SPE 108702
New Insights and Applications of Soft Computing on Analysis of Water Production
From Oil Reservoirs
Leonid Sheremetov, Ana Cosultchi, Ildar Batyrshin, and Jorge Martínez-Muñoz, Instituto Mexicano del Petróleo, and
Sergio Berumen, Pemex Exploración y Producción
Copyright 2007, Society of Petroleum Engineers
This paper was prepared for presentation at the 2007 International Oil Conference and
Exhibition in Mexico held in Veracruz, Mexico, 27–30 June 2007.
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, Texas 75083-3836 U.S.A., fax 01-972-952-9435.
Abstract
During the development of reserves of oil fields high water
production is not uncommon problem. However, the practice
of Oil &Gas production has witnessed many cases with poor
field results and low efficiency of operational efforts to
mitigate the water production problem. One of the most
common reasons is a lack of a precise diagnosis to identify the
problem that considers the analysis of well and reservoir
integrated information. That is why the key factor in the
success of a water control treatment relies in a reliable and
precise diagnostics identifying the origin of the water
production problem. This leads to select an adequate treatment
to effectively mitigate the problem.
This paper describes a new methodology of water shut-off
(WSO). This methodology is implemented through a new
generation computerized tool, SMART-Agua, which uses soft
computing techniques (fuzzy expert systems and perceptual
data mining). The system was designed to manage the
diagnosis of 14 potential problems commonly found both at
wells and reservoirs. The solution analysis phase offers a
technical and an economic analysis of the different chemical
or mechanical applications and evaluates the potential for
additional productivity if the water cut is reduced. Additional
modules for solution design (gels, mechanical solutions and
surface separation facilities) generate the parameters of the
most efficient solutions instead of selecting a particular
product from the vendor. Applications of this new
methodology to several oil field cases were used to validate
the methodology, and results of two such cases are discussed
in the paper. These cases are focused on a carbonate fractured
oil reservoir and a carbonate reservoir under waterflooding.
Introduction
The growing interest and enthusiasm in the petroleum industry
for the intelligent engineering systems like intelligent wells
and intelligent fields have increased significantly the demand
for more powerful, robust and intelligent tools. The interest in
decision support for the operational activities lays in the fact
that in petroleum companies, senior personnel daily have to
solve problems based on extensive data analysis and their
experience gained through years of field work. In recent years,
hybrid intelligent systems (HIS) integrating different artificial
intelligence techniques from expert systems to fuzzy logic
have attracted the attention of the Oil & Gas industry, by their
capacity to handle successfully the complexities of the real
world like imprecision, uncertainty and vagueness [1-4].
In spite of the fact that the petroleum industry is usually
being criticized for moving at a snail’s pace in embracing
information technology, it was one of the first in applying
artificial intelligence technologies. Expert systems like
DIPMETER ADVISOR and PROSPECTOR are often
mentioned as examples of early classic systems. Nevertheless,
an analysis of commercially available intelligent software,
indicates that although there are some applications, this type
of systems have still not been put widely in the market [5]. For
example in the area of water control, two expert systems exist
from service companies, XERO and WaterCase developed by
Halliburton and Schlumberger respectively to provide
assistance during the diagnosis and treatment selection phases
of water control [6, 7]. But these systems are primarily of
internal use for choosing a product from that vendor, and are
not open for scrutiny or modification of the rule bases.
This work describes a hybrid intelligent system that uses
fuzzy logic, the techniques of the expert systems and of data
mining to support the petroleum engineers in the diagnosis of
the problems of water shutoff, one of the challenging
problems of oil production. In many cases, the innovating
WSO technologies can mean a reduction of the costs and an
increase in the hydrocarbon production. In order to properly
diagnose a problem and find most suitable and efficient
solution, one requires the intervention and the experience of
the petroleum engineers. SMART-Agua (Agua stands for
Water in Spanish) makes use of these experiences acquired for
naturally fractured reservoirs (typical in Mexican scenes)
integrated within a hybrid intelligent tool.
Water production problem
Water production represents an imperative and complex
problem affecting the lifespan of most hydrocarbon wells [8,
9] and, consequently, any potential solution and its success
depends on the opportune and reliable diagnosis of the
2 SPE 108702
problem. If one considers that the world-wide water
production is of approximately 210 million barrels per day
which accompanies to the 75 million barrels per day of
petroleum, it would be possible to say that many companies
have practically become water producers [9, 10]. Since the
systems of handling of water are expensive - the cost is
considered between 5 to more than 50 cents per barrel of water
- in a well that produces petroleum with an 80% of water cut,
the cost of the handling of the water can ascend to $4 per
produced petroleum barrel.
Reservoir rocks normally contain both petroleum
hydrocarbons and water but, sources of water may include
flow from above or below the hydrocarbon zone, flow from
within the hydrocarbon zone, or flow from injected fluids and
additives resulting from production activities. This water
becomes produced water when hydrocarbons are produced and
these fluids are brought to the surface as a mixture.
Nonetheless, water-driving mechanism is the beneficial aspect
of an aquifer.
The physical and chemical properties of produced water
vary considerably depending on the field location, the
geological formation, and the type of hydrocarbon being
produced [11]. If water flooding operations are conducted,
these properties and volumes may vary even more
dramatically as additional water is injected into the formation.
Water production may also provoke additional problems
such as: a) corrosion of tubing string, casing or their
connections; b) obstruction of the tubing by scale; c) increase
of hydrostatic pressure, etc. For example, it is challenging to
minimize the amount of produced water by using blocking
devices or chemicals, but the strategies depend upon the
problem location, type of hydrocarbon production (natural or
waterflooding) type of formation (i.e. anisotropy), the results
of previous treatments, etc. This requires synthesizing the
information from different data sources like geological model
of the field, perforation reports, well and production logs,
fluids’ analysis, records of maintenance of oilfield facilities,
etc.
Solution description
The aim of SMART-Agua system is to offer technical support
to field operators in order to diminish the time spent for
diagnosis and to increase the success of the selected solution.
According to the water control process diagram (Fig. 1), in
order to diagnose the problem, the data comprising several
groups such as: borehole features, production histories,
production logs, reservoir characterization, water analysis and
past-treatments data are required. At the diagnostics phase,
there are two stages: a preliminary one and a confirmation
step. The underlying methodology for both of those was
constructed over the experiences of experts in the problems of
water control. The discrimination between the different water
production problems (the preliminary diagnostics) is based on
93 properties and data collected from the field. For a particular
case, usually there is no unique diagnosis but few problems
that can be ranked by score according to their likelihood of
incidence. Hence there lies the need for confirmation of
diagnosis.
Usually, at this second stage, it is required to apply
additional tests, run updated logs and analyze current
produced water. All these activities provide the system with
updated information which will confirm (or discard) problems
found during the preliminary diagnosis.
The stage of solution analysis and selection also consists of
several steps. The first one consists of evaluating the
recommended treatments for the selected problem. Treatment
selection is based upon treatment objectives; these are
classified in three groups: those that minimize the amount of
produced water that reaches the surface, those that recycle or
re-inject produced water, and those that involve disposal of
produced water. Application of the adequate solution
represents challenges and costs which may increase with a
wrong diagnosis. The evaluation of solutions is performed and
adjusted to: the available facilities, characteristics of the
formation, results of water analysis and the objective of the
treatment as stated above.
Upon deciding what solution to apply, it is required a
detailed design as the next step in the process. This task
includes the parametric design of the most efficient solution
along with a review of available products and equipment for
water control and management that meet the requirements of
the suggested solution. Since the justification of a treatment in
any well depends upon the expected increase in the
hydrocarbon production, it is needed to quantify the cost of the
treatment of water control by means of an analysis of risk
components using the methods of quantitative risk analysis.
This analysis is made in the phase of economic evaluation of
the proposed technical solution.
Additionally, the results from each treated case, either
success or failure, are processed in order to feed two
databases: the knowledge base and a historic record of cases
for future reuse.
Statement of Theory and Definitions
In order to develop a straightforward strategy and
methodology for water excess diagnosis and solution analysis
(described in more details in this section), experts from
different problem domains were consulted. These domains
included in particular: reservoir, formation and petrophysical
models, NODAL analysis, pressure transient test, logging
tools, conning simulation, gels and polymers, economic
evaluation of separation facilities and equipment, risk analysis,
etc. Experts experiences along with other additional
information obtained from available bibliographical sources
were structured in order to relate the problems with their
symptoms as explained below. Further on, the relationships
were coded in the form of fuzzy production rules thus forming
three related knowledge bases (preliminary, confirmation and
solutions), one for each stage of the diagnosis and solution
analysis process.
Classification of problems of excessive water production
Most excess-water-production problems are classified using
different criteria depending on the author’s interest [9, 12-19].
That is why, although the water production problems are
widely described in the literature [9], we briefly describe them
emphasizing the considerations used in this application. For
classification purposes there are used: the water entrance
(linear or radial), the water control treatment strategy, or the
type of formation (fractured or not). SMART-Agua uses
SPE 108702
location of the problem as a primary classification parameter
and includes 14 problems. Each of the these problems can be
classified as (Table 1): a) close to the well bottom, b) within
the formation, c) between injector and producers, d) at the
injector and, e) at the well completion. The problems are
defined as follows.
Moving oil-water contact. This problem occurs in wells
producing from homogeneous formation with associated
aquifers, which is a normal situation for all wells where water
is co-produced with oil and the pressure depletion promotes a
slow entrance of bottom water. This problem can be
recognized if the well produces below the critical flow rate.
Coning or cresting. This problem is identified if the well
produces above critical flow rate from homogeneous
formation with associated aquifers. Coning occurs in vertical
producers and depends on the well pressure depletion, while
cresting occurs in horizontal wells.
Water channeling. This case occurs when the formation
has permeability contrasts and impermeable barriers, which
allows water to mobilize towards the producer by flowing
though high permeability strata (channels).
Water channeling through fractures. This problem occurs
in naturally fractured formations which allow water phase
entrance. This problem can occur also for waterflooding. In
such case, it is favored by fracture nets crossing both, injector
and producer, but if such nets are not directly crossing the
well, then water entrance may occur in the future.
Barrier breakdown. Fracturing of naturally impermeable
barriers allows water entrance. Fracturing may be the
consequence of fracturing workovers or pressure excess
during production or completion.
Stimulation out-of-zone. Any stimulation or cleaning
workover may contribute to open channels or fractures into a
fractured formation.
Poor areal sweep. When waterflooding is used in
anisotropic formation containing high permeability strata
(channels) water starts flowing preferentially through these
channels.
Channel through high permeability strata. A similar
problem, but the formation may not necessarily be anisotropic.
The presence of impermeable barriers, however, blocks
crossing flow between strata.
Channel trough high permeability strata with crossflow.
When the impermeable barriers are break down, crossing flow
between strata occurs.
Flow behind casing. A poor adherence of the cement to the
formation side means channels, caverns or fractures, which
allow water to accumulate behind casing. Usually occurs when
casing was poorly cemented or the cement job was damaged
during production or repairing jobs and especially, when the
formation contains sand or clay.
Casing leaks. This problem occurs as a consequence of
corrosion or pipe misalignment when the casing has small or
large aperture breaches.
Casing corrosion in injector well. The problem is similar
to casing leaks, although, in this case, injected water may
displace behind casing instead of the formation front.
Gravity segregation in injector well. This problem happens
in injector wells due to the presence of fractures. Thus,
3
injected water is gravitationally displaced to the well bottom
instead of the formation front.
Problems classified as close to the well bottom are mostly
associated to naturally or artificial lift producer wells:
• Moving oil-water contact
• Coning or cresting
• Water channeling
• Water channeling through fractures
Problems located within the formation are related to
previous workover with negative results on the oil production:
• Barrier breakdown
• Stimulation out-of-zone
Problems located between injector and producer wells are
more common during waterflooding:
• Poor areal sweep
• Water channeling through fractures
• Channel through high permeability strata
• Channel trough high permeability strata with crossflow
Problems located at the Injector well are typical for these
wells:
• Gravity segregation in injector well
• Casing corrosion in injector well
Problems located at the well completion are common to all
producers:
• Casing leaks
• Flow behind casing
Therefore, when the analyzed well is a producer, there are
12 problems to be evaluated, although they are different when
the well produces naturally or through artificial lift than when
it produces by waterflooding. When the well is an injector,
only two problems can be present.
Methodology of preliminary diagnosis
Due to the space limits, the preliminary diagnosis phase is
discussed in more details than the other phases. Different facts
(also called resources of the problem domain) intervene in the
elaboration of the preliminary diagnosis as shown in Table 1.
In SMART-Agua, formation characteristics are obtained
from input data like: mineral composition, permeability,
permeability distribution model (e.g. Dykstra-Parson) and
interpretations of logs (e.g. sonic, gamma ray).
A homogeneous formation is related to two problems; in
comparison, a heterogeneous formation is present in six
problems as follows: a channel-type formation defines the
problem Water channeling, an anisotropic formation is related
to Gravity segregation and Casing corrosion in injector well,
a fractured formation is related to Water channeling through
fractures, Stimulation out-of-zone and Gravity segregation.
Anisotropy and the presence of fractures are involved in the
diagnostics of: Stimulation out-of-zone and Gravity
segregation.
The original and current state of low permeability barriers
helps in diagnosing five problems. Those states are obtained
with data from the Gamma Ray (GR) log, the Formation
Micro-Imaging (FMI) log and the formation lithology – like
the presence of shale strata. Whether the original barriers were
complete and they are currently broken, is an important
evidence for the problem Barrier breakdown. Moreover, it is
possible to discriminate between Channel trough high
4 SPE 108702
permeability strata and Channel trough high permeability
strata with crossflow, as for the first one, barriers integrity
must be complete. Channeling and Poor areal sweep are two
other problems related to a complete integrity.
The location of oil-water contact (OWC) is obtained from
logs and production data. It is used in the diagnostics of eight
problems and represents a very valuable resource for
discriminating between Moving OWC and Coning or cresting.
Water entrance can be either slow or fast. The first case
helps in the diagnostics of four problems whereas the second
case is related to three problems.
Critical production rate is important to distinguish between
Moving OWC and Coning or cresting. It is calculated with the
Muskat equation which works only for a homogeneous
formation.
The results of water analysis, the presence of corrosion
products, as well as water source, type of water and water
trend, help in diagnostics of thirteen problems.
Water source is defined from historic data of total
dissolved solids (TDS) and salinity of the produced water. If
these remain similar through time, it can be concluded that
water comes from the same aquifer and thus, the same source.
In contrast, an unexplainable variation of those elements may
indicate that the produced water comes from another aquifer,
so it has another source.
The same data (salinity and TDS) are used in order to
define the type of water into: fresh (<1000 mg/L), brackish
(1000-10,000 mg/L), saline (10,000-100,000 mg/L) or brine
(>100,000 mg/L). Since this is formation water, the above
classification may differ from the one of the U.S. Geological
Survey [20].
Concentration of Fe ions, H2S and Fe2O, as well as the pH,
the stability index and the type of water are used to determine
whether water trend is: corrosive, scaling or neutral.
Produced water having another source, is a characteristic
related to: Water channeling through fractures, Channel
trough high permeability strata and Channel trough high
permeability strata with crossflow.
Corrosive or neutral water trends are conditions to
consider while diagnosing Stimulation out-of-zone. In contrast,
a scaling trend is a fact related to Barrier breakdown and Flow
behind casing.
Information about cement bond integrity is obtained from
the Cement Bond Log (CBL), the Cased Hole Dynamics
Tester (CHDT) log, the drilling log and the water trend.
Good cement bond becomes important in the diagnostics
of: Moving OWC, Conning or cresting and Water channeling.
On the other hand, a deficient cement bond supports the
possibility of two problems in the completion of the producer
well (Casing leaks and Flow behind casing) and two more in
the injector well (Gravity segregation and Casing corrosion in
the injector well).
Information about casing integrity comes from the images
log of the mechanical state and from the presence of corrosion
products in the produced water. The obtained data help in the
diagnostics of Casing leaks and Casing corrosion in the
injector well.
Data from the drilling log are used in the diagnostics of:
Communication through fractures, Casing corrosion in
injector well, Casing leaks and Flow behind casing.
The results of the most recent workover are used in the
diagnostics of: Water channeling through fractures, Barrier
breakdown, Stimulation out-of-zone and Casing leaks.
Relative mobility of the displacing fluid (water) and the
displaced fluid (oil) is a fact to consider in the diagnosis of
problems between Injector and Producer: Poor areal sweep,
Water channeling through fractures, Channel trough high
permeability strata and Channel trough high permeability
strata with crossflow.
Injection efficiency is calculated internally using
permeability distribution model like Dykstra-Parson. The
result is used in the diagnosis of: Poor areal sweep, Water
channeling through fractures, Channel trough high
permeability strata and Gravity segregation.
Methodology of diagnosis confirmation
At the end of the Preliminary diagnostic phase, we can get one
or more diagnostics, especially, when there is a lot of missing
input data. Thus, a second diagnostic phase is required to
confirm the preliminary diagnostics. The input data, at this
stage, are selected according to a risk level analysis. The risk
level refers to the possibility of an erroneous diagnostics,
which would lead to the application of a wrong treatment or
technology, thus worsening the problem instead of solving it.
The input information used for this phase of diagnostic could
be one from of the following resources: simulators, new logs,
pressure or interference tests or water analysis.
During this phase the confirmation can be made both based
on the results of the additional tests for all pre-selected
problems (in a way similar to preliminary diagnostics) or by
collecting facts for one particular problem even though this
problem was not pre-selected at the previous stage. The latter
mode helps the user to check the reasons why any particular
problem he considers possible was not diagnosed by the
system.
Methodology of the solution phase
Once the diagnosis is established, we move forward to the
WSO treatment selection phase. Additionally, economic
evaluation of the selected technologies can be made.
The water shutoff strategy was designed to manage up to
13 different objectives:
• Water control
• Blocking intervals
• Sealing fractures
• Improving areal sweep
• Retarding oil-water contact advance
• Sealing leaks
• Water deviation
• Water separation
• Creating barriers (hydraulic isolation)
• Modify injection
• Modify separation facilities
• Corrosion or scale control
• Well shut-down
Each objective could be fulfilled by different types of
solutions which offer different technologies as indicated in
Table2.
SPE 108702
Therefore, selection of an effective treatment for the
problem involves the selection of the adequate objective, the
well and formation conditions, produced water composition
and the possibility of critical conditions related to: bottom well
temperature> 145oC, water cut >70%; highly acidic or alkaline
water pH, etc.
Fuzzy expert system framework
The facts associated with the properties of the resources of the
problem domain like the rows of Table 1, are related to the
water production problems by means of fuzzy production rules
of the following form:
R1: If A and/or B then H1, pv(R1),
where A and B are facts (premises or antecedents of a rule),
H1 is a hypothesis to be proved (conclusion or consequent of a
rule) and pv(R1) is the plausibility (sometimes also known as
certainty, truth or possibility) value of the rule measured in
some linearly ordered scale L, for example L=[0,1].
Uncertainty of the expert knowledge is reflected in
linguistic variables, as well as both in the uncertainty of the
rules and facts of the expert system. For example, variables
like “flow type” though possibly measured with numerical
values, are more often defined on linguistic scales like “linear,
bilinear, radial”. Since the expertise is often intuitive, human
perceptions about uncertainties of facts and rules are usually
qualitative: “it is possible that the flow type is radial” or “if
the flow type is linear then it is very likely that there are
fractures”. The model of uncertainty defines the mechanism
used in calculating the overall plausibility value of the
condition-part of the rule, inferring the confidence of the
conclusion part, as well as combining the confidences of
multiple conclusions. The main approaches used so far are
based on Certainty Factors model specifying a continuous
range (0 to 1 or -100 to 100), Bayesian Belief Networks, fuzzy
logic and multi-valued logics (true, false, unknown) [21-23].
Since the reasoning process in the expert system is based on
knowledge that can cover many aspects of a problem, it also
may conclude several possible recommendations in the same
way human experts would. Then, the multiple
recommendations would be arranged in order of their certainty
(possibility) values obtained from the certainty values of the
condition-parts of the rules. The inference engine carries out
this process using some model of confidence.
In order to obtain a diagnosis with high degree of
reliability, a hybrid model representing experts’ knowledge
implementing fuzzy multiset based disjunctive algebra of
strictly monotonic operations is used (a conclusion is
considered as a disjunction of facts) [24-25]. The possibility
values, integrated to the model like numerical values, are
defined on a linguistic scale using finite ordinary scales. Here
is an example of such a scale: Xw= {0.Impossible < 1.Very
Small Possibility < 2.Small Possibility < 3.Average Possibility
< 4.High Possibility < 5.Very High Possibility < 6.Sure}.
To illustrate how the inference engine works based on the
disjunctive algebra of possibilities let us consider an example
from the preliminary diagnosis knowledge base. Suppose we
have a set of three facts and two rules:
5
F1: Formation.Lithology== Carbonate, weight 4 (high)
F2: Formation.Lithology == Dolomite, weight 3 (average)
F3: Water production.Flow type == Bilinear, weight 6
(sure)
R1: IF Formation.Lithology== Carbonate
OR Formation.Lithology == Dolomite THEN
Formation.Type == Fractured, weight 5 (very high)
R2: Water production.Flow type == Bilinear THEN
Formation.Type == Channel, weight 3 (average)
Possibility values of the facts and the rules are represented
as elements of a multiset with seven values (positions) of
uncertainty 0,1,2,3,4,5,6, e.g. the weight 4 (High Possibility)
will be denoted as (0-0-0-0-1-0-0). Then the conclusions of
the rules will obtain the following evaluations of possibility:
R1: ((0-0-0-0-1-0-0) ∨ (0-0-0-1-0-0-0)) ∧ (0-0-0-0-0-1-0) =
(0-0-0-1-1-0-0) (High Possibility) and R2: (0-0-0-0-0-0-1) ∧
(0-0-0-1-0-0-0) = (0-0-0-1-0-0-0) (Average Possibility). The
resulting multi-set based evaluation is affected by both the
weight of the input facts the weight of rules. According to this
example, there is a high possibility of having a fractured
formation and an average possibility of a channel-type
formation.
The set of fuzzy rules forms the model of the problem
domain which is coded in CAPNET knowledge representation
format (KRF) [25]. Fig. 2 illustrates a fragment of the rule
development using CAPNET Knowledge Acquisition Tool
(KAT). At the left-hand side of the screenshot a structure of
rule categories and problem domain resources appear. A
structure of a rule appears in the Rule Text window of Fig. 2.
Antecedents and consequents of the rules are propositions that
relate properties from resources to some value by one of the
following operators: equal to, greater than, greater than or
equal to, lower than or equal to and not equal. The rules are
stored both in internal and external (XML) formats to be used
by the inference engine.
To handle the possibility of facts and rules a fuzzy
inference engine called the CAPNET Expert System Shell,
working with linguistic scales was developed. It implements
three types of algebras: conjunctive, disjunctive and additive;
besides it works under the CAPNET KRF model. With its own
user interface, the Shell can be used for debugging knowledge
bases during the development of expert systems. SMART-
Agua gets access to the embedded inference engine through
the developer’s Application Programming Interface (API).
During preliminary diagnostics stage, the expert system
screens the well based on all of the available data trying to fire
all those rules which antecedents has a certain level of
plausibility. That is why at this stage, forward chaining
method (from the data to the goal) is used by the expert
system. Inferred problems are ordered based on the value of
their possibility values coded in the respective multi-sets. At
the confirmation stage, the expert system also offers the
possibility to screen any particular problem even though this
problem did not appear with high possibility score. For this
purpose backward chaining (from the goal to the supporting
data) is used.
6 SPE 108702
Data mining and pattern recognition for identification of
water entrance shapes
In order to diagnose a water production problem, a lot of
numerical data has to be analyzed both related to the well and
to the asset. Different associations between water-related
parameters can support or discard a diagnosis. When studying
directly numerical data, these associations can simply escape
because of the amount of information, which makes the
manual analysis impractical. In order to solve these
difficulties, the techniques of time series data mining were
used.
As an example of data analysis problem at the well level,
consider the relationship between the patterns of water
production and the presence of certain problems. Several
methodologies for interpreting the behavior of water entrance
have been proposed [12]. In this work, the behavior of the
water oil ratio (WOR) vs. time curve in various time domains
(for example, following the breakthrough) is analyzed against
four typical patterns related to the certain problems of water
production. As we illustrate in the case studies, the real
behavior usually makes hard to relate the water production
curve with these patterns. The method of pattern recognition is
based on the moving approximation transform (MAP),
implemented as a part of Percept-Miner toolbox [26]. The
objective pattern (goal) is represented as a sequence of slope
values of the curve (c1,…, ck). The MAP transform replaces
values of the WOR time series with the sequence of slope
values MAP(y)= (a1,…, aN) of the lines f=ait+b,
approximating the data of the time series using a sliding
window. Being based on the measurement of distance between
sequences of slope values, the method looks for the pattern
(ai,…, ai+k-1) in MAP(y) closest to the searched pattern. We
show the example of the application of this technique in the
case study section.
Data mining for analysis of relations between production
and injection wells
As another example of data analysis problem at the reservoir
level, consider water production for reservoirs with
waterflooding mechanism. Sometimes there is more than one
injector well in the neighborhood of the production web, so to
establish the relationships between the production well under
study and nearby injection wells can be useful for the
diagnosis.
In order to solve this problem the data mining techniques
based on MAP transform along with the partial correlation
analysis of time series were applied. In the latter case, the
program analyses the partial correlations between monthly
average values of water and oil production in wells. For each
pair of considered time series a partial correlation is calculated
as a correlation between time series in the set of time points
(months) where some predefined condition is fulfilled.
Different conditions can be considered. For instance, a non-
zero value of oil and/or water production in compared time
series gives the possibility to take into account only time
points when the wells were open for production. Also another
condition “High level of water injection” was considered in
some tests. For each two time series, the number of time
points (called support) when the considered condition was
fulfilled was used for validation of obtained correlation values
by using statistical t-test. In the program we used support
values greater than 40 and correlation values greater than 0.45
to be sure that correlation is valid. Partial correlation was
applied for all pairs of time series: water injection volumes in
injector wells, water and petroleum production volumes in
production wells. Also correlation between injectors and
producers were calculated separately for different values of
time shifting to analyze the effect of delay between water
injection in injectors and oil and water production in
producing wells. Some results are illustrated below.
Case Studies
The methodology described in this paper was applied to
several real cases. We describe here two of them: case A
belongs to the oilfield from the southern region of Mexico,
whereas Case B belongs to the northern region.
Case Description
Case A (historic case). This is one of the wells selected to
validate the performance of the expert system on the historical
cases. With a total depth of 16,115, production started on
September 1976. The original completed intervals were sealed
almost immediately due to salty water invasion. On June 1991,
the next upper interval had to be closed do to pressure falloff,
a new one had to be shot. On February 1995, the water
fraction reached 40%, so the well was closed on March 1995,
the completed interval was isolated and at the time, this led to
shooting an upper interval from 14,583 to 14,682 ft (Fig. 3).
Water invasion began at the end of 1998 and reached a 75% at
mid 2003. Fig. 4 shows a screenshot with the WOR and the
proximity towards the respective patterns obtained using the
data mining technique described above.
This well produces from a carbonate formation. According
to logs, there is also presence of shale and dolomite in minor
proportions, enough to suggest the presence of low
permeability barriers. Average, horizontal and vertical
permeabilities are 250, 250 and 200 mD respectively. At the
time of the problem, production was around 1,686 BOPD with
a 34 API Gravity and a drawdown pressure of around 1000
psi.
Case B. This well is located in the northern fields of
Mexico. Total depth is 6,362 ft. Production started on July 23,
1991, the type is by waterflooding and artificial lift. There are
5 completed intervals in the KTI formation: 4,469 to 4,531 ft,
4,551 to 4,564 ft, 4,577 to 4,636 ft and 4,682 to 4,711 ft and
one in the JSA formation 5,958 to 6,069 ft. The net pay
interval is mostly limestone with a 5% of shale and average
permeability of 0.51 mD. Production is 1,060 BOPD with a 18
API Gravity and a 335 GOR.
Between September and October 2005, all of the
completed intervals where stimulated with acid and a new lift
equipment was installed. As a result, oil production stayed the
same, but water production increased considerably up to 90%
(Fig. 5). Produced water has a salinity of 31,012 mg/L, which
is much lower than the salinity of original formation water:
250,000 mg/L. This, along with the results of association
analysis described below, indicate that produced water is
similar to the injection water.
SPE 108702
Results of diagnosis and solution analysis
Case A. SMART-Agua gives only one diagnostics with
high possibility: Moving OWC. Among the reasons found by
SMART-Agua for concluding this problem are: the type of
formation, the pattern of water entrance and the drawdown
pressure.
A ‘high possibility’ though reasonable for decision
making, still is a sign of lack of additional data that could tune
the preliminary diagnostics. SMART-Agua recommends for
this case, to provide information from water analysis and
interpretation of logs. However, the given operator’s
diagnostics at the time of the problem, proved affinity to the
results of SMART-Agua.
In order to solve the problem, the system recommended a
series of objectives: Water separation, Retarding oil-water
contact advance and Blocking the interval. Water separation
technology depends in this case, on the capacity of the already
existing facilities. Retarding oil-water advance involves a
chemical solution: permeability modifier. Blocking the
interval implies gel injection.
According to the historical records, the application of a
Halliburton product: MOC/One was attempted to control the
problem, which is one alternative for control of bottom water
production. Unfortunately, the expected results were not
obtained and a second and definitive measure was taken: to
isolate the interval with a TBT Plug.
Plugs are also included in SMART-Agua catalogs,
however, for their application, the system requires more
precise data about the presence and integrity of low
permeability barriers. The suggested gel injection is a
compatible alternative for blocking the interval. Finally, there
is also a necessity of opening an upper interval, which was
part of the adopted measure for case A well.
Case B. The system throws three diagnoses in this case. Water
channeling through fractures (between injector and producer
wells) with very high possibility, Poor areal sweep and
Channel through high permeability strata reached a high
possibility. A screenshot of the SMART-Agua system
showing the diagnosed problems is shown in Fig. 6. Some
missing data that could impact the diagnosis include: oil
relative mobility, variation of salinity and total dissolved
solids in the producer well.
All diagnoses are related to waterflooding problems as
expected. The Water channeling through fractures occurs
when there were already active fractures in trajectory between
injector and producer wells. The problem identified as Poor
areal sweep takes place in reservoirs with anisotropy, thus
opening preferred water flows in zones with higher horizontal
permeability towards the producer. Channel through high
permeability strata occurs when the reservoir has permeability
contrasts besides the presence of low-permeability barriers.
Both features allow preferred water flow through channels in
trajectory to the well.
Though these diagnostics were related to waterflooding
problems, in order to select the most efficient treatment, the
most possible problem should be selected because the
treatment objectives and thus the solutions may be different.
For example, confirmation of Water channeling through
fractures puts Fracture sealing objective in first place, while
7
if one of the other two diagnostics is confirmed, Improving
areal sweep is selected as the first objective.
In order to confirm the first diagnosis, one of the additional
phases of diagnosis confirmation was the association analysis
between injector and production wells of the identified zone
where the analyzed well (W08) is located. Figures 7 and 8
show the results of this analysis. Charts depict time series of
water and oil production, well location in the reservoir in 3D
view, and valid correlations among injectors and producing
wells.
Figure 8 shows high partial correlation between water
injection volume in the well W01 and water production level
in production well W08. It can be caused by water penetration
from well W01 to well W08 through fractures typical for
carbonate formations. Also it was found a high correlation
between water injection volume in injector well W02 and oil
production levels in producer wells W05 and W10. A high
partial correlation between injector W03 and other wells was
not found. As a collateral result of this analysis, it can be
proposed to revise the rationality of water injection in wells
W01 and W03.
However, in all the cases, a third recommendation offered
by the system was to separate water at the surface.
Considering the high water cut and the conditions of the well,
field and reservoir, this solution was selected for a detailed
design and analysis. The objective was to define the best
technical and economically feasible scenario for dehydration
of crude and handling of water. Three types of separators were
considered for dehydration: three-phase, gravitational and
electrostatic; also, there were three alternatives for handling of
water: flotation cells, centrifugation and hydrocycloning.
In total, nine different scenarios were simulated with
interaction between SMART-Agua and the Superpro software
as a support tool. Integrated technical and economic analysis
with projection over fifteen years concluded that, the most
viable solution is an electrostatic separator combined with
hydrocycloning (scenario 1C). This latter technique could be
used for both dehydration and handling (scenario 3C) because
it has the lowest costs of maintenance and operation, but is
technically limited, so the electrostatic separator becomes a
better choice for dehydration. Graphic comparison of the nine
scenarios is shown in Fig. 9.
Conclusions
In this paper we presented a hybrid intelligent system for
water production diagnostics and treatment analysis. SMART-
Agua is available in desktop and Web-based releases. We
described the main features of the system and illustrated its
operation with two case studies for a carbonate fractured oil
reservoir and a carbonate reservoir under waterflooding.
SMART-Agua presents several innovative features. Using
fuzzy algebras, SMART-Agua provides diagnostics for both
the main problem and additional ones taking into account the
change of plausibility of premises in expert systems rules
which use qualitative expert evaluations. As a result,
conclusions on the output of inference procedure can obtain
different plausibility evaluations. The data mining tool allows
detecting with greater reliability the behavior pattern of the
water production through time and identifies the associations
between injector and producer wells. Compared to the existing
8 SPE 108702
commercial solutions, our system has a greater range and
more detailed diagnosis of problem types along with
associated modules for the design of different types of
solutions (chemical, mechanical, surface separation facilities,
etc.). We also connected the data bases of the oilfields
(@DITEP) with SMART-Agua, in order to obtain data in an
automatic way in real time. Finally a number of original tools
like CAPNET Expert System Shell and KAT were used for
the development of Smart-Agua.
The validation process beneficiated from the constant work
with WSO experts from Mexico and abroad and specialists
from PEMEX. Further field testing on the cases from the
fractured reservoirs with the participation of the operative
personnel from PEMEX is carried out.
Acknowledgements
Partial support for this research work has been provided by the
IMP within the project D.00322. Special thanks to all our
colleagues from the IMP who participated in the development
of the system. Our special thanks to Dr. Randall Seright for his
invaluable contribution in validation and improvement of
Smart-Agua. Special thanks to the personnel of PEMEX for
their help in validation of Smart-Agua.
References
1. Agent-Based Hybrid Intelligent Systems: An Agent-Based Framework
for Complex Problem Solving, Z. Zhang and Ch. Zhang (eds.), LNAI
2938, Springer Verlag (2004).
2. Soft Computing and Intelligent Data Analysis in Oil Exploration, M.
Nikravesh, F. Aminzadeh and L. Zadeh (eds.), Elsevier Science (2002).
3. Mohaghegh S. D., Wolhart S., Hill D.: “Increasing Natural Gas
Production using a Hybrid Intelligent System,” Adv. in Sci. Computing,
Comp. Intelligence, and Applications –Mathematics and Computers in
Sci. & Eng., WSES Press (2001) 459-467.
4. Sheremetov L. et al.: “Intelligent Computing in the Petroleum
Engineering, Special Issue,” J. of Petroleum Science and Eng., Elsevier
Science (2005) 47, 1-2, 1-3.
5. Mohaghegh, S. D.: “Recent Developments in Application of Artificial
Intelligence in Petroleum Engineering,” J. of Petroleum Technology
(2005) 86-91.
6. “Oil Information Technology Journal,” oilIT.com,
www.oilit.com/2journal/4php/4_builddoc.php?year=2001&month=9.
7. “Reservoir conformance technology,” Halliburton,
www.halliburton.com/public/pe/contents/brochures/web/h01027.asp.
8. Bailey B. et al.: “Water control,” Oilfield Review, Schlumberger (2000).
9. Seright, R. S., Lane, R. H., Sydansk, R. D.: “A strategy for attacking
excess water production,” paper SPE 84966 (2003).
10. Khatib, Z., and P. Verbeek.: “Water to Value - Produced Water
Management for Sustainable Field Development of Mature and Green
Fields,” J. of Petroleum Technology (2003) 26–28.
11. Birkle P. et al.: “Evolution and origin of deep reservoir water at the
Activo Luna oil field, Gulf of Mexico, Mexico”, AAPG Bulletin (March
2002) 86, 3, 457–484.
12. Yortsos, Y.C., Choi, Y., Yang, Z. and Shah, P.C.: Analysis and
Interpretation of Water/Oil Ratio in Waterfloods, paper SPE 59477,
SPEJ (1999) 4, 413-424.
13. Nishikiori, N. y Hayashida, Y.: “Investigation of fluid conductive faults
and modeling of complex water influx in the Khafji oil field, Arabian
Gulf,” paper SPE 66223, Reservoir Eval. & Eng. (2000) 3, 5, 401-407.
14. Al-Ghasham, Y. Y., Catte, D. R. and Al-Haji A. A.: “Integrated analysis
of down-hole corrosion Logs to investigate casing leaks” paper SPE
93184 (2005).
15. “1997 Annual Report,” PRRC,
baervan.nmt.edu/research_groups/reservoir_sweep_improvement/pages/
Annual%20reports/1997%20Annual/1997artoc.html.
16. “eProduction Solutions Red Eye Water-Cut Meter,” eProduction
Solutions,
www.ep-solutions.com/Solutions/EP/Red_Eye_Products.htm
17. “Reservoir Sweep Improvement,” PRRC, baervan.nmt.edu/randy/.
18. “Water Oil Ratio Tool,” Nan Gall Energy,
www.nangall.com/products/plt/fcap.htm.
19. “Reservoir Engineering Handbok (2nd Edition),” KNovel,
www.knovel.com /knovel2/Toc.jsp?BookID=797.
20. “Water basics glossary,” USGS, capp.water.usgs.gov/GIP/h2o_gloss/.
21. Fuzzy expert systems, A. Kandel, CRC Press, Boca Raton (1991).
22. Kasabov, N.: “Decision support systems and expert systems”, Handbook
of Brain Study and Neural Networks, M. Arbib (ed.), MIT Press, (2002).
23. Probabilistic reasoning in intelligent systems, J. Pearl, Morgan Kaufman
(1988).
24. Sheremetov, L. et al.: “An Uncertainty Model for Diagnostic Expert
System Based on Fuzzy Algebras of Strict Monotonic Operations,”
MICAI 2006: Advances in Artificial Intelligence, A. Gelbukh and C.A.
Reyes-Garcia (eds.), LNAI, Springer Verlag (2006) 4293, 165-175.
25. Sheremetov, L. et al.: “Fuzzy Expert System for Solving Lost
Circulation Problem,” Int. Journal of Applied Soft Computing, Elsevier
Science, (2007) (in press).
26. Sheremetov, L. et al.: “SMART-Agua: a Hybrid Intelligent System for
Diagnostics,” INES 2006 10th IEEE International Conference on
Intelligent Engineering Systems, London (June 2006).
27. Batyrshin I., Sheremetov L.: “Perception Based Time Series Data
Mining with MAP Transform,” Advances in Artificial Intelligence,
LNAI, Springer Verlag (2005) 3789, 514–523
SPE 108702 9

Fig. 1—Water control process. Filled blocks refer to the complementary stages not supported by SMART-Agua.

Fig. 2—A screenshot of the KAT v4.0: rule edition mode. At the left-hand side of the screenshot a structure of rule categories and problem
domain resources appear. A structure of one of the rules for Water channeling through fractures problem (with weight equals 4 – high
possibility) is shown in the Rule Text window.
10 SPE 108702

Casing corrosion in injector well

Gravity segregation in injector
Water channeling through

Water channeling through

Moving oil-water contact
Problems

Stimulation out-of-zone

permeability strata with

Channel through high

Channel trough high

Flow behind casing

Barrier breakdown

permeability strata
Coning or cresting

Water channeling

Poor areal sweep

Casing leaks
Input data

crossflow
fractures

fractures

well
Problem location Close to the well bottom Within the Between injector and producer wells At the Injector At the well
formation well completion

Flow rate vs. Critical X X

flow rate
Formation type X X X X X X X X X X X X X
Impermeable barriers X X X X X

Logs information X X X X X X X X X X X X X X

O/W contact X X X X X X X X
Differential pressure vs X X X
Critical Differential
pressure

Producer well water X X X X X X X X X X X X

analysis
Injector well water X
analysis
Water analysis results X X X X X

Integrity of the casing X X X X X X X

cementing
Casing integrity X X
Drilling report X X X X
Previous workover X X X X
results
Relative oil mobility X X X X
Injection efficiency X X X X

Table 1 — Reservoir and well data used to build a preliminary diagnostics.

SPE 108702 11

Type of
Treatment or Technology
solution
Dual complexion
Production
Bottomhole choke
Management
Water separator
Sand plug
Mechanical
Mechanical plug
solutions
Tubing substitution
Chemical / Polymer plug: Viscosity modifier
Special Polymer plug: Permeability modifier
chemical Silicate plug
solutions Cement plug
Corrosion inhibitors Fig. 4—Graphical output of the tool for data processing and data
mining.
Additional Scale inhibitors
treatments Injector change
New separation facilities

Table 2 — Classification of treatment of technology according to

the type of solution

Fig. 5—Water cut evolution for Case B well.

Fig. 3—Cross section of Case B well.

Fig. 7—Monthly oil and water production for Case B oilfield (the
zone of the case study). The upper three wells are injectors. Time
series for water production for producer wells are marked with
letter “a” (like W04a).
12 SPE 108702

Z2

W 12

W 04
Z

W 08 W 10
W 05
Fig. 9—Net present value comparison for scenarios of water
W 0 3 -I n je c to r separation on surface.
W 0 2 - I n je c to r
W 0 1 - I n je c to r

Z1

X2

Y1

X1

Y
X

Fig. 8—Partial correlations between injector and production wells

in Case B oil field.

Fig. 6—SMART-Agua screenshot (in Spanish). It features the dialog for edition of wellbore data and shows the diagnosed problems in a
sample session for Case B.