You are on page 1of 6

SPE 24997

Society of Petroleum Engineers

A Knowledge-Based Approach to Problem Identification in


Producing Wells
Giacomo Cosenza, Mauro Tambini, * and Giovanni Paccaloni, * AGIP SpA
'SPE Members
Copyright 1992, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the European Petroleum Conference held in Cannes, France, 16-18 November 1992.
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 Librarian, SPE, P.O. Box 833836, Richardson, TX 750833836, U.S.A. Telex, 163245 SPEUT.

Productivity Index (SP!). This is defined, in general, as the


ratio between the output of produced goods and the input
energy. In our case, the system output is the produced rate of
hydrocarbon and the input is the loss of pressure 1. Whenever
the actual SP! is less than the theoretical one a problem exists
and it must be identified. (FIG. I). At this point a
technical/economical risk analysis gives the green-light for
the field remedial job or for the acceptance of the current
situation. On the other hand, when the actual SP! is near the
theoretical one, no production problems exist. Therefore, all
the drilling, completion and workover procedures used can be
taken as "sound practices" for future wells.
The above considerations highlight the importance of a
correct WP A for effective reservoir management. So far, there
are few experts able to completely explore each path of WP A
(FIG. 2), in contrast to the need for rapid responses to
minimize the economical impact of the well production
problems. A greater effort has been made to develop a support
system in order to reach the following aims:

ABSTRACT
This paper presents a knowledge-based system (KBS) for the
identification of problems in producing wells. The KBS is
based on a general Well Problems Analysis (WPA)
methodology that entails three main activities: verification of
a problem's existence, identification of a preliminary set of
possible problems, refinement of the most plausible
problem(s). The relevant experts' knowledge has been
formalized by means of a classification approach based on the
organization of all problems in a hierarchy of classes and, by
associating the corresponding symptomatology with them.
The system, validated at the company headquarters, is
currently in use in the operating districts.
The system is a part of a more comprehensive project, called
PROGRESS (pRoduction Optimization inteGRated Expert
SystemS), aimed at providing a decision support system for
well production optimization.
INTRODUCTION

- to preserve and distribute the strategic know-how inside


the company;
- to improve the rate of success of remedial jobs;
- to support interactively the training of new petroleum
engineers;
- to decrease the experts' involvement in routine problems
allowing them to exploit their expertise in more, and often
unsolved, complex situations.

The adoption and exploitation of an appropriate WPA has


great importance for every petroleum company. Early
identification of the productivity problems is necessary to
optimize the field's production [1] and economics, as is
pointed out in the following considerations:
- uneconomical wells could become economical;
- economical wells producing below their theoretical
potential productivity could result in a significant
production increase;
- correct problem(s) identification allows the choice of
optimal remedial job minimizing cost and risk of
operation failures.

To achieve these objectives, users and experts must have


access to the knowledge and the reasoning process encoded in
the system. This is not directly possible using a conventional
information system. Instead, KBS technology allows the
domain knowledge to be explicitly represented and kept
separated from the other parts of the system (e.g., textual and
graphic interface, processes control mechanisms). Moreover

In the WP A methodology the first step is to ascertain the


existence of production problems. To carry out this activity
the expert uses a basic concept that we call System
References and illustrations at the and of paper

367

A KNOWLEDGE-BASED APPROACH TO PROBLEM IDENTIFICATION IN PRODUCING WELLS

SPE24997

in order to match the mentioned experts' reasoning process


with the knowledge representation and reasoning, a
classification shell has been developed in-house. This shell
architecture permits to structure, browse, modify and easily
update the internal Knowledge Base (KB) in a controlled
way. Such features account for important KBS desiderata:

wells) aimed at quantitatively analyzing the considered well


data. At the end of the WPA the expert would also have
selected the suitable solutions fot the problem(s) identified. If
a remedial job is technically and logistically feasible, all the
information will be collected for an appropriate economk:al
evaluation.

- to justi:f.Y and explain the KBS reasoning;


- to add knowledge regarding future technical innovations;
- to guarantee KBS maintainability.

SYSTEM ARCHITECTURE
Conceptual Model and Reasoning Process

WELL PROBLEMS ANALYSIS

In the KBS life cycle the knowledge acquisition and


representation phase is known as the "bottleneck" of the
application's development Nowadays, this phase tends to be
less time and money consuming thanks to the use of a greater
number of knowledge formalization paradigms and
acquisition methodologies. This trend could be summarized
by saying that a general purpose shell (e.g., rule or frame
based) should be tailored to the requirements of each problem
by creating or choosing an adequate formalization paradigm.
Following this approach we decided to develop a built-inhouse shell, called Classification Toolkit, whose underlying
formalisms are (a) natural to tackle hierarchical classification
problems [3-4], (b) specific to represent the WPA relevant
knowledge and (c) suitable to implement the WPA reasoning
process. The shell has been developed exploiting the features
of a commercial tool already providing rule and frame based
paradigms.
As already stated, in the Well Problem Analysis paragraph,
the conceptual model of the WPA consists of three sequential
steps:

The proposed WPA methodology considers all the well


problems covering 95% of our company field cases. FIG.2
shows some of these problems with their corresponding
remedial jobs. The implementation of the system has been set
up emulating the reasoning process of the domain experts
and taking into account the working practices used in the
operative geographic units.
Domain experts carry out the WPA by comparing the system
production parameters (i.e., well head pressure and rate),
which are monitored daily, with their expected values. In fact
the following general questions must be answered:
- Is the production rate uneconomical at the minimum
wellhead flowing pressure?
- Is the production rate economical, but less than expected?
- Is the wellhead flowing pressure or production rate
abnonnally decreasing?
- Is the wellhead flowing pressure increasing at a constant
production rate?

- verification of a problem's existence;


- identification of a preliminary set of possible problems;
- refinement of the previously identified problems.

When at least one answer is positive, some productivity


problems have arisen in the considered period of time. This
period could range from a few days to several months
depending on the rate of change of the production parameters.
In one example, severe drilling fluid formation damage is
shown immediately after the well start-up, whereas scale
deposit can be detected in a longer well life period. It is
important to underline that domain experts instinctively
separate the well production history in meaningful periods to
reduce both the set of the possible problems and the well data
to be analyzed.' At this point the previously mentioned SPI
analysis (FIG. I) takes place to validate the problem existing
by means of a quantitative evaluation of well data.
Subsequently, the expert collects the relevant information to
carry out a first screening of the potential productivity
problems. After this activity only four or five hypotheses
remain and they will be refined in a successive and more
accurate investigation. The accuracy of this analysis
refinement depends both on the qualitative abduction of
problems from the filed data, and on the use of empirical
formulas (e.g., Turner [2] equation for dewatering in gas

In the first step, the experts ascertain if a productivity


problem really exists for the well being examined. This check
is based on the match between theoretical and actual SP!.
When the actual SPI is less than the theoretical one, the
second step takes place.
Domain experts approach the identification of a preliminary
set of problems using a hierarchical classification method.
Each productivity problem, called diagnostic hypothesis, is
characterized by a set of symptoms. A diagnostic hypothesis
is said to affect the considered well if the well behavior shows
the associated symptomatology. Conversely, and according to
the classification method, a well is a member of a class,
which represents all the wells affected by a diagnostic
hypothesis, if it satisfies the properties (i.e., symptoms
represented in the test) associated with the class. FIG.3 and
FIG.4 illustrate two possible causes of productivity loss:
completion choked by asphaltenes and completion choked by
paraffins. Figures show that these diagnostic hypotheses share

368

SFE 24997

G.COSENZA. M TAMBINI, G.PACCALONI

explanations.

the test parts about production rate and flowing pressure


trends. Therefore it is possible to associate such common
parts with a more general class, as shown in FIG.5. This class
collects, as subclasses, all the diagnostic hypotheses
characterized by a progressive decrease of production
parameters. This technique, called bubble up, helps to
organize the diagnostic hypotheses of the WPA domain in a
hierarchy of classes and subclasses as shown in FIG.6. The
hierarchy's root contains the test to ascertain the existence of
productivity problems. The first level of classes represents the
most general diagnostic hypotheses, whereas subsequent
levels represent more specific ones. Each class contains the
corresponding test. Finally, the terminal classes - classes
without further subclasses - represent the most specific
productivity problems and contain two types of tests: relevant
test (i.e., generic symptomatology) and ref"ming test (i.e.,
more specific symptomatology). The former is aimed at
ascertaining a generic evidence of the associated diagnostic
hypothesis, while the latter is aimed at obtaining a more
detailed identification of the well problem(s). The overall
emulation of the KES reasoning process is summarized in
FIG.7:

To exemplify some CIL features we report the formal


representation of the symptomatology associated with the
problem called "completion choked by asphaltene" (FIG.3):
(AND (OR (the trend of tubing head production
rate is progressively decreasing [1])
(the trend of tubing head production
rate is suddenly decreasing [0.5])
(the trend of tubing head production
rate is floatingly decreasing [0.5])
(the trend of tubing head flowing
pressure is progressively decreasing[l])
(the trend of tubing head flowing
pressure is suddenly decreasing [0.5])
(the trend of tubing head flowing
pressure is floatingly decreasing [0.5]))
(the density of produced oil is <= 22)
... other symptoms .... )

In the previous paragraph we mentioned the evaluation of the


tests associated with each class (i.e., diagnostic hypothesis) of
the hierarchy. The test evaluation consists in matching the
formal representation of a symptomatology with the time
evolution of
the parameters of the considered well.
Symptomatologies are represented in the tests using a formal
language called Classification Toolkit Language (CIL). The
implementation of CTL has been aimed at simplifying the
following system developing activities:

Each clause under the OR operator has an associated


evidence factor. This factor indicates how specific a
symptom is for ascertaining the related diagnostic hypothesis.
Low evidence factors (e.g., 0.1) stand for low specificity,
while high evidence factors (e.g., 1) stand for high specificity.
The clause regarding the density value of the produced oil
(i.e., last clause in the test) is treated in a different way
according to a fuzzy-logic like approach. As illustrated in
FIG.3, a diagram is associated with the oil density parameter.
Its X axis indicates the delta between the actual oil density
and its referenced value (22 API) in the clause. The Y axis
indicates the evidence factor corresponding to delta values.
For example, if the actual oil density is 42 API (delta = 20),
then the evidence factor value is 1 (the maximum). This
means that the higher the density of the oil, the higher the
evidence is that the cause of the productivity problem could
be the asphaltene deposit in the production string. The AND
operator collects the evidence factors returned from all the
clauses calculating their average.
C1L language can be easily extended or modified in order to
treat new operators (e.g., exclusive OR, new average
functions for the AND operator, etc.). Moreover the test
clauses are clear enough to be understood by domain experts
and to be updated by knowledge engineers.
Finally, representation of the tests is used to graphically
display its structure and dynamic firing of clauses (for KES
debugging purpose) as well as to generate textual
explanations of produced results to answer the following
general questions:

- knowledge acquisition and formalization;


- KE extensions and maintenance;
- dynamic generation of reasoning justification and

- Why <diagnostic hypothesis name> could be a problem of


the considered well?
- Why <diagnostic hypothesis name> is not a problem of the

- at step #1, FIG.7a, the WPA system evaluates the test


associated with the hierarchy'S root.
- when the previous test succeeds, the system enters step #2,
FIG.7b, evaluating the first level of the hierarchy classes.
The classes whose test failed will not be considered any
more and neither will their subclasses. Instead, the classes
whose test succeeded will remain active. This process
continues evaluating the subclasses of the active classes
until the terminal classes are reached.
- At this point the second step ends producing as a result the
list of the terminal classes whose relevant tests have
succeeded as shown in FIG.7c.
- Finally, step #3 is undertaken by evaluating the refining
tests associated with the list of the terminal classes
previously obtained, as shown in FIG.7d.
Knowledge Representation

369

A KNOWLEDGE-BASED APPROACH TO PROBLEM IDENTIFICATION IN PRODUCING WELLS

SPE24m

considered well?
CONCLUSIONS
For each type of the above questions the users can choose
different levels of explanation. Each test is associated with a
textual explanation (top level explanations) dynamically
composed by the system from the values assigned to the well
parameters occurring in the test clauses. When the top level
explanations do not satisfy the user, he can ask the system to
produce more detailed explanations exploiting each test
clause until it reaches atomic clauses2 .
Our experience has shown that this explanations module has
a high impact not only on the acceptance of system results by
the users, but also on the system v~dation phase.

1. The adoption of a knowledge-based approach has shown its


usefulness in supporting the proposed WPA methodology.
2. WPA-KBS helps users to ascertain the existence of
productivity problems and to identify their causes.
3. WPA-KBS, validated at the company headquarters, is
currently in use at the operating districts.
4. A built-in-house shell has been developed to approach
hierarchical classification problems which is currently reused
for other KBSs.
5. WPA knowledge base can be easily upgraded thanks to the
en formal language.
6. An explanation module has been implemented to facilitate
both the acceptance of the system results by the user and the
system validation phase.

THE OVERALL PROGRESS PROJECT


The WPA-KBS is the entry point of a more comprehensive
support system, called PROGRESS, aimed at providing
support for well production optimization. The PROGRESS
project architecture has been developed to provide a support
system for each productivity problem considered by WPA
(FIG.2). The WPA-KBS is able to identify almost all the
productivity problems and a corresponding remedial job for
each problem. Currently (see FIG.8), PROGRESS supports
final users for the design of matrix stimulation only. When
WPA-KBS identifies formation damage as one of the well
problems, a specific KBS, called DAMAGE [5], is activated.
This system supports users in diagnosing the formation
damage mechanisms. A correct identification of these
mechanisms (e.g., solids plugging, water blocking, scales,
etc.) is fundamental in order to design a successful matrix
stimulation. Subsequently, DESIGN system uses DAMAGE
results and both the chemicals and equipment databases to
provide an ordered set3 of complete matrix stimulation
designs. Each design in the set contains:

NOMENCLATURE
SPI
q
Llp

= system productivity index


bbl/Dipsi
= production rate
bbl/D
= total system pressure draw down psi

ACKNOWLEDGMENTS
The authors wish to thank the management of AGIP SpA for
permission to publish this paper. We also wish to
acknowledge the contnbution ofL.Tomada to the project.
REFERENCES
[1] Brown, K.E.: The Technology of Artificial Lift Methods,
PennWell Publishing Company, Tulsa, Vol.4.
[2] Turner, RG., Hubbard, M.G., and Dukler, A.E.:
"Analysis and prediction of minimum flow rate for the
continuous removal of liquids from gas wells" J.Pel Tech.,
November 1969
[3] Chandrasekaran, B: "Generic Tasks in Knowledge-based
Reasoning: High-level Building Blocks for Expert System
Design", IEEE Expert, Fall 1986, 23-30
[4] Clancey,W.G.: "Heuristic Classification" Artif.Intell. 27
(1985) 289-350
[5] Matteini, L., Cosenza, G., Paccaloni, G., and Beranger,
A.: "A Knowledge-based approach to Matrix Stimulation",
paper SPE 20966 presented at Europec 90, The Hague,
Netherlands, 22-24 October 1990, 283-287

- the selection of stimulation fluids (i.e., preflush, acid and


additives type and composition);
- the volume of selected fluids;
- the required equipment (i.e., coiled tubing, pumps, etc.);
- the operational procedure description (i.e., how and in
which order the selected fluids must be injected, pumping
pressure and rate schedule, etc.).
The last support system, called RISK, is aimed at analyzing
the probability of success for the designed stimulation job.
RISK takes the technical, economical and logistical feasibility
aspects of the operations into account. For example, if the
operation pay-out does not justify the cost of the best
stimulation design, RISK considers the other designs and
selects the one whose cost is justified by its pay-out, if it
exists.
2Clauses tba1 can't be further decomposed.
3The set includes at least the technically best stimulation

370

SPI Analysis

actual SPI=theoretical SPI

~
I

No Field Operation

~
I

Well Monitor

Well Problems

actual SPI<theoretical SPI

ffI.

r--

Expert

positive

J
\\ Low Permeability

Solution

negative

Formation Damage

System

IProductivity Change

Gas Coning

Water Cut

Problem Identification

WPA

Well Problem Analysis

Solutions

Sanding

I
FIG.1 SPI ANALYSIS

I~

Depletion

+positive

Field Operation

Gas Shut-off Treatments


Water Shut-off Treatments
Matrix Stimulation

...

Hydraulic Fracturing

..

Artificial Lifting

---

Unproper Ccmpl.String

Economical Analysis

Sand Control

Ccmpl.Design Optimlz.

FIG.2 PRODUCTIVITY PROBLEMS

Symptomatology

Symptomatology
Production Rate Trend:

Productton Rate Trend:

progresstvel)' decreasing (1)

progressively decreasing (1)

luddenly decr.aslng (0.5)

sudderly decreasing (0.5)

ftoattng decreastng (0.5)

floating decr.aslng (0.5)


Tubtng Head FICMing Pressure:

Tublng Head Flowing Pressure:

progressivei)' decre.slng (1)

luddenly decr.aslng (0.5)


fto.ttng decr.aslng (0.5)

sudderly decre.slng (0.5)


Itoatlng decre.slng (0.5)
Ott density.

progressively decreasing (1)

>= 22 API gravity

Flowing Temperatl.l'D:

< 248 F.

Flowing Tomp - POll" Point:

< 86 F.

-.{;21
FIG.3 COMPLETION CHOKED BY ASPHALTENE

"

FIG.4 COMPLETION CHOKED BY PARAFFIN

L.... '

Progressive

MlJdInWel:lore

EnUilon In WeIbore

~ 8IIrwinWelbore

ProductMty

Loss

L
I

Corrmun SYfTJloma

Tenrineldesses

Overdlmensloned S01ng

:;:

Uldardlmeoslooed StrIng

-=:::::::::.

Roa:Cless

DewIopmtft Wels

PtwlllljFkMingMonolayef

Paro&t.i FMf'Ig

MJII~

Formaaon o.neg.
P..-meab/ltyorVllCosIIY

~HgIlOlIVlSCOSIty

InsLllldent Permeebiltt

~~"'Chod<edbv""""
C~on Chocked bV IIIspheleoet
CompIedon Chocked bV PeralliOi

Completion

Chocked by

Chocked by

Chocked by

Asphaltenes

Paraffins

Scales

Speclfte Symptoms

Specific Symptoms

Speelfte Symptoms

FIG.5 BUBBLE UP TECHNIQUE


STEP'1

(0)

-=

Completion

Completion

static Pressure
Decline

CompIedon Chocked D;"Sceles

WfllMCoRng

WeterFlnoefing

I
I

Speclftc Symptoms

FIG.6 DIAGNOSTIC HYPOTHESES HIERARCHY


STEPilf:2

Well Problems

Problem Solution

ProbiemEldSI*Q

Diagnostic
Expert
System

WPA
STEPM2

(c)

Chemicals
DataBase

Stimulation

Expert

Design

System

Exp.Sys.

Stimulation

Risk Analysis
Exp.Sys.

P.....

'Ii,..

FIG.7 SYSTEM REASONING PROCESS

FIG.8 PROGRESS PROJECT ARCHITECTURE

Equlpments
DataBase

You might also like