Sunteți pe pagina 1din 10

SPE 77619

Why We Should Stop Using Pseudopressures and Other Good Old Well Test
Interpretation Tools After So Many Years of Good Service
Olivier P. Houz, SPE, KAPPA

Copyright 2002, Society of Petroleum Engineers Inc.

This paper was prepared for presentation at the SPE Annual Technical Conference and Initially, anybody (operating company, service company,
Exhibition held in San Antonio, Texas, 29 September2 October 2002.
university, engineer with some spare time) would write their
This paper was selected for presentation by an SPE Program Committee following review of own well test interpretation program. A typical in-house
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to software would be technically good, user hostile and
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at dreadfully expensive. Cost considerations and downsizing
SPE meetings are subject to publication review by Editorial Committees of the Society of triggered the final move to commercial software. This makes
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is sense, but there are serious consequences in this move that
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 should not be overlooked.
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Software vendors may be reasonably decent and competent
people (?), but they are in a commercial process where they
Abstract have to face any and each request. It is difficult to say No
Some well test interpretation techniques still in use today were when facing a request conditioning a Purchase Order. And
developed more than 30 years ago, to help the engineer at a after all vendors are not and should not be the ones to define
time he/she (generally he) only had graph paper, a hand what the right methodology is. So with time features have
calculator and a set of printed type-curves. Unlike human accumulated, and todays commercial software has, at the
beings, age is not a reason by itself to retire a technique. But, same time, state-of-the-art technology, Star Trek features,
originally invented by paradigm breakers, thanks to our inertia and a living museum integrating whatever was published in
some techniques have become paradigms themselves. They no the last 50 years, the great, and the less great. This paper is
longer bring anything, they confuse the learning of young also a plea to officially allow, or even require software
engineers and worse, sometimes turn out to be incorrect and vendors to do some cleaning.
stand in the way of more rigorous and promising
developments. It may be time to retire them from our industry, From type-curves to computer generated models
acknowledging what they brought us in the past. The author At last the argument seems to be over: the log-log plot is the
respectfully suggests to get rid of type-curves, primary diagnostic plot, where the pressure change and its
pseudopressures1 and pseudotimes2. In each case, the author derivative with respect to superposition time are plotted
will try to explain why these tools were developed, why they together, on a log-log scale. The derivative was originally
were originally useful, but no longer. If the author survives designed as a magnifying glass of the semilog plot to
this one, the following publication will deal with the Horner 3 identify Infinite Acting Radial Flow on the log-log plot. But
plot, equivalent time functions4 and deconvolution 5. Otherwise luckily enough it also carries the signature of numerous well,
there is absolutely nothing new in this paper. reservoir and boundary behaviors.
A foreword on software After diagnosing what the model can be, the engineer
Most pressure transient analyses today are performed using a estimates the model parameters, and the program generates the
computer running dedicated commercial software. The first solution with the same production history as the data. Data and
PC based programs appeared in the 1980s, initially merely models are compared, the user then invalidates the model or
reproducing the manual methods, on a computer. Then a improves the parameter calculations, either by trial-and-error
second generation was developed, with a core methodology or using nonlinear regression. This methodology is the
based on the use of the pressure derivative6, model generation successor of manual type-curve matching, and is still
and parameter optimization using nonlinear regression 7. All sometimes improperly called type-curve matching.
commercial software products today have such capability, but
the old manual techniques were not removed.
2 O. P. HOUZE SPE 77619

The original type-curves consisted of a set of printed or Type-curves may be history and methodology has
stored dimensionless pressure responses, on a log-log scale. changed, but the tools above are still there. In the following,
Real data would be transformed, plotted and slidden in order the author will address todays relevance of pseudopressures
to select and fit one of the type-curves. Physical results would and pseudotimes.
be calculated from the match point and the choice of the curve.
Gas pseudopressures why they exist
But type-curves were far too limited to compete durably In order to benefit for gas of all the good things that were
against computer generated models: (1) They only provided a developed for slightly compressible fluids, the real gas
limited number of combinations of parameters, while software diffusivity equation was repackaged into something that
built-in models could generate the exact solution for any looked like the slightly compressible fluid diffusion 1. This was
combination. (2) They would typically be drawdown done by gathering anything other than the diffusivity into a
responses, when software built-in models would generate the pseudopressure m(p):
exact multirate solution. The integration of the derivative in
the type-curves extended their use, but only to a point, as p p
m( p) = 2 dp .(1)
( p )Z ( p )
derivative type-curves would apply poorly to shut-ins after
0
short/complex productions and to closed systems. (3) For the
initial estimation of each parameter, a software would offer
interactive capabilities that far exceed in accuracy the visual A typical pseudopressure versus pressure response is shown in
selection within a family of curves: You locate exactly the Figure 1. Then the real gas diffusion equation becomes:
double-porosity transition, or the time the derivative takes off
due to boundary effects, and the computer just calculates the m( p ) k
parameter(s). (4) After the initial simulation, computer =A 2 m( p ) ..(2)
generated model parameters can be modified, interactively or t c
by nonlinear regression.
where A is a unit dependent constant. So this looked a lot
Old-style type-curves can still be found here and there, like the slightly compressible diffusivity equation, but the big
usage is seldom but not completely gone. In a quick survey leap of faith was to consider the diffusivity to be constant. As
(2002) run by the author on are type-curves history?, answer shown in Figure 2, the diffusivity is actually anything but
ranged from yes of course to no of course not. If the true constant, and it even tends to zero with the pressure.
answer is no, then they should be retired, as they are
outperformed on all accounts and should be replaced by Once the assumption of constant diffusivity was accepted
computer generated models. This replacement is not just an (and sometimes too quickly forgotten), the inner boundary
evolution, as each method follows a different logic. conditions were adapted and it was possible to match the gas
pseudopressure response on the type-curves generated for
With computer generated models, one will try to keep the slightly compressible fluids. In the calculation, the value of c
original data as much as can be, with the minimum left outside the m(p) definition would be calculated from a
assumptions, and adapt the model to honor the observed carefully chosen reference pressure. The engineer would
response. Models are no longer restricted to the superposition transform pressures into pseudopressures, plot and match the
of dimensionless analytical solutions inherited from type- response on a type-curve, using specific equations to calculate
curves, and they can reach the details of a complete the results from the match point. This was a very reasonable
multiphase simulator run. methodology at the time, as (1) on a typical gas test the
assumption of fairly constant diffusivity would not be too bad,
With type-curve matching, the objective was just the (2) there was no alternative for type-curve matching.
opposite. With such limited flexibility, one would do anything
to correct the data and get closer to the fixed dimensionless Should the flowing pressure be too low (e.g. for tight gas
responses. So new tools and techniques were developed: reservoirs), the variable part of the diffusivity would be
Agarwal equivalent time would be used to project a build-up integrated in a time function called pseudotime2 (to be retired
response on a drawdown type-curve; deconvolution would be later in this paper).
attempted (and would generally fail) to convert a multirate
into a drawdown response, or to reduce wellbore storage Because we are nice people and the numerical range of
effects; pseudopressures would be used to match real gas pseudotimes and pseudopressures were ridiculous, it was also
responses on type-curves generated for slightly compressible decided to normalize them in order to look like time
fluids; for tight gas reservoirs, pseudotimes would be used to and pressure.
correct the early time shut-in response; another pseudotime
function based on the average reservoir pressure would be
developed for limit tests or the use of production data; etc.
WHY WE SHOULD STOP USING PSEUDOPRESSURES AND OTHER GOOD OLD WELL TEST
SPE 77619 INTERPRETATION TOOLS AFTER SO MANY YEARS OF GOOD SERVICE 3

What we do today with pseudopressures The origin of the error was again the diffusivity term left in
When well test interpretation was transferred to computers, the Equation 2. Figure 4 shows the comparison, for a long gas
procedure for gas tests was kept. So here we are today, in production and a short build-up, between a simulation and
terms of agreed methodology: analytical model using pseudopressures. The average pressure
calculation from material balance clearly indicates that the
PVT first. The pseudopressure (and pseudotime) functions are simulator is right. This is not surprising, as a simulator will
defined at the beginning of the interpretation process, from systematically adjust its results at every time step to honor
PVT correlations or tables. material balance.

Extracting a period. When a period (typically a build-up) is In Figure 5 we see the same comparison on a log-log plot
extracted for analysis, pressures are converted to of pseudopressures vs. time. When time goes on, average
pseudopressures, and the log-log plot shows the change of pressure goes down and therefore compressibility goes up,
pseudopressure (normalized or not) and the derivative as a requiring less pressure drop to produce the same fluid. So the
function of the time or the pseudotime (normalized or not). real pressure drop is less than the one simulated
The semilog plot will also show pseudopressure vs. with pseudopressures.
superposition time.
Our answer to the angry call was that, well, the well test
Model generation. From the initial pressure pi, the program packages were wrong but had not really been designed for
calculates an initial pseudopressure mi=m(pi). The model also this. But our engineer did not care less: "I have rates, I have
requires a value of diffusivity, i.e. a reference value of c, that pressures, I want to make an analysis, so I use a well test
will be used in the superposition. The reference pressure may package". Sure...
be calculated by the software (it could be for instance pi or
p*) or controlled by the engineer. The simulation is then First we thought we had just met what we politely call an
performed in terms of pseudopressures. outlier, but indeed the issue of material balance has become
more and more frequent. With the development of downhole
Matching the data. Simulated pseudopressures are extracted permanent gauges, it is legitimate to try to get more
and compared (with derivative) to data pseudopressures. information from this data, in a process between a simple tank
Simulated pseudopressures are converted back to pressures to calculation and a full simulation model. With this in mind,
compare the model and the data on the pressure history plot. using the analytical and numerical tools that were developed
in transient testing makes sense, as long as the assumptions
A typical diagram of the calculation is given in Figure 3. that were the base of our methods are carefully reviewed.
The term analytical engine integrates the dimensionless
solutions (Laplace or real), the conversion to physical values, Material balance with type-curves and pseudotime
the superposition in time due to the well(s) production history This problem and the principle of its solution were known for
(ies), and the superposition in space of the influence of a long time8, but recent papers9 have explicitly addressed
multiple wells. them. The initial reading of these papers would be pretty
depressing for a software vendor, as it would mean writing an
A lot of things are wrong in this process, or at least not umpteenth specialized plot, man-months of additional
optimized. We will see what later, but first we will consider development, complex interface, etc. In essence, the method
the most significant problem: the assumption of constant suggested by these papers is the following:
diffusivity in Equation 2. - Store the model as a type-curve
- Calculate Gi from the estimated size of the reservoir and
Gas material balance and pseudopressures the initial reservoir pressure
Not so long ago we received the first angry call complaining - At any time step, calculate the cumulative production Q(t)
that a well test software would not honor material balance - From p/Z calculation, calculate the average pressure
(material what?). The engineer had entered several years of - In the superposition of the model, use the diffusivity
gas production, matched them with a closed rectangle using corresponding to the calculated average pressure
two different well test packages and a simulator, and got three - At the considered time, stretch the data response with the
different reservoir sizes. normalized pseudotime corresponding to the c at
average pressure
- Reiteratively correct Gi and the model until a suitable
match is found
4 O. P. HOUZE SPE 77619

The method above effectively solves the problem of material And that is all. The comparison between the pressure
balance, but with two important shortcomings: coming from the numerical simulation and the use of material
balance correction will not even be shown here, because there
Model vs. material balance. In these publications, the is just no detectable difference. Problem solved in literally
calculations and corrections for material balance are done four lines of additional source code.
independently of the model content, that is considered as a
mere type-curve (hence not able to integrate the specific Post-mortem on this material balance problem
elements of material balance, i.e. size, PVT and production). Before we get into the final plea to retire pseudopressures, we
This is inevitable when dealing with type-curves, but can use the case of gas material balance to highlight a
absolutely wrong for a computer generated model. In the few points.
interpretation process, we do not know in advance the size of
the reservoir, but each closed system model knows its size, A positive way to present the contribution of
and can make a material balance correction coherent with pseudopressures would be that the pseudotime correction for
itself. In the optimization process, the non linear regression material balance is improving the accuracy of the solution
can then change the reservoir size (hence Gi) with no risk of when average pressure declines. Actually, it is wrong. The
inconsistency, as every intermediate model will be coherent pseudotimes (or the material balance correction of the model)
with its own Gi, hence honor material balance. are in fact correcting an error created by the pseudopressures.
This can be seen in Figures 4 and 7.
Over-complexity. All this methodology amounts to
considering that, at a given time, we should use the average In Figure 4, we see that the pressures generated by the
pressure as the reference pressure for the model calculation. simulator are fairly linear at late time, similar to the pseudo-
This statement is the kernel of the solution, and it indeed steady state of a slightly compressible fluid. It is the analytical
solves the problem. But there was no need to implement this model created from pseudopressures that is bending
complex and user hostile type-curve technique, while it could downwards. Why? The answer is in Figure 1. The
just be integrated as a simple correction in the computer pseudopressures behave like p at low pressure. When the
generated model, at a fraction of the computation and pseudopressures simulated by the model reach Pseudo-Steady
development cost, and with no need for specific State, they become linear with time, and then the
post-processing. corresponding pressures will inherit the bending of Figure 1
and will not follow a straight line.
Material balance correction within a model
In the following, a model is not just a dimensionless (tD,pD) It would be OK if there were a physical explanation for
type-curve, but the physical model that solves (analytically this. There is none, this behavior is a pure artifact coming
here, numerically in other cases) the problem as defined by the from the pseudopressures. In the real diffusion equation solved
user (e.g. a rectangle of this size, a reservoir of this by the simulator, the non linearity of the terms grouped in the
permeability, fluid with this PVT, etc). So the model is what pseudopressures are mostly compensated by the non linearity
the analytical engine defined above can produce. As for the of the c left outside, and globally the pressure behavior is
type-curve technique, the core idea is to re-evaluate the close to a linear response.
average pressure, at every time step.
In Figure 7, we see the same log-log responses for a
In 1996 Bourgeois et al10 suggested to add to a closed system simulation and an analytical model without material balance
solution (using pi as the reference pressure) an additional correction, but this time we are displaying pressures, not m(p).
pseudopressure change for material balance correction. This The (true) simulated pressures follow a unit slope, while the
correction would compensate the gap between the analytical analytical model without material balance correction
and the numerical models in Figure 4. This is indeed correct, bends upwards.
but could be even simplified, by using at any time step the
average pressure (calculated from material balance) as the new In conclusion, what we did was not to improve the
reference pressure. approximation of pseudopressures, but really to correct an
error that was created when we split the real gas diffusion
The new calculation diagram is shown in Figure 6: equation into two groups, the pseudopressure and the
- At the beginning of the simulation, the model considers its diffusivity, solved for one while keeping the other one
size, the initial pressure and the PVT, and calculates Gi constant. Then all the bright things we have done were to
- At every time step, the model calculates the cumulative correct this initial error.
production Q(t), and gets the average pressure from p/Z
- For the calculation of the simulated pressure at this time,
the corresponding average viscosity and compressibility
are used
WHY WE SHOULD STOP USING PSEUDOPRESSURES AND OTHER GOOD OLD WELL TEST
SPE 77619 INTERPRETATION TOOLS AFTER SO MANY YEARS OF GOOD SERVICE 5

Why we should retire pseudopressures Incoherence with numerical models. With the development
The author sees four main reasons for this: of numerical well test models, we end up with a ridiculous
process. The simulator will solve the real gas diffusion
Non linearity. First let us dare challenge a statement equation, with of course no need for pseudopressures. The
everybody accepts: the pseudopressures have not been PVT used may be much more refined than what is input in the
designed to linearize the real gas diffusion, and they are not m(p) calculation. If we use the pseudopressure as the default
doing that. Equation 2 is anything but a linear equation, as the diagnostic scale, then, to compare the data and the numerically
factor left outside is anything but a constant. The equation simulated pressures, we will have to transform the simulated
becomes linear when you assume that the diffusivity is pressures into pseudopressures, potentially unrelated to the
constant, and the above shows how wrong this assumption can PVT used in the simulator! A bit like scratching your left ear
be. Another way to illustrate this is by considering Figure 1, with your right toe.
i.e. the relation between p and m(p). At high pressures, it is a
linear relation, and therefore Equation 2 does not add any Suggested new methodology
linearity to the problem. The only case where m(p) would The essence of the suggested methodology is:
eventually linearize the diffusion would be in the curvy part of - Replace m(p) by p in the diagnostic plots
Figure 1, i.e. at low pressure. Bad news, it is exactly where the - Transfer the PVT information to the model
pseudopressures turn out to be completely incorrect. - Use m(p) in the analytical engine, nowhere else
No, as stated before pseudopressures were designed to
formulate the diffusion equation in a way that would look The suggested workflow for the generation of an analytical
like the slightly compressible fluid equation, hence establish model is provided in Figure 8. The physical description of the
the connection with the type-curves developed for these model, the parameters and the PVT used are integrated in the
slightly compressible fluids. Later, the same process was used model description. An analytical model becomes, as in a
to allow analytical models to be used on gas tests. It did, and numerical model, a black box that gives pressure vs. time.
still does, the job relatively well. But this is done in the
calculation process, the analytical engine not accessible to Because the PVT is now part of the analysis (and not part
the user. of the pre-processing), the user may actually try and compare
several PVT options, exactly as he/she would compare
No specific relevant behavior. Using m(p) instead of p in the different models or model parameters. In a given analysis, the
diagnostic plots brings absolutely nothing. Scales apart, at user might change the PVT as he changes the other
high and medium pressure the responses look strictly identical, parameters, or duplicate the analysis, then change the PVT,
and at low pressures, when the relation is not linear, the and with nonlinear regression checks how the refinement of
pressure is right and the pseudopressure is wrong. In other the PVT affects the results, or ultimately the whole diagnosis.
words, if we were showing pressures and not pseudopressures, The problem of receiving a new PVT description months after
this would not change anything in our diagnosis process. the initial analysis then becomes easy to handle.

We do not match de data. When you transform pressures A multiple analysis of the same data would then look like
into pseudopressures, you are actually modifying the data Figure 9. Under this new logic, and with the absence of the
assuming a PVT. There is a philosophical issue here. When constraints coming from pseudopressures, the well test
you transform data, whatever error you are doing in the package can receive, with no specific processing, the solutions
process will be irreversible and will add to the uncertainty of generated by its own built-in analytical and/or numerical tools,
the final interpretation. Of course, we need at a point to do or import the result of a third party simulation or home made
some assumptions on the PVT, but doing it upfront means that models, via files or D-COM type connection.
you are stuck with it for the rest of the interpretation. If later Pseudopressures subsist within the number crunching part of
you change your PVT assumption, then you will change the the built-in analytical models, but they have disappeared from
pseudopressures, hence your target data. In a typical gas well the user interface.
test, the first interpretation will be done with a very rough
PVT knowledge (separator gravity and a choice of Partial retirement, then
correlation). If, a few weeks later, a refined PVT analysis has So, actually, the author lied in his abstract. In the suggested
been done, then the only alternative will be to change the methodology, the pseudopressures are not completely retired,
pseudopressures and to do the analysis again. There is a much as they stay in the internal calculations of the analytical model.
better solution (see later). But this is transparent to the user, a little like image wells are
internally used for boundary models without being
explicitly displayed.
6 O. P. HOUZE SPE 77619

Nothing prevents us displaying the pseudopressures as a Wellbore pressure correction. Typically for tight gas
side facility, but they would be kept out of the core reservoir, the assumption of constant wellbore storage turns
methodology, as complementary information. out to be incorrect. As the gas compressibility would
dramatically change in the wellbore, it was decided to
We could even question keeping pseudopressures inside integrate the c in the time function in order (mainly) to
the analytical model calculations. After all, the constraint to account for changing storage, and therefore stretch the real
map the slightly compressible fluid equation is gone with the pressure response, on the time scale, to give it back the shape
type-curves, and we have seen that the grouping of of a constant wellbore storage type-curve. The pressure used
pseudopressures is not really ideal to linearize the equations. in the pseudotime calculation was the wellbore pressure.
So we might re-consider the overall algorithms used in the
analytical engine and completely remove pseudopressures. But So both have the same goal of reducing the inaccuracy of
this may not be worth the research effort. the analytical model, but the first one is a high frequency
correction to handle wellbore effects, the other one is a low
For any practical purpose, when we compare numerical frequency correction, for limit tests or the analysis of
and analytical models, both match pretty well except at late production data. And both are incorrect, because
time for this material balance issue. After material balance
correction there is no detectable difference. Of course there are There is a qualitative difference between the
some differences. Even after correction the analytical model pseudopressures, that are only a function of pressure
does not rigorously solve the nonlinear diffusion. But, (applicable at any point of the reservoir), and pseudotimes,
considering the signature of the data we interpret with these that are not a function of time only, but of both time and
models, the approximation is pretty reasonable. So the pressure. The correction, that uses a single pressure at a given
pseudopressures may as well stay in the analytical model. time, is not valid at any point in the reservoir, because
pressure is not uniform. So the pseudotime is a local
It is also the authors conviction that, for real gas tests, correction, an approximation that may have side effects. You
analytical models will be quickly phased out and replaced by cannot be right everywhere, and there is no magic way to
numerical models, where no assumption is made, and non transform our analytical model into a rigorous solution.
linearities overlooked for years are at last taken into account.
The gridding technology is ready, and these models are fast The alternative to material balance pseudotime has already
and reliable enough. Problems remain with multiphase flow, been developed at length in this paper. For wellbore storage
but no more for single-phase. So, why should we stay pseudotime, we see two alternatives: (1) seamlessly integrate
analytical when solving a nonlinear problem? the pseudotime correction in the analytical model, but keep the
time in the model output, (2) when the use of pseudotime is
Why we should retire pseudotime aimed at turning changing wellbore storage into constant
First, if we retire pseudopressures, it would look pretty silly wellbore storage, just use a pressure related wellbore storage
keeping pseudotimes Beyond this, it could be interesting to model, strictly linked to the gas compressibility.
review why pseudotimes were introduced, and how the
problems they addressed could be handled under the Figure 10 shows a typical tight gas response, the transform
suggested methodology. of this response using pseudotime, and the match with a
changing storage solution with no pseudotime correction.
First, we have noticed confusion between the traditional
pseudotime and the pseudotime introduced to handle material Other changing wellbore storage models simulate a time
balance problems. These methods have a completely different related transition between early and late storage. They are fast
range of application, but both were developed to handle the c and can match anything, but they have no physical basis,
left outside the pseudopressure equation when the changes and they are time related. In Figure 11, we consider the
cannot be ignored. pressure related wellbore storage of a gas well. We call C1 the
wellbore storage calculated for the last flowing pressure, and
Average pressure correction. For long-term production, the C2 the wellbore storage calculated at final shut-in pressure.
goal was to permanently adjust the same c, from a p/Z The build-up response will show C1 at early time and C2 at
calculation, in order to honor material balance. late time. If you take the sequence of a drawdown and a build-
up, the build-up may be perfectly matched with C1 then C2,
but the drawdown simulation will also show C1 then C2,
while, physically, it is the opposite. This does not rule out time
related storage models, but they must be used with caution.
WHY WE SHOULD STOP USING PSEUDOPRESSURES AND OTHER GOOD OLD WELL TEST
SPE 77619 INTERPRETATION TOOLS AFTER SO MANY YEARS OF GOOD SERVICE 7

An objection to the removal of pseudotime is that we are Constant composition assumption. This is certainly the most
loosing any chance to restore a nice looking constant wellbore serious question. As there is a unique relation between p and
storage response. Actually, from experience, pseudotimes do m(p), this implies that we have the same fluid composition at
not compensate fully the strange shape of the early time any time and any point of the reservoir. For a given pressure,
response. And, after all, it is the REAL response, and we there may be several phases present, but it will always be the
might as well use the model to simulate this strange but same relation. The different relative perms integrated in the
real shape. pseudopressures have an effect on the total mobility, but an
analytical model with pseudopressures will not be able to
Multiphase pseudopressures simulate the cumulative effect of these different mobilities,
The origin of multiphase pseudopressures11 is the same as for i.e. the segregation of the components within the reservoir. For
single-phase: an attempt to get closer to the slightly example, a gas condensate pseudopressure will not simulate a
compressible fluid equation. In the process, there is a condensate banking. It will just show a change in the
factorization with respect to the reference phase (gas for dew mobility at every point where the pressure is below dew point.
point fluids, oil for bubble point fluids). The reservoir
permeability is also factorized, while the phases relative So we have to be very careful with these pseudopressures.
permeabilities are integrated in the pseudopressure. For a They may work great (especially to combine a multiphase
given pressure, the program will calculate the different surface production into a single-phase downhole rate), but
saturations, hence the relative permeabilities of each phase and they carry too many approximations to just use and forget.
their physical properties. This leads to the calculation of a An additional user-related problem is that they require very
pseudopressure function that will integrate all phase detailed PVT and relative permeability information, and one
components. Again, only the total diffusion is left in the worrying risk is that the user takes the default of whatever
equation, the rest is packaged in the definition of the he/she does not know
pseudopressure. Assuming again that the diffusion is a
constant, we can now solve the problem as for a slightly The authors position is that software vendors should
compressible fluid. remove them from their default installations, unless and until
they are eventually formally accepted and required by the
A typical pseudopressure response is shown in Figure 12, industry. There is no general agreement on them, and software
in the case of a gas condensate. Above dew point and at low vendors are not there to take the risks of spreading a
pressure, the behavior is the same as the single-phase m(p). potentially improper methodology. The trouble is that there is
When flow gets below dew point condensate drop-out occurs no reliable alternative at present, and for example the
evidenced by a bending of the pseudopressure curve. interpretation of gas condensate tests remains an
open question.
The principle of these multiphase pseudopressures has
been published in the litterature, but never really accepted by Conclusions
the industry as a standard. Their implementation in
commercial software is more an initiative from a couple of Type-curve matching is technically history, and was
software vendors than the result of a consensus in the industry. replaced by matching computer generated models on data, a
method still improperly called type-curve matching. With
The reason why these pseudopressures were developed type-curves, we would modify the data to fit the shape of the
makes sense: until we have an automatic, reliable and fast curve. Now we refine the model to match the observed data.
numerical solution for multiphase well tests (and we are not Methods that change the data upfront, such as the use of
there yet), all we have so far is Perrines method12, that carries pseudofunctions, should now be questioned.
numerous approximations. The purpose of the multiphase
pseudopressures is to reduce the number of these Gas pseudopressures do not linearize the real diffusion
approximations. These pseudopressures will for sure be as bad equation but merely package it in a way that looks like a
or less bad than Perrines. The question is: is it enough, and, if slightly compressible fluid problem. Assumptions are wrong,
it is not enough, is the whole process relevant? but they are partly compensated by using pseudotime.

Of course, the same limitations as for single-phase gas But these pseudofunctions uselessly constrain the analysis
pseudopressures will apply. We will not develop them again to one PVT model, they do not bring any significant signature
here. We see two additional major problems: that pressure would overlook, and they are wrong at low
pressure. When numerical simulations are run in parallel,
Sensitivity. They are very sensitive to the relative converting simulated pressures to pseudopressures is an
permeability tables. But this after all is not a specific issue, as irrelevant process. So pseudofunctions should be removed
any numerical models will have the same problem. from the diagnostic plots.
8 O. P. HOUZE SPE 77619

In the analytical model generation, pseudopressures may References


be internally kept. Pseudotimes may go for good: when used 1. Al-Hussainy, R., Ramey, H.J., Jr., and Crawford, P.B. : "The
to address material balance issues they are advantageously Flow of Real Gases Through Porous Media", J. Pet. Tech., May,
replaced by a material balance correction within the model, 1966, pp 624-636
and a pressure related changing wellbore storage model could 2. Agarwal, R.G. : Real Gas Pseudo-Time A New Function for
Pressure Buildup Analysis of MHF Gas Wells, SPE 8279, paper
replace the traditional pseudotime. presented at the 1979 SPE Annual Technical Conference and
Exhibition held in Las Vigas, NV., September 23-25, 1979.
Multiphase pseudopressures were developed with the same 3. Horner, D.R. : Pressure Buildups in Wells, Proc. Third World
idea to get a slightly compressible fluid equation. They have Pet. Cong., E.J. Brill, Leiden 2 , 1951, pp 503-521.
the same limitations as the gas pseudopressures, and much 4. Agarwal, R.G., : A New Method to Account for Producing Time
more (sensitivity, constant composition). Only specialists Effects when Drawdown Type Curves are Used to Analyze
should use them, and with caution. Unfortunately there is no Pressure Buildup and Other Test Data, SPE 9289, paper
real alternative. presented at the 1980 SPE Annual Technical Conference and
Exhibition held in Dallas, TX, September 21-24, 1980.
5. Bostic, J.N., Agarwal, R.G., and Carter, R.D.: Combined
The original relevance of these good old tools is Analysis of Postfracturing Performance and Pressure Buildup
absolutely not questioned. But we are, collectively, very slow Data for Evaluating an MHF Gas Well, Jour. Pet. Tech., Oct.
to assess when a new method should replace an older one and 1980, pp 1711-1719.
not just needlessly accumulate tools for the sake of it or 6. Bourdet, D., Ayoub, J.A., and Pirard, Y.M. : Use of Pressure
through inertia. Derivative in Well-Test Interpretation, SPE 12777, paper
presented at 1998 California Regional Meeting held in Long
Todays interpretation engineers are multi-disciplinary, Beach, April 11 14, 1984.
and they use computers. There is apparently no risk of massive 7. Rosa, A.J and Horne, R.N. : Automated Type-Curve Matching
power failure that will shut down all computers and will force in Well Test Analysis Using Laplace Space Determination of
Parameter Gradients, SPE 12131, paper presented at 1983 SPE
them to go back to millimeter paper and printed type-curves. Annual Technical Conference and Exhibition held in San
So we must turn the page, and focus on what is important: (1) Francisco, CA, Oct. 5-8, 1983.
an optimized methodology, adapted to todays tools and needs, 8. Blasingame, T.A. and Lee, W.J. : The Variable-Rate Reservoir
cleaned from obsolete techniques, (2) proper training and (3) Limits Testing of Gas Wells, SPE 17708, paper presented at
common sense. 1988 SPE Gas Technology Symposium held in Dallas, TX., Jun.
13-15, 1988.
Acknowledgement 9. Gardner, D.C., Hager, C.J., Brown, T., and Agarwal, R.G.:
The author wishes to acknowledge Chih Chen, Kevin Siggery Incorporating Rate-Time Superposition into Decline Type Curve
and Olivier Allain for helping in the review of this paper. He Analysis, SPE 62475, presented at the 2000 SPE Rocky
Mountain Regional/Low Permeability Reservoirs Symposium,
also wants to acknowledge A, B and C, who also helped but
Denver, CO., March 12-15, 2000.
would not want to ruin their reputation. 10. Bourgeois, M.J. and Wilson, M.R. : Additional Use of Well Test
Analytical Solutions for Production Prediction, SPE 36820,
Nomenclature paper presented at 1996 SPE European Petroleum Conference
held in Milan, Italy, Oct. 22-24, 1996.
A = conversion ct, unit dpendent 11. Jones, J.R. and Raghavan, R. : Interpretation of Flowing Well
c = total compressibility, Lt2/m, kPa -1 [psi-1] Response in Gas-Condensate Wells,, SPE 14204, paper
m(p) = gas pseudopressure, m/Lt3, kPa2/cp [psi2/cp] presented at the 1985 SPE Annual Technical Conference and
p = pressure, m/Lt2, kPa [psi] Exhibition held in Las Vegas, NV, Sept. 22-25, 1985.
12. Perrine, R.L. : Analysis of Pressure Buildup Curves, Drilling
t = time, t, hr
and Production Practice, American Petroleum Insitute, 1956, pp
Z = gas compressibility factor 482-509.

Greek
= Gradient operator
= Differential
= Viscosity, m/Lt, cp
= porosity, fraction
WHY WE SHOULD STOP USING PSEUDOPRESSURES AND OTHER GOOD OLD WELL TEST
SPE 77619 INTERPRETATION TOOLS AFTER SO MANY YEARS OF GOOD SERVICE 9

pi
p from p/Z
analytical
simulation

wrong
0

Figure 1 Typical Pseudopressure vs. Pressure Figure 4 Material balance correction (linear plot)
pressure vs. time

analytical
simulation

Figure 2 Typical Diffusivity vs. Pressure Figure 5 Simulation vs. analytical with no M.B. correction
log-log plot pseudopressure vs. time

pi pref linear
linear
p(t)model PVT match
match m(pi)
PVT PVT pi table
table table Gi
p(t)model
PVT p(t) c(t)
table Q(t)
m(pi) (c)ref m(t)model
t analytical
analytical log-log engine
t m(t)model
engine match log-log
match

Figure 3 analytical model generation, using pseudopressures Figure 6 analytical model generation, using pseudopressures
without Material Balance correction with Material Balance correction
10 O. P. HOUZE SPE 77619

analytical original data (pressure & derivative)


simulation data after pseudo-time correction
changing wellbore storage model

Figure 7 Simulation vs. analytical with no M.B. correction Figure 10 Tight gas reservoir response with and without
log-log plot pressure vs. time pseudotime, and changing wellbore storage model

Gas PVT tables, initial pressure C2


C1
well(s) production(s) C2
wrong
well(s)+reservoir+boundary(ies) model

log-log
init m(pi) + (closed system) Gi match
+
semi-log
match
material balance +
correction linear
match
closed
system analytical wrong
t m(t)model p(t)model
engine C2
C1
C1

Figure 8 Suggested analytical model workflow

Figure 11 Pressure related wellbore storage


Errors with time related wellbore storage models

Analysis 1 m 1 (p)
analytical - PVT1 - model1 - parameters1
p1(t)

m 2 (p)
Analysis 2
analytical - PVT2 - model2 - parameters2 p2(t) log-log
dew
match
+ liquid
Analysis 3 m 3 (p)
semi-log
t analytical - PVT3 - model3 - parameters3 p3(t) match
drop-out

+
Analysis 4 linear
numerical - PVT4 - description4 p4(t) match

Analysis 5
numerical - PVT5 - description5 p5(t)

Figure 9 Multiple analytical / numerical analyses

Figure 12 Gas condensate pseudopressure vs. pressure

S-ar putea să vă placă și