Sei sulla pagina 1di 10

SPE 142963

Practical Aspects of Gas Material Balance: Theory and Application


Phil Diamond, SPE, Prescience Consulting Ltd, and Jonathan Ovens, SPE, Improved Recovery Consulting Ltd

Copyright 2011, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE EUROPEC/EAGE Annual Conference and Exhibition held in Vienna, Austria, 23–26 May 2011.

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 have not
been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum
Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper 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 SPE
copyright.

Abstract

Gas material balance is often considered one of the more elementary areas of Reservoir Engineering. However, the simplicity
of the straight line p/Z plot is often masked by many geological realities: partially sealing faults, aquifer influx, heterogeneity
and low permeability. The purpose of this paper is to present a more realistic but still practical theory of gas material balance.

The paper begins by reviewing the underlying theory behind gas material balance and shows how an average reservoir
pressure can be calculated that accounts for flux between reservoir compartments and by using this average reservoir pressure,
water influx volumes can be estimated in compartmentalised reservoirs. The technique is applied to three real field cases; in
the first case water influx was suspected but using this analysis it was possible to show, convincingly, that there was no water
influx and previous deviations from a straight line p/Z trend were due to partially sealing faults. In the second case, the field
clearly did suffer from aquifer influx and we show how this technique was used to estimate the invaded volumes and the
material balance analysis was validated by core flood results.

Finally, we show a case where the material balance technique has been linked to an uncertainty analysis. For each GIIP
estimate, a corresponding water influx volume was calculated, which in turn was converted to an equivalent displacement of
the gas-water contact (GWC) depth. We show how the "most likely" GIIP and water influx estimates for the field were
validated by production and log data from two wells within the field.

Introduction

Gas material balance (GMB) is at once the simplest phenomena to describe, but in practice can be the hardest to analyse in a
convincing and unique manner. A simple linear relationship is predicated between cumulative production and average
reservoir pressure, based on the gas law pV = ZNRT, the so called ideal “tank”. However, this ideal behaviour can often be
modified by aquifer influx, low permeability, compartmentalisation, layered or heterogenous reservoirs or incomplete data.
The correct recognition of these often competing phenomena is the key to correctly diagnosing the true reservoir behaviour.

Furthermore, as noted by Cason1, straight line p/Z plot behaviour does not necessarily of itself guarantee that a reservoir is
behaving like a true tank. Conversely, a curved p/Z plot does not always mean the reservoir is experiencing aquifer influx,
despite the fact that most of the literature on gas material balance seems to suggest that aquifer influx is the prime cause of
curved p/Z plots. Therefore, given this lack of diagnostic capacity of the p/Z plot, the Reservoir Engineer should always seek
to set his analysis within the context of the known geology and where the material balance interpretation leads to an
ambiguous analysis this difficulty should not be ignored simply to derive a unique GIIP value.

This paper, which is based on both recent work and work going back twenty years, seeks to show how different interpretations
of gas material balance can arise when compartmentalized or water-drive reservoirs are encountered.
2 SPE 142963

Background

There is abundant material on gas material balance in the literature. Pletcher2 provides a comprehensive summary of the
theory behind GMB, while Payne3 and Hagoort4 have shown how numerical techniques can be used to explore the impact of
compartmentalisation and low permeability on GMB.

Despite the abundance of literature in this area, our impression is that the issue of compartmentalisation tends to be ignored in
favour of aquifer influx. The papers of Collins et al5 provide exact solutions to the p/Z behaviour of a compartmentalised
reservoir and for us these solutions have provided significant insights into the behaviour of several real world gas reservoirs.
Collins’ solution, as illustrated in Figure 1, show that at early time the p/Z plot reflects the volume of the directly drained
volume, while at late time the line curves over to a trend that reflects the total drained GIIP. Moreover, the effective GIIP
drained by a well at any point in time is reflected by the slope of the p/Z plot and not by the extrapolation of the p/Z trend to
the Gp axis, as is almost universally carried out when analysing p/Z plots.

Production
P/Z

SLOPE relects total volume
Observed P/Z
If constant production
rate Abandonment

Total Volume
Directly connected tank

Cumulative Production

Fig. 1 – Idealised p/Z behavior based on Coliins’ two tank model (see inset) assuming the production rate is constant. At early time
the slope of the p/Z plot reflects the direct drained GIIP, while at late time the slope reflects the total GIIP.

Example A – Compartmentalised Reservoir without Aquifer Influx

While carrying out a volumetric review of a Southern North Sea gas field, deviations away from a linear trend were observed
in individual platform p/Z plots, suggesting a possible water influx that had occurred over the 20 year producing lifetime,
assuming each platform drained an isolated volume. Such a conclusion was important in the GIIP determination because most
of the wells had been drilled high in the crest of the field and so water influx would not show up in any logs or as excessively
high water production. Since the field consisted of a series of upthrown fault blocks, there seemed to be limited chance of
direct connection to an aquifer, but this possibility could not be ruled out ab initio.
P/Z and F/Eg
P/Z and F/Eg

Cumulative Production Cumulative Production
Alternate p/Z plots from field data. Left hand plot (Fig. 2) used production rate weighting and a more limited number of wells; right
hand plot (Fig. 3) used the slope of the p/Z plot for each well as a weight and included additional wells. The left hand plot suggested
water influx that is not apparent in the right hand plot.
SPE 142963 3

The left hand plot above (Figure 2) shows the p/Z and F/Eg trends for a single platform. The rising F/Eg trend was interpreted
as indicating water influx. The right hand plot (Figure 3) shows plots for the same platform area but also includes additional
wells that could be in communication through partially sealing faults. Although there is some initial rise in F/Eg, the trend
stabilizes after the fifth point, suggesting that the apparent influx was due to gas leaking and not water influx.

Careful examination of the early pressure data suggested that small decreases in pressure could have been generated by
production from neighbouring platforms. By grouping wells and platforms into several distinct groups it was possible to
define a weighted average pressure for several sections of the field, following the method outlined above. The new p/Z plot is
shown below and indicates a nearly perfect linear trend.

Average P/Z

Cumulative Production

Fig. 4 – Field wide p/Z plot based on weighted production from all wells. The pressure range is from pri to pri/4. The line is essentially
linear and shows no evidence of aquifer influx.

Although the new p/Z analysis pointed away from water influx, the Cole plot (illustrated below) added weight to the
compartmentalized reservoir interpretation. The data shown below consists of a series of superposed Cole plots for each of the
platform areas, plotted using the total field production. At early time the first two platforms (circles) are only seeing a limited
GIIP, with the rising F/Eg values symptomatic of some form of influx. Later platforms (triangles) show a falling trend,
suggesting that gas is being drawn away from these platform areas by some external sink i.e. production from the earlier
platforms. Ultimately the F/Eg plot converges to the true total GIIP that is consistent with the volumetric GIIP.

Fig. 5 – Cole Plot for all compartments, showing how at early time the initial wells saw only a limited GIIP which then increases to a
steady late time volume that was consistent with the volumetric GIIP. The falling trend in the later developed wells indicates that gas
is flowing from these compartments
4 SPE 142963

The conclusion is that in the case of a compartmentalized reservoir, the late time Cole plot reflects the total GIIP, while the
early time response only reflects the directly connected GIIP.

Example B – Compartmentalised Reservoir with Aquifer Influx

In this example, water influx, as evidenced by direct and increasing water production, was observed after several years’
production, indicating the field clearly had aquifer support. In addition, a wide spread of pressures was seen in different wells,
indicating that the reservoir was also compartmentalized, with partially sealing faults causing different areas of the field to
experience limited pressure communication.

Following the methods outlined above, an average reservoir pressure was computed, taking into account the lack of
equilibrium between the various compartments in the reservoir. The reservoir was divided into two broad sections, east and
west, with wells in each section showing a roughtly similar pressure history. One well, located in an isolated area to the west
of the field, always showed a significantly higher pressure compared to the other western area wells, indicating limited
communication between this well and the rest of the field.

Water influx trends were established using the average pressure trend and the timing of the observed water breathrough in
various wells, and a variety of different methods were used to provide a range of estimates for the GIIP. An uncertainty
analysis on the GIIP estimates suggested a range of 5% in certainty of GIIP.

Conclusions on Cole plots

The examples shown above suggest that non constant F/Eg trends can arise from two sources: aquifer influx and/or
compartmentalisation. In the literature the latter option is almost never mentioned, but should always be kept in mind as a
possible cause of a rise in F/Eg. Example Cole plots should include the compartmentalised option, as illustrated below.

For instance, what conclusions might have been drawn if the falling F/Eg trend observed in some of the platform data in
Example A had been omitted?

Aquifer
F/Eg

Volumetric

Compartmentalised

Cumulative Production

Fig. 6 – Suggested type behavior for the Cole plot

It should also be noted that the estimate of GIIP is diametrically opposed between these two situations. For the
compartmentalized reservoir the true drained GIIP only shows up at late time, while in the aquifer situation the drained GIIP is
estimated be extrapolation to early time. Thus external information is required to determine which of these situations is most
correct for the reservoir at hand.
SPE 142963 5

Probabilistic Material Balance Analysis

The previous sections have demonstrated how an understanding of reservoir architecture and drive mechanism is vital to the
reliable estimation of initial gas in-place (GIIP) in a gas reservoir. However, even when the prevailing drive mechanism has
been correctly identified, and a material balance model has been established which accounts for this, there typically remains a
degree of uncertainty to the input variables in the GIIP calculation. Unlike in a conventional material balance analysis, where
the cumulative gas production, fluid properties and reservoir pressures are known to a reasonable degree of certainty, the
additional contributory variables in a more complex analysis (such as fault transmissibility or aquifer strength) are rarely
known to a sufficient level of accuracy. This gives rise to the possibility of non-unique solutions where, for example, multiple
combinations of GIIP and aquifer strength parameters will give similar responses. The example below illustrates how this has
been addressed by applying a stochastic technique to the solution of the extended material balance equation, using Monte
Carlo simulation to generate a large number of solutions and hence a probabilistic GIIP estimate range for the field.

The field in this case is a mature Central North Sea field producing gas from a Palaeocene sandstone reservoir at about 9,000 ft
TVDSS. A consistent reservoir monitoring program had ensured that a comprehensive production and pressure database had
been established for the field. Although the field was by now relatively mature and off plateau, there still remained
considerable uncertainty as to the precise field extent and the ultimate level of reserves. Specifically, the far north of the field
is quite sparsely drilled and the northern limit of the reservoir not very well resolved. Similarly the limit of the stratigraphic
pinchout to the east is poorly defined. The volumes of gas contained in these regions of the field, and their contribution to
current production was hence unclear. It was also not known whether the depletion of the gas reservoir is being supported in
any way by influx from the aquifer adjoining to the north and west. Figure 7 shows the p/Z plot for the example field: a linear
regression and extrapolation of the data yields an apparent GIIP estimate, Gapp. However a diagnostic Cole plot for the same
data (Figure 8) reveals a deviation away from the horizontal which is indicative of aquifer support.

1400
5000

p/Z Data 1300


4500
Apparent GIIP
1200
4000
Reservoir Withdrawal, F/Eg (Bscf)

1100
3500

1000
3000

900
2500
p/Z (psia)

800
2000

700
1500

600
1000
500
500
400
0 0 100 200 300 400 500 600 700 800
Cumulative Gas Production, Gp (Bscf)
0 200 400 600 800 1000
Cumulative Gas Production, Gp (Bscf)
1200
Gapp
1400

Fig. 7 – Linear extrapolation of p/Z data Fig. 8 – Cole plot for observed pressure data

A spreadsheet-based model was built with a modified form of the material balance equation to account for aquifer influx:

p pi ⎛ G p ⎞ ⎛ (We − W p ) Bw Ei ⎞
= ⎜1 − ⎟⎟ ⎜⎜1 − ⎟⎟
Z Z i ⎜⎝ G ⎠ ⎝ G ⎠

Unlike the other variables in the equation, the water influx volume, We, cannot be directly measured in the field. Instead we
have to estimate this quantity by applying a theoretical paradigm for natural water influx to the material balance model. In this
case, the most appropriate is the unsteady-state theory Hurst & van Everdingen7. This theory expresses cumulative water
influx at any point in time as a function of a number of factors, including reservoir pressure depletion, aquifer extent, and fluid
properties such as viscosity and compressibility. Since these aquifer properties are not known to any great degree of certainty,
the model was constructed from the outset to incorporate uncertainty in all the input variables. Indeed, as well as the aquifer
itself, one can also extend the application of uncertainty to the primary input parameters of pressure and production volumes
(accounting for possible errors in gas rate measurements, datum correction to pressures, fluid property variation, etc.)
6 SPE 142963

For this extended set, there are a total of fifteen different input variables to the material balance equation, each of which can
have an element of uncertainty associated with it:

• Reservoir Pressure: pr
• Fluid Properties: Z
• Produced Volumes: Gp, Wp
• Aquifer Properties: k, h, φ, Tr, μw, Bw, cw, cf, ro, reD, θ

In the model, the equation was solved probabilistically using an in-built Monte Carlo simulation algorithm which uses random
numbers to select input values to the equation and then solves for GIIP using an “aquifer fitting” process that matches the
calculated pressure response to the observed data. The simulation algorithm is summarised graphically in Figure 9:

1. Random numbers select values from each input


parameter range

x1 x2 x3

2. Material balance model predicts p/Z response for


given “first guess” GIIP (using Hurst & van Everdingen
Repeat 1000 times

theory to account for any aquifer influx)

3. Excel “Solver” utility solves for GIIP by minimising the


total error between predicted and observed p/Z data

4. Algorithm checks that total RMS error is below user-


input threshold value RMS Error < Threshold ?

No. GIIP Error


1 1000 1.35%
2 1050 1.43%

5. Solution stored for GIIP and magnitude of error 3


4
975
1020
1.29%
1.41%
: : :
: : :
: : :

6. Cumulative probability distribution (S-curve) plotted


for GIIP, plus associated “error space” graph

Fig. 9 – “Aquifer fitting” algorithm for Monte Carlo simulation

Having established the deterministic field GIIP, the model was then run in full probabilistic mode with 1000 trials in a Monte
Carlo simulation. All fifteen input variables were allowed to vary randomly between their minimum and maximum end-point
values. The resulting output S-curve for GIIP is shown in Figure 10.

Note from Step 6 of the algorithm shown in Figure 9, that the output from the model is not only an S-curve for GIIP but also
an associated “error space” plot. This is a plot of the normalised RMS error from the “aquifer fit” to the observed p/Z data as a
function of GIIP. The idea of this plot is that it will act as a guide to the value of computed GIIP that gives the closest match
to the observed p/Z decline: the lower the error, the better the match. Figure 11 gives the “error space” plot associated with the
S-curve of Figure 10. All 1000 results of the Monte Carlo simulation are recorded in the “error space” plot, forming a cloud of
data below the threshold error limit (in this case, 1.5%). It can be seen that, as might be expected, there are fewer data points
at the extremes of the GIIP range and generally the errors are greater there, suggesting that although solutions can be found
here, the fit to the observed p/Z data is poorer than for the mid-range GIIP solutions. Indeed the error distribution is
reasonably symmetrical and best solutions do appear to coincide more or less with the P50 GIIP estimate.
SPE 142963 7

1.00 1.8

0.90 1.6
GP10
0.80 1.4

0.70
1.2

Normalised RMS Error (%)


Cumulative Probability

0.60
1.0
0.50
GP50
0.8
0.40
0.6
0.30

GP90

GP50

GP10
0.4
0.20
RMS Error
0.10 0.2
GP90 Threshold
0.00 0.0
800 900 1000 1100 1200 1300 1400 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350
GIIP (Bscf) GIIP (Bscf)

Fig. 10 – Cumulative probability S-curve for GIIP Fig. 11 – “Error space” plot from Monte Carlo simulation

Figure 12 shows the p/Z plot for the P50 GIIP estimate, GP50. The calculated data from the “aquifer fit” is shown together with
the original field p/Z data. As can be seen, the fit to the observed data is generally good throughout the production history.
The red dashed line on the plot shows the linear trend that would have been followed if the reservoir were depleting purely
volumetrically (i.e. without the support of aquifer influx), while the blue dashed line shows the earlier result of a simple linear
regression and extrapolation of the deterministic observed p/Z data. The “apparent GIIP” from this extrapolation, Gapp, was
some 18% higher than the “true GIIP”, GP50, for this case. This graphically illustrates the danger in always assuming
straightforward volumetric depletion when trying to estimate GIIP from a p/Z plot.

5000

p/Z Data
4500
Aquifer Fit
Apparent GIIP
4000
True GIIP
3500

3000

2500
p/Z (psia)

2000

1500

1000

500

0
0 200 400 600
Cumulative Gas Production, Gp (Bscf)
800 1000
GP50 1200
Gapp1400

Fig. 12 – P50 aquifer fit to p/Z data


8 SPE 142963

Verification of Water Influx Volumes

The GIIP estimates described above were all predicated on the assumption of a degree of aquifer influx into the reservoir, but
the actual volume of water influx had not yet been independently verified. Figure 13 shows the cumulative water influx
calculated by the material balance model, as a function of GIIP. Again, since there were many solutions output from the
Monte Carlo simulation, we have a cloud of data but for the P50 GIIP, the range of calculated water influx volume was
approximately 150 – 300 MMrb, which equates to a surface volume of 145 – 290 MMstb.

450

400

350

300
Water Influx (MMrb)

250

200

150

100
GP50

50

0
800 900 1000 1100 1200 1300 1400
GIIP (Bcf)

Fig. 13 – Water influx as a function of Monte Carlo calculated GIIP

Direct field measurement of water influx is not possible so we cannot make a comparison between these modelled volumes
and any field data. However, influx of water into any gas reservoir will usually manifest itself as a rise in the gas-water
contact (GWC). Hence, although the water influx volume itself cannot be measured directly, the resulting change in contact
depth can be measured in the field by means of PLT surveys or other forms of logging. In order to make a comparison
between the modelled influx and field observations, a relationship had to therefore be established between water influx volume
and GWC rise.

Using the available static geo-cellular model as a calibration tool, cumulative aquifer influx was converted into an equivalent
rise in GWC depth by calculating mobile hydrocarbon pore volume (HCPV) as a function of depth. Figure 14 shows GWC
rise as a function of influx volume for an assumed trapped gas saturation, Sgt, of 0.30.

Production well “A” in the field watered out some 4½ years after first gas and, from the material balance model, the
cumulative water influx at this point in time was estimated to be in the range 49 – 122 MMrb, with a P50 value of 87 MMrb.
According to Figure 14, this gives a range of 44 – 73 ft for the rise in GWC, with a P50 value of 59 ft. The bottom perforation
of well “A” is at a depth of 9,128 ft TVDSS, some 63 ft above the original GWC depth of 9,191 ft TVDSS. This figure is
therefore in excellent agreement with the P50 estimate of 59 ft derived from the material balance model.

Similarly, a PLT conducted in well “B” two years later, showed a possible free-water level (FWL) at 9,090 ft TVDSS, some
101 ft shallower than the original depth. The material balance model calculated a P50 aquifer influx volume of 201 MMrb for
the same point in time which, from Figure 14, equates to a contact rise of 104 ft. Again there appears to be very good
agreement between the depletion performance and influx predicted by the material balance model, and that observed from
field measurements.
SPE 142963 9

400

350

300

250
Rise in GWC (ft)

200

150

100

50

0
0 100 200 300 400 500 600 700
Cumulative Water Influx (MMrb)

Fig. 14 – Rise in GWC level as a function of water influx

Conclusions

1. The correct identification of reservoir drive mechanism is crucial in arriving at an accurate estimate of in-place
volumes. This requires a full integration of all available data, including seismic, geology and production data.

2. The technique of grouping wells and then calculating a weighted-average pressure can be used to compute the
dynamically drained GIIP.

3. Ignoring the possibility of aquifer influx can lead to a significant over-estimation of GIIP, but at the same time
ignoring the possibility of a compartmentalised reservoir could significantly underestimate GIIP.

4. Type examples of Cole plots should include the rising effect of a compartmentalised reservoir.

5. Accounting for aquifer influx into the material balance equation introduces a greater degree of uncertainty in the
calculation of GIIP, in which case a probabilistic approach is appropriate whereby a large number of solutions can be
calculated and a confidence range applied to the resulting GIIP estimates.

6. The identification of weak aquifer drive in the reported example was verified by independent field observations
showing an apparent rise in the GWC caused by aquifer influx.
10 SPE 142963

Nomenclature

Bw = water formation volume factor (rb/stb)


cw = water compressibility (psi-1)
cf = formation compressibility (psi-1)
Ei = 1/Bgi = gas expansion factor at initial conditions (scf/rcf)
F = cumulative reservoir voidage (Brcf)
Gapp = apparent gas initially-in-place (Bscf)
GP50 = P50 estimate of gas initially-in-place (Bscf)
Gp = cumulative gas production (Bscf)
h = reservoir thickness (ft)
k = permeability (mD)
pr = reservoir pressure (psia)
pri = initial reservoir pressure (psia)
reD = dimensionless aquifer radius
ro = reservoir radius (ft)
Sgt = trapped gas saturation (fraction)
Tr = reservoir temperature (ºF)
We = cumulative water influx (MMstb)
Wp = cumulative water production (MMstb)
Z = gas compressibility factor
φ = porosity (fraction)
μw = water viscosity (cp)
θ = aquifer encroachment angle (degrees)

References

1. Lloyd D Cason Jr: “Waterflooding increases gas recovery”, paper SPE 12041 (1989)
2. Pletcher, J.L.: “Improvements to Reservoir Material Balance Methods,” paper SPE 75354, SPE Reservoir
Engineering & Evaluation (February 2002).
3. Payne D.: “Material Balance Calculations in Tight Gas Reservoirs: The Pitfalls of p/Z plots and a More Accurate
Technique”, SPE Reservoir Engineering (November 1996)
4. Hagoort J., Sinke J., Dros B., and Nieuwland F.: “Material Balance of Faulted and Stratified, Tight Gas Reservoirs”,
paper SPE 65179 (2000)
5. Howler T., Collins R.: ”Detecting Compartmentalization in Gas Resevoirs Through Production Performance”, paper
SPE 19790 (1989)
6. Ehlig-Economides, C.A.: “Applications for Multiphase Compartmentalized Material Balance”, paper SPE 27999
(1994)
7. Van Everdingen, A.F. and Hurst, W.: “The Application of the Laplace Transformation to Flow Problems in
Reservoirs,” paper SPE 949305-G, Petroleum Transactions, AIME, Volume 186, (1949).

Potrebbero piacerti anche