Professional Documents
Culture Documents
Improvements for Flow Correlations for Gas Wells Experiencing Liquid Loading
N. Kumar, Murphy Oil Corp.
Programmed Correlation. A working version of original The change in volume of gas during pressure reduction is
Gray correlation was developed using Fortran 90 as the represented by ∆Vwp and the change in volume of gas due to
programming language. The program has used equations as reduction in temperature is represented by ∆VwT. Figure 1 and
described for Gray correlation in API-RP14B2. The program Figure 2 give values of ∆Vwp and ∆VwT as function of
has as inputs gas, oil and water flow rates, API gravity of the reservoir temperature and pressure. The formation volume
condensate if any produced, the gas specific gravity, the water factor of water may be computed from these values using the
specific gravity, the surface and bottom hole flowing following equation:
temperatures, recorded surface and bottom hole pressures,
depth, size and number of tubing strings and their lengths and B w = (1 + ∆ V wp )(1 + ∆ V wT ).......... .......... .......... ........ 1 . 2
depths, the roughness value desired to be used in calculating
the friction pressure drop, and the angle (for vertical wells the Where Bw is the formation volume factor. As shown in
angle is 90 degrees). The calculation starts with dividing the the next section, the effect of the water density correction on
total depth into a fixed number of equal intervals. PVT the pressure traverse was limited as might be expected.
parameters are calculated at each interval. The program is
capable of calculating the pressure traverse from top down or Correction and modeling for water condensation. The
bottom up as is typical of most industry correlations. The original Gray correlation did not account for the condensation
necessary equations to describe the model are summarized in or vaporization of water in the tubing. It was correlated to data
Appendix A. The output from program includes pressure that had condensate up to 154 bc/MMscf/D and water up to 10
versus depth, pressure drop, elevation, friction and bw/MMscf/D. Thus, the form of the relationship for the
acceleration gradients and flow rates. If measured test data is holdup was developed to represent the holdup that would
input then output includes differences between predictions and match data for wells in which condensates are the major
measurements. contributor to liquid loading. Actually only water and not
condensate is present in many loaded wells. Using existing
Modifications to Original Gray correlation investigated. correlations19 in literature on modeling condensing water, the
The original Gray correlation has some limitations as original Gray correlation has been modeled for condensing
discussed in Appendix A, where a detailed summary of the water in a tubing downhole.
original Gray correlation has been provided. Some A downhole pressure survey19 may show where there is
modifications which could be made on original Gray condensation—the pressure gradient can nearly equal a water
correlation to improve pressure traverse prediction for liquid gradient in the condensation region, which is usually in the top
loaded gas wells are discussed here. A correction for gas portion of a well where the flow stream begin to cool. Figure 3
density affects13 due to depth and PVT conditions down hole shows a gradient in the condensation region where the psi/ft
has been investigated. The effect of this correction is small but value of the pressure gradient is almost as high as 100% water.
visible. The original Gray correlation has also been modified This particular gradient was found to oscillate and this could
to include predications of when input water is in the vapor or be because the condensed liquids are forming slugs and
free liquid state using Bukacek’s14 solubility chart. By using surfacing as slug flow showing intermittent pressure at a given
the Griffith’s15,16 and wallis’s17 bubble flow and Hagedorn18 location. If water is condensing, the additional liquid where
and Brown’s slug flow holdup equations, flow regimes were condensation is occurring can be found by calculating Qc,
introduced into the original single phase Gray correlation with where Qc is the barrels of water condensing at a cooler portion
some success. Using a regression technique, pressure profiles of the tubing as found using the dew point chart for water21.
from the “Modified24” Gray correlation were improved Figure 4 shows a heavy pressure gradient near the surface
relative to the original Gray correlation by adjustment of two and also at the bottom of a well bore. At the surface the heavy
input constants that allow the user to modify the correlation flowing pressure gradient is due to water condensation while
from input of these constants. The modified Gray correlation at the bottom it is because of liquids accumulating. Water
is discussed in Appendix B. The effects of each of these tested from condensation will add to the pressure drop in the tubing
modifications are described below. by adding to the liquid holdup where the water breaks out. If
the velocity in the tubing is low enough, then water will fall
Detailed Description of Corrections to Original Gray back in the tubing and accumulate at the bottom of the tubing
Correlation to add to the pressure drop. This has been shown in a test23 of
using heat from an electrical resistance heated cable to prevent
Correction for Change in Water Density with Depth. water condensation. The initial pressure drop was found to be
Gray’s work considers a constant water density for the entire high where the water condensed in the tubing and then a high
pressure traverse in a well. Where both water and condensate pressure drop area was found at the bottom of the tubing.
are present, Gray correlation takes a weighted average density: When heat was added, both of the high pressure gradients
ρ q + ρ wqw present were greatly reduced.
ρl = o o .......... .......... .......... .......... ......... 1 . 1
qo + qw The solubility curves for increasing pressure and
Density at reservoir conditions is determined by dividing temperatures in Bukacek’s14 work were converted into data
the density at standard conditions by the formation volume points and stored in separate files. The program reads the
factor of the water at reservoir conditions13. pressure and temperature as input and calls the subroutine to
find out water solubility at the given temperature and pressure.
SPE 92049 3
With increasing temperature and pressure, more and The use of coefficients found from regression should be
more surface reported water production is in a vapor state at more representative of an improved prediction technique as
depth. In other words, the model can predict that, at reservoir the program handles more and more data.
conditions, water may exist only in vapor form. The program
can then take the surface reported water and predict the Discussion of Results
distribution of water vapor and free water at down hole See Appendix C for a discussion of errors using the various
conditions. If only fresh water is reported at the surface it is an modifications. Table 2 shows the output obtained from
indication that only condensed water is coming to the surface. original Gray correlation. Note that except for one well, the
If salt water is reported, then it’s an indication that produced pressures for the data set are being over predicted when using
free water is coming into the well at bottom hole conditions. the original Gray correlation. Wells B, H and L are the closest
Use of the water solubility charts allows one to calculate if matches to Gray generated predictions. Table 3 shows some
free water and/or water vapor are present at bottom hole production data which will be used as results are discussed.
conditions as well. Note that nine wells have total barrels per MMscf/D below the
Figure 5 shows, with 100°F water at 30 psig bottom hole 130 bbls/MMscf/D, which was the limit of the data set
conditions. 6 bbls/MMscf/D of water as vapor can exist at available to correlate the original Gray model. Also, four wells
these conditions. Since some low pressure gas wells produce are producing below 0.12 MMscf/D, i.e. four wells are outside
less than 6 bbls/MMscf/D just providing low pressure from of Gray’s original data set. The results obtained in terms of
use of a compressor can often be an adequate artificial-lift absolute average error, average error and standard deviation
technique to eliminate the effects of free water on the pressure for the pressures obtained at bottom hole by using the various
drop in the tubing. The graph in Figure 6 shows an expanded investigated modifications for the entire data set are
view of dew point effects on determining if water is in the summarized in Table 4.
vapor phase.
Results from adding the water density calculation. The
Adding Flow Regimes. The original Gray correlation is a two errors in Table 4 and Table 5a / 5b show that effect of the
phase correlation. The default flow regime for original Gray density correction to original Gray correlation is limited but
correlation is mist flow which is gas flow with some droplets. quantitatively expressible.
This suggested a need to include more flow regimes than the
original Gray correlation provided and investigate any impact Results from Original Gray Correlation Modified for
on the accuracy of the predictions. A description of various Water Condensation. Table 4 and Table 5a / 5b suggests that
standard flow regimes22 can be found in Figure 7. A flow modeling the condensation of water in the tubing show
regime selection criterion given by Orkiszewski8, often used improvement in some wells when comparing predicted
with multiphase flow correlations has been used in this work. pressures to measured data. This effect was found to be a
Other flow regime selection maps are available, but major contributing factor for at least three wells
Orkiszewski’s work takes into consideration the existing well Although the water condensation modification was
developed hold-up equations and selection criteria8 available applied to nine wells, significant effects can be seen in only
in industry and puts them together into his flow map. three of them. This is because the rest of the wells were
If a well is flowing above critical rate and associated flowing at very low gas flow rates. For these wells, the
velocity as per Turner21, then the flow regime is mist and the reservoir energy is very low and the problem essentially
holdup is calculated using original Gray’s criteria. If gas demands use of a better gas holdup equation, and not just a
velocity is below critical velocity then it is directed to the correction for free water present and average density.
Orkiszewski’s flow regime subroutine which decides if the
flow is slug, or bubble flow. If a flow map determines slug Predicted Results from Original Gray Correlation
flow regime, then Hagedorn and Brown’s18 slug holdup is Corrected for Water Condensation and Density
evaluated. If the flow map determines the flow condition is Correction. Results from Table 4 and Table 5a / Table 5b
bubble flow, Griffith’s15,16 criteria is used to determine the suggests that combined density and water correction effects
liquid holdup. The holdup calculated from respective flow produced better results than obtained by water correction by
regimes are then passed on to the pressure gradient equation. itself, though the correction is slight.
Modified Gray. A regression program was introduced, which Modified Gray Results. Effects of regression to find values
finds the value of two input constants in the modified Gray of two holdup adjusting factors have been applied with
holdup expression, which will then yield the minimum success using the modified Gray correlation. Note that when
average error between the predictions and the measured data regression was applied without any modifications to the Gray
for pressure drop in the tubing considering the input constants correlation, the pressures predicted for at least one well (well
values. The modified Gray equations are described in A) had a large error relative to the measured well test data.
Appendix B and are referenced in reference 24. The explanation is the fact that the coefficients being input and
In the regression program, an initial guess is made for regressed are based on all well data and well A is outside the
maximum and minimum values for the two constants .The original Gray data set. As more and more data sets are added
program makes the regression calculations to determine the to the available data, it is hoped that the coefficients predicted
two coefficients in the Modified Gray correlation that will by modified Gray would be more representative of a universal
minimize the average error. data set and will not be affected by a single well which may be
4 SPE 92049
somewhat of an exception. However more likely use of the the droplet model for most critical velocity expressions would
modified Gray correlation may make the correlation more no longer be applicable. At present this amount of liquid flow
suitable for a large group of particular wells with particular vs. the gas flow where droplets begin to coalesce is not
conditions. incorporated into a critical velocity expression such as in
From Table 4, regression of input constants was able to Equation 1.3.. If it were predicted, better predications might be
bring the average error down to 0.25% and bring down the available for critical velocity.
absolute average error to 24.28 %. This regression technique
bring the average error to zero on the entire data set.
Definitely, it can be said that the coefficients provided by the Critical Velocity and Gas Holdup. Figure 22 shows22 that
program for different corrections are the best available the flowing pressure gradient (or holdup) is related to the
coefficients for this or a particular set of data. The coefficients superficial velocity of gas
obtained are shown in Table 6. Also note that with water Since holdup is related to gas velocity, somewhere on the
condensation and water density corrections added to Modified velocity axis, the critical velocity will have been reached. As
Gray, the coefficients tend to move towards the Gray such it seems it might be possible to relate holdup to how
evaluated constant value of -2.314 ( Coeff1). Table 5a and 5b close or how far away the actual velocity is to the critical
show the percentage error for each well obtained using the velocity as mentioned above. A better course of action might
modified Gray correlation. As can be seen, use of the modified include studying the modifications presented here to a larger
Gray correlation and the regression technique has in general, data set for additional verification.
provided far better flowing well bottom hole pressure
estimates compared to the original Gray correlation. Conclusions
• Some few but high qualicty flowing gas well data
Addition of Flow Regimes to Original Gray and Modified with produced liquids is presented.
Gray Correlations. The addition of additional flow regimes • Modifications were made to existing correlations to
was considered to better model water accumulating at the better fit the measured data.
bottom of the tubing in liquid loaded wells. The flow regime • The modifications are expected to match gas well
additions when applied to wells D and E (Figures 11 and 12), data producing some liquids more accurately in
yielded better results than the original Gray / better results general than results from the original Gray
than any other correction applied to original Gray correlation. correlation.
Both these wells produce high water rates (991 and 2594 • Gray’s correlation is a good choice of the correlations
bbls/day). However, In general, addition of flow regimes leads considered here to calculate pressure traverses for gas
to over predicting of pressures when the entire data set is wells producing some liquids, provided the well
considered. Table 4 and Table 5a / 5b suggest that work with conditions are similar to those mentioned in Table
flow regimes was not in general successful comparing to the A.1.
limited data available. • The Modified Gray correlation can be used to
distribute errors over an entire data set. The larger the
Well Pressure Traverses. Figures 8 thru 21 show the pressure data set available, the more generalized this approach
traverses obtained for each well using different corrections. may become.
Results from these plots have already been discussed in the • The modification of adding Orkiszewski’s flow map
tables described above. to the existing Gray correlation in determining flow
regimes and the holdup for liquid loaded gas well was
Possible Directions for Future Work to Predict met with limited success. It was hoped to better
Performance of Liquid Loaded Wells Critical Velocity model accumulation of liquids at the bottom of the
Critical velocity is often used in the literature to determine well.
when a well is accumulating liquids (below critical gas
• An increase in pressure drop arising from liquid hold-
velocity) and the rate is high enough to lift droplets of liquids
up becomes more pronounced in low gas flow rate
from the tubing (gas velocity above critical velocity). So it
wells. A new holdup equation relating liquid holdup
would seem logical that as gas velocity drops to the critical and critical gas velocity may lead to better solutions
velocity and below, the holdup would begin to become larger.
for gas wells experiencing liquid loading problems.
Critical velocity given by Turner21 et al. is given by the
• The suggested modifications to allow an existing
following equation:
multiphase flow computer program to better compare
1.59τ 1/ 4 (ρl − ρg )1/ 4 with measured data are presented for consideration.
Vc = ..................................1.3 However although the data is high quality, the
ρ g 1/ 2 number of data sets is limited so the suggested
modifications must be tested and perhaps adjusted
So for vertical flow pressure drop correlations, it is
before general use. Testing the suggested
suspected that the liquid holdup could be correlated as a
modifications to a larger data set could provide
function of the critical velocity. Conversely the current critical
additional insight.
velocity expressions are not a function of the amount of
liquids flowing. It is suspected that for larger flow rates of
liquids, liquid droplets of liquids would begin to coalesce and
SPE 92049 5
Acknowledgements 7. Duns, H., Jr. and Ros, N. C. J.: “Vertical Flow of Gas
The authors thank BP America Inc. and Tom Brown Inc. for and Liquid Mixtures in Wells,” Proc. Sixth World
obtaining the field data and for assistance in some of the study Pet. Congress, Frankfurt (Jun. 19-26, 1963) Section
presented here. Although this research was an integral part of II, Paper 22-PD6.
PhD dissertation at Texas Tech University for the principal 8. Orkiszeweski, J.: “Predicting Two Phase Pressure
author, he is obliged to Murphy Oil Corporation for financial Drops in Vertical Pipes,” J. Pet. Tech. (June, 1967)
support and for allowing him to work on this research while 829-838.
working full time for them. 9. Beggs, H. D. and Brill, J. P.: “A Study of Two-Phase
Flow in Inclined Pipes,” J. Pet. Tech. (May 1973)
Nomenclature 607-617.
Bw Formation volume factor, dimensionless 10. Dukler, A.E.: “Gas-Liquid Flow in Pipelines, I.
D Pipe diameter, inches Research Results,” AGA-API Project NX-28 (May
∆Vwp Volume change of gas during pressure reduction, psia 1969).
∆VwT Volume change of gas during Temp reduction, oF 11. Brill, J.P., and Mukherjee H.: “Multiphase flow in
ρg Gas density, lb/ft3 wells”, Henry L. Doherty Memorial Fund of AIME,
SPE Inc, Richardson, Texas, 1999.
ρl Liquid density, lb/ft3 12. Aziz, K.,Govier, G.W. and Fogarasi, M.: "Pressure
drop in Wells Producing Oil and Gas," Journal of
ρo 3
Oil density, lb/ft
Canadian Petroleum Technology, (July-Sept. 1972)
ρw Water density, lb/ft3 38-48.
ETOT Total error 13. Mc.Cain, William D. Jr.: “The Properties of
ft Friction factor for two phase flow Petroleum Fluids”, Book, second edition, pg. 446-
FBHP Flowing bottom hole pressure, psia 448, PennWell Books, Tulsa, OK, 1990
PREF Reference pressure, psia 14. Bukacek, R.F.: “Equilibrium Moisture Content of
PWTA Pressure from well test, psia Natural Gases,” Research Bulletin, Institute of Gas
qo Oil flow rate , bbls/day Technology, Chicago, 1955
qw Water flow rate , bbls/day 15. Griffith Peter.: “Multiphase Flow in Pipes”. SPE
T Temperature in Fahrenheit 12895, distinguished author series, JPT March 1984.
τ Surface tension for mixture, dyne/cm 16. Griffith P., Wallis, G.B.: “Two Phase Slug Flow,”
Vc Critical velocity for gas, ft/sec ASME Journal of heat transfer, (August 1961), 307.
Vg Gas phase velocity, ft/sec 17. Wallis, G.B.: “One Dimensional Two Phase Flow”
Vt Turner velocity, ft/sec McGraw Hill book Co. Inc, New York, NY, 1969.
Z Real gas compressibility factor, dimensionless 18. Hagedorn, A.R. and Brown, K.E.: “Experimental
Study of Pressure Gradients Occurring During
ξ In situ volume fraction gas (gas hold-up) Continuous Two Phase Flow in Small Diameter
Vertical Conduits” J. Pet. Tech. (April, 1965) 475-
References 484.
1. Gray, H.E.: “Flowing Pressure Calculations for 19. David A. Simpson and J. F. Lea.: “Coal Bed Methane
Gas/Condensate Wells”, EPR Report 855, Shell Oil Production”, paper SPE 80900 presented at SPE
Corporation, 1955. Production and Operations Symposium held in
2. Gray, H.E.: “Vertical flow correlation in gas wells.” Oklahoma City, Oklahoma, U.S.A., 23–25 March
In User manual for API 14B, Subsurface controlled 2003.
safety valve sizing computer program, App. B. (June 20. Sutopo and Pudjo Sukarno.: “Modification of
1974). Flowing Gradient Equation in Gas Wells with Liquid
3. Cullender, M.H. and Smith, R.V.: “Practical Solution Present”, paper 92-23.11, Twenty First Annual
of Gas-Flow Equations for Wells and Pipelines, Convention of Indonesia Petroleum Association,
Trans. AIME, 1956, p 281-287. October 1992.
4. Poettman F.H and Carpenter, P.G.: “The Multiphase 21. Turner, R. G.; Hubbard, M. G. and Dukler, A. E.
Flow of Gas, Oil and Water Through Vertical Flow “Analysis and Prediction of Minimum Flow Rate for
Strings with Application to the Design of Gas Lift the Continuous Removal of Liquids from Gas Wells”,
Installations”, Drilling Production Practices, API J. Petroleum Technology, Nov. 1969. pp.1475-1482.
(1952), p 257. 22. Lea J.F., Nickens, H.V. and Wells M.: “Gas well
5. Bexendell, P.B. and Thomas, R.: “Calculation of Deliquification,” Elsevier Press, first edition, 2003.
Pressure Gradients in High-Rate Flowing Wells”, 23. Pigott, M.J., Parker, M.H., Vincente, D., Dalrymple,
Journal of Petroleum Technology (October 1961) p L.V., Cox, D.C., and Coyle, R.A., “Wellbore Heating
1023. to Prevent Liquid Loading,” SPE 77649, presented at
6. Fancher,G.H. & Brown, K.E.: “Prediction of Pressure the SPE Annual Technical Conference and
Gradients for Multiphase Flow in Tubing”, Trans Exhibition, San Antonio, TX, September 2, 2002.
AIME 140, 1963. 24. PerformTM, Well Performance Analysis Manual, IHS
Energy group. Version 4.0, December 2001.
6 SPE 92049
steady state, vertical, two phase flow for compressible qo = 0.681qoa βI1So − I 2 − (0.0593− 0.07S0 )(t − 710)....a.2
fluid is:
The parameters used in this equation are defined below.
This is an empirically obtained equation which has been used
2 2
⎛1⎞
dp =
g
(ξρg + (1 − ξ )ρl )dh + f t G dh − G d ⎜⎜ ⎟⎟...a.1 to describe condensate behavior at any depth and pressure.
gc 2g c Dρ mf g ⎝ ρi ⎠ Here, So is the specific gravity of oil and t is temperature at
that specific depth. The value of So is evaluated:
Where: D Conduit traverse diameter, inches :
ft Friction factor for two phase flow
g Gravitational constant, lbm-ft/(lbf-s2) 141.5
So = ...................................................a.3
G Mass velocity, lbm-ft/sec API o + 131.5
h Depth, ft
ξ In situ volume fraction gas (gas hold-up)
Gray has defined β as follows:
ρ Density, m: mixture, lbm/ft3 2
i inertial effects ⎛ P − 1.8 × 10 5 ⎞
β = 1 − ⎜⎜ ⎟ .............................................a.4
5 ⎟
⎝ Pd − 1.8 × 10 ⎠
The above equation can be solved to obtain flowing pressures
provided that:
Where Pd is given by
¾ Suitable definitions of mixture densities ρmf and ρmi
exist. ( )
Pd = 144 1000 + α ............................................a.5
¾ The state of the fluid as a function of pressure and
temperature can be determined and α is provided using the following equation:
¾ Independent information on the variation of
temperature with either pressure or depth is available [
α = A1 S o − A2 − 100 t − 585 + B ' (S o − 0.69)2.05 ..a.6 ]2
For gas specific gravity, the model takes into The only undefined parameters in equation a.1 are I1 and
consideration a mass interchange between gas and condensate. I 2.
However, there is no correction for water gravity. As S o > 0.765
temperature increases, water density decreases while increase
in pressure is always accompanied by increase in density. As I1 = 13.03
such, the liquid density has also been taken to be independent
of pressure and temperature13.
I 2 = 9.57...........................................................a.10
in data set is in two input parameters; the gas rate in MMscf/D Appendix B: Modified Gray
and free water and condensate production. In this work, all The modified Gray24 correlation is an attempt to improve
wells having input parameters beyond original Gray’s data from the accuracy compared to the original Gray correlation
range are referred as outside of the original Gray’s data set predictions. In original Gray holdup equation, a constant was
range. used. Gray determined the value for this constant to be -2.314,
based on experimental data from 108 well tests. As discussed
Effect of Input Parameters outside of the Original Gray in Appendix A, original Gray’s correlation has its limitations.
Data Set on Predicted Pressure Traverses. The effect The Modified Gray Program is used where two coefficients
parameters outside of the Gray data set can be observed by can be used to adjust the holdup expression with the idea to
taking two cases from Table A2. Consider well H which has better fit to the field data. Since Gray correlation originally
all its input parameters within original Gray’s range of data set was for primarily condensate flow the adjustments used in this
and well E which has most of its input parameters as an outlier discussion to better fit data may realistically be adjusting the
to original Gray’s data range. The actual values of input holdup expression to accommodate gas flow with water
parameters are shown in Table A3. instead of originally having the holdup be for representing
The absolute average error seen in predicting pressures condensate flow. Gray’s model has been modified to the
using original Gray correlation on data set from H well Modified Gray Model using a direct search regression
compared to the actual or well measured test data is barely technique discussed in the paper.
1.88% at bottom hole conditions. The pressure traverse is Below are listed the equations from Modified Gray
given in Figure A1. Note that the match between predicted routine where ξ is the holdup for gas and 1.0 – ξ is the liquid
values from original Gray and the actual measured pressure is holdup. The two coefficients that are determined by regression
excellent throughout the tubing. On the other hand, Figure A2 to field measured data to minimize the average error of
shows the pressure traverse for well E, which has a free water predictions to measured values are shown in Equation b.1.
ratio of 1314 bbls/MMscf/D. The absolute average error seen
in predicting pressures using original Gray correlation on data B
⎛ ⎛ ⎞⎞
set from E well compared to the actual or well measured test ⎜ R + EXP⎜ COEFF1⎡⎢ N ⎛⎜1 + 205.0 ⎞⎟⎤⎥ ⎟ ⎟
v⎜ ⎟
data is 40.9% at bottom hole conditions. Also, Figure A1 ⎜ ⎜ N D ⎠⎦ ⎟⎠ ⎟
⎝ ⎣ ⎝
shows a mismatch for predicted pressure traverse from top to ξ=⎝ ⎠ . + COEFF 2.......b.1
bottom of the well. It can be concluded from these two plots R + 1.0
that the original Gray correlation works very well for wells
like well H (ideal input parameters) while application of Where B and Nv and R are Gray evaluated experimental
original Gray correlation to wells like well E (outside input values and are described in Appendix A in detail. For the
parameters) may (but not necessarily) lead to and erroneous original Gray correlation the coefficients are COEFF1 = 2.314
pressure traverse calculation. Thus, there seems to be a need and COEFF2 = 0.0. As per Perform24, commercially available
for a new correlation which will provide better pressure drops software which has the Modified Gray correlation, suggested
for liquid loaded wells like well E where the input parameters values for COEFF2 are between 0 and 1 and COEFF1 varies
are outside of the data originally used by Gray to correlate as in a narrow range close to -2.314.
described in Table A1.
Appendix C: Discussion on Errors and standard
Correlations other than original Gray: Are better deviation
correlations generally available other than original Gray? In this section, equation describing errors have been discussed.
This question is answered with the help of Figure A2. It is a The absolute average error, average error and the standard
comparison plot of the pressure traverse obtained using deviation were found for the output obtained from each model
various correlations for well H. As can be seen, there is a large discussed so far. The errors were calculated as below:
spread in predicted pressures by the various correlations used.
However, for this well, which meets all the conditions under ⎛ 100(POUT − PREF1) − (PWTA− PREF1) ⎞
ERROR= ⎜⎜ ⎟⎟............c.1
which original Gray correlation was correlated, it can be ⎝ (POUT − PREF1) ⎠
observed that the original Gray correlation predicts with the
least error compared to measured data. The closest prediction ETOT = ERROR + ETOT .......... .......... .......... .. c . 2
to original Gray’s output is that of Cullender3 and Smith,
which is a single phase gas flow correlation. This plot justifies Where POUT = Output pressure from program, psia
the fact that original Gray correlation is the best correlation to PWTA is the well test pressure data in psia obtained from
be used for gas wells producing liquids of the correlations industry for wells used here and PREF1 is the surface
tested here. Other correlations exist but are not tested here pressure, psia
some of which require compositional input. Models using The absolute average error was found by dividing the
correlations and not using compositional input are examined absolute value of sum of all errors ETOT by the total number
here.. Good comparison to other models is one of the reasons of errors calculated by the program NDAT. The average error
that API RP 14B recommends the original Gray correlation to was found by dividing the sum of all errors ETOT by the total
be used for gas wells producing some liquids. number of errors calculated by the program NDAT.
SPE 92049 9
∑
i=n
( x i − x av ) 2
σ= i =1
.......................................................c.3
N
FIGURES
12 SPE 92049
SPE 92049 13
14 SPE 92049
SPE 92049 15