Sie sind auf Seite 1von 13

SPE 57452

Virtual Well Testing


S.S.Dakshindas, SPE, T.Ertekin, SPE, and A.S.Grader, SPE, The Pennsylvania State University, University Park, PA

Copyright 1999, Society of Petroleum Engineers Inc.


This paper was prepared for presentation at the 1999 SPE Eastern Regional Meeting held in
Charleston, West Virginia, 2122 October 1999.
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
Well test analysis represent an essential component of
reservoir engineering analysis and management. It provides
estimates of reserves, heterogeneties, permeabilities, and
information on the status of wellbore conditions. A well
testing program can become prohibitively costly due to
equipment and personnel costs and lost production. It may not
be possible to design an effective well testing program that
includes all the important wells in a specific field. In this
paper, a novel Virtual Well Testing approach is presented.
The new method uses artificial neural networks (ANNs) to
generate virtual transient pressure data. The proposed method
does not eliminate the need for traditional well tests, but
provides a basis to extract more information from the existing
sets of transient pressure data. In virtual well testing, transient
pressure data from selected wells in the field are used to train
a neural network that is capable of predicting transient
pressure responses at other well locations where well tests
have not been conducted. The proposed methodology can also
be used for designing well tests and for completing the
partially recorded pressure measurements due to malfunction
of testing equipment.
This paper provides guidelines for selecting actual well
test locations, for choosing input parameter neurons and for
training the ANN. Several simulated case studies are used to
train and evaluate virtual well testing capabilities of the ANN.
The effects of the number of wells tested, well locations and
production histories are evaluated. The ANN is successful in
predicting interference effects and reservoir permeabilities.
The ANN generates reliable virtual transient pressure data in a
matter of seconds and is an effective reservoir engineering
analysis and management tool.

Introduction
Well testing is an important engineering tool in the petroleum
industry.
However, well testing can be prohibitively
expensive as it may demand a large amount of time,
equipment, and manpower. The work presented in this paper
deals with the application of neural networks to well testing.
The main objective of this work is to develop an artificial
neural network, which is capable of predicting transient
pressure responses without conducting an actual test. The
virtual pressure data generated by the neural network can be
used for the determination of reservoir characteristics with
traditional analysis procedures. The investigation presented in
this paper lays emphasis on providing guidelines for selecting
input to the network, and for selecting wells in the field for
conducting actual well tests. Parameters and well locations
are presented and the predictive capabilities of the network are
demonstrated.
The artificial neural network
A neural network is a non-algorithmic, non-digital, and
intensely parallel information processing system. A
backpropagation network (BPN)1 is used in this paper. A
backpropagation network is multilayered and information
flows from the input to the output through at least one
hidden/middle layer. Each layer contains neurons that are
connected to all neurons in the neighboring layers. The
connections have numerical values (weights) associated with
them. During the training phase, the weights are adjusted
according to the generalized delta rule1. Training is completed
when the network is able to predict the given output.
A simple BPN with one middle layer is shown in Fig. 1.
In a BPN, the input activity is transmitted forward while the
error is propagated backwards. The neurons in the BPN use a
transfer function that is sigmoid or S shaped. A key feature of
the sigmoid function is that it has a minimum value of 0 and a
maximum value of 1 and is differentiable everywhere with a
positive slope. The derivative of the transfer function is
required to calculate the error that is backpropagated and the
derivative of the sigmoid function is easy to calculate. The
sigmoid transfer function1 is given by
1
f (I ) =
.(1)
1 + e I
where I is the input to the neurons.

S.S. DAKSHINDAS, T.ERTEKIN, A.S.GRADER

The learning constant is another parameter used in the


generalized delta rule. It determines the rate at which the
network learns and is always between 0 and 1, where 0
indicates that the network does not learn at all while a value of 1
indicates a rapid learning. A value of 0.2-0.3 is a reasonable
learning constant, but is case specific. A variable learning
constant may be also applied where an initial value of 0.3 is
exponentially reduced during training2. Momentum1 is used in
addition to the learning constant to increase the learning rate
of the network. It accelerates the learning process and
prevents the network from converging to a local minima.
The pressure data generator
Analytical and numerical simulators were used to generate the
pressure data for training the ANN. In the analytical
simulator, the infinite-acting line source solution and the
principle of superposition3,4, were used to generate pressuretime responses. For wells with different flow rates and
different flowing times, simultaneous superposition in space
and time was used. For heterogeneous cases, the pressuretime data were generated using a numerical simulator.
Case studies
Various case studies in which the ANN was trained with
different input parameters are presented. Two classes of wells
are identified in this work: training and prediction wells. In a
training well, an actual well test was conducted and pressure
transient data recorded. These data are then subsequently used
in training the ANN. In a prediction well, actual pressure data
were not recorded and the ANN is used to generate the
pressure-time response.
Case I: Homogeneous, six wells, identical flow rates,
shut-in times and production times
An infinitely large field (Fig. 2) with homogeneous property
distribution and six active producing wells is considered. All
wells have exactly the same production history and are shut in
at the same time.
The network is trained with the pressure-time data
generated using an analytical simulator. There are five
training wells and one prediction well (pressure data are
available from wells 1, 2, 3, 5, 6). The ANN is trained using
the data from the five wells and the pressure response for well
4 is predicted.
The well configuration is axi-symmetrical yielding three
pairs of wells with identical responses (wells 1 and 5, wells 2
and 6, and wells 3 and 4). This symmetry must be recognized
by the ANN. The structure of the ANN for this case study is
shown in Fig. 3. The input parameters of the ANN are the x
and y coordinates of each well, inter-well, dimensionless time,
and their functional links. When well 1 is introduced to the
ANN, the input comprises of the x and y coordinates of well 1,
distances of well 1 from all other wells, a functional link of the
distance, dimensionless time tD1 at which pressure is known,
and functional links of dimensionless time (Fig.3). When this
input is introduced to the ANN the output generated is the

SPE 57452

dimensionless pressure at well 1 at time tD1. This is done for


all times at all of the training well locations.
Significant interference effects start at tD=107 and all the
wells are shut in at tD=108. The ANN is designed to predict
pressure responses up to tD=1010 (two log cycles after shut in).
Two separate networks are trained to predict pressure
responses, before and after the shut in. Results from both
ANNs are combined and a complete pressure profile is
generated. The reason for using two networks is because a
BPN can be used only if the function is monotonically
increasing or decreasing. The network used in training for
after the shut-in has 14 input layer neurons, 7 middle layer
neurons, 1 output layer neuron, a learning constant of 0.3 and
a momentum of 0.7. The network used in training for before
the shut-in has a similar structure with the exception of having
10 neurons in the middle layers and uses the same training
parameters.
A comparison between the ANN and the simulation
responses for training and prediction wells are shown in Figs.
4a and 4b, respectively. The match of the training well
response is satisfactory (Fig. 4a). However, the drawdown
match of the prediction well (Fig. 4b) can be improved. This
improvement was obtained by removing the x-y coordinates
and by introducing the inter-well distances in an order of
increasing magnitude. The order of the inter-well distances
assists the network to rank interference effects. The improved
results for training and prediction wells are shown in Figs. 5a
and 5b, respectively.
The results of the first case study infer that the input is
important not only in content but also in the manner in which
it is introduced to the ANN. In designing the input structure
one needs to ensure that the selected input entries contain
information that is useful to the ANN. The x and y
coordinates of the wells did not convey any vital information
to the ANN and the output was not adversely affected when
they were removed. The presence of unnecessary information
can contribute to a prolonged training time.
Case II: Homogeneous, 20 wells, different flow rates,
identical shut-in times and production times
A schematic diagram of the field is shown in Fig. 6. There are
11 training wells (1,3,4,5,6,7,8,11,13,15 and 18) and nine
prediction wells (2,9,10,12,14,16,17,19 and 20). The field is
infinitely large and homogeneous and its properties are
summarized in Table 1. The wells are put on production at
the same time and are shut in 100 hours later. The flow rates
of the wells are given in Table 2. In this study, the
permeability value is to be determined from ANN predictions
and the dimensional pressure and time are used as network
parameters.
Pressure transient data are generated by the analytical
model and are used to train the ANN. In the previous case,
pressure was generated for periods before and after shut-in.
From this case study onwards only shut-in pressure data are
considered by a single BPN.
Input parameters are distances between a well and its
surrounding wells and functional links, flow-rate of the well

SPE 57452

VIRTUAL WELL TESTING

and functional links, time and functional links and Horner


time, (tp+t)/t, and functional links. The definition of
distance between wells is modified to incorporate the flowrates of the surrounding wells. A new distance related
parameter is calculated using the following equation.
dist(well adjwell)
....(2)
Dmod = q(well)
q(adjwell)
In Equation (2), Dmod is the modified distance based on flowrates, q(well) is the flow-rate of the well under consideration,
q(adjwell) is the flow-rate of the adjacent well and dist(well-adjwell) is
the distance between the two wells.
The flow-rates and Horner times are included in the input
set. Inclusion of flow-rates in the input list is necessary as it
changes from one well to another. The Horner time becomes
useful if the generated pressure data are used for permeability
determination. The output from the ANN is the pressure. The
network has 25 input layer neurons, 33 middle layer neurons,
1 output layer neuron, a learning constant of 0.1, and a
momentum of 0.8.
The ANN was trained using the pressure transient data
from 11 wells and was used to predict the pressure transients
for the remaining 9 wells. The results for training well 1 and
prediction well 2 are shown in Figs. 7a and 7b, respectively.
The pressure data were later analyzed for permeability
estimations using the Horner method. Fig. 8 shows the
estimated permeability values at nine prediction well locations
where virtual buildup tests are conducted.
The ANN output and the simulator output show a good
match indicating that the ANN was capable of generating the
pressure transient data quite accurately.
Although the
prediction results shows a slight deviation from the simulator
output, the deviation is not significant considering the
increased complexity of the problem. The virtual data closely
follows the simulator pressure profile. This indicates that the
nature of the problem is well understood by the ANN. The
reservoir permeabilities calculated from the virtual pressure
data are in reasonable agreement with the original
permeability, within an error range of 8-18%. The majority of
the prediction wells (8) have less than 15% error in the
estimated permeability (Table 3). Inclusion of the flow-rates
and the new definition of the distances enabled the ANN to
understand the intricacies of the interference effects and
generate high quality pressure transient data.
Case III: Homogeneous, 20 wells, different flow rates
and shut-in times, identical production times
The field studied in this case is the same as in Case II with
each well having the same designations (training or
prediction) and flow-rates as in Case II. The new complexity
in this case is that the wells are shut in at different times
(Table 4).
Pressure data generated using the analytical solver are
used to train the ANN. The input parameters are the distances
between wells (calculated using Equation 2), flow-rates and
functional links, Horner time and functional links, shut-in
times, and the constant group, (qB)/h. The network now has

28 input layer neurons, 39 middle layer neurons, 1 output


layer neuron, a learning constant of 0.05, and a momentum of
0.9.
In some well test operations it is possible that pressure
transient data are incomplete due to malfunction of downhole
equipment. The ANN was trained with incomplete pressure
data at some of the training wells and then used to predict the
missing pressure data. Training well 1, which was shut-in after
100 hours, had a malfunction after 30 hours and could not
record pressure data for the last 20 hours of the test.
The results for training well 1 (with incomplete pressure
data) and prediction well 10 are shown in Figs. 9a and 9b,
respectively. The virtual pressure data generated using the
ANN and the simulator generated pressure data are in good
agreement. The analysis of the pressure data for permeability
yielded values which are in good agreement with the original
permeability of 70 md (Fig.10). The error range for the
permeability values is between 5 and 20%. Table 5 shows
that 6 of the 9 prediction wells have an error of less than 10%
in the estimated permeability values.
Case IV: Homogeneous, 20 wells, different flow
rates, shut-in times and production times
The same field as in Cases II and III is still under
consideration. In this case, the wells are put on production
and shut in at different times and have different flow-rates
(Table 6). Hence, most of the simplifying assumptions
imposed on the previous cases are removed.
The input parameters includes the distances between
wells, flow-rates of the wells and relevant functional links,
Horner time and related functional links, the parameter
(qB)/h and functional links, shut-in times of wells and
functional links, and the times at which the wells are put on
production. The network now has 32 input layer neurons, 47
middle layer neurons, 1 output layer neuron, a learning
constant of 0.05 and a momentum of 0.9.
The results for this case for well 1 (training well) and well
2 (prediction well) are shown in Figs. 11a and 11b,
respectively.
The pressure data were analyzed for
permeability displayed in Fig. 12. The simulator results and
the ANN results show a good agreement and so do the
estimated permeabilities. The inclusion of the times at which
the wells are put on production makes the input set more
complete. The analysis of the virtual pressure data yielded
permeability values, which were 5-30% in error of the original
values. Table 7 shows that 7 of the 9 prediction wells have an
error less than 15% in the estimated permeability.
The input set developed for this case study generated the
most comprehensive ANN structure. The input neurons
consist of the following parameters and their functional links:
Inter-well distances (function of flow-rates)
Flow-rates
Shut-in times
Time when the well is put on production
Horner time
Parameter (qB)/h.

S.S. DAKSHINDAS, T.ERTEKIN, A.S.GRADER

This generated an input structure which could be used for


simpler cases which can be considered as the subsets of the
case IV. To verify this postulation the same ANN structure
was tested on some of the previous cases. The network
architecture was maintained by fixing the network parameters.
Case III had a field with wells having different flow rates,
and shut-in times but with the same production time. The
network was trained with the generalized input set and
network architecture. The results show a good match between
simulator output and training output. The plots for training
well 1 and prediction well 2 are shown in Figs. 13a and 13b,
respectively. Case II had a field with wells having different
flow-rates, but the same shut-in times and flowing times. The
results of Case II also show good agreement with the
simulator output. The plots for training well 1 and prediction
well 2 are shown in Figs. 14a and 14b, respectively. A field
similar to Case IV, but with wells having same flow-rates,
shut-in times and flowing times, was considered. The results
are in good agreement with the simulator output. The plots for
training well 1 and prediction well 2 are shown in Figs. 15a
and 15b, respectively.
The developed network structure can be used for a variety
of cases from simple ones to more complex ones with a good
degree of accuracy. This provides an independent check on
the predictive capability of the ANN since a simple case is a
subset of a complex (more general) case.
Case V: Heterogeneous, 30 wells, identical flow
rates, shut-in times and production times
In this case a field of heterogeneous permeability with 30
wells is studied. The wells have the same flow rates, shut-in
times and flowing times, but they are located in different
permeability regions. Two maps of permeability distribution
and well locations are shown in Fig. 16. The light shades
indicate high permeability regions and the dark shades mark
the low permeability regions. The range of permeabilities is
between 20 md and 115 md. Table 8 summarizes the field
properties.
The reservoir is finite and surrounded by no flow
boundaries. A reservoir simulator generated the pressure
transient data for this field. The wells are put on production at
the same time and are produced for 400 days. After that, they
are shut-in for 24 hours and then brought back on production.
Pressure transient data are available from 15 wells (Fig. 16).
The input to the ANN includes the distance between
surrounding wells and its functional links, distance to the
nearest boundary and its functional links, and Horner time and
its functional links. In the previous cases, the distance from
the boundaries did not matter as the field was considered to be
infinitely large. In this case, the distance between the well and
the boundary is included due to the significance of boundary
effects.
The results for training well 6 and prediction well 8 are
shown in Figs. 17a and 17b, respectively. The pressure data
are analyzed for permeability and the estimated permeability
values are compared against the permeability values input to
the simulator (Fig. 18).

SPE 57452

The pressure transient data for training wells are in good


agreement with the simulator, indicating that the training was
successful. The predicted pressure values show a deviation
from the simulator output. This deviation is of the order of 3
to 4 psig, but overall trend of the predicted pressure response
is similar to the simulator response.
The predicted
permeability values are in reasonable agreement with the
actual values (except for a couple of well locations, with a
maximum deviation of 42% at well 30). Table 9 summarizes
how the error is distributed in the estimated permeability
values of the prediction wells.
Guidelines for the selection of training wells
An ANN is a universal, non-linear interpolator and hence it
cannot extrapolate outside the range it is trained with. It is
therefore necessary that the training wells cover the range of
possible locations in the field. The following factors must be
considered when selecting training wells.
Interference effects: In Case I, the field was homogeneous,
infinite acting, with wells having the same flow rate and
production schedule. In this case, the effects of interference
between the wells were studied since all other parameters were
held uniform. Inter-well distances are most important for
interference and they are part of the network input. A full
range of inter-well distances were used. Arranging these
distances in the ascending order improved the ANN structure
and yielded good training and prediction capabilities.
Flow rate effects: The next important factor of the input set is
the flow-rate. In Case II, the flow-rates of the wells were
different. A complete range of flow-rates were used in the
input set. The definition of inter-well distances was modified
to incorporate the effects of the flow-rates of surrounding
wells. Hence, the ANN was informed about the coupled
effects of inter-well distances and flow-rates.
Effects of shut-in time: In Case III the wells were shut-in at
different times. The shut-in time of each well was included as
a parameter in the input set. Different shut-in times were
assigned to the training wells so that the prediction wells were
within the range of these shut-in times. Training well 5 was
assigned a maximum shut-in time of 135 hours and training
well 1 was assigned a minimum shut-in time of 100 hours. All
other shut-in times are within this range. This alignment of
shut-in times is crucial for obtaining good prediction results
from the ANN.
Effects of production time: The next parameter that was
studied was the time at which the wells are put on production.
In Case IV, wells were put on production at different times.
The training wells were assigned flowing time values which
determined the range. Training well 1 was put on production
at 2 hours which is the earliest, whereas training well 13 was
put on production at 80 hours making it the last well to be put
online. Other training and prediction wells are within this
production time range. Training well 11 was put on

SPE 57452

VIRTUAL WELL TESTING

production at 2 hours and shut-in at 130 hours making it the


well which had the largest flow time prior to shut-in, similarly
the training well 15 was put on production at 70 hours and
shut-in at 100 hours making it the well with the shortest
production interval prior to shut-in. These two wells provide
the range of the interpolation bracket for the ANN.
Boundaries and heterogeneity effects: In Case V, a
heterogeneous field with no flow external boundaries is
studied. In this case the effects of the boundaries and the
various permeability regions need to be captured by the
behavior of the training wells. Wells from all permeability
regions are used in training. Training well 12 is in the high
permeability region and training well 1 is in a low
permeability region (Fig.16). Similarly, other training wells
are distributed in various permeability regions, all transferring
vital information to the ANN. The effects of boundaries on
pressure responses are significant and need to be incorporated
in the input set. Training well 10 is located close to the
boundary and training well 28 is positioned away from the
boundary (Fig.16). Wells that are close to each other are also
included in training to ensure the incorporation of interference
effects.
In summary, the following guidelines are recommended
for selecting wells for conducting actual well tests.
Select well locations that span the inter-well distance
range
Include wells with a wide range of flow rates
Include wells with a wide range of production periods
(different shut-in times and different starting times)
Include wells located in different regions of the reservoir,
near and way from boundaries.
The number of training wells is case specific. In this
study, the ratio of training wells to prediction wells is kept
around 1. The number of training wells or the ratio of training
wells to prediction wells cannot be predetermined, as it is
controlled by the complexity of the specific field. Although it
is desirable to reduce the number of training wells to a
minimum, such a strategy will eventually increase the risk of
jeopardizing the quality of predictions.
Conclusions
This study introduces a novel virtual well testing methodology
and provides guidelines for selecting locations for the actual
well tests. All the cases studied were instrumental in
identifying the major parameters necessary for developing a
well-trained network with good prediction capabilities. The
ANN must be exposed to a wide variety of scenarios, such that
physical responses of the reservoir are conveyed to the ANN.
The variety of the scenarios also determines the working range
for the ANN. After such a working range is set for the ANN
during training, it is important to stay within this range during
the prediction phase to ensure reliable results.
The results of the case studies presented in this paper
indicate that the ANN provided reasonable prediction

capabilities. Although the virtual pressure data were analyzed


and permeability values estimated, it should be emphasized
that the main objective of this work was to generate virtual
pressure transient data and not to estimate the permeability
values.
The proposed virtual well testing methodology has other
applications as well. It can be used to complete pressure data
that are missing due to malfunction of downhole equipment.
The ANN can also be used for designing more informed well
tests. Virtual well testing provides a cost effective solution to
a variety of problems faced in the operations of a field and has
the potential of becoming a valuable tool for the petroleum
industry. Virtual well testing is a novel, fast, reliable and
economic way of generating simulated well test data. It does
not and cannot eliminate the need for traditional well tests, but
provides a solid basis to extract more information from the
existing sets of pressure transient data.
Acknowledgements
The authors wish to thank the Consortium for Virtual
Operations Research (CVOR) housed within the Pennsylvania
State University and West Virginia University for its support
of the work presented in this paper. The authors also wish to
thank Dr. Mirna Urquidi MacDonald of the Engineering
Science Department at Penn State for her help in this project.
References
1.
Caudill, M. and Butler, C., Understanding Neural
Networks, Part I and II, MIT Press, Cambridge, USA,
1994.
2.
H. Doraisamy, T. Ertekin, A.S.Grader, Key Parameters
Controlling the Performance of Neuro-Simulation
Applications in Field Development, Proceedings
Eastern Regional Meeting, Pittsburgh, Pennsylvania,
SPE Paper 51079, 1998.
3.
Lee, John, Well Testing, Society of Petroleum
Engineers of AIME, New York, USA, 1982.
4.
ERCB, Alberta, Canada, Gas Well Testing, Theory and
Practice, fourth edition, 1979.
Nomenclature
I
q(well)
q(adj-well)
tD

h
B
tp
t
Dmod
Dist(well-adjwell)

=
=
=
=
=
=
=
=
=
=
=

Network input
Flow-rate of the well, bpd
Flow-rate of the adjacent well, bpd
Dimensionless time
Oil viscosity, cp
Formation thickness, ft
Formation volume factor, RB/STB
Producing time, hours
Time after shut-in, hours
Modified distance
Distance between well and its adjacent
well

S.S. DAKSHINDAS, T.ERTEKIN, A.S.GRADER

Table 1: Reservoir properties for Case II


Permeability
70 (md)
Porosity
20%
Oil viscosity
1.3 (cp)
Oil compressibility
5x10-6
Initial reservoir pressure
3000 (psig)
Formation volume factor
1.2
Formation thickness
25 (ft)
Wellbore radius
0.25 (ft)
Table 2: Flow rates for Case II
Flow rate
Well
Flow rate
Well
(BPD)
(BPD)
1 (t)
400
11 (t)
500
2 (p)
500
12 (p)
400
3 (t)
100
13 (t)
100
4 (t)
250
14 (p)
300
5 (t)
150
15 (t)
500
6 (t)
700
16 (p)
400
7 (t)
400
17 (p)
600
8 (t)
200
18 (t)
100
9 (p)
600
19 (p)
150
10 (p)
300
20 (p)
600
t: training well
p: prediction well
Table 3: Permeability error distribution for Case II
Error %
Number of wells
within the error range
0-10
2
10-15
6
15+
1
Table 4: Shut-in times of wells for Case III
Well
Shut-in times
Well
Shut-in times
(hrs)
(hrs)
1 (t)
100
11 (t)
130
2 (p)
110
12 (p)
120
3 (t)
120
13 (t)
110
4 (t)
130
14 (p)
110
5 (t)
135
15 (t)
100
6 (t)
100
16 (p)
110
7 (t)
100
17 (p)
130
8 (t)
110
18 (t)
135
9 (p)
110
19 (p)
130
10 (p)
130
20 (p)
100
t: training well
p: prediction well
Table 5: Permeability error distribution for Case III
Error %
Number of wells
within the error range
0-10
6
10-15
1
15+
2

Well
1 (t)
2 (p)
3 (t)
4 (t)
5 (t)
6 (t)
7 (t)
8 (t)
9 (p)
10 (p)
11 (t)
12 (p)
13 (t)
14 (p)
15 (t)
16 (p)
17 (p)
18 (t)
19 (p)
20 (p)

SPE 57452

Table 6: Well parameters for Case IV


Flow rate
Time when put
Shut-in time
(BPD)
on production (hrs)
(hrs)
400
2
100
500
40
110
100
66
120
250
32
130
150
26
135
700
9
100
400
10
100
200
50
110
600
30
110
300
20
130
500
2
130
400
60
120
100
80
110
300
65
110
500
70
100
400
9
110
600
15
130
100
35
135
150
50
130
600
20
100
t: training well
p: prediction well

Table 7: Permeability error distribution for Case IV


Error %
Number of wells
within the error range
0-5
1
5-10
1
10-15
5
15+
2
Table 8: Field properties for Case V
Permeability
15 to 120 (md)
Porosity
20%
Reservoir grid dimensions
31 x 24
Grid block
1000 (ft) x 1000 (ft)
dimensions
x 50 (ft)
Well flow-rates
100 (bpd)
Formation depth
5000 (ft)
Oil viscosity
1 (cp)
Oil density
42 (lb/cu.ft.)
Initial reservoir
5500 (psig)
pressure
Shut-in time
400 (days)
Table 9: Permeability error distribution for Case V
Error %
Number of wells
within this error
0-10
5
10-20
5
20-30
3
30+
2

Das könnte Ihnen auch gefallen