EOR

1. Copyright 2002, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the SPE 13th European Petroleum Conference
held in Aberdeen, Scotland, U.K., 29–31 October 2002.
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. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
The Venezuelan National Oil Company, PDVSA, has
dedicated a sustained effort to adapt EOR/IOR technologies to
rejuvenate a large number of its mature fields. The first step
towards achieving this objective was to select cost-effective
technologies suited for conditions of Venezuelan reservoirs.
The current strategy for screening EOR/IOR applications is
based on the Integrated Field Laboratory philosophy, where a
representative pilot area of a number of reservoirs is selected
to intensively test EOR/IOR methods, such as WAG injection
(water alternating gas) and ASP (alkali polymer surfactant),
currently underway. Two problems with this approach are the
lack of objective rules to define a reservoir type and the
project completion time. In general, the trouble with using
expert opinion is that it tends to be too biased by operational
experience. It is known that the success of a given EOR/IOR
method depends on a large number of variables that
characterize a given reservoir. Therefore, the main difficulty
for selecting an adequate method is to determine a relationship
between reservoir characteristics and the potential of an
EOR/IOR method. In this work, data from worldwide field
cases have been gathered and data mining was used to extract
the experience on those fields. Here, a space reduction method
has been used to facilitate the visualization of the needed
relationship. Machine learning algorithms have been utilized
to draw rules for screening. To illustrate the procedure, several
Venezuelan reservoirs have been mapped onto the extracted
representation of the international database.
Introduction
Primary and secondary recovery methods generally result in
recoverable reserves between 40 and 50%. The latter depends
on reservoir complexity and reservoir conditions, field
exploitation strategy and is greatly affected by economics.
Tertiary recovery or Improved Oil Recovery (IOR) methods
are key processes to replace or upgrade reserves, which can be
economically recovered, beyond conventional methods.
Therefore, the application of IOR methods offers opportunities
to replace hydrocarbon reserves that have been produced in
addition to those coming from exploration and reservoir
appraisal1,2
. In this work, we concentrate on screening of EOR
processes, rather than IOR, but no real limitation for the
method presented here is known at the present time.
PDVSA, the Venezuelan National Oil Company, holds a
long history of oil and gas production, with all its E & P assets
located in Venezuela. This history brings along a large number
of mature, near abandonment, reservoirs. PDVSA operates a
variety of accumulations, most of them in sandstone
formations, with wide spread in API gravity, from bitumen
and heavy oils, to volatile oil and condensate reservoirs.
Exploitation plans have often yielded low recovery factors,
that in average amount to 30% for waterflooding and 40% for
gas injection, and lower values for primary recovery in most
cases.
One of the major difficulties to manage such a portfolio of
opportunities relates to numerous reservoirs under dissimilar
conditions and the long list of Enhanced Oil Recovery (EOR)
technologies available. As expected, screening/ranking of
these processes can become a daunting task. Two constraints
limit the use traditional evaluation techniques in PDVSA’s
case:
1. Maturing reservoirs have short life span, hence time is
quite limited for the decision-making process.
2. Reservoir characterization is far from complete in a
large portion of the portfolio. Although integrated
studies are underway, many reservoirs lack enough
financial performance to justify information or data
gathering.
PDVSA-Intevep, PDVSA´s R&D division, has embarked
the development and adoption of EOR methods that are
suitable for Venezuelan reservoirs. The latter requires
techniques for visualization of opportunities with good grasp
of the risks involved. This is a consequence of uncertainties,
due to incomplete information, a constant in the E & P
business. Methods for analysis should be designed to enable
SPE 78332
Selection of EOR/IOR Opportunities Based on Machine Learning
Vladimir Alvarado, SPE, PDVSA-Intevep, Aaron Ranson, PDVSA-Intevep, Karen Hernández, PDVSA-Intevep, Eduardo
Manrique, SPE, PDVSA-Intevep, Justo Matheus, PDVSA-Intevep, Tamara Liscano, FUNDATEC, Natasha Prosperi,
PDVSA-Intevep

2. 2 V. ALVARADO, A. RANSON, K. HERNÁNDEZ, E. MANRIQUE, J. MATHEUS, T. LISCANO, N. PROSPERI SPE 78332
iteration during evaluations, including progressively more
details, such that they allow us to refine as the opportunity
becomes more attractive and data gathering (reservoir
characterization mostly) turns out justifiable. Along with these
ideas, PDVSA has developed the concept of the Integrated
Field Laboratory (IFL) to facilitate testing field technologies
and their deployment in a number of exploitation units.
A data mining strategy applied to a collated database of
international project results is used here for knowledge
extraction on applicability of EOR processes. Statistical
analysis of the data yields importance of variables, in terms of
how they influence clustering of reservoirs. A small number of
these variables, representing average values for each reservoir
are used to rank EOR processes and extract rules. It is
important to notice that, as mentioned previously, the process
of inquiring the database does not end at the first
representation of the data, which means that further refinement
is necessary, until a decision can be made or information is
exhausted from the extraction process. This differs from
traditional analysis in the sense that several iterations of the
screening/ranking process are not only possible, but also
necessary.
The paper is organized as follows. After this introduction,
a summary of the IFL strategy is summarized. Then, reference
to artificial intelligence methods and several EOR screening
methodologies is carried out. The proposed methodology is
then explained, followed by the results section. Closing
remarks and recommendations are provided at the end. This
work does not pretend to be comprenhensive, but rather
intends to show a first view of a whole strategy thought of for
these purposes.
Integrated Field Laboratory
The idea behind the IFL philosophy is the speedy
evaluation and incorporation of technologies to field
operations3
. However, finding the best technology for
individual reservoirs would represent an endless task. Here
comes in the idea of grouping reservoirs by type, i.e. by
analyzing together reservoirs with similar characteristics.
Applicability criteria for EOR technologies and reservoir types
are a motivation for investing in these advanced pilot test
areas. One of the weakest aspects in IFL projects, has been the
method for extrapolation of learned strategies from a pilot area
to a large set of reservoirs. Several EOR technologies are
under scrutiny in the IFL projects: Optimized Water flooding3
(OWF), Alkaline-Surfactant-Polymer (ASP) flooding4-6
,
Continuous Steam Floding (CSF) in heavy oil7
and Water-
Alternating-Gas (WAG) injection8-10
. Fig. 1 illustrates the
typical workflow for evaluation in an IFL, for the VLE
example.
Carvajal et al.3
describe the management approach for
technology evaluation in the IFL's as follows:
• Advanced reservoir characterization, simulation and
visualization.
• Drainage strategies, combining EOR methods and
well architectures.
• Speedy well construction, with minimum damage and
cost.
• IOR technologies to deal with injectivity and
productivity enhancement.
• Advanced monitoring technologies.
• Production fluid handling technologies.
• Risk evaluation and mitigation.
Eight pilot tests have been planned since 1996. As seen in
this section, intensive application of technologies in field
operations is an important part of the objective in the IFL
strategy. However, technology evaluation of that level of
detail is only possible in pilot areas. Screening criteria that
include the possibility of portafolio analysis is an answer to
extrapolation of lessons learned in pilot tests. Alternative
approaches, based on artificial intelligence, are now describe
to introduce the context of this work.
Artificial Intelligence
Artificial Intelligence (AI), specifically Neural Networks,
Fuzzy Logic, and Expert System, have been often proposed
and used for supporting E&P operations. Their use varies
depending on the specific problem. In the case of neural
network and fuzzy logic, they have been proposed for data
filtering11
(smoothing), or as modeling tools. All the potential
of these information processing systems are used to build non-
linear models for oil production forecast, log interpretation to
identify total porosity as well as lithofacies12
, or reservoir
property related estimations13
. In the case of Expert system is
mostly used for knowledge representation in the form of IF-
THEN rules, where specific Know-How from Experts are used
to build schemes that would be automated and used for
modeling an Expert reasoning. Some expert system have been
developed recently14,15
, for different disciplines of E&P, which
included drilling areas16,17
, well bore simulation18,19
, well
testing and logging20,21
, EOR and fluid property
predictions22,23
. It is important to mention that combinations of
these different tools are also possible, i.e. fuzzy rules are used
to increase the capability of the expert system to deal with
uncertainties. Also, fuzzy activation functions are used in
combination with Neural Network topologies so that some
form of regression techniques to adjust the fuzzy set rules are
possible. A recent new player in these AI solution suites is
Machine Learning (ML)24
. One possible realization of this
technique is the combination of Clustering techniques25,26
and
rule extraction algorithms27
. In this approach, all the data
available is used to extract implicit and explicit process or
business rules hidden within the data, whether heuristic or
not28
. In all these cases, AI has showed an excellent
performance as well as simplicity in the final solution. With
these techniques, the probabilities of success are strongly
related to type of information available, the amount/quality
and the existing knowledge experience. Therefore, the
application prerequisite for these technologies or any
deterministic model requires a comprehensive and detail
analysis of the problem and the available information. A rule
of thumb is that AI techniques are preferred when decision-
making cases are based on bulk heterogeneous and incomplete

3. SPE 78332 SELECTION OF EOR/IOR OPPORTUNITIES BASED ON MACHINE LEARNING 3
information, where deterministic modeling is rather difficult,
but each case must be evaluated in case-by-case basis.
EOR Screening
During the past 20 years, screening criteria have been
employed to evaluate a number of reservoirs for the
applicability of different IOR processes in a simple way,
before any detailed evaluation is done. This is especially
helpful when a large number of reservoirs needs to be
analyzed. Screening criteria have evolved through the years
and and they are now well established thanks to more field
experiences as well as laboratory and numerical simulation
studies29-31
. Furthermore, several computer programs or
analytical model have been developed to select feasible IOR
methods and predict their oil recovery performance based on
reported screening criteria32-35
. On the other hands, since early
90´s computer technology has improved the application of
screening criteria through the use of artificial intelligence
techniques to select and design IOR processes and even to
perform IOR Project Risk and Economic Analysis29,36
. Typical
selection criteria are shown in Table 1.
Oil and gas reservoirs represent a complex system with
high degree of uncertainty, starting with the definition of the
important parameters, finishing with the data availability and
then quality. Hence, a first order screening of EOR/IOR
methods applicable under particular reservoir conditions is
important such that it is possible, in early stages, to establish
development scenarios.
A primary goal of this work is to propose and develop an
AI frame work based on ML, where it is possible to identify,
based on a reduced set of characteristics reservoir variables,
reservoir clusters, that heuristically will be called reservoir
typology. The combination of the reducted space
representation with machine learning approach28
opens a
different way of screening EOR methods. It is also possible to
visually identify reservoir types, despite the subjective nature
of the resulting 2D data representation. In this sense, it is
observed that certain types of reservoirs tend to group in
specific areas of the maps and these reservoirs have in
common the EOR methods that have been applied to them.
The proposed approach allows us to carry out fast and clean
screening of the EOR methods bases on the “reservoir pseudo
typology”. The information used can be handled at different
levels of granularity, which means that it guides data
capturing, which in turn allows is to request data only in cases
where we have foreseen that more information will indeed add
more value to the decision-making process.
Results
The database used to generate the results contains
information from EOR projects carried out around the world,
which would allow to compare the results with those from
other projects in more than one country or continent. Most of
these EOR projects have been completed in the USA and
Canada; the remaining projects have been carried out in many
other countries from several continents, including Asia,
Europe and Latin America. The database includes a list of
more than 20 reservoir and fluid variables, although some
records are missing information for some of those variables.
The main sources of information were biannual reports
from 1972 through 2000 published by the “Oil & Gas
Journal”, articles, books and reports from the SPE, and
databases created in PDVSA-INTEVEP.
The database contains information from a total of 290
cases, 70% of which correspond to miscible drive and the
remaining 30% of the processes use inmiscible drive. Fig. 2
summarizes the statistics of all methods represented in the
database.
Table 2 shows the range of values for each of the
parameters used for the different methods found in the
database. The database contains information ranging from
extra-heavy oils to condensates, and HP/HT reservoir
conditions. Missing conditions are very deep reservoirs, where
pressure and temperature exceeds 11.000 psi and 325 o
F,
respectively, and Tarsands, common in Canadian
underground.
The first step after database collation and quality control
are carried out, is to process the data to generate a knowledge
map. Although more than 20 variables were initially
considered, to be able to have a large number of records
available for the analysis, 6 variables were selected to generate
the maps. The selection was based on importance of those
variables to form well-defined clusters, but also reduction of
redundant information, based on correlation analysis. Fig. 3
shows a projection of the different reservoirs that make up the
international database. This type of projections does not intent
to represent two axes, as might be interpreted from a 2-D
representation, but it instead is a compact representation of a
combination of the six variables. After applying cluster
algorithm to the projection, six clearly defined clusters can be
determined, representing six mixed reservoir typologies, that
is, each cluster is made out of different reservoirs to which
different EOR methods have been applied.
Two examples will be used here to illustrate the use of the
information obtained from projections. The way we proceed
goes as follows. If a new reservoir, not originally found in the
international database, is projected on Fig. 3, and the new
reservoir is, for instance, located in cluster 5, this will mean
that the newly incorporated reservoir has similar
characteristics to those in cluster 5. Analyzing the statistics per
method, and per cluster, we can show that most recovery
methods in cluster 5 correspond to thermal ones; therefore, if a
reservoir is located within cluster 5, experience shows that for
this particular typology, thermal methods are frequently used,
and perhaps adviced. The method statistics that can be applied
to the reservoirs in cluster 5 are shown in Table 3.
In addition to the methods statistics that can be applied to
each cluster, we have developed a set of rules that allow us to
characterize each one of the clusters. The set of rules
associated to clusters shown in the projection of Fig.3 are
listed in Table 4. It is important to highlight that not all six
variables are in principle used to define a rule automatically.
This is a consequence of the algorithm employed for this
purpose. However, it has been already forseen that rules can
be, and will be, complemented by expert opinion, based on
observation of the maps.

4. 4 V. ALVARADO, A. RANSON, K. HERNÁNDEZ, E. MANRIQUE, J. MATHEUS, T. LISCANO, N. PROSPERI SPE 78332
To find reservoir typologies, representative candidates for
specific processes, that is, clusters made out of different
reservoirs to which the same EOR method has been applied,
each of the six clusters in Fig. 3 are reanalyzed (the same
applies to other clusters). This is part of the possible
refinement of the data analysis, mentioned before. The latter
means that the same methodology and cluster algorithms can
be used over the subset of data integrated for each one of the
reservoirs that conform cluster 5 or any other cluster (see Fig.
3). For example, looking closely at cluster 5 (Fig. 4), we see
once again a new classification (six clusters A, B, C, D, E and
F). We also observe that pure typologies, those related to one
method only, are better defined. Cluster B and cluster C are
clear examples of pure typologies of thermal methods within
global cluster 5. Since the new analysis is performed on a
reduced set of data of the international database, new results
and a different set of rules are obtained.
Figure 5 shows a projection of the international database
that includes two Venezuelan reservoirs. For instance,
Reservoir A (a relatively shallow light oil reservoir in southern
Venezuela- indicated by the dashed Brown Circle) is located
in cluster 4. On the other hand, reservoir B (a deep extra-
heavy oil reservoir in western Venezuela – marked with the
dashed Red Circle) is located in cluster 5. For reservoir B, the
set of points located within the dotted red circle corresponds to
the sensitivity values for some of the process variables in the
same reservoir. This leads to the possibility that sensitivity
analysis on reservoir history can be explored with the
proposed methodology
Reanalyzing cluster 4, which contains the Reservoir A
(Fig. 5), it can be concluded that the statistics and list of
methods that can be applied to this reservoir are those listed in
Table 6.
If we take a look at reservoir B (Fig. 6 - Dotted Red
Circle), we can see that the reservoir belongs to cluster 5 in the
general projection map. This means that within a global scope,
the reservoir has conditions for thermal methods. However,
notice that the original reservoir and its sensitivity values are
located on the boundaries of the cluster. However, a zoom of
cluster 5 (Fig. 7) shows that reservoir B and its sensitivity
values are close to cluster F, but out of the region that defines
the cluster. The latter means that by making operational
changes such as decreasing the reservoir’s pressure, reservoir
B can move into cluster F.
The statistics associated with this reservoir are as shown
by Table 6 and the set of rules defined for cluster 5 are those
in Table 7. The whole procedure applied for the examples
discussed here can be used if more variables were considered,
as the clustering algorithms are flexible enough to complete
this task. Rules derived from the automatic extraction method
should , from our point of view, be complemented by using
expert opinions, guided in turn by the reduced representation.
Conclusions
1. Space reduction techniques have been applied
successufully to produce bidimensional maps that clearly
show reservoir types by using 6 reservoir variables.
2. The generated maps allowed us to establish applicability
criteria or selection rules, based on international
experience on EOR processes.
3. Several Venezuelan reservoirs were mapped, based on
their reported average reservoir variables, and sensible
conclusions on applicability of EOR processes were
drawn from the analysis method proposed here.
4. The outcomes of this work drives further development of
the techniques proposed here to refine the
screening/ranking criteria based on detailed analysis on
the available data.
5. Firmer rules and conclusions can be drawn as the
gathered experience, represented in the database, is
enlarged. This, however, would require collaboration
from oil companies, as the results of the application of
EOR methods are not often found in the open litterature.
Acknowledgements
We would like to thank PDVSA-Intevep for permission to
publish this paper.
References
1. Goodyear, S. G. and Gregory, A. T.: “Risk Assessment and
Management in IOR Projects” Paper SPE 28844 presented at the
1994 European Petroleum Conference, London, October 25-27.
2. Thompson, M. A. and Goodyear, S. G.: “Identifying Improved Oil
Recovery Potential: A New Systematic Risk Management”
Paper SPE 72103 presented at the 2001 Asia Pacific Improved
Oil Recovery Conference, Kuala Lumpur, October 8–9.
3. de Carvajal, G., Velasquez, A., Graterol, J., Ramirez, F., and
Medina, M.: “Lagomar's Integrated Field Laboratory for
Intensive Evaluation of Technologies” Paper SPE 53984
presented at the 1999 SPE Latin American and Caribbean
Petroleum Engineering Conference, Caracas, April 21–23.
4. Manrique, E., de Carvajal, G., Anselmi, L., Romero, C., Chacón,
L.: "Alkali/Surfactant/Polymer at VLA 6/9/21 Field in
Maracaibo Lake: Experimental Results and Pilot Project
Design" Paper SPE 59363 presented at the 2000 SPE/DOE
Improved Oil Recovery Symposium held in Tulsa, April 3–5.
5. Hernández, C., Chacón, L J., Anselmi, L,, Baldonedo, A,, Qi, J.,
and Pitts, M. J.: " ASP System Design for an Offshore
Application in the La Salina Field, Lake Maracaibo" Paper SPE
69544 presented at the 2001 Latin American and Caribbean
Petroleum Engineering Conference, Buenos Aires, March 25–
28.
6. Hernández, C., Chacón, L., Anselmi, L., Angulo, R., Manrique,
E., Romero, E., de Audemard, N., Carlisle, C.: “Single Well
Chemical Tracer Test to Determine ASP Injection Efficiency at
Lagomar VLA-6/9/21 Area, C4 Member, Lake Maracaibo,
Venezuela” Paper SPE 75122 presented at the 2002 Improved
Oil Recovery Symposium, Tulsa, April 13–17.
7. Lopez, E, Rojas, L., Mata, T., Mendoza, H., and Briñez, A.: "
Integrated Laboratory Field Application or Thermal Recovery
Process" Paper SPE 53983 presented at the 1999 SPE Latin
American and Caribbean Petroleum Engineering Conference,
Caracas, April 21–23.
8. Manrique, E., Calderón, G., Mayo, L., and Stirpe, M. T.: "Water-
Alternating-Gas Flooding in Venezuela: Selection of Candidates
Based on Screening Criteria of International Field Experiences"
Paper SPE 50645 presented at the 1998 European Petroleum
Conference, The Hague, October 20-22.
9. Manrique, E. , Padrón, R., Surguchev, L., De Mena, J., McKenna,
K.: "VLE WAG Injection Laboratory Field in Maracaibo Lake"

5. SPE 78332 SELECTION OF EOR/IOR OPPORTUNITIES BASED ON MACHINE LEARNING 5
Paper SPE 65128 presented at the 2000 European Petroleum
Conference, Paris, October 24-25.
10. Alvarez, C., Manrique, E., Alvarado, V., Samán, A., Surguchev,
L., and Eilertsen, T.: "WAG pilot at VLE field and IOR
opportunities for mature fields at Lake Maracaibo" Paper SPE
72099 presented at the 2001 Asia Pacific Improved Oil
Recovery Conference, Kuala Lumpur, October 8–9.
11. Balch, R. S., Hart, D. M., Weiss, W. W., and Broadhead R. F.:
“Regional Data Analysis to Better Predict Drilling Success:
Brushy Canyon Formation, Delaware Basin, New Mexico”
Paper SPE 75145 presented at the 2002 Improved Oil Recovery
Symposium, Tulsa, April 13-17.
12. Weiss, W. W., Balch, R. S., and Stubbs B. A.: “How Artificial
Intelligence Methods Can Forecast Oil Production” Paper SPE
75143 presented at the 2002 Improved Oil Recovery
Symposium, Tulsa, April 13-17.
13. Surguchev, L. and Li, L.: “IOR Evaluation and Applicability
Screening Using Artificial Neural Networks” Paper SPE 59308
presented at the 2000 SPE/DOE Improved Oil Recovery
Sympsium held in Tulsa, April 3-5.
14. Bergen, J. K. and Hutter, J.E. 1986. The Mudman Service-an
artificial intelligence aid for drilling. DrillIng and Production
Technoly Symposium PD-vol. 4 (Book No. 100203), The
American Society of Mechanical Engineering.
15. Peden, J.M. and Tovar, J.J.: “Sand prediction and exclusion
decision support using an expert system” Paper SPE 23165
presented at the 1991 Offshore European Conference, Aberdeen,
September 3-6.
16. Einstein, E.E. and Edwars, K.W. 1990. Comparison of an expert
system to human experts in well-log analysis and interpretation.
SPE Form. Eval. March: 39-45.
17. Allain, O. and Houze, O.P.: “A Practical artificial intelligence
application in well testing interpretation” Paper SPE 24287
presented at the 1992 European Petroleum Conference,
Stavanger, May 25-27.
18. Hutchin, L.A., Burton, R.K. and Macintosh, D.J.: “An expert
system for analyzing well performance” Paper SPE 35705
presented at the 1996 Western Regional Meeting, Anchorage,
May 22-24.
19. Alegre,L., Morokooka, C.K. and Rocha, A.F.: “Intelligence
Diagnosis of rod pumping problems” Paper SPE 26516
presented at the 1993 Annual Technical Conference, Houston,
October 3-6.
20. Patricio, A.R., Rocha, A.F. da and Morooka, C.K.: “Seplant: an
expert system for process plant and gas lift well” Paper SPE
28238 presented at the 1994 Petroleum Computer Conference,
Dallas, July 31-August 3.
21. Corpoven, M.V.O.: “Real Time Expert System (R.E.T.S) for rod
pumping optimization” Paper SPE 34185 presented at the 1995
Petroleum Computer Conference, Houston, June 11-14.
22. Gharbi, R.B. 2000. An Expert System for Selecting and
Designing EOR Processes. Accepted for Publication, J.Petrol.
Sci.Eng.March 2000.
23. Dharan, M.B., Turek, E.A. and Vogel, J.L.: “The fluid properties
measurement expert system” Paper SPE 19134 presented at the
1989 Petroleum Computer Conference, San Antonio, June 26-
28.
24. Mitchell, T. M., “Machine Learning”, Mc Graw Hill, 1997.
25. Ranson, A.; Hernandez, K.; Matheus, J.; Vivas A.: “Monitoring
and knowledge extraction in real time multivariable dynamical
processes”. ANNIE 2001. 2001.
26. Rujano, R.: Implementation and evaluation of clustering
algorithms using non-deterministic search algorithms to find
the optimal number of classes, Thesis PDVSA INTEVEP, 2002.
27. Quinlan, J. R : C4.5 Programs for Machine Learning, Morgan
Kaufman Publishers, California 1993.
28. Ranson, A., Hernández, K.Y., Matheus, and Vivas A.A.: “A New
Approach to Identifying Operational Conditions in
Multivariable Dynamic Processes Using Multidimensional
Projection Techniques” Paper SPE 69523 presented at the 2001
Latin American and Caribbean Petroleum Engineering
Conference, Buenos Aires, March 25–28.
29. Joseph, J., Taber, F., David, M. and Seright R. S.: “EOR
Screening Criteria Revisited” Paper SPE/DOE 35385 presented
at the 1996 SPE/DOE Symposium on Improved Oil Recovery,
Tulsa, April 21-24.
30. Thomas, S., and Farouq, A.: “Field Experience with Chemical Oil
Recovery Methods”, Chemical Abstract, Vol. 3, 1995, P:45-3 to
49-3..
31. Rao, D: Gas Injection EOR a New Meaning in the New
Millennium. The Journal of Canadian Petroleum Technology,
Feb. 2001, Volume 40, No. 2.
32. Dobitz, J. K. and Prieditis J.: “A Stream Tube Model for the PC”
SPE/DOE 27750 presented at the 1994 Symposium on
Improved Oil Recovery, Tulsa, April 17-20.
33. K. Thukral and M. Karuppasamy. Hydrocarbon Development -
Simulation of EOR Applications. Energy, Vol. 16, No. 9, pp.
1207-1212. 1991.
34. T. Okazawa, P. E. Bozac, A. C. Seto, G. R. Howe. Analytical
Software for Pool-wide Performance Prediction of EOR
Processes. The Jorunal of Canadian Petroleum Technology,
April 1995, Vol. 34, No. 4.
35. PRIze : Analytical Model for Evaluating the EOR Potential of
Petroleum Reservoir. 1994.
36. Yu, J. P., Zhuang, Z., and Watts, R. J: “Microcomputer
Applications in Economic Assessment and Risk Analysis of
CO2 Miscible Flooding Process” Paper SPE 19318 presented at
the 1989 Eastern Regional Meeting, West Virginia, Oct. 24-27.
37. Basnieva, I. K., Zolotukhin, A. B., Eremin, N. A., and Udovina.
E. F.: “Comparative Analysis of Successful Application of EOR
in Russia and CIS” Paper SPE 28002 presented at the 1994
University of Tulsa Centennial Petroleum Eng. Symp, Tulsa
August 29-31.
38. Chung, T.-H., Carroll, H. B, and Lindsey, R.: “Application of
Fuzzy Expert Sistems for EOR Project Risk Analysis” Paper
SPE 30741 presented at the 1995 Annual Technical Conference
& Exhibition, Dallas, October 22-25.
39. Gharbi, R. B. C.: “An expert system for selecting and designing
EOR processes”. Journal of Petroleum and Engineering, 27
(2000) 33-47.
40. Abou-Kassem, J. H. Screening of Oil Reservoirs for selecting
Candidates of Polymer Injection. Energy Sources, 21: 5-16,
1999.

6. 6 V. ALVARADO, A. RANSON, K. HERNÁNDEZ, E. MANRIQUE, J. MATHEUS, T. LISCANO, N. PROSPERI SPE 78332
Table 1. Summary of Screening Criteria for Polymer and CO2 Flooding29,35,40
Parameter Polymer
Flooding
CO2
Flooding
Oil Gravity (°API) > 22 > 25
Oil viscosity(cp) < 100 < 15
Crude Oil Composition NC High % C5-C12 fraction
Oil saturation (% PV) > 50 > 25
Water salinity (ppm) < 100000 NC
Water hardness (ppm) < 5000 NC
Mobility ratio 2 - 40 NC
Reservoir temperature (°F) < 200 NC
Rock type Sandstone preferred Sandstone or carbonate
Permeability (mD) > 50 NC
Depth (ft) < 9000 > 2500
Net thickness NC Wide range
Minimum Miscibility Pressure NC < Original pressure
Drive mechanism No gas cap and no bottom
water drive
No gas cap
Table 2. Range of Values for each Parameter in the entire database.
Parameter Interval Mean
Porosity (%) 5.5-37 17.60
Temperature (°F) 60-325 132.89
Pressure (psi) 20-10800 2044.77
Permeability (mD) 0.2-10500 450.04
Gravity (°API) 8.5-55 32.48
Viscosity (cP) 0.07-5000 72.95
Table 3. Statistics for methods associated with cluster 5.
Method %
Air 41.38
Steam 27.59
CO2 Immisc. 10.34
Polymer 8.62
WAG CO2 Inmisc. 5.17
Water Flooding 5.17
N2 Inmisc. 1.72
Table 4. Set of Rules defined for International Data Base Projection.
Cluster Number RULE
1 IF POROSITY <= 15.05 && TEMP <= 120.5 && VISC > 3.35 &&
VISC <= 7.6 Then Cluster 1
2 IF VISC <= 7.6 && POROSITY <= 15.05 && TEMP > 120.5 && TEMP
<= 255 && PRESSURE > 1976 Then Cluster 2
3 IF POROSITY <= 15.05 &&TEMP <= 120.5 && VISC <= 3.35 Then
Cluster 3
4 IF VISC <= 7.6 && POROSITY > 15.05 &&TEMP <= 145 &&PERM >
375 Then Cluster 4
5 IF TEMP <= 255 && VISC > 7.6 && PERM > 81.25 && POROSITY >
9.75 Then Cluster 5
6 IF TEMP > 255 && API > 40.75 Then Cluster 6

7. SPE 78332 SELECTION OF EOR/IOR OPPORTUNITIES BASED ON MACHINE LEARNING 7
Table 5. Statistics of recovery methods associated with Cluster 4-C (see Fig. 6).
Method %
Polymer 33.33
CO2 Misc. 16.67
Steam 16.67
Air 16.67
Water flooding 16.67
Table 6. Statistics of methods associated with cluster 5-F of Fig. 6.
Method %
CO2 Inmisc. 40
N2 Inmisc. 20
Wag CO2 Inmisc. 20
Polymer 20
Table 7. Set of Rules defined for New Analysis Cluster 5 ( Reservoir B).
Cluster Number RULE
1 IF IF VISC <= 135 && API > 24.5 && PRESS <= 489.85 Then Cluster 1
2 IF API > 20.05 && VISC > 135 Then Cluster 2
3 IF API <= 20.05 && TEMP <= 107.5 Then Cluster 3
4 IF API <= 20.05 && TEMP > 107.5 &&PERM > 314.75 &&VISC > 95
Then Cluster 4
5 IF VISC <= 135 && API > 20.05 &&API <= 24.5 Then Cluster 5
6 IF API <= 20.05 && TEMP > 107.5 &&PERM <= 314.75 5 Then Cluster
6

8. 8 V. ALVARADO, A. RANSON, K. HERNÁNDEZ, E. MANRIQUE, J. MATHEUS, T. LISCANO, N. PROSPERI SPE 78332
Large field scale
application Successful
Simulation studies
and Pilot Design
Definition & Performance
Laboratory
evaluation
Project performance
and
Monitoring
New drainage
strategies
SPE
65128
Planning & Conceptualization
Main
Screening
Criteria
Preliminary
Economy
Evaluation
(EE)
SPE 50645
SPE
72099
75259
Project
Evaluation
and
EE
Unsuccessful
Fig. 1. Typical planning for IFLs, illustrated through the VLE example. SPE papers reflect the progress of this particular case.
Water Flooding
30%
Polymer
15%
Air
10%
CO2 Misc.
9%
Steam
7%
Wag CO2 Misc.
7%
CO2 Inmisc.
5% N2 Misc.
5%
N2 Inmisc.
4%
Wag CO2 Inmisc.
1%
Wag HC Inmisc.
1%
Wag N2 Misc.
1%
Wag HC Misc.
5%
Water Flooding
Polymer
Air
CO2 Misc.
Steam
Wag CO2 Misc.
CO2 Inmisc.
N2 Misc.
Wag HC Misc.
N2 Inmisc.
Wag CO2 Inmisc.
Wag HC Inmisc.
Wag N2 Misc.
Fig. 2. Distribution of EOR Methods reported in the collated database. Waterflooding abunts in the database, followed by polymer flooding.
Scarce data were available for processes such as Nitrogen injection

9. SPE 78332 SELECTION OF EOR/IOR OPPORTUNITIES BASED ON MACHINE LEARNING 9
CO2 Inmisc.
CO2 Misc.
N2 Inmisc.
N2 Misc.
Polymer
Steam
Wag CO2 Inmisc.
Wag HC Inmisc.
Wag CO2 Misc.
Wag HC Misc.
Air
Water Flooding
Cluster 1
Cluster 5
Cluster 6
Cluster 2
Cluster 3
Method %
Water Flooding 29.17
CO2 Misc. 20.83
Polymer 18.75
N2 Inmisc. 6.25
Steam 6.25
WAG HC Misc. 6.25
CO2 Immisc. 4.17
WAG CO2 Misc. 4.17
N2 Misc. 2.08
WAG N2 Misc. 2.08
Method %
Water Flooding 38.46
WAG CO2 Misc. 13.46
WAG HC Misc. 13.46
N2 Misc. 9.62
CO2 Misc. 7.69
N2 Inmisc. 7.69
Polymer 5.77
Air 3.85
Method %
Water Flooding 48.28
Polymer 25.29
WAG CO2 Misc. 12.64
CO2 Misc. 10.34
N2 Inmisc. 1.15
WAG HC Misc. 1.15
Steam 1.15
Cluster 6
Method %
N2 Misc. 42.86
N2 Inmisc. 21.43
WAG N2 Misc. 14.29
Water Flooding 14.29
WAG HC Misc. 7.14
Cluster 4
Method %
CO2 Immisc. 22.58
Air 12.90
Water Flooding 12.90
CO2 Misc. 9.68
Polymer 9.68
WAG HC Inmisc. 9.68
N2 Misc. 6.45
WAG HC Misc. 6.45
N2 Inmisc. 3.23
Steam 3.23
WAG CO2 Misc. 3.23
Method %
Air 41.38
Steam 27.59
CO2 Immisc. 10.34
Polymer 8.62
WAG CO2 Inmisc. 5.17
Water Flooding 5.17
N2 Inmisc. 1.72
Fig. 3. International DataBase Projection.
CO2 Inmisc.
N2 Inmisc.
Polymer
Steam
Air
Water Flooding
Wag CO2 Inmisc.
Cluster A
Cluster C
Cluster D
Cluster B
Cluster F
Method %
Steam 37.5
Air 25
Polymer 25
Water Flooding 12.5
Method %
Air 52.94
Steam 47.06
Method %
Air 70
Steam 30
Method %
CO2 Inmisc. 33.33
Air 22.22
Wag CO2 Inmisc. 22.22
Water Flooding 11.11
Polymer 11.11
Method %
Air 44.44
Steam 22.22
CO2 Inmisc. 11.11
Polymer 11.11
Water Flooding 11.11
Cluster E
Method %
CO2 Inmisc. 40
N2 Inmisc. 20
Wag CO2 Inmisc. 20
Polymer 20
Fig. 4. New Analysis of Cluster 5 of the International Data Base Projection.

10. 10 V. ALVARADO, A. RANSON, K. HERNÁNDEZ, E. MANRIQUE, J. MATHEUS, T. LISCANO, N. PROSPERI SPE 78332
CO2 Inmisc.
CO2 Misc.
N2 Inmisc.
N2 Misc.
Polymer
Steam
Wag CO2 Inmisc.
Wag HC Inmisc.
Wag CO2 Misc.
Wag HC Misc.
Air
Water Flooding
Reservoir A
Reservoir BCluster 1
Cluster 5
Cluster 6
Cluster 2
Cluster 3
Method %
Water Flooding 29.17
CO2 Misc. 20.83
Polymer 18.75
N2 Inmisc. 6.25
Steam 6.25
WAG HC Misc. 6.25
CO2 Immisc. 4.17
WAG CO2 Misc. 4.17
N2 Misc. 2.08
WAG N2 Misc. 2.08
Method %
Water Flooding 38.46
WAG CO2 Misc. 13.46
WAG HC Misc. 13.46
N2 Misc. 9.62
CO2 Misc. 7.69
N2 Inmisc. 7.69
Polymer 5.77
Air 3.85
Method %
Water Flooding 48.28
Polymer 25.29
WAG CO2 Misc. 12.64
CO2 Misc. 10.34
N2 Inmisc. 1.15
WAG HC Misc. 1.15
Steam 1.15
Cluster 6
Method %
N2 Misc. 42.86
N2 Inmisc. 21.43
WAG N2 Misc. 14.29
Water Flooding 14.29
WAG HC Misc. 7.14
Cluster 4
Method %
CO2 Immisc. 22.58
Air 12.90
Water Flooding 12.90
CO2 Misc. 9.68
Polymer 9.68
WAG HC Inmisc. 9.68
N2 Misc. 6.45
WAG HC Misc. 6.45
N2 Inmisc. 3.23
Steam 3.23
WAG CO2 Misc. 3.23
Method %
Air 41.38
Steam 27.59
CO2 Immisc. 10.34
Polymer 8.62
WAG CO2 Inmisc. 5.17
Water Flooding 5.17
N2 Inmisc. 1.72
Fig. 5. Venezuelan Reservoir map in the international data base projection, located in clusters 4 and 5 (see dashed circles).
CO2 Immisc.
CO2 Misc.
N2 Immisc.
N2 Misc.
Polymer
Steam
WAG-HC Immisc.
WAG-CO2 Misc.
WAG-HC Misc.
air
Water flooding
Reservoir A
Cluster A
Method %
Aire 20.0
Co2 Inmi 20.0
Polymer 20.0
Wag-HCInmi 20.0
Water Flooding 20.0
Cluster B
Method %
Aire 28.6
Co2 Mis 14.3
N2_Miscible 14.3
Wag-HCInmi 14.3
Wag-HCMisc 14.3
Water Flooding 14.3
Cluster C
Method %
Polymer 33.3
Water Flooding 16.7
Steam 16.7
Co2 Mis 16.7
Aire 16.7
Cluster D
Method %
Co2 Mis 16.7
N2_Miscible 16.7
Wag-Co2Misc 16.7
Wag-HCInmi 16.7
Wag-HCMisc 16.7
Water Flooding 16.7
Cluster E
Method %
Co2 Inmi 85.7
N2_Inmiscible 14.3
Fig. 6. New Analysis of Cluster 4 for ReservoirA, clearly indicated by the dashed circle in cluster C.

11. SPE 78332 SELECTION OF EOR/IOR OPPORTUNITIES BASED ON MACHINE LEARNING 11
CO2 Inmisc.
N2 Inmisc.
Polymer
Steam
Air
Water Flooding
Wag CO2 Inmisc.
Series20
Cluster 1
Cluster 3
Cluster 4
Cluster 2
Cluster 6
Method %
Steam 37.5
Air 25
Polymer 25
Water Flooding 12.5
Method %
Air 52.94
Steam 47.06
Method %
Air 70
Steam 30
Method %
CO2 Inmisc. 33.33
Air 22.22
Wag CO2 Inmisc. 22.22
Water Flooding 11.11
Polymer 11.11
Method %
Air 44.44
Steam 22.22
CO2 Inmisc. 11.11
Polymer 11.11
Water Flooding 11.11
Cluster 5
Method %
CO2 Inmisc. 40
N2 Inmisc. 20
Wag CO2 Inmisc. 20
Polymer 20
Fig.7. New Analysis of Cluster 5 for Reservoir B, indicated by the dashed brown ellipse. Several points indicate sensitivity analysis.