifer size are presented
in Section 2. The application of the method to the Nam-Su FBR reservoir ⎛NO (
∑j obs )2 ⎞12
simulation model is presented in Section 3. The resulting model is ⎜ pj,i − psim
j,i ⎟
⎜ ⎟
compared to the manual history-matched model of the field operator. EP,j = ⎜ i=1 ⎟ (5)
⎝ NOj ⎠
The comparison shows considerable improvement between simulation
results and observed field data.
Minimizing the objective function calculated by equations (1)–(5)
2. Method description means reducing the mismatches between measurement and calculation
of all the wells and at all measurement times.
The method has been developed for simultaneously calibrating rock
compaction coefficients and aquifer size. In this approach, the optimi­
zation algorithms are used to locate the minima of the objective func­ 2.2. Adjusted parameters
tion, which quantifies the mismatch between measurement and
calculation of the two types of well data: water rate and bottom-hole For oil and gas reservoirs, rock compaction has a significant influ­
shut-in pressure. ence on reservoir pressure during production. The effect of rock
compaction causes a reduction in the gross reservoir volume and in­
crease in the reservoir solids volume due to the increase of effective

3
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

overburden pressure (Pk) or decline in Pres.


W1 = λ2 .W1,i ; W2 = λ2 .W2,i ; …; Wn = λn .Wn,i (12)
In the reservoir simulation, rock compaction effects are modeled by
parameterizing the pore volume change (multipliers on pore volume)
Where n is the number of aquifers; W1 , W2 , …, Wn are the aquifer vol­
and the change of permeability (multipliers on transmissibility). The
umes after adjustment; subscript (i) refers to initial value; and λ1 , λ2 , …,
most commonly used modeling method to calculate pore volume change
λn are multiplying factors that will be adjusted.
with reservoir pressure is the value of rock compressibility. Rock
By using the above parameterization techniques, simultaneous ad­
compressibility is defined as the change in pore volume concerning a
justments of rock compaction coefficients and aquifer size can be per­
change in reservoir pressure. In this modeling approach, the porosity
formed by solving the following optimization problem:
change is approximated as follows:
Determine the values of parameters: a, b, c, d, λ1 , λ2 , …, λn in order to
Φ minimize the function:
= eCr (Pres − Pres,i )
(6)
Φi
E = E(a, b, c, d, λ1 , λ2 , …, λn ) (13)
where Cr is the rock compressibility, Φis the porosity, Pres is the reservoir
where the value E is calculated from equation (1). The function (13) is
pressure, and the subscript (i) refers to the initial value.
determined from reservoir simulation.
However, rock compressibility clearly is not independent of reservoir
pressure. For example, compressibility may decline as the pressure first
declines below the initial pressure because of grain rearrangement. 2.3. 3 Optimization algorithms
Therefore, several other modeling methods have been developed for
rock compaction. Many optimization algorithms can be used to minimize the function
In our study, the compaction model proposed by Pettersen (2005) (10) in our adjustment method described in Section 2. Seven different
has been used. In this model, porosity change is approximated as conventional optimization algorithms have been incorporated into our
follows: computer program, namely:

Φ 1. Steepest descent algorithm (see, e.g., Fletcher, 1987; Press et al.,


2
= ea(Pres − Pres,i )+b(Pres − Pres,i )
(7)
Φi 1992).
The following similar equation approximates permeability change: 2. Gauss-Newton algorithm (see, e.g., Fletcher, 1987; Press et al.,
1992).
k 2
3. SPSA algorithm (Spall 1992, 1998).
= c.ec(Pres − Pres,i )+d(Pres − Pres,i )
(8)
ki 4. SIMPLEX algorithm (Nelder and Mead, 1965).
5. Direction set algorithm (see, e.g., Fletcher, 1987; Press et al., 1992).
where k is the permeability, and the subscript (i) refers to the initial
6. Conjugate gradient algorithm (see, e.g., Fletcher, 1987; Press et al.,
value.
1992).
By using equations (7) and (8), four parameters a, b, c and d will be
7. Variable metric algorithm (see, e.g., Fletcher, 1987; Press et al.,
adjusted to determine the effect of rock compaction on pore volume and
1992)
transmissibility.
Different methods can be applied to assess the aquifer support of
The detailed descriptions of the listed optimization algorithms can be
petroleum reservoirs. There are several popular aquifer models in the
found in the references.
ECLIPSE reservoir simulator. The Fetkovich model (Fetkovich, 1971) is
used in the existing Nam-Su reservoir simulation model (model OLD). In
3. Application to Nam-Su fractured basement reservoir (Nam-Su
the Fetkovich model, the water influx rate, qw , can be calculated as
FBR)
follows:
qw = Jw (p − pw ) (9) 3.1. Geological setting of the field

where Jw is the productivity index of the aquifer, p is the average aquifer The Nam Su field is located offshore Vietnam in an area known as the
pressure and pw is the boundary pressure of the reservoir. The maximum Cuu Long Basin. There are both fractured/weathered basement, and
encroachable water Wei can be calculated from the material balance as overlying Miocene/Oligocene plays in the Cuu Long. Fig. 3 shows a
follows: schematic of the two plays as well as hydrocarbon source rocks and oil
Wei = ct .W.pi (10) migration pathways. Fig. 4 shows a generalized stratigraphic column for
the Cuu Long Basin. The fractured basement, which consists of granite
where ct is the total compressibility of the aquifer, W is the water volume and granodiorite batholiths, is considered to be Late Jurassic to Early
of the aquifer, and pi is the initial aquifer pressure, the average aquifer Cretaceous in age. The granites consist mainly of microline, oligoclase,
pressure can be calculated as follows: quartz, and smaller amounts of biotite.
Five stages of development are generally accepted for the creation of
pi
p= − We + pi (11) the basement reservoirs in the Cuu Long Basin. The first stage of
Wei
development was the emplacement of the intrusive bodies (mostly
where We is the cumulative outflow of the aquifer. granites) in existing country rock from the Late Triassic through the
Based on equations (9)–(11), according to the Fetkovich model in the Early Cretaceous periods. In the second stage of development, during the
ECLIPSE simulator, the aquifers modeling requires water volume, pro­ Middle Cretaceous, tectonic processes began deforming the basement
ductivity index, total compressibility, initial pressure, and water prop­ rock leading to significant faulting, fracturing, and the creation of
erties. Only volumes of aquifers have been estimated during the manual breccia zones. During the third stage of development, the Late Creta­
history matching process for the reservoir model performed by the field ceous and Early Paleogene, significant weathering and erosional pro­
operator. Other parameters of all aquifers are given the same as the cesses occurred, which exposed the basement granites and initiated
reservoir or as default by the simulation software. meteoric circulation. With influence from a hydrothermal activity from
Adjustment of aquifer size in our method is performed by adjusting below the granites, this meteoric circulation initiated the enhancement
multiplying factors for the volume of all aquifers: of the fracture systems within the granites. The process of fracturing and
solution enhancement was further reinforced by rifting and uplift of

4
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

Fig. 3. Cuu long basin petroleum system.

et al., 1992). Three dominant fracture types (bounding fractures,


discrete fractures, and drilling-induced fractures) at Nam-Su FBR have
been characterized using core, logs, and/or seismic data. Bounding
fractures exhibit lengths from meters to more than a kilometer, and the
larger types can be seen on seismic (Fig. 5). They provide the dominant
storage and flow paths in the productive parts of the reservoir. Discrete
fractures are background fractures that provide potential storage.
Drilling-induced fractures are present only in cores and wellbore images,
are a direct result of drilling operations, and do not provide storage
within the reservoir.
Porosity determination for an FBR is a critical issue for reserves
estimation and well placement. In the granite basement rocks, matrix
porosity is essentially zero (“fresh granite”) or very small (mineral-filled
dikes, intrusives), and any reservoir porosity is entirely made up by
fracture volume (Li et al., 2004). To ensure and enhance the productivity
of the wells, they are always designed to penetrate perpendicular to the
seismic fracture plane and hit as many fractures as possible. As such, a
complete characterization of the fractures system is essential to deter­
mining reservoir porosity.
The fracture characterization for the Nam-Su field has been pro­
cessed by all available information, including the well logging, mud loss,
production logging, and dynamic data collected during production. The
data confirms that reservoir property (porosity or facture density) is
constrained only to fracture systems. As typically shown in Fig. 6 and
Fig. 7, for any type of fracture (reverse or normal or strike-slip faults),
the associated reservoir property has a decreasing trend with i)
increasing distance from a seismic-scale bounding fracture and ii)
increasing depth from the top of the basement. It can be explained for
this decreasing trend by the reduction of rock damage (or reduction in a
number of smaller-scale fractures) with the distance away from the
Fig. 4. Generalized Cuu long basin stratigraphic column.
bounding fracture or deeper from the top of the reservoir. Based on this
observed trend of reservoir property distribution, and with limited geo-
basement blocks. In total, some five significant phases of deformation mechanic data, a normalized, simple “Halo” model has been applied
and fault/fracturing development are known to have occurred in this (Darmadi et al., 2013). For Nam-Su’s Halo model, relationships were
third stage of development. The fourth stage of developing the Cuu Long developed between the larger bounding-type features seen at seismic
Basin reservoirs was the reburial of the basement and the development scale and those smaller sub-seismic features. The size of the Halo ranges
of the regional seal. The fifth and final stage of development was the from 200 m to 400 m at the top of the reservoir (at a depth of − 3000
generation and migration of hydrocarbons in the Neogene and Pleisto­ mSS), tapering to only a few meters at the base (around assumed OWC at
cene periods. − 4000 mSS), reflecting decreasing fracturing with increasing depth as
seen in Fig. 8. Porosities follow the same trend, with Φmax = 7% at the
3.2. Geological model top of the faults where the Halo is widest, down to 0.001% near the base
of the faults as seen in Fig. 7. The Operator used the relationship be­
Unfractured/unweathered basement rock has essentially no porosity tween k and Φ given by Jones (1975): k = kmax [Φ/Φmax ]3 with the
and no permeability. What gives basement reservoir properties is the correction to match the well data for the reservoir model. The initial
degree of faulting/fracturing and the weathering along these faults and distributions of porosity and permeability for the Nam-Su FBR simula­
fractures to provide storage volume (Dmitriyevskiy et al., 1993; Areshev tion model can be observed in Fig. 9.

5
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

Fig. 5. Cross-section view of major faults along Nam-Su FBR.

Fig. 7. Nam-Su normalized porosity vs. distance from a seismic fracture.


Fig. 6. Nam-Su porosity vs. depth relationship.

developed and produced. The model was developed based on the


This Halo model has been validated with actual data. The prediction ECLIPSE black oil reservoir simulator. The updated reservoir model has
of existing and planned wells’ performance is reasonably satisfied. Using the following main inputs and assumptions:
history matched model, the Operator has successfully carried out
reservoir management and development activities. • Through 12 years of production, historical data comes from 16
A notification is that following the Halo concept, with very small production wells and two water injection wells (Fig. 9).
porosity below OWC at the depth − 4000 mSS, it is hard to explain the • The grid dimensions are 149 × 38 × 37 (104,168 active cells) in the
existence of any big aquifer at the bottom of the reservoir. In fact, based
x, y, and z directions, respectively.
on the analysis results of produced water isotope data, it has been • OWC assumed at structural spill point of − 4000mSS.
identified that the aquifer locates in the clastic sediments deposited in
• Minimum WHFP: 250 psi
the flanks of Nam-Su FBR. The reservoir pressure depletion might be the • Maximum WHIP: 2500 psi
reason for the water influx through the fault systems.
• Eight aquifers of the field were invoked for history matching reser­
voir energy. Fetkovich analytical model was chosen to simulate
aquifers (Schlumberger, 2009a, b). The positions of aquifer supports
3.3. Existing Nam-Su FBR simulation model
can be seen in Fig. 10. One of the aquifers (AQ2) assumed provides
support from the southern part of the reservoir with a water volume
Based on the geological model, the reservoir simulation model of
of 3.0 BSTB; Four aquifers (AQ3, AQ6, AQ7, and AQ8) assumed to
Nam-Su FBR has been constructed repeatedly as the field has been

6
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

• The history matching process is based on matching well bottom-hole


pressures and water cut while controlling production and injection
rates as seen at each existing wells.
• History matching results of the existing Nam-Su reservoir simulation
model (model OLD) can be shown in Fig. 11.

3.4. Adjustment results and discussions

A computer code has been developed based on the adjustment


method presented in Section 2 with seven optimization algorithms as
mentioned above. The computer code was used to adjust rock compac­
tion coefficients and volumes of eight aquifers determined in the existing
Nam-Su FBR simulation model.
In order to see how the forecasts of adjusted models match the
observed data, history production data of only the first nine years
(January of 2003 to April of 2012) are used to adjust the model. The
history production data of the last three years (May of 2012 to April of
2015) are used to evaluate the forecasting ability of simulation models.
Fig. 8. Fracture damage zone, the “halo” concept. The objective function value calculated by equation (1) has been
calculated using monthly measured and simulated data of water rate and
provide support from the eastern edge of the reservoir with water bottom-hole pressure of the first nine years of all production wells. The
volumes of 2.0 BSTB, 2.0 BSTB, 5.5 BSTB, and 0.9 BSTB respectively. value of the weighted factor αR and αP are taken so that the contribution
Three aquifers (AQ1, AQ4, and AQ5) were assumed to provide sup­ of produced water rate mismatch (between measurement and simula­
port from the western part of the reservoir with water volumes of 2.0 tion) in the objective function is equivalent to the contribution of
BSTB, 2.5 BSTB, and 3.0 BSTB, respectively. These assumed sizes of bottom-hole pressure mismatch. The iterative process of the method
the aquifers had been evaluated during the manual history matching begins with starting values of rock compaction coefficient a, b, c, and
process for the reservoir model performed by the field operator. OWC d equal 9.0 × 10− 06, 0.0, 0.0, and 0.0, respectively. The starting values of
of all aquifers is assumed at − 4000 mSS; Datum depth of all aquifers multiplying factors for all aquifers equal 1.0.
is assumed at − 2800 mSS; Initial pressures and productivity index of At first, to compare optimization algorithms and select a suitable one
all aquifers are set by default of the simulation software, which for the application, we used all these optimization algorithms to mini­
means they are calculated to place these aquifers as nearly as mize objective function (1) for a ‘shorten’ model of the existing Nam-Su
possible in equilibrium with the reservoir; Rock compressibility and reservoir simulation model. The simulation is only performed for the
water properties in all aquifers are set to be the same as those in the first two years in the shortened model. The objective function (1) is
reservoir. calculated from simulated results and historical production data for the
• Cr = 9 × 10− 06 psi− 1; Swc = 15%, Sor = 40%. first two years. Application of seven algorithms to the stated adjustment
• One set of kr curves and zero Pc is used. problem has been performed. In each application, with the same starting
above rock compaction coefficients and aquifer parameters, we ran a
corresponding algorithm with 100 function evaluations (each function

Fig. 9. Distributions of porosity and permeability in the existing reservoir simulation model of Nam-Su FBR.

7
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

evaluation requires one time of reservoir simulation for the first two
years). The reduction rates of the objective function (1) obtained by
these different algorithms have been compared. The percentage de­
creases of the objective function after 100 function calls corresponding
to seven optimization algorithms above are shown in Table 2. Fig. 12
shows the obtained minimum value of the objective function depending
on the number of function calls corresponding to these different algo­
rithms. It can be seen that the maximum reduction rate of the objective
function is associated with the SPSA algorithm. Therefore, the SPSA
algorithm has been selected to apply the adjustment method with full
history production data.
With a tolerance of 0.001 for the iterative process of the SPSA al­
gorithm, adjusting rock compaction coefficients and aquifer size takes
about five days computing by a PC 3 GHz. The obtained adjusted values
of rock compaction coefficient a, b, c, and d are equal to 10.92228 ×
10− 06, 8.616952 × 10− 10, 2.194425 × 10− 06, and − 1.425004 × 10− 10
respectively. These adjusted results mean that the following correlations
present the rock compaction effect in the changes of porosity and
permeability:
Φ 2
= e10.92228E− 06(Pres − Pres,i )+8.616952E− 10(Pres − Pres,i )
(14)
Φi
Fig. 10. Assumed aquifers and their volumes of existing Nam-Su reservoir
k 2
model (OLD).
= e2.194425E− 06(Pres − Pres,i )− 1.425004E− 10(Pres − Pres,i )
(15)
ki
The modeling of rock compaction effects on pore volume (multiplier
on pore volume) and permeability (multiplier on transmissibility) before
and after adjustment can be seen in Fig. 13. Fig. 13 shows the multipliers
on pore volume and transmissibility as functions of reservoir pressure
drop (Pres,i-Pres) before adjustment (OLD) and after adjustment (NEW).
As mentioned in Section 3.2, the Cr is equal to 9.0 × 10− 06 psi− 1 and
independent of reservoir pressure in the existing Nam-Su FBR simulation
model. After adjustment, the Cr changed in the range from 10.92 ×
10− 06 psi− 1 at initial pressure to 3.68 × 10− 06 psi− 1 at bubble point
pressure. At the initial condition, the Cr varied from 6.79 × 10− 06 psi− 1
at the top of FBR to 11.62 × 10− 06 psi− 1 at the assumed OWC. This result
is reasonable and is supported by a previous study (Son et al., 2007),
which found that reservoir rock should be divided vertically (from top of
FBR down to OWC) into several compaction regions. The rock at OWC
depth that has a four (4) times smaller porosity than the rock at the top
of the reservoir (Fig. 6) would have higher Cr (inverse proportional
correlation between Φ and Cr has been published by some authors).
Our computed value is compared to previous works on the deter­ Fig. 11. History matching results of existing Nam-Su reservoir simulation
mination of Cr. The Operator used Conventional Laboratory Methodol­ model (OLD).
ogy (CLM) and Earth Loading Methodology (ELM) to determine Cr for
the Nam-Su FBR. Due to the difficulty in core sampling and recovery, pressure associated with fluctuations in earth loads, such as tidal and
there is no core available for the fracture system. Using CLM, rock barometric forces. The Cr values for the Nam-Su reservoir using the ELM
compressibility is measured from an almost “fresh granite” core, which method range from 6.5 × 10− 6psi− 1 to 13.3 × 10− 6psi− 1. It can be seen
is the matrix rock with very low permeability and porosity. With an that our computed values for Cr are very similar to the ones determined
overburden pressure (Pob) gradient in the Nam-Su area of approximately by the ELM method.
one psi/ft and under Pi of 4450 psia at datum depth, Cr is measured at ca. As presented, along with rock compaction coefficients, the volumes
30 × 10− 6psi− 1. The ELM is an in situ method and relies on highly ac­ of the eight aquifers are also adjusted in our methodology. The process
curate downhole gauges for measurements of fluctuations in reservoir uses aquifer size in the existing Nam-Su model as a starting point, and
obtained results show the values of adjustment factors of all aquifers
(multiplying factorsλ1 , λ2 , …, λn ) are in the range from 0.5 to 0.6. The
Table 2
Percentage decreases of the objective function after 100 function calls ob­ adjustment factors for the eight aquifers from AQ1 to AQ8 (see Fig. 10)
tained by different optimization algorithms in method testing with the are: λ1 = 0.542142, λ2 = 0.505435, λ3 = 0.513051, λ4 = 0.538488, λ5 =
shortened model. 0.574720, λ6 = 0.557024, λ7 = 0.542624 and λ8 = 0.538769, respec­
tively. The values of aquifers’ volumes after adjustment are also sum­
Optimization algorithm Percentage decrease
marized in Table 1 together with their old values.
Steepest descent algorithm 2.40
The mismatch between measurements and calculations of the ob­
Gauss-Newton algorithm 2.73
SPSA algorithm 2.85 tained reservoir simulation model using adjusted rock compaction co­
SIMPLEX algorithm 2.60 efficient and aquifer volumes (model NEW) is compared with one of the
Direction set algorithm 0.22 current manual history-matched models of the Operator (model OLD).
Conjugate gradient algorithm 2.05 Table 3 shows the overall standard deviations of water rate (ER ) and
Variable metric algorithm 2.71

8
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

calculation presented in Table 3 indicate that model NEW is statistically


better in matching with water rate measurement compared to model
OLD. A quantitative comparison between two models in matching with
measured water rate at well-level have been made by evaluating and
comparing the values of the standard deviation between measurement
and calculation of produced water rate ER,j for all production wells.
These values are calculated from Eq. (3). The maximum, minimum,
mean, and median values of ER,j are presented in Table 4 for two models.
Table 4 shows that the model NEW gives smaller maximum, mean, and
median values of ER,j than that of model OLD. Only the minimum value
of ER,j of model NEW is greater than that of model OLD. It also can be
seen that the model NEW gives a smaller standard deviation with
measurement of water rate ER,j in a larger number of wells compared to
model OLD.
Comparison of two models for each well matching the measured
water rate data may be observed on the cumulative water and water rate
plots versus time. Because of the large number of production wells, only
two plots for two wells are presented here in Figs. 14 and 15. Fig. 14
shows one of the cases in which the value of standard deviation with
Fig. 12. Performances of different optimization algorithms in method testing measurement of water rate ER,j of model NEW is greater than that of
with a shortened model. model OLD. On the other side, Fig. 15 shows the case in which the value
of standard deviation with measurement of water rate ER,j of model
NEW is smaller than that of model OLD. The comparison plots for other
wells can also be observed. From these plots and Table 4, it is clear that
the model NEW gives the agreement of water rate better in a larger
number of wells than that of model OLD. The comparison plots also
indicate that the forecasting ability for the production water rate of the
model NEW is better than that of model OLD.
For well bottom-hole pressure, the values of overall standard devi­
ation between measurement and calculation EP in Table 3 also indicate
that model NEW is statistically better in matching well bottom-hole
pressure than that of model OLD. The better agreement with measure­
ment of model NEW can also be shown in plots of well bottom-hole
pressure versus time. For illustration, the plots for two wells are pre­
sented here in Figs. 16 and 17. Fig. 16 shows the plot for the first pro­
duced well of the field. The plot of the well having the most rate
measurements is presented in Fig. 17. The plots of bottom-hole pressure
comparison for other wells can also be observed. More quantitative
comparison between the two models can also be obtained by evaluating
the values of the standard deviation between measurement and calcu­
Fig. 13. Multiplier on pore volume and transmissibility vs. reservoir pressure lation of well bottom-hole pressure EP,j obtained by these two models.
before adjustment (OLD) and after adjustment (NEW). These values are calculated from Eq. (5). The maximum, minimum,
mean, and median values of EP,j are presented in Table 5. It can be seen
from Table 5 that the model NEW gives smaller maximum, minimum,
Table 3
mean, and median values of EP,j in comparison with model OLD. The
Overall standard deviations between measurement and calculations obtained by
two models. result of the comparison also shows that the value of EP,j of model NEW
is smaller than that of model OLD in a larger number of the production
Overall standard deviation Model OLD Model NEW
wells. The comparison plots of well bottom-hole pressure also show the
Overall standard deviation of water rate ER (STB/ day) 2418.64 2368.82 better forecasting ability of model NEW.
Overall standard deviation of pressure EP (psi) 326.57 258.41 The comparisons above of history match of calculated results be­
tween two models for all production wells have shown the agreement
with measurement of model NEW is statistically better than that of
bottom-hole pressure (EP ) between measurement and calculations ob­
tained by these two reservoir simulation models. The calculations of ER
and EP in equations (2) and (4) respectively have been performed with
Table 4
all monthly measurement times and all production wells of the field. The Comparison of well water rate standard deviations between measurement and
water rate is measured in 16 production wells. The bottom-hole pressure calculations obtained by two models.
is measured in 14 production wells of the field. It can be seen from
Statistical quantity Model Model
Table 3 that the overall standard deviations with measurement of both OLD NEW
water rate and bottom-hole pressure of model NEW are smaller than that
Maximum value of ER,j (STB/day) 108.39 134.01
of model OLD. The overall standard deviation with the measurement of
Minimum value of ER,j (STB/day) 4887.95 4565.84
the water rate obtained by model NEW is 3.5 percent smaller than that
Mean value of ER,j (STB/day) 1876.32 1754.88
obtained by model OLD. The overall standard deviation with measure­
Median value of ER,j (STB/day) 1877.25 860.78
ment of bottom-hole pressure obtained by model NEW is 19.7 percent
Number of wells having smaller ER,j in comparison with 6 10
smaller than that obtained by model OLD.
the remaining model
The values of overall standard deviation between measurement and

9
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

Fig. 17. Comparison of bottom-hole pressure between measurement and sim­


ulations for well P4.
Fig. 14. Comparison of produced water rate between measurement and sim­
ulations for well P1.

Table 5
Comparison of well bottom-hole pressure standard deviation between mea­
surement and calculations obtained by two models.
Statistical quantity Model Model
OLD NEW

Maximum value of EP,j (psi) 60.63 57.21


Minimum value of EP,j (psi) 671.53 552.50
Mean value of EP,j (psi) 284.08 230.41
Median value of EP,j (psi) 226.13 215.17
Number of wells having smaller EP,j in comparison with 3 11
the remaining model

Fig. 15. Comparison of produced water rate between measurement and sim­
ulations for well P2.

Fig. 18. Comparison of produced water rate between measurement and sim­
ulations for Nam-Su FBR.

matched in comparison with model OLD at the field level. It can also be
seen that the model NEW has better forecasting ability than the model
OLD at the field level.
In summary, the comparison between the two models has shown a
considerable improvement of the Nam-Su FBR simulation model by
Fig. 16. Comparison of bottom-hole pressure between measurement and sim­ applying our proposed method in determining rock compaction co­
ulations for well P3. efficients and adjusting aquifers’ volumes. Besides introducing the
newly developed method as mentioned above, the key finding in this
study is that using a single value of rock compressibility is not suitable
model OLD. Along with well-level evaluations, the comparisons of
agreement with the measurement between two models have also been for the simulation of FBRs with high thickness. Using a rock compaction
approach is likely the solution for this kind of reservoir in Vietnam. In
implemented at the field level. The measured and calculated total field
cumulative water production is presented in Fig. 18. The comparisons of the time to come, further investigation of methods to determine the
range of this parameter (Cr) as well as the aquifer size based on addi­
measurement and model simulations on produced water rate of the field
can be seen in Fig. 19. It indicates that model NEW is better history tional studies on geomechanics, reservoir, and aquifer characterization

10
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

Acknowledgement

The research was supported by state project code 001.19.CNKK.QG


of Vietnam. We also appreciate the comment by anonymous reviewers,
which improved the paper.

Abbreviations

AQ Aquifer
Bach Ho Bach Ho fractured basement field in Vietnam
CLM Conventional Laboratory Methodology
ELM Earth Loading Methodology
FBR Fractured Basement Reservoir
FOPR Field oil production rate
FOPRH Field oil production rate history
FWCT Field water cut
Fig. 19. Comparison of cum. Water production between measurement and FWCTH Field water cut history
simulations for Nam-Su FBR.
FWIR Field water injection rate
FWPR Field water production rate
is still needed to solve the remaining questions such as whether aquifer FWPRH Field water production rate history
is ten times bigger than OOIP, or the wide range of Cr, etc. FWPT Field water production cumulative total
FWPTH Field water production cumulative total history
4. Conclusions GOR Gas Oil Ratio
Model OLD Existing model created by the Operator of the field
The current modeling of fractured basement reservoirs (FBRs) faces Model NEW NE-SW New model created for this study by the authors
problems with the application of uncertain parameters. Rock compac­ Northeast - Southwest
tion effect with the variation of rock compressibility by depth should be OOIP Oil originally in place
considered for these thick, high-relief reservoirs, but the direct deter­ OWC Oil Water Contact
mination of the compaction coefficient is practically impossible due to SPSA Simultaneous Perturbation Stochastic Approximation
the lack of representative cores. In addition, aquifers’ volume is another Algorithm
parameter that strongly affects reservoir energy but is not adequately SS Subsea
determined. ST Sidetracked
This study has developed a methodology to adjust rock compaction PermX Permeability value in the X-direction
and aquifers’ sizes from production data. The adjustments were per­ TVT True vertical thickness
formed automatically using optimization algorithms. The developed WBHP Well bottom hole pressure
method and accompanied computer tools were applied for the Nam-Su WBHPH Well bottom hole pressure history
FBR model, and a considerable improvement in matching with WHFP Wellhead flowing pressure
observed data was obtained. Therefore, the proposed method could be WHIP Wellhead injection pressure
considered the most suitable approach in determining and calibrating WWPR Well water production rate
the rock compaction and aquifer parameters for this reservoir type. WWPRH Well water production rate history
WWPT Well water production cumulative total
SI metric conversion factors WWPTH Well water production cumulative total history

bbl × 1.589874 E − 01 = m3 References


ft × 3.048 E − 01 = m
Areshev, E.G., et al., 1992. Reservoirs in fractured basement on the continental shelf of
ft3 × 2.831685 E− 02 = m3
southern Vietnam. J. Petrol. Geol. 15 (4), 451–464. https://doi.org/10.1111/j.1747-
md × 9.86923 E− 16=m2 5457.1992.tb01045.x.
nD × 9.86923 E− 22=m2 Chavent, G., Dupuy, M., Lemonnier, P., 1975. History matching by use of optimal theory.
SPE J. 15 (1), 74–86. https://doi.org/10.2118/4627-PA.
psi × 6.894757 E+03=Pa.
Chu, L., Komara, M., Schatzinger, R.A., 2000. An efficient technique for inversion of
reservoir properties using iteration method. SPE J. 5 (1), 71–81. https://doi.org/
Authors credit statement 10.2118/60224-PA.
Dang, C.T.Q., Chen, Z., Nguyen, N.T.B., Bae, W., 2011. Improved oil recovery for
fractured granite basement reservoirs: historical lessons, successful application, and
Le Ngoc SON: Formal analysis (Lead), Investigation (Lead). Meth­ possibility for improvement. In: Paper Presented at the SPE European Formation
odology (Lead). Writing – original draft (Lead). Phan Ngoc TRUNG: Damage Conference. https://doi.org/10.2118/144148-MS. Noordwijk, The
Netherlands, June 2011, SPE-144148-MS.
Conceptualization, Supervision (Lead), Writing – review & editing. Darmadi, Y., Harahap, A., Achdiat, R., 2013. Reservoir characterization of fractured
Yoshihiro MASUDA: Conceptualization (Lead), Formal analysis (Equal), basement using seismic attributes, dayung field case study, south sumatra Indonesia.
Writing – review & editing. Sumihiko MURATA: Conceptualization, In: Proceedings of the Indonesian Petroleum Association Thirty-Seventh Annual
Convention and Exhibition. https://doi.org/10.29118/IPA.0.13.G.155. May 2013.
Supervision, Writing – review & editing. Sunao TAKAGI: Writing – re­
Dmitriyevskiy, A.N., et al., 1993. Hydrothermal origin of oil and gas reservoir in
view & editing (Lead). Nguyen The DUC: Data curation, Project basement rock of the south Vietnam continental shelf. Int. Geol. Rev. 35 (7),
administration (Lead), Software (Lead), Validation. Ahmad Ghassemi 621–630. https://doi.org/10.1080/00206819309465547.
Review and Technical Editing. Fasanino, G., Molinard, J.E., de Marsily, G., Pelcé, V., 1986. Inverse modeling in gas
reservoirs. In: Paper Presented at the SPE Annual Technical Conference and
Exhibition. https://doi.org/10.2118/15592-MS. New Orleans, Louisiana, October
1986, SPE-15592-MS.
Declaration of competing interest Fetkovich, M.J., 1971. A simplified approach to water influx calculations—finite aquifer
systems. J. Petrol. Technol. 23 (7), 814–828. https://doi.org/10.2118/2603-PA.

The authors have no conflict of interest.

11
L.N. Son et al. Journal of Petroleum Science and Engineering 210 (2022) 109894

Fletcher, R., 1987. Practical Methods of Optimization, second ed. John Wiley & Sons, Oliver, D.S., Chen, Y., 2011. Recent progress on reservoir history matching: a review.
New York. Comput. Geosci. 15 (1), 185–221. https://doi.org/10.1007/s10596-010-9194-2.
Gao, G., Li, G., Reynolds, A.C., 2004. A stochastic optimization algorithm for automatic Ouenes, A., Doddi, R.S., Cunningham, G., Saad, N., 1994. A new approach combining
history matching. SPE J. 12 (2), 196–208. https://doi.org/10.2118/90065-PA. neural networks and simulated annealing for solving petroleum inverse problems. In:
Gómez, S., Gosselin, O., Barker, J.W., 2001. Gradient-based history matching with a Conference Proceedings, ECMOR IV - 4th European Conference on the Mathematics
global optimization method. SPE J. 6 (2), 200–208. https://doi.org/10.2118/71307- of Oil Recovery. https://doi.org/10.3997/2214-4609.201411185 cp-233-00044 Jun
PA. 1994.
Guttormsen, J.J., Wade, J.M., Achita, R., 2008. Domain based geologic modeling of a Pettersen, O., 2005. Sandstone compaction modelling and reservoir simulation. In:
fractured granite reservoir. Indonesian petroleum association. In: Proceedings of Conference Proceedings, 67th EAGE Conference & Exhibition. https://doi.org/
32nd Annual Convention & Exhibition. May 2008, IPA08-G-207. 10.3997/2214-4609-pdb.1.I002. Jun 2005, cp-1-00347.
Hung, N.D., Le, H.V., 2004. Petroleum geology of Cuu Long Basin - offshore Vietnam. In: Press, W.H., et al., 1992. Numerical Recipes in Fortran: the Art of Scientific Computing,
Proceedings of AAPG International Conference. Barcelona, Spain, September 21–24, vol. 402. Cambridge University Press, New York.
2004. Rwechungura, R., Dadashpour, M., Kleppe, J., 2011. Advanced history matching
Jacquard, P., Jain, C., 1965. Permeability distribution from field pressure data. SPE J. 5 techniques reviewed. In: Paper Presented at the SPE Middle East Oil and Gas Show
(4), 281–294. https://doi.org/10.2118/1307-PA. and Conference. https://doi.org/10.2118/142497-MS. Manama, Bahrain,
Jones, F.O., 1975. A laboratory study of the effects of confining pressure on fracture flow September 2011, SPE-142497-MS.
and storage capacity in carbonate rocks. J. Petrol. Technol. 27 (1), 21–27. https:// Schlumberger, 2009a. ECLIPSE, Reference Manual 2009, vol. 1. Schlumberger Editor,
doi.org/10.2118/4569-PA. 2009.
Li, R., Reynolds, A.C., Oliver, D.S., 2003. History matching of three-phase flow Schlumberger, 2009b. ECLIPSE, Technical Description 2009, vol. 1. Schlumberger
production data. SPE J. 8 (4), 328–340. https://doi.org/10.2118/87336-PA. Editor, 2009.
Li, B., Guttormsen, J., Hoi, T.V., Duc, N.V., 2004. Characterizing permeability for the Son, L.N., Jamiolahmady, M., Questiau, J.M., Sohrabi, M., 2007. An integrated geology
fractured basement reservoirs. In: Paper Presented at the SPE Asia Pacific Oil and and reservoir engineering approach for modelling and history matching of a
Gas Conference and Exhibition. https://doi.org/10.2118/88478-MS. Perth, Vietnamese fractured granite basement reservoir. In: Paper Presented at the
Australia, October 2004, SPE-88478-MS. EUROPEC/EAGE Conference and Exhibition. https://doi.org/10.2118/107141-MS.
Liu, N., Oliver, D.S., 2005. Critical evaluation of the ensemble kalman filter on history London, U.K., June 2007, SPE-107141-MS.
matching of geologic facies. In: Paper Presented at the Reservoir Simulation Spall, J.C., 1992. Multivariate stochastic approximation using a simultaneous
Symposium. https://doi.org/10.2118/92867-PA. Houston, Texas, 31 January - 2 perturbation gradient approximation. IEEE Trans. Autom. Contr. 37 (3), 332–341.
February 2005, SPE 92867. https://doi.org/10.1109/9.119632.
Makhlouf, E.M., Chen, W.H., Wasserman, M.L., Seinfeld, J.H., 1993. A general history Spall, J.C., 1998. Implementation of the simultaneous perturbation algorithm for
matching algorithm for three-phase, three-dimensional petroleum reservoirs. SPE stochastic optimization. IEEE Trans. Aero. Electron. Syst. 34 (3), 817–823. https://
Adv. Technol. 1 (2), 83–91. https://doi.org/10.2118/20383-PA. doi.org/10.1109/7.705889.
Mantica, S., Cominelli, A., Mantica, G., 2001. Combining global and local optimization Wen, X.H., Chen, W.H., 2007. Some practical issues on real-time reservoir model
techniques for automatic history matching production and seismic data. In: Paper updating using ensemble kalman filter. SPE J. 12 (2), 156–166. https://doi.org/
Presented at SPE Reservoir Simulation Symposium. https://doi.org/10.2118/66355- 10.2118/111571-PA.
MS. Texas, 11–14 February 2001, SPE 66355-MS. Wu, Z., 2001. A Newton-raphson iterative scheme for integrating multiphase production
Mittermeir, G.M., et al., 2004. Automated determination of aquifer properties from field data into reservoir models. SPE J. 6 (3), 343–351. https://doi.org/10.2118/62846-
production. In: Proceedings of 9th European Conference on the Mathematics of Oil MS.
Recovery, Cannes, France, 30 August - 2 September 2004, cp-9-00037. Yang, P.H., Watson, A.T., 1988. Automatic history matching with variable-metric
Nelder, J.A., Mead, R., 1965. A simplex method for function minimization. Comput. J. 7 methods. SPE Reservoir Eng. 3 (3), 995–1001. https://doi.org/10.2118/16977-PA.
(4), 308–313. https://doi.org/10.1093/comjnl/7.4.308. Zhang, F., Reynolds, A.C., 2002. Optimization algorithms for automatic history matching
Nguyen, N.T., Dang, C.T.Q., Bae, W., 2011. Geological characteristics and integrated of production data. In: Conference Proceedings, ECMOR VIII - 8th European
development plan for giant naturally fractured basement reservoirs. In: Paper Conference on the Mathematics of Oil Recovery. https://doi.org/10.3997/2214-
Presented at the Canadian Unconventional Resources Conference, Calgary. https:// 4609.201405958. September 2002, cp-104-00042.
doi.org/10.2118/149510-MS. Alberta, Canada, November 2011, SPE-149510-MS.

12

You might also like