Vous êtes sur la page 1sur 33

Use of Deconvolution for Designing

and Interpreting Well Test

Perapon Fakcharoenphol, PTTEP


Contents
ƒ Well test analysis
9 Principle
9 Convolution and deconvolution
ƒ Deconvolution
9 Brief development
9 Deconvolution in well test analysis
ƒ Benefits
9 Radius of investigation
9 Shape of derivative
ƒ Cautions
9 Initial pressure
9 Data selection
ƒ Practical uses
ƒ Conclusions
Principle of Welltest Analysis

Rate Recorded
Reservoir
Change Pressure

Input System Output


Convolution
Rate Equivalent Rate
q q1
q1

t3
q2

= q2-q1
t1 t3
q3=0
t2 q3-q2
0 t1 t3
t2 t3

Recorded Pressure P P(q1,t)

P
P P(q2-q1,t)

Convolved Pressure
P P(q3-q2,t)

0
t1 t2 t3
t1 t2 t3
Convolution and Deconvolution

Duhamel’s principle, pressure drop is the convolution of flow


p D' as function of time
rate q and the reservoir impulse response PD
T
∆p (t ) = pi − pw(t ) = ∫ q (τ ) pD' (t − τ )dτ
0
pD is a reservoir response function with constant rate assumption

Solving the above equation for PDpD function is amount to invert


the convolution integral, this method is called “Deconvolution”

Q: Why solve for PD ?


A: We have sets of known solutions to constant-rate assumption (shapes
of derivative in model catalogue)
Deconvolution
Rate Constant Rate
q
q1

q2

q3=0
0 t1 t2 t3 0 t1 t2 t3

Recorded Pressure

P Deconvolution P

Convolved Pressure
Deconvolved Pressure
0 0
t1 t2 t3 t1 t2 t3
Brief Development

• The concept of Deconvolution has been existed for more


than 40 years

• Many attempts has failed to cope with the typical range of


measurement errors, both in rate and in pressure. For
instance, Baygun et al. ‘s deconvolution works well incase
that errors are lower 5% in pressure and 1% in rate

• Practical algorithm was presented in 2001 by Thomas von


Schroeter et al. The authors claim the new algorithm can
tolerate error up to 0.5 % in pressure and 10% in rate, SPE
71574

• It was implemented in a commercial WTA software in 2006


Deconvolution in Welltest Analysis
Pressure and
Input Rate Data

Conventional Advance
Estimate Derivative
Estimate
Using Derivative using
“Convolved Pressure”
“Deconvolved Pressure”

Geological Knowledge Pattern Recognition


Identify Model Process, Compare derivative
& Other Information
to model Catalogue

Verify model using


Match Model
a simulation mode

Output Model Parameters


Benefits
Radius of Investigation (ROI)
Rate Equivalent Rate
q Top View of a Reservoir
q1 Fault
q1
q2
Tested
= t3 Well
q3=0
0 t1 t2 t3 q2-q1
t1 t3
p Recorded Pressure

q3-q2
t2 t3
Convolved Pressure Deconvolved Pressure
0 t1 t2 t3
(Conventional) (Advance)
Convolved Pressure Deconvolved Pressure
Give small ROI Make use all information,
PD, P’D
PD, P’D
ROI = f (total history time)

f(t3-t2) f(∆t) t3-t2 t2-t1 t3-t1 ∆t


t2 t3
Case Study 1: Well Test Design

1750

Data used in
deconvolution
3 hrs PBU
1550

24 Hrs Flowing 26 Hrs Shut-in


1350

10000

5000

0 10 20 30 40 50

H is to ry p lo t (P re s s ure [p s ia ], G a s R a te [M s c f /D ] v s T im e [hr])
Case Study 1: Well Test Design

3 hrs 26 hrs

1E+8

1E+7
Convolved derivatives

1E+6
1E-3 0.01 0.1 1 10

Lo g-Lo g plo t: dm(p) and dm(p)' [psi2/cp] vs dt [hr]


Case Study 1: Well Test Design
Well B Deconvolution.KS3 - Deconvolution
Well B.KS3 - Full Data (ref) Convolved derivative (26-hr PBU data)
26 hrs
Deconvolved derivative (3-hr PBU data)
1E+8

1E+7

Effect of superposition
1E+6
time function
1E-5 1E-4 1E-3 0.01 0.1 1 10

Compare files: dm(p) and dm(p)' [psi2/cp] vs dt [hr]


Case Study 1: Well Test Design
Well B Deconvolution.KS3 - Deconvolution
Well B.KS3 - Full Data (ref)
Convolved derivative (26-hr PBU data)
26 hrs
Deconvolved derivative (3-hr PBU data)
1E+8

PBU of a multi-rate test can be


1E+7 significantly reduced using the
deconvolution technique
1E+6
1E-5 These two responses can be perfectly matched
1E-4 1E-3 0.01 0.1 1 10

with Compare
the identical model
files: dm(p) and andvsparameters
dm(p)' [psi2/cp] dt [hr]
Shape of Derivative
1.E+05 Convolved Deconvolved
Pressure Pressure

Rate History 1
Case 1 Case 2

Rate History 2
1.E+04
Case 3 Case 4
Convolved Pressure Deconvolved Pressure
(Conventional) (Advance)

Rate history affects shape of Case 2&4 of rate history


No effect
1.E+03
derivative
Case 1

Case 3

1.E+02
1.E-02 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04
Case Study 2: Well Test Analysis

2140

2100

Analyzed Data
2060

2020

5000

0 1 2 3 4 5 6 7 8 9 10 11

H is to ry p lo t (P re s s u re [p s ia ], G a s R a te [M s c f /D ] v s T im e [h r])
Case Study 2: Well Test Analysis
Constant Pressure
Radial Comp
Close

1E+7

Convolved derivatives

1E+6

1) Closed Boundary
2) Constant Pressure
3) Radial Composite
1E+5
1E-4 1E-3 0.01 0.1 1
Case Study 2: Well Test Analysis

Convolved derivative
Arthit14-10xBU-6 DEC.ks3 - Model identification
Arthit14-10xBU-6 DEC.ks3 - Deconvolution Pi=2137 (ref)

Deconvolved derivative

1E+7 A unit-slope line


Deconvolution reduces
indicatesambiguity
the closed in model
identification process
boundary(in some cases)
model

1E+6

1) Closed Boundary Distortion due to


superposition time
2) Constant Pressure function effect
3) Radial Composite
1E+5
1E-4 1E-3 0.01 0.1 1 10

Compare files: dm(p) and dm(p)' [psi2/cp] vs dt [hr]


Cautions
Cautions
1. Deconvolution of single period pressure data is
sensitive to initial reservoir pressure

Ref: Dynamic Flow Analysis, Kappa Engineer


Cautions
1. Deconvolution of single period pressure data is
sensitive to initial reservoir pressure

Ref: Dynamic Flow Analysis, Kappa Engineer


Cautions
2. Poor data selection for deconvolution will yield erroneous
results
2.1 Deconvolution of multi periods containing different parameter
value, such as rate dependent skin in different flow period

Deconvolve all data together


47 0 0

42 0 0

37 0 0

32 0 0

30 0 00

20 0 00

10 0 00

1E+10 0

0 40 80 12 0 160 200 2 40

H is to ry p lo t (P re s s ur e [p s ia ], G a s R a te [M s c f /D ] v s T im e [h r])

1E+9

Convolved derivative
sk in-2.k s3 - Analysis 1
sk in-2.k s3 - Analysis 2 (ref)

1E+8
Deconvolved derivative
1E-3 0.01 0.1 1 10 100

Co mpare files: dm(p) and dm(p)' no rmalized [psi2/cp] vs dt


Cautions
2. Poor data selection for deconvolution will yield erroneous
result
2.2 Deconvolve periods containing non-reservoir response

2140

2100

Well A.ks3 - Model identification Deconvolve all data together


2060
Well A.ks3 - Deconvolution Pi=2129 (ref)

2020

5000

0 1 2 3 4 5 6 7 8 9 10 11

1E+9 H i s to r y p lo t ( P r e s s u r e [ p s ia ] , G a s R a te [M s c f /D ] v s T i m e [h r ] )

1E+8

Convolved derivative
Deconvolved derivative
1E+7
1E-4 1E-3 0.01 0.1 1 10

C fil d ( ) d d ( )' li d[ i2/ ] dt


Benefits and Cautions
Benefits Cautions
ƒReduce shut-in period for well ƒAccurate initial reservoir pressure
test operation. This offers is required to deconvolve single
significant cost reduction (23 hrs PBU data
in the case study)
•Poor data selection results
ƒReduce ambiguity in model distortion in deconvolved derivative
identification process (in some
cases)

ƒIncreaseinformation extracting
from measured pressure
Practical Uses

ƒ The deconvolution method does not replace


conventional (convolved pressure) method, but it adds
on information to the conventional analysis

ƒ Evaluation of time and rate dependent parameters such as skin


should be performed by the conventional method
ƒ Deconvolution is appropriate for analyzing of reservoir and
boundary responses because information gained from
deconvolution is typically in reservoir or boundary responses

ƒ Cross-check; use conventional analysis to cross-check


deconvolution result. “The identical model should be
able to match both convolved and deconvolved
derivatives”
Conclusions

ƒ Deconvolution is successfully used to assist well test


interpretation and design

ƒ It extracts more information than that obtained from the


conventional method

ƒ The duration of build up period in a multi-rate test can be


minimized using deconvolution technique. This offers a
considerable cost reduction.

ƒ It can be utilized as a model verification to eliminate


ambiguity of derivative shape due to superposition time
effect
Q&A
Superposition vs. Deconvolution
Superposition Deconvolution
(Conventional)

Radius of Underestimates ROI Make use all information,


(ignore information from ROI = f (total time history)
investigation (ROI)
prior FPs)

Shape of Distorts characteristic Only shape of the 1st


Derivative features. Many features DD
look different in BU/DD
data

FP=flow period, DD=Drawdown, BU=Build-up


Case Study: SPE93988

Final BU No final BU

w/o final BU
WBS

Slope =1
Deconvolution Development

Authors Year Main Feature Major Failure


Coats & al 1964 -Time domain Works well for data without noise
- linear programming but that fails in the present of
even small amounts of noise
Kuchuk & al 1990 -Time domain Reasonable results for error up
- Lease squares to 2% in pressure,

Bougeois & 1993 -Laplace space Inherit error of Laplace


Horne - Direct method, linear transforms taken over finite
programming interval

Baygun & al 1997 -Time domain Reasonable results for error up


- non-linear lease squares to 5%, give un-interpretable
interpolation (pressure and result if error in rate exceed 1%
curvature)
New Algorithm

1. Deconvolve for derivative rather than pressure


dpD
= t. pD' avoid sign constrain in the
d ln t algorithm

2. New error object function

E= ε +ν δ + λκ (z )
2 2 2
2 2

δ
introduce rate error in the error
is error of rate match,
ε
2

2 is error of pressure match, and objective function


κ (z ) is measure of the curvature
ν is deconvolution weight for rate match
λ is deconvolution weight for the roughness penalty
Deconvolution
Input Output

Rate history Deconvolved


derivative and
pressure

Selected pressure i.e


PBU(s), PBU +DD
1.E-02
Conventional Derivative
#(1-6)[6]{1.43229E+03}5781.00
#(1-6)[6]{1.43229E+01}5781.00
#(1-6)[6]{1.43229E+02}5781.00
No Data
No Data Coverage
Coverage
#(1-6)[6]{1.43229E+04}5781.00
Deconvolved Derivative

#(1-6)[6]{1.43229E+05}5781.00
#(1-6)[6]{1.43229E+06}5781.00
#(1-6)[6]{6.50937E+03}5781.00
1.E-03

Core Calibrated Wireline


Stabilisation
1.E-04
1.E-03 1.E-02 1.E-01 Elapsed Time hrs 1.E+00 1.E+01 1.E+02
5750 100000
Giv en Rate

nls d#( 1-6)[6]{ 1.43229E+03} .c pInitP=5781. 16 nodes


5700 90000 nls d#( 1-6)[6]{ 1.43229E+01} .c pInitP=5781 16 nodes
Pressure Data and Pressure Match

nls d#( 1-6)[6]{ 1.43229E+02} .c pInitP=5781, 16 nodes


5650 80000 nls d#( 1-6)[6]{ 1.43229E+04} .c pInitP=5781, 16 nodes

d at a nls d#( 1-6)[6]{ 1.43229E+05} .c pInitP=5781, 16 nodes

5600 nls d#( 1-6)[6]{ 1.43229E+06} .c pInitP=5781, 16 nodes


nls d # ( 1- 6 ) [ 6 ] { 1.4 3 2 2 9 E+0 3 } .c p Init P=578 1. 16 no d es Rate (Mscfd) 70000
nls d#( 1-6)[6]{ 5.78809E+03} .c pInitP=5786,16 nodes
nls d # ( 1- 6 ) [ 6 ] { 1.4 3 2 2 9 E+0 1} .cp Init P=578 1 16 no d es nls d#( 1-6)[6]{ 6.50937E+03} .c pInitP=578, 31 nodes
5550 60000
nls d # ( 1- 6 ) [ 6 ] { 1.4 3 2 2 9 E+0 2 } .c p Init P=578 1, 16 no d es

5500 nls d # ( 1- 6 ) [ 6 ] { 1.4 3 2 2 9 E+0 4 } .c p Init P=578 1, 16 no d es 50000


nls d # ( 1- 6 ) [ 6 ] { 1.4 3 2 2 9 E+0 5} .cp Init P=578 1, 16 no d es
5450 40000
nls d # ( 1- 6 ) [ 6 ] { 6 .50 9 3 7E+0 3 } .c p Init P=578 , 3 1 no d es

5400 30000

5350 20000
23.5 24 24.5 25 25.5 0 5 10 15 20 25 30
T im e (h r) Tim e (hr)

Vous aimerez peut-être aussi