SPE

Society of Petroleum Engineers

SPE 17791

EOR Screening With an Expert System

by D.R. Guerillot, lnst. Franc;ais du Petrole

SPE Member

Copyright 1988, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Symposium on Petroleum Industry Applications of Microcomputers held in San Jose, California
June 27-29, 1988.

This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the
author(s) . Contents of the paper, as presented, have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the
author(s) . The material, as presented, does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Papers
presented at SPE meetings are subject to publication review by Editorial Committees of the Society of Petroleum Engineers. Permission to copy is
restricted to an abstract of not more than 300 words. Illustrations may not be copied . The abstract should contain conspicuous acknowledgment of
where and by whom the paper is presented. Write Publications Manager, SPE, P.O. Box 833836, Richardson, TX 75083·3836. Telex, 730989 SPEDAL.

Then, the architecture of the system is described, and the

ABSTRACT
The application of an Artificial Intelligence (AI) technique
to assist in the selection of an Enhanced Oil Recovery process
(EOR) is described. The aim of this Expert System (ES) is to
provide reasoned comments on the applicability of such
processes on the basis of reservoir characteristics.
The knowledge base has been developed using a professional
inference engine. To be closer to the type of reasoning used
by experts, fuzzy logic concepts have been introduced in the
knowledge representation.
This approach leads to a methodology for selecting EOR
processes and for intproving know-how by checking the
criteria used by comparison with practical experience, and it
helps to transfer the expert's knowledge to the users of the
system. Moreover, estimations of additional field cases makes
it possible to continuously refine the screening procedure.
INTRODUCTION
method for formalizing the knowledge is indicated using
. 11
f uzzy l oglC concepts .
THE DIAGNOSIS PROBLEM
EOR methods or processes have as their objectives to
increase recovery from reservoirs which would not respond
. al floodi . . . 12, 13, 16, 18,20,2 ,24
to convention water ng or gas InJeCtiOn .
The choice of enhanced recovery processes is based on
12 24
technical and economic criteria - • The problem involved is
to find all the EOR processes applicable to the oil field
concerned, or to check the suitability of a particular process
in the light of the information available about the reservoir.
In this application, we considered only the technical criteria
of process application, since economic criteria are too subject
to change from one company to another, but they could be
easily added taking into account each petroleum company
strategy.

Another very important aspect in the choice of a process

Expert systems (ES) are sophisticated computer programs
is the possibility of obtaining precise documentation on
that manipulate knowledge to solve problems efficiently and
reservoirs similar to the reservoir studied.
effectively in a narrow problem area. Like real human
experts, these systems use symbolic logic and
heuristics/rules-of-thumb to find solutions 1.2. This kind of Why an Expert System?
software, also called Knowledge-Based System (KBS), today
constitutes the most operational aspect of AI methods. An Expert System to solve this problem has been chosen
in the light of the difficulties following:
Various problems occurring in production are solved using
3 10
a KBS approach - • This article discusses the construction of • Decision-making rules are numerous. These rules are
an advisor system for selecting possible EOR processes in related to the reservoir data, which themselves are
account of reservoir characteristics. numerous. The type of reservoir, its depth, its pressure,
the permeability and porosity of the rock, etc., must be
First, the diagnosis problem is defined and some of its known. Moreover, we do not always need the same data
difficulties are stressed. The type of reasoning is illustrated to assess the applicability of a particular process. For
by the example of carbon dioxide (C02) flooding.
example, the acid number of the oil in place is a
References and illustrations at end of paper. screening criterion only for certain chemical processes.
137
2 EOR Screenina With An
• Few experts who know about all the processes are
available. The multidisciplinary knowledge is very vast,
and few persons have the background required for a
critical evaluation of all the mechanisms involved in
EOR processes. This knowledge is both theoretical
(physical laws of multiphase flows, thermodynamic laws
of interphase exchanges, etc.) and practical (surface
facilities, injection and production conditions, etc.). An
attempt is made here to obtain a diagnosis, even an
approximate one, of the overall processes rather than a
detailed diagnosis of one of the processes.
• This area is in a constant state of development. In
the past twenty five years, scientists have investigated
methods for recovering more hydrocarbons from
depleted reservoirs which often still contain more than
half of the hydrocarbons originally in place. Many
experiments performed on oil fields are now available,
renewing the criteria for application of the processes.
• Field data are poorly known. They are obtained by
various means: geological surveys, petrophysical
measurements, core displacement tests, well tests,
numerical modelling. They are often imprecise due to
measurement difficulties.
• Expertise rules are not strict rules such as may be
found, for example, in problems of diagnosing machine
breakdowns. The latter are related to an area of "hard"
knowledge. These are fields in which the reactions to an
external action can be predicted exactly and
deterministically. Here, the knowledge is "soft" such as
in the case of a medical diagnosis, for example. In fact,
when we say that the permeability of the rock must be
2
greater than 10 mD (0.010 tJm ) for C02 flooding, we
cannot categorically exclude a reservoir with a
2
permeability of 8 mD (0.008 tJm ).
• Expertise is not completely formalized. Much of the
literature offers rather general rules for process
application, but in making a choice on an actual scale,
it emerges that experts actually use many rules which
are not clarified in the literature. Moreover, these rules
are closely linked to the experience of these experts
with a few processes, and can sometimes lead to a lack
of objectivity. We need not only the conditions for the
application of each process but also comparative studies
of the effectiveness of a process compared with the
other processes.
The idea of developing an expert system for this problem
23
started at the end of 1983. • A first version using an
26 27
inference engine developed at IFP ' ("lnstitut Fran~ais du
Petrole ") in a LISP dialect has shown the feasibility of this
way of helping to select EOR processes. This prototype has
enabled us to specify an industrial system with a better
28
knowledge representation , with refined expertise rules and
with more processes considered. This operational system is
described here.
138
Exnflrt Svstem SPE 17 791
Brief Description of EOR Processes Selected
The term "Enhanced Oil Recovery" covers all reservoir
treatment processes designed to recover the "trapped" fraction
of the oil in place, namely the fraction that has not been able
to move under the natural effect of the pressure gradient,
imbibition and expansion of the dissolved gas, pressure
maintenace by water or gas injection.
Here, the rule base has been developed for nine processes,
classified in three families:
• four thermal processes: hot water injection, steam
stimulation, steam injection, in situ combustion;
• two dissolution processes: carbon dioxide injection,
hydrocarbon gas injection;
• three chemical processes: polymer injection, injection
of a surfactant solution followed by polymers, injection
of an alkaline solution.
The influence of reservoir chara~teristics on the choice of
a process can already be indicated qualitatively:
. for a carbonate reservoir, polymer and surfactant
solutions cannot be injected without precautions;
. the reservoir depth is limited for steam injection for
various reasons (surface equipment, heat losses in wells);
. the reservoir pressure must be sufficient for carbon
dioxide injection in order for phase exchanges with the oil to
have significant effects;
. polymer and surfactant solutions cannot withstand high
temperatures, etc.
Let us detail expertise rules for a specific process, i.e.
carbon dioxide injection.
Rules and Reasoning for Carbon Dioxide Injection
A simplified case is described here illustrating the
procedure of evaluation of carbon dioxide (C02) injection into
a hydrocarbon reservoir.
The term C02 injection applies to the following two
processes:
1) C02 injection without dynamic miscibility which concerns
heavy oils and takes benefit of oil swelling and viscosity
reduction associated to C02 dissolution into the oil,and
2) C02 injection with dynamic miscibility wich can be
reached with light oils and permits very high displacement
efficiencies to be achieved.
The primary criterion is pressure. Pressure must be
sufficient for carbon dioxide to have a significant effect on
the oil in place. It can be estimated that below 725 psi (5.
6
10 Pa), C02 injection is ineffective, giving rise to the
following rule:
SPE 17 791 D. GUERILLOT
6 2. An aquifer may exist which maintains the pressure.
IF the pressure is lower than 725 psi (5. 10 Pa), THEN
C02 may not be injected.
The layer treated must not be too thick. It can be stated
that, for both C02 injection processes, the thickness must be
less than 60 ft (20 m).
IF the thickness is higher than 60ft (20m), THEN C02
may not be injected.
To distinguish between the foregoing two C02 injection
processes, it is also necessary to know the Minimum
Miscibility Pressure (MMP). Nevertheless, without knowing
this pressure, one can already sort out the extreme cases with
the following two rules:
IF the pressure is lower than 1450 psi (1. 1d Pa) and IF
the oil gravity is lower than 25 oAPI (specific gravity higher
than 0.904), THEN immiscible C02 injection is feasible.
IF the pressure is higher than 3600 psi (2.5 1d Pa) and
IF the oil gravity is higher than 40 oAPI (specific gravity
lower than 0.825), THEN miscible C02 injection is feasible.
In other cases, the MMP is calculated, as described
hereafter.
The MMP is rarely known for the field to be investigated.
However, laboratory data are available for estimating the
MMP versus temperature for different oils. Correlation~have
29
been published by Holm and Josendal and extended by
30
Mungan to predict the MMP of C02 -oil systems, knowing
the molecular weight of the C5+ fraction · of the oil and the
temperature. If the molecular weight of the C5 + fraction of
12
the oil is unknown, other correlations can be used to obtain
it from the oil gravity. Two cases are distinguished according
to the relative values of the MMP and the reservoir pressure,
yielding the following rules:
IF the MMP is lower than the reservoir pressure, THEN
C02 injection with dynamic miscibility is feasible, and IF
NOT, can this pressure be varied?
It may be of great interest to increase reservoir pressure
to reach dynamic miscibility pressure. It must be emphasized
that in this case the reasoning is carried out in relation to
conditions which are not those of the reservoir at the time of
the diagnosis. However, increasing pressure raises several
problems such as:
1. Fracturing the layer to be treated must be avoided. To
check this, the MMP can be compared with the initial
pressure as follows:
IF the initial pressure is higher than the MMP, THEN
there is no risk of fracturing, and IF NOT, the user is
warned of this risk and, without further user's
recommendation, immiscible C02 injection is considered.
139
3. The gas cap must not be too large in consideration the
high compressibility of the gas.
Finally, for the variation in reservoir pressure, the
following rule is applied:
IF there is no risk of fracturing the rock and IF nothing
else opposes the increase in reservoir pressure, THEN the
quantity of fluid to be injected to reach MMP is estimated.
The quantity of fluid to be injected can be estimated by
using an average compressibility factor.
Moreover, for immiscible as well for miscible C0 2
flooding, several rules are applied that are essentially
extracted from tables for selecting the main enhanced
recovery methods C'screenin~ criteria"). Many authors have
12 16 18 2
furnished such tables ' ' ' •
This example illustrates the type of reasoning involved in
this problem. In the rules written above, difficulties also arise
from the fact that it is very difficult to draw clear frontiers
between admissible values and nonadmissible ones for a
parameter for a given process. The terms "higher than" or
"lower than" are not well suited. To avoid this jump between
"good" and "bad" values, determination functions have been
introduced to ~odulate this jump. Exgertise ~ules. have been
implemented usmg "fuzzy" set theory . Modtficatwns of the
inference engine have been made for this better knowledge
representation. This technical point will be detailed in
Appendix A.
DESCRIPTION OF THE SYSTEM
Software from outside
The software, called SARAH (for "Systeme d' Aide en
Recuperation Assisree d'Hydrocarbures"), is a diagnosis
expert system. Using reservoir characteristics, it assesses all
the processes considered, or a subset of these processes.
It is an interactive system. The user is queried on the
value of some parameters (different unit systems could be
used), but could himself interrogate the system to know the
whole set of possible answers, the justification of a question
and the correlation used to determine the parameter. It is also
possible to obtain, at any time, the history and current value
of a parameter.
Consultation is very quick. Answers are given nearly
without delay. A whole interactive session takes less than half
an hour of dialog with the system and costs only a few
seconds of CPU time with a microcomputer.
The diagnosis determined by SARAH is argued. It is
justified by a report generated automatically at the request of
the user, including:
 EOR Screening With An Expert System
4 SPE 17 791
1. A review of the values of the parameters considered for
the expertise, expresse 1 in the unit system chosen by
the user.
2. For each process examined, a general text on the
enhanced recovery mechanism and then a summary of
the judgment criteria compared to the consultation data.
An effort has been made to make this part "dynamic".
3. Possibly, bibliographical references on reservoirs having
similar characteristics. Indeed, a data base is simulated
that contains the characteristics of well-known reservoirs
on which at least one of the processes being investigated
has been tried out, successfully or not.
4. A general bibliography on EOR methods and on the
SARAH system itself.
This report contains formatting orders for text processing.
The user merely has to print out the formatted file, which
amounts to an average of about 20 pages.
The Inference Engine
The expert system has been developed with a professional
31
she11 • Because it is written in the C programming language,
it enables the same knowledge base to be used through
various computing environments (PCs, workstations and
mainframes).
Knowledge is organized in the form of
"subject-attribute-value" triplets. Classes can be defined and
possibly grouped in types of classes, attributes belonging to a
class or a type of class and hierarchies of values.
This inference engine does not fall into any standard
classification (engine using 0 or 0+ proposal logic, or
first-order logic). It contains a variable concept represented
by instances of classes, but the syntax of its rules is close to
that of a 0 + engine.
It operates mainly by backward chaining. A rule is
triggered by an attempt to determine one of its conclusions.
Rules can be structured in packets, so that the search can be
guided and limited to the set of candidate rules. An attribute
can be determined by triggering a rule containing this
attribute in its concluding part, by execution of a procedure
or by questioning the user. At any time in the query, the
justification can be requested of a question, the value of an
attribute, its history, etc.
The declarative part is assisted and supervised by a
consequential procedural part. Procedures can be defined to
check the order for triggering rules and determining
attributes.
It also has a developed user interface for associating a
message with each attribute (its meaning in natural language)
and for displaying messages and results in a user-friendly
way.
140
It accepts the attachment of outside functions written in C
language to enrich the structures available to the interface
(writing in a file, consulting of tables or data bases, etc.).
Notice that other shells including the features described
above could also be used for our problem. The advantage to
use a professionnal shell to develop expert systems are
numerous: a better reliability (it is used for several
applications), a good portability (the shell builder generally
proposes it for many computers), a good documentation, a
support course, etc.
Some reservoir data concern the rock, others the oil in
place, etc. All parameters required for consulting have thus
been grouped by concepts defining the classes: field,
reservoir, rock, oil, formation water, gas, process. An
example of the definition of CLASS is given in TABLE 1.
A set of real, entire, Boolean or text attributes is assigned
to each of these concepts. These attributes will be determined
during the query, either by a question asked to the user or by
being deduced from other attributes (mathematical formula,
correlation, deduction by testing the values of the attributes).
As in any expert system, an attempt is made to make the
dialogue as user-friendly as possible. For example, the system
will only ask useful questions on physical parameters of the
reservoir (depth, thickness, etc.), of the oil (viscosity, specific
gravity, etc.), of the water and gas, etc. An example of the
definition of ATTRIBUTE is given in TABLE 2.
There are formulas definitively linking some parameters.
For example, the transmissivity is computed from layer
permeability and thickness and oil viscosity. When this
parameter has to be determined, the engine triggers the rule
determining the transmissivity and does the computing. An
example of the calculation of an attribute by a mathematical
formula is given in TABLE 3.
For complex formulas or ones requmng high arithmetic
precision, it may be useful to make use of outside functions.
Experiments have determined relations among different
parameters to give an approximate value to a given factor.
For example, the temperature inside the reservoir can be
estimated from the depth of the reservoir. But this value has
to be checked by being confirmed by the user. These
attributes may thus be determined either by a rule, if the
value is accepted, or by questioning the user if it is not
accepted.
Reasoning
Expertise is performed by examining the following
parameters:
• for the reservoir: depth, lithology (nature), bubble
pressure, fracturing pressure, etc.;
SPE 17 791 D. GUERILLOT
• for the rock: thickness, porosity, permeability, oil
saturation, etc. ;
• for the fluids in place: oil properties (viscosity, specific
gravity, acid number, etc.), presence of a gas cap, of
an aquifer, formation water (salinity, viscosity), etc.
These parameters are acquired by the methods described
above.
A query of the expert system is thus divided into several
parts. Each part corresponds to the forming of an opinion
about the process. The expert system can be queried for one
or several processes. Since chaining is of the backward type,
the inference engine tries to evaluate the processes. It
searches in the rule base for the rules that would enable it to
assess the pertinence of a process. To trigger the premises
for these rules, it begins by ascertaining whether it already
has the information required. If it does not, it tries to obtain
it by a new rule, a correlation, a formula or, as a last resort,
by asking the user a question. Hence there is almost never
the same dialogue with the system. Each investigation may
lead to an examination of data suited to the situation. This is
an important difference compared to an interactive system,
which systematically asks the same questions while adapting
itself with great difficulty to the case at hand.
To assess the pertinence of a process, an original method
has been developed. This method prevents the number of
conditions for which a process is applied from having
consequences on the result of the diagnosis.
In the first phase, a numerical coefficient between -1 and
+ 1 is determined, characterizing the compatibility of the
reservoir, rock and fluid characteristics with the judgement
criteria for the process. The same importance is assigned to
each family of characteristics.
From these three real numbers, a new coefficient is
determined between - 1 and + 1, representing the feasibility
of the process for the reservoir examined. Then from this
"grade," an assessment is determined which is the result sent
to the user. A schematic diagram representing the reasoning
is given in Figure 1. Two different composition rules are
used to obtain these coefficients.
These two stages in this "grading" method for a process
are described in Appendix A.
Rule Base
Most rules test the value of several pertinent attributes to
judge the reservoir, rock or fluids for a process. They are
expertise rules such as the rules described above.
Other rules are established and grouped by packets, so that
calculations and unit conversions can be performed. These are
utilitarian rules for determining the value of an attribute by a
mathematical formulation or correlation. (For example, the oil
content of the rock is computed from the porosity and initial
oil saturation).
141
Simulation test of reasoning by analogy
We have tried to represent the reasoning by analogy of an
expert by simulating a data base for well-known reservoirs.
At the end of a query, we look to see whether there is a
reservoir in memory having similar characteristics to the
reservoir being examined. The originality of our approach is
to use production rules to find these reservoirs. Note that this
rules of proximity between the reservoir being examined and
the reference reservoirs depends on the EOR process.
At present, about ten typical reservoirs have been input for
each EOR process, but this proportion can very easily be
increased. We could also connect on already existing data
bases. In this case, the expert system itself queries the data
base to obtain useful information.
Validation and example of consulting
The rule base was tested on many cases taken
from published articles as well as from IFP experience. To
obtain a knowledge system capable of performing the given
diagnosis problem, three important activities are depicted in
figure 3: knowledge acquisition, knowledge system
construction, and knowledge system execution and testing:
• Knowledge acquisition involves a dialogue between the
expert and the knowledge engineer. The knowledge
engineer uses also written information (books, articles,
etc.) to learn problem solving principles specific to the
task.
• Knowledge system construction is the process of
expressing the knowledge acquired in a knowledge base.
• Knowledge system execution and testing involve running
the system on cases to test its performance and the
validity of the answer. The methodology followed here
was to compare the expert and the system diagnosis on
well known cases. After analysis, differences between
these two answers allowed us improving the knowledge
representation of the expert system. Now, the system
gives good results, especially for thermal processes. For
other processes, tests still remain to be performed,
because there are not yet enough actual data.
An example of consulting is given in figure 4. It illustrates
some features of the expert system:
• The consultation phase is initiated by issuing the
start.consulation command. In order to start a
consultation, you must have loaded a knowledge base,
and this knowledge base must be consistent;
• The name of the field is used to give a name of the
paper report file (question 1);
• The choice between the three process families
considered is done using the procedural part of the
inference engine (question 2).
6 EOR Screening With An Expert System
• Three unit systems could be chosen (question 3). This
choice modifies the units written in the question (see for
instance "ft" in question 5).
• Parameter values are proposed using correlations wich
can be commented (question 6). Here, reservoir
pressure is correlated with the depth of the reservoir.
Display is requested, and the value is accepted;
• Having enough information, the system can now
conclude after question 15 for the "steam drive"
process.
CONCLUSIONS
This operational system of about 300 expertise
rules has led to the following results:
1. Confirming the feasibility of an expert system for
assisting in the selection of EOR processes.
2. Enabling quick screening criteria to be determined on
the basis of reservoir characteristics for editing a
reasoned paper report.
3. Instructing the expertise rules used for choosing an
EOR process and helping to transfer the expert's
knowledge to the user of the system.
4. Giving references to similar reservoirs.
5. Enabling the expert to constantly refine process
selection procedures.
6. Justifying sensitivity tests to different parameters.
7. Being of particular interest when there are numerous
cases to be examined.
ACKNOWLEDGMENTS
The author thanks the management of the Institut Fran~ais
du Petrole (IFP) for permission to publish this paper. He also
thanks Murielle Roussel for her contribution to the
development of this knowledge-based system as well as all the
IFP experts in EOR for their continuous help.
REFERENCES
1 Hayes-Roth, F.,
Waterman, D.A. and Lenat, D.B.:
Building Expert Systems,
Addison-Wesley Publishing
Company Inc., Massachusetts, (1983) 445.
2 Waterman, D.A.: A Guide to Expert Systems,
Addison-Wesley Publishing Company Inc., Massachussets,
(1985) 420.
3. Affieck, N. and Zamora, M.: "PC-Based Expert System
Aids Optimum Mud Selection," Petroleum Engineer
International (Jan. 1987) 38-42.
142
SPE 17 791
4. Courteille, J.M., Fabre, M. and Hollander, C.R.: "An
Advanced Solution: The Drilling Adviser," JPT (Aug. 1986)
899-904.
5. Gani, R. and Fredenslund, A.: "Thermodynamic of
Petroleum Mixtures Containing Heavy Hydrocarbons: An
Expert Tuning System," Ind. lng. Chern. Res . V.26, No.7,
(July 1987) 1304-1312.
6. Hoffmann, F.C. and Valentin, E.P.: "OPUS: An
Integrated Assistance System for Oil Production", Expert
Systems V.4, No.4, (nov. 1987) 242-250 .
7. Kuo, T.-B.: Well Log Correlation Using Artificial
Intelligence, PhD. Thesis, Texas A&M Univ. (Jan. 1987)
150.
8. Kuo, T.-B. and Startzman, R .A.: "Field-Scale
Stratigraphic Correlation Using Artificial Intelligence,"
Geobyte V.2, No.2, (1987) 30-35.
9. Leblanc, L.: "Integrated Data Aids Rig Management,"
Offshore V.47, No.4, (Apr. 1987) 36-38 .
10. Startzman, R.A. and Kuo, T.-B.: "A Rule-Based System
for Well Log Correlation," SPE Formation Evaluation (sep.
1987) 311-319.
11. Turner ,R.: Logics For Artificial Intelligence, Ellis
Horwood Ltd., Chichester (1984) 121.
12. Bailey, R .E. and Curtis, L.B.: Enhanced Oil Recove1y,
National Petroleum Council (June 1984).
13. Bardon, C.: "Recuperation assistee du petrole - Injection
de gaz carbonique", (in French), Course at the "Ecole
Nationale Superieure du Petrole et des Moteurs" (Jan. 1985).
14. Burger, J. and Champion, D.: "How to estimate
production cost by steam drive?," Petroleum Engineer
International (June 1983) 66-70.
15. Burger, J. and Champion, D.: "How to estimate in situ
combustion cost?" Petroleum Engineer International (Nov.
1983) 32-44.
16. Burger, J., Sourieau, P . and Combarnous, M.: Thermal
Methods of Oil Recovery, Editions Technip, Paris, (1985).
17. Chauvel, A., Franckowiak, S. and Vacelet, 0.:
"Recuperation assistee du petrole par le C02 : disponibilites et
coOts en Europe de l'Ouest," (in French), Revue de l'Institut
Fran~ais du Petrole, No.1, (Jan./Feb. 1984).
18. Dafter, R.: Scrapping the barrel - The worlwide potential
for enhanced oil recovery, The Financial Times Business
Information Ltd. (1980).
19. Denoyelle, L. and Champion, D .: "Study show cost of
EOR by C02 , " Petroleum Engineer International (July 1985)
46-54.
SPE 17 791 D. GUERILLOT 7

20. Haynes, H.J., Thrasher, L.W., Katz, M.L. and Eck,
T .R.: 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 (Dec.
1976).
21. Latil, M.: Enhanced Oil Recovery, Editions Technip,
Paris, (1980).
22. Simandoux, P., Burger, J., Chauveteau, G., Champion,
I., Combe, J., Denoyelle, L. and Labrid, J.: "Etude
economique des prmclpaux procedes de recuperation
assisree," (in French), IFP report No. 31872 (Feb. 1984).
23. Simandoux, P., Bardon, C., Denoyelle, L. and Vacelet,
0.: "Recuperation assisree des hydrocarbures par injection de
C0 : aspects techniques et economiques", (in French), Revue
2
de l'lnstitut Fran~ais du Petrole, No. 4, (July/Aug. 1984).
Appendix A: Fuzzy Logic for Assessing the System on a
Process
Assessing the Influence of a Parameter for a Process
For each numerical parameter, there is a range of
unfavorable values for the choice of the process. This range
is difficult to quantify. A strict threshold cannot be
determined in relation to which a given factor can be said to
be favorable or unfavorable for the process in question. This
is why we decided to deal with the problem by a fuzzy logic
approach.
Two cases are to be considered: 1) numerical parameter,
2) non-numerical parameter.
Case 1:

24. Van Poollen, H.K. and Associates, Inc.: Fundamentals of
Enhanced Oil Recovery, PennWell Publishing Company
(1980).
25. Guerillot, D.: "Maquette d'un systeme en recuperation
assisree des hydrocarbures," (in French), IFP note
No.7740/1399 (Dec. 1984).
26. Guerillot D.: "Maquette d'un systeme de diagnostic en
recuperation assisree d'hydrocarbures," (in French), IFP
report No. 33511 (Sep. 1985).
Let us then define three types of intervals in the range of
acceptable values:
• a range in which the parameter would be favorable to
the process;
• a range in which the parameter would be moderately
favorable to the process;
• the remaining part of the interval in which the
parameter would be unfavorable.

27. Guerillot, D. and Bessis, F.: "Sarah-Diezol: un systeme For a given process, a real value between -1 and + 1 is
de diagnostic en recuperation assistee d 'hydrocarbures"' (in associated with each parameter:
French), Revue de 1' Institut Fran~ais du Petrole, V .41 , Na. 6,
(Nov./Dec. 1986) 759-771. -1: unfavorable
from - 1 to + 1: moderately favorable
(with nuances between these two values),
28. Roussel, M.: "Etude par un systeme expert de la
+ 1: favorable.
pertinence de procedes de recuperation assistee
d'hydrocarbures pour un gisement," (in French), IFP report
No. 35425 (Aug. 1987). Example: For the "polymer flooding" process, the
temperature factor is:
29. Holm, L.W. and Josendal, V.A.: "Mechanisms of oil
displacement by carbon dioxide," JPT (Dec. 1974). . favorable if it is less than 340 °K,
. moderately favorable if it is between 340 and 360 °K,
30. Mungan, N.: "Carbon Dioxide Flooding Fundamentals," . unfavorable if it is over 360 °K;
JPT (Jan./Mar. 1981).
Let ~ be the estimation function of the temperature T (fig.
31. Framentec: S.l Reference manual, (1986) 2):
~: T-. ~(T) e [ -1,+1]

so that:
~(T)= + 1 for T<340°K,
~(T)= -T/10+35 for 340°K~T<360°K:
linear segment joining the two points (340, 1)
and (360, -1) in the (T ,R) axes,
~(T)= -1 for 360°K~T.

This function is continuous, defined for all values of the

parameter, and represents the judgment corresponding to the
parameter for a process.

143
8 EOR Screening With An Expert System SPE 17 791

Case 2: IF the ci values all have the same sign,

In the same way, an estimation function could be defined
for non-numerical parameters. In this case, the set of
admissible values is reduced to the two values { - 1, + 1 } .
Example: For the process "surfactant solution", sandstone
rock is a favorable factor.
Let ~ be the estimation function for the "rock nature"
parameter r:
~: r E {sandstone, sand, etc.} .... ~(r) e { - 1, + 1}
so that:
~(r)= + 1 if r=sandstone,
~(r)= -1 if not.
n
THEN K = (p/n) * Ii=l Ci'
IF NOT, let c 1,c 2 , .. ,cm be all the negative values,
m
THEN K = (p/n) * Ii= 1 Ci.
This rule is interpreted as follows: if all the criteria have
the same sign, their arithmetic mean is taken; if not, the sum
of negative criteria divided by p is taken. This factor enables
the scale to be enlarged at interval ( -p, +p).
Here, we take p=2.
We then decide to associate a judgment with the numerical
coefficient K, which gives an opinion about the system for
the process considered and the reservoir in question, obtained
by the following convention:

from -2 to -1.5, impossible

How to use these estimation functions? from -1.5 to -0.5, unfavorable
from -0.5 to +0.5, possible
The judgment of a process for a reservoir is done in two from +0.5 to + 1.5, favorable
stages (fig. 1):
from + 1.5 to +2, very good.

First stage: estimation criteria corresponding to the
reservoir, the rock or the fluids.
For each premise, there is an evaluation g1vmg a real
number between - 1 and + 1. The concepts of true and false
have thus been extended to a real interval bounded by values.
It then suffices to combine these values to obtain an
evaluation of the premise part. The combination rule adopted
is that of the minimum of all the values. Indeed, a single
factor tending toward the unfavorable estimation suffices to
reject the process.
This convention is very flexible. By taking p = 3, we could
have another scale varying from - 3 to + 3 with the
corresponding values for impossible, very bad, unfavorable,
possible, favorable, very good, excellent.
We felt it preferable to adopt such an assessment system
for parameters rather than conventional plausibility
coefficients. Indeed, our coefficients do not represent a degree
of certainty but rather a degree of suitability for the process
in question. It was absolutely necessary for a rule to be
triggered, even with a premise "evaluated" at -1.

A judgment on the reservoir, the rock and the fluids in

place is obtained from the parameters characterizing them.

Second stage: general assessment of the system for a

process.

There are three elementary criteria for judging the process

proposed. Let us combine them to obtain the general
criterion, called K. It does not suffice to take the simple
arithmetic mean. Indeed, a value of + 1 for the rock criterion,
for example, will not suffice to compensate for a -1 value
for the fluid criterion, and the process must thus be rejected.
These values must be combined so that, if there is at least
one negative criterion, it will be preponderant. On the other
hand, if all the criteria are positive, the process will be
judged to be applicable. The original combination rule
proposed is as follows:

Let n be the number of elementary criteria to be created,

and let ci be the value of criterion i, then 1 <i<n (here,
n = 3: reservoir, rock and fluids), we calculate the general
criterion K with the rule hereafter:

144
 SPE 17 7 91

TABLE 2 - EXAMPLE Of DEfiNITION Of ATTRIBUTE

DEfiNE ATTRIBUTE thickness.extern

: :DEfiNED.ON !:layer
::TYPE real
::LEGAL.MEANS query.user
TABLE 1 - EXAMPLE Of DEfiNITION Of CLASS ::OETERMINATION.MEANS query.user
::CHECKING.fUNCTION check.bound
DEfiNE CLASS crude.oil ::PROMPT
::NUMBER.INSTANCES 1 "What is the average thickness of the layer
::PRINT.ID "crude.oil II" (in" ! lmail.unit(l) ! ")?"
::CLASS.TRANSLATION "a crude.oil" ::REPROMPT
::PLURAL.CLASS.TRANSLATION "the crude.oils" "The thickness must be in ("!
::BLANO.INSTANCE.TRANSLATION "the crude.oil" make.text(round(restore(l,
END.DEfiNE "lmail",unit.system(l)),l)) ! "," !
make.text(round(restore(JOO,
"lmail",unit.system(l)),l)) ! "). Thickness=?"
::TRANSLATION "the average thickness of the layer"
END.DEfiNE

TABLE 3 - EXAMPLE Of CALCULATION Of AN

ATTRIBUTE BY A MATHEMATICAL fORMULA.

DEfiNE RULE RULE.TRANSMISSIVITY

::APPLIED.TO !:layer
::PREMISE already.existing(o:crude.oil I
permeability(!) known
and viscosity(o) known
and thickness(!) known)
::CONCLUSION transmissivity(!)=
permeability(!) * thickness(!) I viscosity(o)
END.DEfiNE

145
 SPE 1 7 7 91

Estimation Function
cp (T}

+1~------------------,

0 Temperature (in °K)

RESERVOIR ROCK FLUIDS

-1

•
ESTIMATION CRITERIA WITH
+
11
FUZZY LOGIC
11
+
APPROACH Fig. 2-Estimation function of the temperature parameter.

K E /-2,+2/

GENERAL ASSESSMENT FOR A PROCESS Acquire and Refine

Knowledge

SESSION REPORT I Encode the

Knowledge
Knowledge System
Fig. 1-Schematic reasoning diagram.

Fig. 3-The knowledge engineering loop.

146
 SPE 1 7 7 91

-> start.consultation

This will be case nu~ber 1

1: What is the name of this field study?
1> Bellevue
You can choose the processes you want to study:

a THERMAL PROCESSES

b MISCIBLE PROCESSES

c CHEMICAL PROCESSES

Type any co•binaison of a,b,c to select a class of processes, abc for all

2: Your choice:
2> a
J: What is your unit system?
J>?
Enter one of the following: 51, PfU, PAU, CGS
J: UNIT SYSTEMS:
51 for 'International System unit'
PfU for 'Practical french Units'
PAU for 'Practical Anglo-American Units'
CGS for 'CentiMeter-Gram-Second'
J> pau
4: What is the nature of the reservoir?
4> sand
5: What is the depth of the reservoir (a positive value in ft)?
5> 400
I propose the value 176.83 psi for the pressure of the reservoir
6: Do you agree with this value? y(es)/n(o)/e(xplanation)
6> e
I used the correlation (in 51):
if depth <= 1800 : pressure = (depth * O.l)*l.E05
if depth > 1800 : pressure = (depth * 0.2 + 180) *l.E05
-- reference: "fORMULAIRE DU PRODUCTEUR, Technip, PARIS, (1970)
7: Do you agree with this value? y(es)/n(o)/e(xplanation)
7> y
I propose the value 459.758 psi for the fracturing pressure of the reservoir
8: Do you agree with this value? y(es)/n(o)/e(xplanation)
8> y
9: What is the average thickness of the layer (in ft)?
9> 90
10: What is the average absolute permeability of the layer (in 11D)?
10> 500
11: What is the viscosity of the crude oil (in cp)?
11> 500
12: What is the average porosity of the layer (in %)?
12> 38
lJ: What is the oil saturation before the EOR process (in ")?
lJ> 52
14: What is the gravity of the crude oil (in °API)?
14> 19.5
*************************************************************************
STEAM.DRIVE PROCESS:

The reservoir properties are FAVORABLE

The layer properties are POSSIBLE
The fluids properties are fAVORABLE

The acreening criteria are POSSIBLE for the STEAM.DRIVE PROCESS

*************************************************************************

etc ••• for the other processes.

Fig. 4-Example of consulting.
147