Professional Documents
Culture Documents
SPE
Society of Petroleun Engineef's of AIME
"Member SPE·AIME
tNow employed by Harper Oil Company.
The best hyperbolic curve is numerically fit to user he can then use his own judgement to accept or
production decline data from selected oil wells. A adjust this prediction.
non-linear optimization computer program finds the
three best values of the general hyperbolic para- This investigation is based on the premise that
, a, and b using least squares regression. most declining oil production follows a line repre-
By residuals, better history matching sentable by a hyperbolic equation. A drastic change
is obtained than with previous least squares methods. of conditions during the primary production of ewell
The future rate-time behavior of a well with only a would upsPt this trend. Such changes include re-
few months of production is predicted by an analogy completion in a ne\-l zone or mechanical problems.
with other similar wells.
Confidence intervals show the probable range of
This production extrapolation method is super- errors of an extrapolation of the fit from the actual
ior to graphical or type curve procedures since it future production. This interval is based on the
does not require engineering judgement until the un- deviation of the known production from the calculated
biased curve fit has been calculated from observed fit and the regression calculations which derive the
production data. The program calculates consistent hyperbolic curve.
results based only on production values. Then, the
evaluation engineer can use his judgement and experi- Comparisons of predictions and confidence inter-
ence to modify the prediction. vals of »'ells in the same area show similarities.
Wells in the same geographical area and producing from
INTRODUCTION corresponding geological formations will have similar
characteristics in their hyperbolic fits. For example
The extrapolation of decline curves is a common some fields consist of wells which decline at a con-
method of predicting the performance of a producIng stant rate of 12% per year; others decline steady at
oil well or a group of similar wells. 1 ,2 However, first, then level out after the first year. These
these methods usually involve some manipulation of the similarities can be used to make more reliable pre-
data, such as adjusting scales with type curves dictions for newly completed wells.
(Slider)l. Some introduce bias into the data previous
to the extrapolation, such as graphically choosing THEORY
one or two parameters to represent trends of widely
varying data. These and other decline analysis tech- The general hyperbolic equation for oil produc-
niques are discussed by Ramsay3, Gentry4, McC ray 5, tion rate (q) as a function of the (t) can be express-
and Fetkovich 6 , ed as
1
By using an efficient curve-fitting computer q = qo (1 + a b (1)
program, a general hyperbolic curve can be fit to the
raw monthly oil production data. A continuation of The parameter q is the initial oil production rate at
this hyperbolic curve beyond the period of known pro- time zero and hgs the same units as q. The parameter
duction gives an extrapolation based only on the a is the initial decline rate and has the same units
production values, and not on any interpretation by as t. The third parameter, b, has no units and con-
the evaluation engineer. After this curve and its trols the degree of curvature of the hyperbolic line.
If b=O, there is an exponential decline of q with
respect to t: and if b=l, the relationship is termed
Reference and illustrations at end of paper.
2 STATISTICAL DECLINE CURVE ANALYSIS SPE 10179
harmonic. The general hyperbolic form was chosen since properly fit a known decline curve with unique para-
it incorporates the other common decline curves. meters was tested by fitting such data corrupted with
normally distributed errors and properly ex-
The hyperbolic equation was used to represent the ponential, harmonic and more general hyperbolic curves.
declining trends of eight Oklahoma oil wells. Monthly
oil production was known through abandoment for all The continuation of the hyperbolic line describ-
examples. The wells produce at low gas-oil ratio (GOR) ed by the optimum parameters is taken to be a predic-
from the Mississippi limestone (4 wells), the Oswego tion of the well's performance beyond its known pro-
limestone (3 wells), and the Dutcher sandstone (1 well) duction history. The reliability of this extrapola-
tion depends on the goodness of fit to the known pro-
The hyperbolic parameters were calculated by a duction. A standard indicator of this is the stand-
non-linear regression technique. The hyperbolic eq- ard deviation (OR)' defined as
uation which best fits the production data is found by
minimizing the variance between calculated values (<1)
SUM OF SQUARES ~
and observed values (qT) for each month of known pro- (DEGREES OF FREEDOM)
duction, as shown in Figure 1.
mation.
In the following example, three Mississippi
limestone wells (A,B and C) have been producing for
several years. Another nearby well (D) has only been a Initial decline rate - second hyperbolic
producing for six months. Well D does not have enough parameter, months
data to allow confident prediction of future produc- b Curvature exponent third hyperbolic
tion. Fits other of the three wells' production are parameter, unit1ess
given in Table 1. n Total number of production history values
n-4 Degrees of freedom
A fit for the new Well (D) would expected to q Oil flow rate. barrels per month
have an "a" of 3.85 to 4.00 and "b" from 0.70 to 1.4. qo First hyperbolic parameter, barrels per
Values chosen to be average for the geographical area month
T
of these wells could be a 3.90 and b 1.00. The q Actual oil flow rate, barrels per month
very low value of "a" for Well (C) is ignored since q Calculated oil flow rate, barrels per
this is probably a function of the initial flow rate; month
Fell (D) produced only 3,000 barrels of oil in its SQSUM Sum of squares
first and highest month. An unbiased value of q t Time since start of production, mont£s
3,746 bbl/mo is found by fitting the first six m8nths OR Standard deviation of the ratio (q-q )/q
of data with a and b held constant at 3.90 and 1.00,
respectively.
ACKNOWLEDGEMENTS
This prediction is compared with the fit after
six years in Figure 5. The estimated fit gave a
The first author acknowledges the financial
standard deviation, (OR), of 0.261 after 72 months; support of the U.S. Department of Health, Education
the actual fit after 72 months had a oR of 0.205.
This indicates that decline curve analysis is still a and Welfare. Both authors than the University of
useful prediction method for new wells with short pro- Oklahoma Computing Services for the resources pro-
vided by it.
duction histories.
The pattern search optimization routine was used 1. Slider, H.C., "A Simplified Method of Hyperbolic
to fit the best hyperbolic curve to each well at diff- Decline Curve Analysis," Journal of Petrolemn
erent points in their producing lives. Figures 6 and March 1968, pp 38-41.
7 show curves which were determined using production
records up to 24, 48 and 72 months and near abandon- 2. Arps, J.J., "Analysis of Decline Curves,"
ment for wells A and D. It can be seen that the 1945, v. 160, pp. 228-247.
character of decline in the first few years reflects
the later behavior of a well; a hyperbolic curve 3. Ramsay, H.J., Jr., "The Ability of Rate-Time
which fits earlier years is a good representation of Decline Curves to Predict Future Production
future production. Empirically, consistent reliabil- Rates", M.S. Thesis, University of Tulsa (Tulsa,
ity is probable with 24 or more months of data. Oklahoma), 1968.
10. McNulty, R.R., "Statistical Decline Curve 11. Snedecor, G.W. and Cochran, W.G., Statistical
Analysis", M.S. Thesis, University of Methods, Iowa State University Press (Ames,
Oklahoma (Norman, Ok1ahema), 1981. Iowa), 6th Ed., 1978.
~
"I CALCULATED PRODUCTION RATE,
~6---t;1 __
W
~
CY
Z
o
t; "
:J
o
o
c::!:
Q.
..J
o
HISTORY
.-J
o
::3
---
W
0:::
0:::
<{
III
10'2
------- ------
10'
o
10'3
::I:
t-
Z
0
:l:
a:::
w
(L
....I
0
(f)
.-J
101:
W
0:::
0:::
<t
rn ---
TIME MONTHS
I FIT
.....
Z
0
::2
cr::
W
0..
....J
0
------------- ---- --
(f)
....J
IO~
W
cr::
cr::
4
ro
10'
o
10'3
I
I-
Z
0
::2
cr:
W
0..
(f)
..J
W
cr::
cr:
101- -- ---
«
ro
10'
o
TIME I MONTHS
Fig. 5 Comparison of actual 72-month fit and analogy estimate for Mississippi limestone
Well D.
24~MONTH FIT
48~MONTH FIT
10?
72-MONTH FIT
::t:
91-M:
d
I-
Z
o
::!:
a::
lIJ
a..
....I
5
~ 10'2.
-~
--- ----
----
w
a::
a::
d:
rn
o 24 48 72 96
TIME , MONTHS
10"!>
::t:
I-
Z
0
::!:
ct:
W
(L
....I
0
(j) 101-
....I
W
a::
a::
d:
co
o 24 72 96
TIME MONTHS