Vous êtes sur la page 1sur 15

Journal of Petroleum Science and Engineering 27 2000.

3347
www.elsevier.nlrlocaterjpetscieng

An expert system for selecting and designing EOR processes


Ridha B.C. Gharbi )
Department of Petroleum Engineering, College of Engineering and Petroleum, Kuwait Uniersity, P.O. Box 5969, Safat 13060, Kuwait
Received 21 April 1999; accepted 25 February 2000

Abstract
The importance of enhanced oil recovery EOR. processes in the petroleum industry makes its understanding and
prediction critical in the decisions on the applicability of certain recovery techniques. A large number of EOR projects have
been applied in many regions around the world. This has lead to an increased understanding of the selection and applicability
of certain oil recovery projects. In this paper, an artificial intelligence AI. technique has been applied to assist in the
selection and design of EOR processes. The expert system developed in this study is able to perform the following
consultations: 1. select an appropriate EOR process on the basis of the reservoir characteristics, 2. prepare appropriate
input data sets to design the selected EOR process using the existing numerical simulators, and 3. make optimization studies
on several key parameters selected by the user. in order to optimize oil recovery from the selected EOR process. q 2000
Elsevier Science B.V. All rights reserved.
Keywords: artificial intelligence; expert systems; enhanced oil recovery; porous medium

1. Introduction
During the last few decades, the petroleum industry has experienced a rapid increase in the number of
enhanced oil recovery EOR. projects being conducted. This has occurred because most of the worlds
giant fields have already been drained and are in an
advanced state of depletion from conventional production methods. These methods include primary
production, pressure maintenance, and water flooding. A larger amount of oil is yet to be recovered

Tel.: q965-481-1188; fax: q965-563-2994.


E-mail address: ridha@kuc01.kuniv.edu.kw R.B.C. Gharbi..

using EOR techniques compared to exploration of


new fields.
The major EOR processes include thermal, miscible, and chemical methods. Thermal recovery methods add heat to the reservoir by the use of steam or
in-situ combustion. to reduce oil viscosity. Reservoir
depth is limited because of the heat loss associated
with wells. Miscible recovery methods are based on
injecting solvents that will mix with the oil under
reservoir conditions to dissolve and displace more
oil. Examples of these solvents include carbon dioxide CO 2 ., light hydrocarbon gases, and nitrogen.
Unfortunately, these methods involve injecting fluid
that is much more mobile than the resident oil. The
resulting adverse mobility ratio, coupled with reservoir heterogeneity, may cause early breakthrough of

0920-4105r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 0 - 4 1 0 5 0 0 . 0 0 0 4 9 - 8

34

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

the injected fluid and poor oil recovery. In order to


have miscibility of the solvent with the reservoir
fluids, the reservoir pressure must be higher than the
minimum miscibility pressure MMP.. Correlations
are available in the literature to estimate the MMP of
solvents versus temperature for different crude oil
systems Mungan, 1981; Holm and Josendal, 1974..
Chemical EOR methods are designed to improve the
sweep efficiency by adding chemicals polymer or
surfactant. to the injected water in order to reduce
interfacial tensions or create a favorable mobility
ratio, and thus enhance oil production.
Unfortunately, heterogeneity and the interaction
of several forces viscous forces driven by adverse
mobility ratios, capillary forces from interfacial
forces between immiscible fluids, gravity forces
driven by fluid density gradients and dispersive forces
caused by concentration gradients between miscible
fluids. can make many EOR displacement processes
unstable and difficult to predict. Existing methods
for the prediction of fluid flow in porous medium
include laboratory core flood experiments and numerical simulation techniques.
During the last 15 years, the petroleum industry
all over the world has experienced a rapid increase in
the number of expert systems being developed because of greater availability of human experts and
publication of a larger number of case studies. One
of the early expert systems is DIPMETER ADVISOR which was developed to perform well log
analysis Baker, 1984.. Since then, a number of
other expert systems were developed in various disciplines of petroleum engineering such as drilling
Bergen and Hutter, 1986; Thompson and Dunlap,
1985; Mabile et al., 1989; Fenoul, 1989; Chiu et al.,
1993; Kulakofsky et al. 1992; Heinze, 1992; Amara
and Martin, 1990; Cayeux and Overti, 1992; Martinez, 1992; Tangen and Baleix, 1992; Fear et al.,
1994; Jellison and Klementich, 1990; Peden and
Tovar, 1991., stimulation Cram and Hendrickson,
1986; Holditch et al., 1993; Blackburn et al., 1990;
Van Domelen et al., 1992; Chavane and Perthuis,
1992., formage damage Vitthal et al., 1989; Thomas
et al., 1995., well logging Einstein and Edwards,
1990; Soto and Soto, 1995; Walsh et al., 1993., well
testing Al-Kaabi et al., 1990; Allain and Houze,
1992., and production Darwich et al., 1989;
Schirmer et al., 1991; Cordier et al., 1991; Hutchin

et al., 1996; Ramirez et al., 1990; Keating et al.,


1991; Derek et al., 1988; Martinez et al., 1993;
Corpoven, 1995; Patricio et al., 1994; Greffioz et al.,
1993; Alegre et al., 1993.. There have been applications in the area of EOR screening Guerillot, 1988;
Shindy et al., 1997; Chung et al., 1995; Maksimov
and Tetelbaum, 1992; Elemo and Elmtalab, 1993.,
simulators input data preparation Khan et al., 1992.
and fluid properties Dharan et al., 1989.. The objective of this paper is to develop an expert system
that combines EOR screening, simulator input preparation, and parameter selection that optimizes the
recovery from selected EOR processes.

2. Artificial intelligence (AI)


AI studies are divided into two main categories:
studies that try to mimic the operations of human
brains and studies that understand and apply thinking
methodologies. The first is the artificial neural networks ANNs. and the second is the classical AI.
Since AI techniques became aligned with conventional computer hardware architecture in the middle
1980s, their economical applications to petroleum
exploration and production have become available
Alegre, 1991.. Fuzzy logic systems, neural networks and expert systems are three AI technologies
having a major impact in the petroleum industry
McCormack and Day, 1993..
Fuzzy logic, invented in 1964, is an approach to
reasoning where the rules of inference are approximate rather than exact Zadeh, 1979.. It is useful for
manipulating information that is incomplete, imprecise, or unreliable. Traditional set theory defines set
membership as a Boolean predicate e.g., tall
means being greater than some specific height, and
either you are tall or you are not.. Fuzzy set
theory represents set membership as a possibility
distribution the greater the numeric value assigned
to your height, the more likely you are to be tall..
Once set membership has been redefined in this way,
you can define a reasoning system based on techniques for combining distributions. Fuzzy logic has
applications in control theory. When you are programming things to function in a complex environment, fuzzy rules may be easier to derive and faster

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

to use than explicit formulae. Since fuzzy logic is


used mainly for efficiency, some people think that it
is doomed by the emergence of massively parallel
processing.
ANNs, a biologically inspired computing methodology, have the ability to learn by imitating the
learning method used in human brain. It is an interconnected assembly of simple processing elements,
units, or neurons, whose functionality is loosely
based on the brain neuron. The processing ability of
the network is stored in the inter-unit connection
strengths, or weights, obtained by a process of adaptation to, or learning from, a set of training patterns.
Neural networks are well suited to complex problems. They generally have large degrees of freedom,
thus they can capture the non-linearity of the process
being studied better than conventional regression
methods. ANNs are relatively insensitive to data
noise, as they have the ability to determine the
underlying relationship between model inputs and
outputs, resulting in good generalization ability. A
neural network model can be subjected to additional
training in order to adapt itself to new situations at
which its inputoutput performance is inadequate.
Expert systems, also known as knowledge-based
systems KBS., are programs that contain a knowledge base and a set of algorithms or rules that infer
new facts from knowledge and from incoming data.
An expert system uses the knowledge base of human
expertise to provide expert advice and aid in solving
problems. The degree of problem solving is based on
the quality of the data and rules obtained from the
human expert. Expert systems allow the human expertise to be accumulated and stored in a computer.
Once stored, the expertise can be retrieved at any
time and used to solve problems in a specialized
area. Expert systems, once successfully developed,
provide a permanent knowledge base. This base can
be easily transferred in a concise and economical
way. Expert systems also can be used to ameliorate
the performance of individuals because of incorporation of knowledge from other experts in the field.
An expert system program contains three parts: a
knowledge-base; an inference engine and user interface. The knowledge base is the first part of an
expert system which contains all the knowledge or
expertise in the form of rules and facts. This step
requires the acquisition, or gathering of available

35

knowledge, and then storage of this knowledge using


a knowledge representation language in a form recognizable by a computer. The expert system develops answers by running the knowledge base through
an inference engine a software program that interacts with the user and processes the results from the
rules and data in the knowledge base.. Thus, the
inference engine provides the path that directs one
toward the solution. The user interface is the part
that establishes the communication between the user
and the expert program. It requests input from the
user and presents the results obtained from the expert
system to the user. Some expert systems require
running large external programs, which is the case in
this study. If this is the case, an expert system can be
directly linked to such programs so that their results
can be used in the inference process.
Expert systems can be developed using either an
AI language, or an expert system shell. AI languages
are more flexible compared to expert system shells,
but they require much more programming. These
languages are different from the conventional programming languages such as FORTRAN, BASIC,
PASCAL, etc. Each of the AI languages works on a
different paradigm and thus offers different features
Russo and Peskin, 1987.. The expert system shells
contain a built-in inference engine which has a
knowledge representation language and pre-specified
control strategies. Several commercial expert system
shells, each with different features, are available in
the market. Expert system shells are more convenient
to use than AI languages; however, expert system
shells offer limited or no capabilities for modification of the inference engine. In this study, the expert
system was developed using the expert system shell:
Nexpert Object Arcidiacono, 1988..
Nexpert Object, developed by Neuron Data, supports both rule-based reasoning and object-oriented
knowledge representation. This makes it more a
powerful tool for knowledge representation compared to strictly rule-based shells that do not allow
any object-oriented features. Because it is written in
C programming language, the knowledge base can
be used on a large number of computing platforms
personal computers, workstations, and mainframes..
It includes a user-friendly graphical user interface,
and it has the ability to run external programs.
Furthermore, Nexpert Object has the following addi-

36

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

tional features: 1. modification and retrieval of


databases, 2. flexibility to dynamically control the
inference engine, and 3. interactive debugging. The
rules of Nexpert are set up using the IFTHENDO
statement. The IF part contains the conditions which
have to be satisfied, in order to execute all the
actions in the DO part. These actions may include:
1. running external programs, 2. retrieval or modification of a database, 3. performing numerical
calculations, and 4. assigning or setting parameter
values.

3. Description of the simulators


The two simulators used in this study are
UTCHEM and UTCOMP, both of which developed
at The University of Texas at Austin Saad, 1989;
Chang et al., 1990..
UTCHEM is a three-dimensional chemical flood
simulator. The solution scheme is analogous to implicit pressureexplict saturation IMPES., where
the grid block pressure is solved for implicitly, but
component concentrations, rather than phase saturations, are solved for explicitly. An energy balance
equation is solved explicitly for reservoir temperature. The energy balance equation includes heat flow
between the reservoir and the over- and under-burden
rocks. The simulator is capable of modeling up to
three-phase flow, which are: 1. aqueous phase, 2.
oleic phase and 3. microemulsion phase. It can
handle multiple components, namely, water, oil, surfactant, polymer, anions, divalent cations, alcohols,
tracers and gel. The program code is fully vectorized, thus it is very efficient when running on vector
machines. The major physical phenomena modeled
in the simulator include dispersion, dilution effects,
adsorption, interfacial tension, capillary trapping,
cation exchange, aqueous reactions, dissolutionrprecipitation, and in-situ generation of surfactant from
acidic crude oil.
UTCOMP is a three-dimensional, equation-ofstate EOS. compositional simulator. UTCOMP is
able to model up to four phases, i.e., 1. an aqueous
phase, 2. an oleic phase, 3. a vapor phase, and 4.
a second non-aqueous phase. Phase behavior is calculated using the PengRobinson EOS or a modified

version of the RedlichKwong EOS. A rigorous


Gibbs stability test is performed before all flash
calculations to determine the number of co-existing
phases present, and a phase identification is test is
done after all flash calculations to consistently label
each phase for subsequent property calculations Perschke, 1988.. Several relative permeability models
are available as options.
A finite-difference method that is third-order correct in space is used in both simulators to reduce
numerical dispersion and grid orientation effects
when solving the material balance equations. An
iterative solver based on the Jacobi-conjugate-gradient JCG. is used to solve the pressure equation. The
JCG method is strongly recommended when using
fine-mesh simulation with a large number of grid
blocks. The physical dispersion in the two simulators
is modeled using a full dispersion tensor, and capillary and gravity effects are included. Molecular diffusion and mechanical dispersion contribute to the
elements of the dispersion tensor. Different values of
longitudinal and transverse dispersivities can be used.
The two simulators have been used for many years
by many oil companies, and have been subjected to
rigorous testing. More detailed descriptions on
UTCHEM and UTCOMP are given in Saad 1989.
and Chang et al. 1990., respectively.

4. Expert system development


As mentioned previously, Nexpert Object supports both object-oriented and rule-based reasoning
for knowledge representations hybrid approach..
This capability was fully used to develop the knowledge base. The development of the knowledge base
was initiated by creating several class-object data
structures for various input parameters. There are
two types of hierarchical data structures: classes and
objects. Each class and object can have subclasses
and sub-objects, respectively. These data structures
were then integrated into several sets of rules to form
the knowledge base. The rules in a particular set
constitute chunks of knowledge called knowledge
islands. A set of meta-rules then governs the control
strategy. The methodology implemented in the ex-

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

pert system is analogous to using a control blackboard Carrico et al., 1989.. Depending on the event,
both forward and backward chaining control strategies are used.
The system was formulated by representing the
acquired knowledge using the conditional IF
THENDO statements. These statements are represented by either numeric values or characters. For
example, the range of certain parameters e.g., API
gravity. are represented by numeric values, whereas,
others e.g., type of formation. are represented by
characters i.e., sand or carbonates.. Some other
statements may require yesrno answer e.g., swelling
clay content..
The delivery system for Nexpert Object called
Nexpert Forms was used to develop a customized
user interface. The Nexpert Form allows the creation
of interactive questionranswers windows in a character-based environment. It also allows the display of
messages, user help, and results during and after
consultation. This is done by developing what are
called text files. An example of a form used in the
expert system is shown in Fig. 1. In order to configure the expert system to interact with various processes, a run time definition RTD. file was developed. This file is the driver program for the expert
system. It invokes Nexpert Forms by executing the
main routine in the expert system code, loading the
knowledge base, and starting the consultation. The

37

inference engine of the Nexpert Forms runs the


expert system while interacting with several other
processes. Details on how to develop forms, text
files, and RTD files are given in the Nexpert Forms
Users Guide 1995.

5. Discussion
The expert system developed in this study is able
to perform the following consultations: 1. select an
appropriate EOR process on the basis of the reservoir characteristics, 2. prepare appropriate input
data sets to design the selected EOR process using
the existing numerical simulators, and 3. make sensitivity studies on several key parameters in order to
optimize the oil recovery from the selected EOR
process. Only technical screening guides are used in
the expert system and no economic evaluation is
performed. The economic evaluation of the selected
EOR process must take place before any decision is
made. Fig. 2 shows a flow chart for the expert
system. Numerous parameters should be carefully
studied and selected in order to, first, assess the
applicability of various EOR methods, and second,
to design and optimize the oil recovery obtained
from these methods.
The most difficult part when developing an expert
system is knowledge acquisition. The source of

Fig. 1. An example of a form used in the expert system.

38

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

Fig. 2. Flow chart for the expert system.

knowledge used to develop the expert system includes the following: 1. literature review, 2. case
histories, 3. simulation experts, andror 4. experimental data. The first phase carried out in the development of the expert system was a comprehensive
literature review of the successful EOR processes
around the world. During the last 3 decades, several
investigators have studied different methods both
experimentally and numerically. for recovering more
oil from depleted reservoirs. These reservoirs often
still contain more than 50% of original oil in place.
Many of these studies are now available in the
literature, thus providing criteria for application of
these different methods. Ten EOR methods are considered in step 1. of the expert system. These
processes are: steam flooding, carbon dioxide CO 2 .
flooding, miscible hydrocarbon injection, polymer
flooding, surfactant flooding, alkaline flooding, nitrogen flooding, hot water injection, in-situ combustion, and microbial flooding. The first four EOR
processes are the most common ones. It should be
noted that the user has the option of selecting water
flooding prior to step 1. of the expert system. If this

is the case, then step 1. of the expert system is not


performed and, depending on the users selection,
the program will produce input data files for either
UTCOMP or UTCHEM to design water flood.
There are many parameters affecting the selection
of these EOR processes: 1. depth, 2. pressure, 3.
temperature, 4. rock type, 5. formation thickness,
6. porosity, 7. permeability, 8. API gravity, 9. oil
viscosity, 10. dipping angle, 11. connate water
saturation, 12. swelling clay content, 13. crude oil
type, 14. acid number, and 15. water salinity.
These parameters were used in the expert system for
selection of EOR process suitable in a particular
reservoir. This represents the largest number of parameters that have been considered for EOR selection. The other parameters were not included in the
expert system because their effects on EOR processes are yet to be fully investigated. Several rules
have been extracted from previous studies Khan et
al., 1992; Burger et al., 1985; Premuzic and Woodhead, 1993; Taber and Martin, 1983; Slider, 1983;
Schumacher, 1980; Bailey and Curtis, 1984; Dafter,
1980; Haynes et al., 1976; Lake, 1993. and are used
in the knowledge base of step 1. of the expert
system. The knowledge base is continuously updated
to include new data when available.
Step 1. of the expert system is to provide the
user with all the feasible EOR processes as well as
the most suitable one for a particular reservoir. This
is done in two steps: 1. eliminate any processes that
may not be feasible for a particular reservoir, and 2.
find out from the remaining feasible processes if
more than one., the one that has more probability of
a success. Once step 1. of the expert system is
executed, the user is then asked to either stop here or
continue with step 2. of the expert system. The
objective of step 2. is to prepare the appropriate
input data sets to design the selected EOR process
from step 1. using the numerical simulators. A
substantial simulation experience has been gained
using the two reservoir simulators UTCHEM and
UTCOMP.. This serves as a basis for the simulation
aid development. These two simulators were briefly
described in a previous section. It should be noted
that step 2., and consequently step 3., will not be
performed if the selected EOR process is one of the
following: steam flooding, microbial flooding, or
in-situ combustion. The reason is that the two numer-

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

ical simulators used in this study do not handle these


applications. Study is underway to implement other
simulators that can simulate these EOR processes.
Several sources for acquiring knowledge have been
used for input parameter selection. Such sources
include the extensive simulation experience gained
by the use of these two simulators, individuals
opinion, and the users guide for UTCHEM and
UTCOMP. The two simulators have evolved over
many years and have been used extensively by others over a long period.
The input data set for the chemical or compositional flood simulator include a large number of
parameters to be selected. These parameters include:
fluid description data, reservoir description data, numerical aspects data, and inputroutput control options. Fig. 3 and 4 show an overview of UTCHEM
and UTCOMP input parameters, respectively. An
example input data set used for UTCOMP simulator
is shown in Appendix A. These input parameters
must be carefully chosen in order to make accurate
and efficient simulations. The conventional approach
to select an input data set is through the use of users
Guide. This approach can be time consuming and a
slight mistake may cause serious errors. Khan et al.
1992. have developed an expert system to prepare
input data sets for UTCOMP to simulate miscible
flooding. This experience is extended to cover input
data sets not only for UTCOMP, but also for
UTCHEM the chemical flood simulator.. Some of
the input parameters do not require an expert advice,

Fig. 3. Overview of UTCHEM input data.

39

Fig. 4. Overview of UTCOMP input data.

such as reservoir temperature and initial pressure.


Other input parameters, however, such as numerical
dispersion control options and values of dispersivities, are selected by the user and do require an expert
advice. The expert system can provide this advice,
and can be used for the following:
selection of pseudo-components and their properties for crude oil systems,
selection of time-step size control parameters,
selection of dispersivities values i.e., longitudinal
and transverse.,
selection of the matrix solver which can be used
for this simulation,
selection of EOS for a given crude oil system for
UTCOMP.,
selection of appropriate relative permeability and
capillary pressure curves.
The procedure used to develop the input data files
is performed through a step-by-step procedure. The
user is asked several questions and unnecessary
questions are eliminated based on the users answers.
Fig. 5 shows a flow chart for step 2. of the expert
system. During the creation of the input file, short
simulation runs one- or two-dimensional runs. may
be performed in order to estimate some of these
parameters such as the optimum slug size. The ex-

40

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

mance from the EOR process selected in step 1. of


the expert system. The appropriate resolution of
these issues is a key to economically apply the
selected EOR process in a given field. These issues
include:

selection of well type horizontal or vertical.,


selection of horizontal well length,
selection of well pressures andror rates,
selection of injectors and producers,
recommendation of additional wells to be drilled
their types and locations.,
selection of concentrations in the injected fluids,
selection of optimal slug sizes,
selection of optimal WAG ratio.

Fig. 5. Flow chart for step 2. of the expert system.

pert system provides recommendations for several


input parameters and the user has the choice whether
to use them or override them.
The user can also choose whether to use geostatistical techniques in assigning permeability and porosity values to various grid blocks, or assign values for
these parameters. Two statistical packages are included in the expert system that can be used to
generate geostatistical data for a given problem. One
of the statistical packages uses the turning band
methods TBM., and the other uses the matrix decomposition method MDM.. Additional details on
these packages can be found in a reference by Yang
1990.. At the end of consultation of step 2. of the
expert system, an input data file is generated and can
be readily used with UTCHEM or UTCOMP, depending which EOR process was selected in step 1.
of the expert system. If the user selected a geostatistical technique for the permeability field, then an
additional file is generated which contains the permeability values to the grid blocks.
Step 3. of the expert system involves studying
several key issues to optimize the recovery perfor-

The user may choose from a table of options one


or more of these parameters to be investigated. Once
parameters are chosen, the expert system will
perform all the necessary simulations in order to
investigate the effect of these parameters on the
performance of the EOR process. Based on these
simulations, the expert system recommends the value
of the parameter that gave the best oil recovery from
the selected EOR process. For example, the user may
choose to study the effect of changing some of the
existing vertical wells to horizontal wells. In this
case, the expert system automatically makes the
corresponding changes needed in the input file and
simulations are carried out to study the performance
of the new horizontal wells. The user may also
choose to study the effect of injection rates on the
flood performance. In this case, the injection rates of
all the injectors are varied systematically and simulations are repeated in order to optimize the flood
performance. The expert system then displays the
wells and their corresponding new injection rates.
Step 3. of the expert system is divided into two
options. Option 1. is used when the user wishes to
study the effect of several parameters on the oil
recovery one at a time. New value of the selected
parameter is assigned and simulation is repeated with
this new value. The corresponding simulation results
are then displayed and the user may wish to either
stop or continue by assigning another value to the
same parameter or study the effect of another parameter on the flood performance. The second option, or
option 2., is available when the user wishes to make

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

an optimization study for one or more parameters. In


this case, the user may choose several parameters to
be investigated at the same time. These parameters
are then systematically varied over a range and
simulations are repeated for all necessary combinations. The output of the simulator is fed into the
expert system in the form of data representing the
history recovery curves. The expert system finally
recommends a combination of these parameters that
optimizes the flood performance. The user may also
choose to study the effect of heterogeneity on the
flood performance. In this case, new permeability
fields can be generated using the statistical package.
Simulations are then repeated using the new generated data and results are displayed. Among the three
steps in the expert system, step 3. requires the
largest amount of computational effort depending on
the number of parameters selected to be investigated.

6. Case study
To check the applicability of the expert system, a
reservoir whose properties are summarized in Table
1, is considered as a case study. The reservoir investigated here is assumed to have already been produced to its economic limit and is a potential candidate for an EOR process or otherwise subject to
abandonment. Step 1. of the expert system yielded
the following feasible EOR processes for this reservoir: 1. carbon dioxide flooding, 2. miscible hydrocarbon injection, and 3. nitrogen flooding. In addition, the results also show that miscible hydrocarbon
Table 1
Reservoir properties used in the case study
Depth
Temperature
Rock type
Formation thickness
Porosity
Permeability
API gravity
Oil viscosity
Dipping angle
Clays
Formation water
Acid number

3800 ft
1408F
carbonate
50 ft
15%
120 md
378
high percentage of light hydrocarbons
8 cP
35
present large quantity.
25,000 ppm chlorides
0.1 mg KOHrg of oil

41

Fig. 6. Permeability map of the first x z slice.

injection is the most suitable EOR process for the


reservoir. Reasons for this answer can be summarized by listing some rules used by the expert system.
1. The presence of high amount of clays is undesirable for surfactant and polymer flooding. Clays
increase polymer adsorption.
2. High chloride concentration in the formation water 25,000 ppm. is not suitable for surfactant
flooding.
3. Polymer, surfactant, or alkaline flooding is not
preferred for carbonate reservoirs.
4. The oil API gravity range for alkaline flooding is
13348 API.
5. For alkaline flooding, the oil should have an acid
number greater than 0.2 mg KOHrg of oil.
6. Carbonates are avoided for alkaline flooding because they often contain anhydrite or gypsum that
interact adversely with the caustic chemical.
7. In-situ combustion is used for heavy crudes and
in sandstone with high porosity.
8. Steam flooding is applicable to viscous oil oil
viscosity generally greater than 100 cP. and is not
normally used in carbonate reservoirs.
9. Deeper reservoirs are needed for nitrogen and
CO 2 flooding generally greater than 4500 ft. in
order to develop miscibility with the reservoir oil.
The investigation was further carried out to complete step 2. of the expert system. The objective of
step 2. is to create an appropriate input data set to
design the selected EOR process using UTCOMP.
The reservoir considered is a symmetric element of a

42

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

line-drive pattern that has a well spacing of 10 acres.


A three-dimensional mesh simulation with xyz
grid of 60 = 60 = 15 was used. The three-dimensional permeability field of the reservoir was assigned geostatistically to each grid block using the
MDM method with a spherical model of autocorrelation. The field is log-normally distributed in accordance with observations in sedimentary rocks. Fig. 6
shows the permeability map of the first xz slice.
The z-permeability was set at 1r10 the horizontal
permeability in each grid block. The resulting input
data set created by step 2. of the expert system is
shown in Appendix A.
To minimize simulation cost, only two parameters
one at a time. was chosen to be investigated in step
3. of the expert system. The first parameter is the
effect of the well combination on the performance of
the EOR process in this reservoir. After the selection
of this option, the expert system made all the necessary changes in the input file and simulations were
repeated for different combinations of vertical and
horizontal wells. These well combinations consisted
of vertical injectorvertical producer, horizontal injectorhorizontal producer, vertical injectorhorizontal producer, and vertical producerhorizontal injector. Results from step 3. show that the performance of horizontal producer in conjunction with
vertical injector yielded the best oil recovery for this
reservoir.

Fig. 7. Effect of amount of cross flow on the oil recovery after 1


PVI.

The effect of the amount of cross flow on the


displacement behavior can be further investigated.
The ratio of the z-permeability to horizontal permeability k vrk h . was systematically varied by the
expert system over a range of 0.0010.6, and simulations were repeated for several values. Fig. 7 shows
the results of this analysis. The figure shows that, for
this reservoir, the displacement performance of horizontal wells is better than that of vertical wells if
k vrk h is greater than 0.1. Furthermore, the performance of horizontal producervertical injector
yielded better recovery than either two horizontal
wells or two vertical wells for k vrk h greater than
0.05.

7. Conclusion
An expert system was developed to perform three
consultations: 1. selecting an appropriate EOR process on the basis of the reservoir characteristics, 2.
preparing appropriate input data sets to design the
selected EOR process using the existing numerical
simulators, and 3. making optimization studies on
several key parameters selected by the user. in order
to optimize the oil recovery from the selected EOR
process. Ten EOR methods were considered in step
1. of the expert system. Two reservoir simulators
are connected to the expert system that can simulate
EOR processes. In addition, two statistical packages
are also connected to the expert system to generate
geostatistical data for permeability or porosity distributions.
Nomenclature
AI
artificial intelligence
ANN artificial neural network
EOS
equation-of-state
IMPES implicit pressureexplict saturation
KBS
knowledge-based systems
JCG
Jacobi-conjugate-gradient
MMP minimum miscibility pressure
MDM matrix decomposition method
RDT
run time definition
TBM turning band method
UTCHEM chemical flood simulator
UTCOMP compositional flood simulator

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

Appendix A. Input data file for UTCOMP

43

44

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

References
Al-Kaabi, A.U., McVay, D.A., Lee, W.J., 1990. Using an expert
system to identify a well test interpretation model. J. Pet.
Technol. 42 5., 654661, May.
Alegre, L., 1991. Potential applications for artificial intelligence in
the petroleum industry. J. Pet. Technol. 43 11., 13061309,
November.
Alegre, L., Morooka, C.K., Rocha, A.F.de., 1993. Intelligent
diagnosis of rod pumping problems. SPE 26516. 68th Annu.
Tech. Conf., Houston, TX, October 36.
Allain, O., Houze, O.P., 1992. A practical artificial intelligence
application in well testing interpretation. SPE 24287. Eur.
Petrol. Conf., Stavanger, Norway, May 2527.
Amara, M.H., Martin, B., 1990. The offshore directional drilling
advisor: an expert system for directional drilling optimization.
SPE 20419. 65th Annu. Tech. Conf., New Orleans, LA,
September 2326.
Arcidiacono, T., 1988. Expert system on-call. PC Tech. J., 112
135, November.
Bailey, R.E., Curtis, L.B., 1984. Enhanced Oil Recovery. National
Petroleum Council, June 2.
Baker, J.D., 1984. Dipmeter advisor: an expert log analysis system at Schlumberger. In: Winston, P.H., Prendergast, K.A.

45

Eds.., The AI Business: Commercial Uses of Artificial Intelligence. MIT Press, Cambridge, MA, pp. 5165.
Bergen, J.K., Hutter, J.E., 1986. The Mudman Service an
artificial intelligence aid for drilling, Drill. Prod. Technol.
Symp. PD-vol. 4 Am. Soc. Mech. Eng, Book No. 100203..
Blackburn, C.R., Abel, J.C., Day, R., 1990. An expert system to
design and evaluate matrix acidizing. SPE 20337. 5th Pet.
Comput. Conf., Denver, CO, June 2528.
Burger, J., Sourieau, P., Combarnous, M., 1985. Thermal Methods
of Oil Recovery. Gulf Publishing, Houston, TX.
Carrico, M.A, Girard, J.E., Jones, J.P., 1989. In: Building Knowledge Systems: Developing and Managing Rule-Based Applications. McGraw-Hill, New York, pp. 10112.
Cayeux, E., Overti, P.O., 1992. ODDA: an expert system for
planning and conducting directional drilling. SPE 24274. Eur.
Pet. Comput. Conf., Stavanger, Norway, May 2527.
Chang, Y., Pope, G.A., Sepehrnoori, K., 1990. A higher-order
finite difference compositional simulator. J. Pet. Sci. Eng. 5
1., 3550, November.
Chavane, C., Perthuis, H.G., 1992. A fluid selection expert system
for matrix treatments. SPE 24995. Eur. Petrol. Conf., Cannes,
France, November 1618.
Chiu, T.-J., Caudell, E.A., Wu, F.L., 1993. Development of expert
systems to assist with complex fluid designs. SPE Comput.
Appl., 1820, February.

46

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347

Chung, T.-H., Carroll, H.B., Lindsey, R., 1995. Application of a


fuzzy expert systems for EOR project risk analysis. SPE
30741. Annu. Tech. Conf., Dallas, TX, October 2225.
Cordier, B., Gokana, D., Huynh, C.T., 1991. SCOOTER: a realtime expert system for process control application to gas
dehydration on offshore platforms. SPE 22306. 6th Pet. Comput. Conf., Dallas, TX, June 1720.
Corpoven, M.V.O., 1995. Real time expert system R.E.T.S.. for
rod pumping optimization. SPE 30185. Pet. Comput. Conf.,
Houston, TX, June 1114.
Cram, R.S., Hendrickson, A.R., 1986. An investigation into the
application of expert systems to matrix treatment design. SPE
15602. 61st Annu. Tech. Conf., New Orleans, LA, October
58.
Dafter, R., 1980. Scrapping the Barrel The Worldwide Potential for Enhanced Oil Recovery. Financial Times Business
Information.
Darwich, T.D., Toral, H., Archer, J.S., 1989. An expert system for
multiphase measurement and regime identification. SPE 19136.
4th Pet. Comput. Conf., San Antonio, TX, June 2628.
Derek, H.J., Jennings, J.W., Morgan, S.M., 1988. Sucker rod
pumping unit diagnostics using an expert system. SPE 17318.
Permian Basin Oil Gas Recovery Conf., Midland, TX, March
1011.
Dharan, M.B., Turek, E.A., Vogel, J.L., 1989. The fluid properties
measurement expert system. SPE 19134. 4th Pet. Comput.
Conf., San Antonio, TX, June 2628.
Einstein, E.E., Edwards, K.W., 1990. Comparison of an expert
system to human experts in well-log analysis and interpretation. SPE Form. Eval., 3945, March.
Elemo, R.O., Elmtalab, J., 1993. A practical artificial intelligence
application in EOR projects. SPE 26248. Pet. Comput. Conf.,
New Orleans, LA, July 1114.
Fear, M.J., Meany, N.C., Evans, J.M., 1994. An expert system for
drill bit selection. SPE 27470. IADCrSPE Drill. Conf., Dallas, TX, February 1518.
Fenoul, R.G., 1989. Enhanced drilling software and integrated
advisor expert systems. SPE 19133. Pet. Comput. Conf., San
Antonio, TX, June 2628.
Greffioz, J., Olver, A.J., Schirmer, P., 1993. TOPEX: an expert
system for estimating and analyzing the operating costs of oil
and gas production facilities. SPE 25315. Asia Pac. Oil Gas
Conf., Singapore, February 810.
Guerillot, D.R., 1988. EOR screening with an expert system. SPE
17791. Symp. Pet. Ind. Appl. Microcomput., San Jose, CA,
June 2729.
Haynes, H.J., Thrasher, L.W., Katz, M.L., Eck, T.R., 1976.
Enhanced Oil Recovery An Analysis of the Potential for
Enhanced Oil Recovery from Known Fields in the United
States 1976 to 2000. National Petroleum Council, December.
Heinze, L.R., 1992. CHES: casing hydraulic expert system. SPE
24420. 7th Pet. Comput. Conf., Houston, TX, July 1922.
Holditch, S.A., Xiong, H., Rueda, J., Rahim, Z., 1993. Using an
expert system to select the optimal fracturing fluid and treatment volume. SPE 26188. Gas Technol. Symp., Calgary,
Alberta, Canada, June 2830.

Holm, L.W., Josendal, V.A., 1974. Mechanisms of oil displacement by carbon dioxide. J. Pet. Technol., December.
Hutchin, L.A., Burton, R.K., Macintosh, D.J., 1996. An expert
system for analyzing well performance. SPE 35705. West.
Reg. Meet., Anchorage, Alaska, May 2224.
Jellison, M.J., Klementich, E.F., 1990. An expert system for
casing string design. SPE 20328. 5th Pet. Comput. Conf.,
Denver, CO, June 2528.
Keating, J.F., Laine, R.E., Jennings, J.W., 1991. Application of a
pattern-matching expert system to sucker-rod, dynamometercard pattern recognition. SPE 21666. Prod. Oper. Symp.,
Oklahoma City, OK, April 79.
Khan, S.A., Pope, G.A., Sepehrnoori, K., 1992. An expert system
for miscible gasflooding. SPE 24431. 7th Pet. Comput. Conf.,
Houston, TX, July 1922.
Kulakofsky, D.S., Wu, F.-L., Onan, D.D., Wohleb, J.B., 1992.
Development and application of a knowledge-based expert
system for cement slurry design. SPE 24417. 7th Pet. Comput.
Conf., Houston, TX, July 1922.
Lake, L.W., 1993. In: Enhanced Oil Recovery. Prentice-Hall, New
York, pp. 20128.
Mabile, C.M., Hamelin, J.-P.A., Amaudric du Chauffaut, B., La
Fonta, J.-G.M., 1989. An expert system helps in formation
recognition. SPE 19132. Pet. Comput. Conf., San Antonio,
TX, June 2628.
Maksimov, M.M., Tetelbaum, Y.I., 1992. Expert system for simulation of oilfield development: EKRAMON. SPE 24280. Eur.
Pet. Comput. Conf., Stavanger, Norway, May 2527.
Martinez, E., 1992. Directional drilling expert system. SPE 23664.
2nd Lat. Am. Petrol. Eng. Conf., Caracas, Venezuela, March
811.
Martinez, E.R., Moreno, W.J., Castillo, V.J., Moreno, J.A., 1993.
Rod pumping expert system. SPE 26246. Pet. Comput. Conf.,
New Orleans, LA, July 1114.
McCormack, M.D., Day, R., 1993. How artificial intelligence
impacts E&P productivity. World Oil, 8187, October.
Mungan, N., 1981. Carbon dioxide flooding fundamentals. J. Pet.
Technol., JanuaryMarch.
Nexpert Forms Users Guide. Neuron Data, Palo Alto, CA.
Patricio, A.R., Rocha, A.F.de., Morooka, C.K., 1994. Seplant: an
expert system for process plant and gas lift well. SPE 28238.
Pet. Comput. Conf., Dallas, TX, July 31August 3.
Peden, J.M., Tovar, J.J., 1991. Sand prediction and exclusion
decision support using an expert system. SPE 23165. Offshore
Eur. Conf., Aberdeen, September 36.
Perschke, D.R., 1988. Equation of state behavior modeling for
compositional simulation. PhD Dissertation. The University of
Texas at Austin, 13276.
Premuzic, E.T., Woodhead, A., 1993. Microbial enhancement to
oil recovery recent advances, Proceedings of the International Conference on Microbial Enhanced Oil Recovery. Elsevier.
Ramirez, R., Haor, M., Backhurst, M., Lugo, L., 1990. DIANA:
an expert system for fault diagnosis in gas compression plants.
SPE 20335. 5th Pet. Comput. Conf., Denver, CO, June 2528.
Russo, M.F., Peskin, R.L., 1987. Knowledge-based systems for
the engineer. Chem. Eng. Progress, 3843, September.

R.B.C. Gharbir Journal of Petroleum Science and Engineering 27 (2000) 3347


Saad, N., 1989. Field scale simulation of chemical flooding. PhD
Dissertation. The University of Texas at Austin, 15211.
Schirmer, P., Gay, J.C., Toutain, P., 1991. Use of advanced
pattern-recognition and knowledge-based system in analyzing
dynamometer cards. SPE 22305. 6th Pet. Comput. Conf.,
Dallas, TX, June 1720.
Schumacher, M.M., 1980. In: Enhanced Oil Recovery of Residual
and Heavy Oils. Noyes Data Corp., Park Ridge, NJ, pp. 759.
Shindy, A.M., Darwich, T.D., Sayyouh, M.H., Osman, A.A.,
1997. Development of an expert system for EOR method
selection. SPE 37708. Middle East Oil Show, Bahrain, March
1518.
Slider, H.C., 1983. In: Worldwide Practical Petroleum Reservoir
Engineering Methods. PennWell Books, Tulsa, OK, pp. 648.
Soto, R.B., Soto, C.T., 1995. An object oriented expert system to
enhance the log analysis of the Colombian basins. SPE 30196.
Pet. Comput. Conf., Houston, TX, June 1114.
Taber, J.J., Martin, F.D., 1983. Technical screening guides for the
enhanced oil recovery of oil. SPE 12069. Annu. Tech. Conf.,
San Francisco, CA, October 58.
Tangen, K., Baleix, J.M., 1992. Application of a pressure analysis
expert system during exploration drilling. SPE 25000. Eur.
Petrol. Conf., Cannes, France, November 1618.

47

Thomas, L.G., Albertsen, M., Perdeger, A., Knoke, H.H.K.,


Horstmann, B.W., Schenk, D., 1995. Chemical characterization of fluids and their modeling with respect to their damage
potential in injection on production processes using an expert
system. SPE 28981. Int. Symp. Oilfield Chem., San Antonio,
TX, February 1417.
Thompson, D.R., Dunlap, L., 1985. Computer system controls
mud during kick kill. Oil Gas J. 83 47., 9298, November.
Van Domelen, M.S., Ford, W.G.F., Chiu, T.J., 1992. An expert
system for matrix acidizing treatment design. SPE 24779. 67th
Annu. Tech. Conf., Washington, DC, October 47.
Vitthal, S., Gupta, A., Sharma, M.M., 1989. A rule-based system
for estimating clay distribution, morphology, and formation
damage in reservoir rocks. SPE Form. Eval. 4 4., 621626,
December.
Walsh, J.W., Brown, S.L., Asquith, G.B., 1993. Shaly sand
advisor: creating a program to provide expert help. SPE
Comput. Appl., 35, February.
Yang, A.P., 1990. Stochastic heterogeneity and dispersion. PhD
Dissertation. The University of Texas at Austin, 10115.
Zadeh, A.L., 1979. A theory of approximate reasoning. In: Hayes,
J. Ed.., Mach. Intell. 9 Halstead Press, New York.

Vous aimerez peut-être aussi