91 / 743
Dynamic Data Analysis – v5.12.01 - © KAPPA 1988-2017
Chapter
3 – P ressure Transient Analysis (PTA)- p91/743
3.D.7
Conclusion
The recent deconvolution is a nonlinear regression on a model derivative, without knowing the
model. The main unknown are these derivative points on a loglog scale. We curve is bent,
integrated and superposed in order to best fit the data of interest, generally consecutive and
reasonably consistent build-ups. Additional unknown may be the initial pressure and some
slack on the rate history. Additional constraints include a minimization of the curvature and of
the amplitude of the rate changes.
This new deconvolution provides, in one run, a synthesis of information present in different
parts of the pressure / rate history. When you combine two build-ups that are far apart in
time, you do not only check their coherence but you integrate, in the deconvolved response,
the information of depletion that took place between these two build-ups.
There is nothing really that a conscientious and trained engineer could not do before having
this tool, but all this information is calculated in a single run, early in the interpretation
process, instead of being a trial-and-error, last minute analysis following the interpretation of
the individual build-ups.
So the tool is useful and interesting, but we must emphasize its uses and limitations:
For any practical purpose, today it only works on shut-ins.
It should not be used as a black box. It is only an optimization process, and
optimization processes are typically impressive when they work and pathetic when
they do not. Any error in the deconvolution process will be carried over during the rest
of the interpretation.
Deconvolution should be considered a complement to, not an alternative to, standard
build-up analysis. It can be useful, early in the interpretation process, to run a
deconvolution and see if it carries any information that could be integrated in the
engineer thinking process, but the reference and ultimate decision should come from a
match on the real data (typically the history match and the individual build-ups), not
only the match on the product of an optimization process that carries some
assumptions.
To work at once it requires us to bundle together coherent, typically build-up,
responses. When build-ups are incoherent, it is still possible to run individual
deconvolutions based on the same value of pi, and modify pi until the different
deconvolutions are coherent.
Deconvolution only works if superposition works and if the system does not change in
time. Superposition works if the running equations are linear. If the flow is nonlinear
(e.g., material balance depletion, nonDarcy, multiphase, etc) deconvolution will fail
and may be misguiding. Same will apply, even if the diffusion is perfectly linear, if
other wells are interfering with the analyzed pressure response (multi-well case).
So, as with any new tool there is the ‘buzzword’ syndrome. It should not be oversold nor
overbought. Deconvolution is NOT the silver bullet that will provide your reserves that we
could not see before.
Still, it is great to have something new in an area where the last really innovative theoretical
tool was the 1983 Bourdet derivative.