Sie sind auf Seite 1von 30

A Better Way to Forecast Production in

Unconventional Gas Reservoirs


John Lee
Texas A&M University
7 October 2010
We Have a Problem
Forecasting methods we use in
conventional reservoirs may not work
well in
Tight gas
Gas shales
Unconventional gas resources generally
What Can We Do About It?
Understand limitations of conventional
methods
Support efforts to improve
Understanding of basic physics controlling
stimulation outcomes, production mechanisms
Modeling methods based on correct physics
Reservoir characterization (model parameters)
Until verified theoretical models available, use
most appropriate empirical models (e.g.,
decline curves)
Decline Curves: Approaches
Major categories
Arps empirical model
As originally proposed
With terminal exponential decline imposed
With a priori terminal b value imposed
Recent empirical models
Valk Stretched-exponential model
Ilk et al. Augmented Stretched-
exponential model
10
100
1000
10000
0 100 200 300 400 500 600
Time, months
R
a
t
e
,

S
T
B
/
m
o
Exponential
Hyperbolic
Harmonic
Critique of Arps Model
Requires stabilized (not transient) flow for
validity
Transient flow likely
for most, possibly all,
life of well in ultra-low permeability reservoirs
Best-fit b values almost always >1 for recent gas
wells
Extrapolation to economic limit with high b value
leads to unrealistically large reserves estimates
Reserves as rate 0 (time ) for b 1
) / 1 (
) 1 (
1
b
i
i
t bD
q q
+
=
0
200
400
600
800
1000
1200
1400
1600
0 100 200 300 400 500 600
Time, months
C
u
m

P
r
o
d
,

M
S
T
B
Exponential
Hyperbolic
Harmonic
Arps: Keeping Reserves Estimates
Reasonable
Common method: Use best-fit b until
predetermined minimum decline rate reached;
then impose exponential decline
Problems
Any extrapolation with best-fit b unrealistic
apparent best b decreases continually with time
Appropriate minimum decline rate based on
observed long-term behavior in appropriate analogy
usually unavailable in resource plays
Leaves too many degrees of freedom, inevitably
leads to subjective judgment
Minimum Decline From Analogy?
qi 5000 STB/Day
ADR = 55%
qt 200 STB/Day
ADR = 15%
mADR = 10%
mADR = 7%
mADR = 5%
Courtesy Ryder Scott Company
Arps: Forcing Appropriate b
Set b to perceived proper terminal value,
based on theory or experience
Determine q
i
, D
i
by extrapolating trends from
history and fixing terminal b-value
Result sensitive to specific value of the
forced b
Result questionable for wells that never
stabilize
More validation required for ultra-low
permeabilities, horizontal wells with multi-
stage fractures
Terminal b Improves Forecast
(Cheng et al., SPE 108176)
100
1,000
10,000
100,000
0 50 100 150 200 250 300
Time, months
G
a
s

r
a
t
e
,

M
S
C
F
/
m
o
Actual data
b=0.6, new method, error=-2.97%
b=1, constraint b1, error=-47.18%
b=2.65, best fit, error=36.50%
Stretched-Exponential ( ) Decline Model
Empirical model
Model parameters
q
i
initial rate (e.g., mscf/month) taken as peak
rate, usually in second month of production
characteristic time (e.g., months)
n exponent (dimensionless)
Advantages
Conservative (finite EUR at zero rate, infinite time)
Easily applied straight-line plot to estimate reserves
(recovery potential plot)
(
(

\
|
=
n
i
t
q q

exp
Example Recovery Potential Plot: All US Gas Wells Completed in
2000-2004 Having At Least 5 Years Production History: 45,506
Wells (n : 0.36 and q
i
: 3.34 tcf/mo)

5.0 10
9
1.0 10
10
1.5 10
10
2.0 10
10
2.5 10
10
3.0 10
10
0.5
0.6
0.7
0.8
0.9
1.0
Q
r
p
40yr 1.14 10 1.14 10
Mean 40 yr EUR:
1.14 bcf/well
Defining differential equation of
the model
Rate expression as function of
time stretched exponential
Dimensionless rate expression (t
D
and q
D
)
Dimensionless cumulative
production expression
Dimensionless EUR expression
Recovery potential calculated
from dimensionless rate
t
q t
n
dt
dq
n
|

\
|
=

-
(
(

\
|
=
n
t
q t q

- exp ) (
0
( ) | |
n
D D
t q - exp =
( )
)
`

=
n
D D
t
n n
Q ,
1 1
n

=
n
EUR
D
1
n

=
D
q
n
n
rp ln ,
1
1
1

| |
inf
0
1
/
/
,
:
D D
t
D D D
i D
D
z
t a
Q EUR
dt q Q
q q q
t t
dt e t z a
where
D
=
=
=
=
=

EUR
Q
rp
where
=1
SPE 109625
Rushing-Blasingame Study: 42 Simulated Cases
Base Case Stretched-Exponential Model
(based on 5-yr production history)
Out[930]=

0 500 1000 1500


0
500
1000
1500
2000
days
Q
,
m
m
c
f
Stretched- exp model n:0.25, t :25.4 days qi:11.7 mmcf d

0 500 1000 1500


0.5
1.0
2.0
5.0
10.0
days
q
,
m
m
c
f

d
Stretched- exp model n:0.25, t :25.4 days qi:11.7 mmcf d
Base Case Arps Model
(based on 5-yr production history) Just as good a fit

0 500 1000 1500


0
500
1000
1500
2000
days
Q
,
m
m
c
f
Arps model b:1.5, D:0.007 1days qi:4.823 mmcf d

0 500 1000 1500


0.5
1.0
2.0
5.0
10.0
days
q
,
m
m
c
f

d
Arps model b:1.5, D:0.007 1days qi:4.823 mmcf d
Comparison: 50 yr Forecasts Based on 5-Yr Prod History

0 10 20 30 40 50
0
1000
2000
3000
4000
5000
6000
yrs
Q
,
m
m
c
f
Red: Arps b1.5, Blue: Stretchedexp n0.25
Conclusion:
While the limited span of data can be described
equally well with the traditional and the new model,
the extrapolation to 50 yrs yields different results
(the new model being more conservative and nearer
to the actual value known in this case.)
Forecasting Ability of SE Model Much Better
Years of
History
Matched
Best Fit,
Arps b
Arps: Error
in
Remaining
Reserves,
%
SE: Error in
Remaining
Reserves,
%
2 2.66 145 36.1
5 1.91 104 23.9
10 1.51 30.6 6.73
25 1.20 7.9 0.21
50 1.14 N/A N/A
Statistics for 42 Rushing-Blasingame Cases
50 yr forecast based on production history available for various years
Stretched exponential model with fixed n = 0.25
Based on yr 2 5 10 20 50
Mean abs error %
(Stand. abs err.
%)
11.3
(16.2)
6.0
(7.4)
5.6
(4.6)
3.1
(2.1)
0
(0.002)
Field/Reservoir/Formation Group Analysis:
The Data-Driven Approach
Is it better to try to match individual
wells accurately
Or
Match groups of wells in given area and
derive individual well performance
project from group-average parameters?
Some Problems with Individual Wells
Changes in technology during life of well
Restimulation
Reactions to changes in gas prices
Variations in field pressures
Available slots
But, for statistically valid sample
Changes may average out over lives of
individual wells
Example: Member of
Group (n=0.3)
Evidence Indicates Data-Driven Approach
Preferable
Applied to gas wells completed in 2000-
2004 and having at least 5 years
production history examples:
Barnett Shale
Carthage
Haynesville
All US
Barnett Shale
Carthage Field
Haynesville
SE Analysis of Groups
Group Barnett
Shale
Carthage Haynesville All US
wells in group 2,849 1,126 1,629 46,506
Mean current
cumulative
0.63 bcf 0.63 bcf 1.15 bcf 0.79 bcf
Model-par n=0.16
=0.019 mo
n=0.32
=3.71 mo
n=0.36
=2.6 mo
n=0.36
=3.7 mo
Mean 40yr
forecast
1.4 bcf 1.08 bcf 1.54 bcf 1.14 bcf
Conclusions
Forecasting in resource plays uncertain
Understanding of basic physics
incomplete
Ability to model hypothesized controls
on production limited by incomplete
data, difficulty in validating models due
to limited duration well histories
Identifying and applying appropriate
empirical models necessary
Conclusions
Arps empirical model inappropriate
Best fit b changes (decreases) continuously
with time
Fit of data at given time can be excellent,
at least as good as fit with SE model
However, best fit b values > 1 lead to
unreasonably large reserves estimates
when used for extrapolation
Conclusions
Stretched exponential model more
appropriate
Fits both transient, stabilized flow data with
unchanged parameters (n, )
Reserves estimates bounded as rate 0
Particularly appropriate for large groups of
wells smoothes noise due to operations
decisions, identifies characteristic formation
parameters
A Better Way to Forecast Production in
Unconventional Gas Reservoirs
John Lee
Texas A&M University
7 October 2010

Das könnte Ihnen auch gefallen