Summary
With the common use of highly sensitive electronic pressure recorders and the widespread use of reservoir characterization by type-curve
pressure-derivative techniques, preprocessing data has now become
an important prerequisite to pressure-transient analysis. In almost all
cases during a well test, the electronic pressure recorders measure a
multitude of effects other than the intended reservoir pressure transients. For example, multiphase, geotidal, microseismic, changing
liquid levels, recorder drift, recorder plugging, etc. can mask the reservoir transient. If not properly identified, these effects could easily
be misinterpreted as reservoir characteristics, such as pressure depletion, boundaries, or dual porosity. The preprocessing of the digital
data, starting with the comparison of pressures from multiple recorders and ending with data filtration to remove the noise associated with
the signal, is a significant component of well-test interpretation and
is a prerequisite to pressure-transient analysis.
From experience with thousands of well tests, we have chosen
several examples to illustrate the need for preprocessing the data.
The observed pressures must be corrected to remove the nonreservoir effects before pressure-transient analysis is attempted. Often,
diagnosis of the observed data can only be done with the help of information other than the pressure trace.
Introduction
During the 1980s, two major developments took place in the field
of pressure-transient analysis. These were the (1) common use of
very sensitive electronic pressure recorders and (2) sophisticated
analytical solutions available at the analysts fingertips.
1. Some electronic recorder manufacturers claim a pressure resolution of 1 part per million. This is equivalent to saying that pressure
changes of less than one hundredth of a psi are readily detectable with
many of the high-end recorders. The reliability of these electronic
recorders has improved to the point where the chances of failure are
no greater than by using mechanical gauges. Some electronic gauges
have a surface read-out of the data, which makes the number of data
points available virtually limitless (a reading every second if desired).
Most memory gauges can record 20,000 and many 40,000 or more
time/pressure/temperature data points. Compare this to the approximately 100 data points obtainable from mechanical gauges. Electronic recorders present a plethora of tens of thousands of pressure data
with unprecedented resolution and at affordable costs.
2. Many sophisticated analytical and semianalytical solutions have
been developed for analyzing complex reservoir situations, such as
multilayers, finite-conductivity hydraulic fracture with skin on the
fracture face, dual-porosity systems, horizontal wells, etc.
The dramatic increase in computer power has kept pace with, and
often promoted, the development of these complex reservoir models.
With the proliferation of personal computers, every well-test analyst
can have ready access to these sophisticated methods of analysis.
These two phenomena have synergized each other to the point
where many people think that the use of very accurate electronic
gauges, coupled with complex modeling, can identify hitherto undefinable reservoir characteristics. This drive towards complex
analysis has also been fueled by the trend in advanced reservoir
engineering towards reservoir characterization.
Copyright 1996 Society of Petroleum Engineers
Original SPE manuscript received for review Nov. 4, 1992. Revised manuscript received April
9, 1996. Paper peer approved April 10, 1996. Paper (SPE 24729) first presented at the 1992
SPE Annual Technical Conference and Exhibition held in Washington, DC.
120
the liquid level moves down from RRD to MPP. After t2 , the liquid
level is below MPP.
In a situation like this, the RRD pressure must be corrected to
MPP before it can be used for pressure-transient analysis. Before
time t1 , the correction is 100 m the liquid gradient (100 10
+1000 kPa). After time t2 , the correction is 100 m gas gradient
(100 1+100 kPa). If these corrections are applied, the RRD pressures become equal to the MPP pressures, and what was a crooked
pressure trace becomes a straight line and validly analyzable by
pressure-transient analysis.
Does this really happen, or is this an intellectual exercise? Experience with thousands of well tests has shown that approximately
three quarters of the wells flowing a mixture of gas and liquid exhibit this phenomenon. For example, many oil wells that flow oil (and
errors arise out of the internal voltage change and component temperatures, as well as interpolation routines used with external calibration table.
There are many causes for inconsistencies between recorders.
Some are related to wellbore hydraulics or known recorder characteristics, such as resolution and drift. However, many others remain
unexplained. An example is Fig. 15, which is Problem #4 from the
SPE Petroleum Showcase at the 1992 Petroleum Computer Conference in Houston. Whereas the two recorders track each other within
of a psi throughout most of the test, there is an alarming difference
in the trend during the last 4 hours of shut-in. The bottom recorder
pressure is increasing at a rate of 10 times that of the top recorder.
The cause of this rather sudden change at 168 hours is undetermined.
Veneruso et al.8 have studied the performance of electronic
gauges using transfer functions. Some of the effects they discussed
were short term drift; long term drift; errors owing to temperature
gradient; noise; calibration and aging of tranducer; hysteresis owing
to temperature and pressure cycling; signal processing errors; and
adiabatic sensitivity. They also give some examples of comparative
gauge performance and discuss pressure gauge overshoot caused by
a step change in pressure.
External Factors
Other than transients imposed at the well during a test, there are
many external factors that affect the recorded pressure. The most
documented one is that of geotidal effects in which the rise and set
of the moon is observed in these very sensitive gauges. An example
is shown in Fig. 16. This is part of the buildup test in a high permeability oil well. We have observed other cases in which the peak to
trough magnitude of the geotidal effect was 1 psi (7 Kpa).
There are numerous cases, however, for which there is no known
explanation. Such an example is shown in Fig. 17, which shows two
124
recorders in an observation well as part of a pulse test. Several observations are in order.
1. One is a strain gauge and the other a quartz gauge, yet the responses are identical in every detail. This is a glowing tribute to the
accuracies and resolutions of these two gauges.
2. The general declining trend does not correlate with the pulse
test procedure and has to be attributed to external factors (such as
a declining regional aquifer pressure?).
3. The observed spikes correspond in every detail between the
two gauges. Therefore, these effects actually took place and are not
gauge malfunctions.
4. Some of the spikes are of the order of 60 kPa (10 psi) and are
orders of magnitude larger than the resolution of the two gauges.
caused by a reservoirs well-test transient. This follows from the engineering gut feeling that the largest effect will be observed first, and
the strength of the transient decreases with time. Figs. 18 and 19
show synthetic data for two modelsa dual porosity reservoir, and
a homogeneous reservoir with wellbore phase redistribution. Figs.
18A and 19A show the strip chart. A simple visual inspection of Fig.
19A shows the wellbore dynamics in action at 10 to 15 hours. Figs.
18B and 19B, which portray the derivative (semilog derivative), do
not demonstrate any shape difference in the derivative between the
two effects. However, the PPD7 (dp/dt) shown in Figs. 18C and 19C
clearly identifies the wellbore dynamics by the increasing PPD.
3. By its very nature, the reservoir acts as a filter for a signal. This
is confirmed by the behavior of the diffusivity equation. This means
Fig. 18CPPD
125
Fig. 19CPPD.
sometimes this remedy can be worse than the disease. In the vicinity
of abrupt changes, some of the smoothing algorithms can get out of
hand. Fig. 22 shows the raw data and the smoothed data supplied to
us by a service company. Their algorithm did a fine job of smoothing
the data everywhere except in the vicinity of the shut-in time. Many
smoothing algorithms work very well for most well-behaved data
but none work automatically for all the data. Smoothing of data involves a lot of judgement and indeed forms part of the well-test interpretation process. It is only another step in a series of judgement
calls. It is not sufficient to use purely statistical procedures for interpreting practical well-test data. The use of the well-known least
squares method for minimizing errors can sometimes lead to incorrect analysis. For example, Fig. 23 shows how a few erroneous, outlying, data points can skew the least squares fit away from the true
observed trend.
Some effects that have a periodicity can be filtered using Fourier
transform or similar procedures. In other instances, some people
have developed equations or tables to calculate the geotidal effects
and thus correct the pressure data accordingly. Often a simple logarithmically spaced sampling of the data results in sufficiently
smoothed pressures for pressure-transient analysis. Many computer
programs also have procedures for smoothing the derivatives. In the
end, it should be the analysts judgement that determines whether a
particular procedure has achieved the desired smoothing.
Some oil companies are now aware of the inconsistencies between gauges and the limitations of electronic recorders. They have
issued Recorder Performance Requirements in which they detail the
quality of acceptable pressure data and some criteria that the recorders must meet. For example, they are requesting plots showing the
difference between recorders, overlay plots of all recorders, exploded plots to show the resolution of each gauge, comparison of apparent to specified resolution, calibration details, etc.
A minimum of two gauges must be used during a test. Some companies will use three gauges on a routine basis. In some tests we have
even had ten gauges in the wellbore! There are many companies that
run two gauges but only look at one. The author has known this to
happen even for some very high-profile wells.
Conclusions
Data processing has become a significant component of well-test interpretation and must be viewed as an integral part of the process.
This processing must incorporate all relevant external information
and involves its own level of judgement and interpretations. Data
filtration and reduction is an interpretive process and can affect the
results of the subsequent pressure-transient analysis.
1. Well-test interpretation is more than just pressure-transient
analysis.
2. Wellbore dynamics are constantly at work in conjunction with
reservoir transients.
3. The wellbore configuration can affect the shape of the response
curve recorded on pressure recorders.
4. Insist on being supplied with the raw, not smoothed, data. If the
data needs to be smoothed, do it yourself, using your own judgement
and knowledge of pressure transient behavior, as well as wellbore
dynamics.
5. If it happens abruptly, it is not a reservoir effect.
SPEFE
127