Basic HTML Version
OH – ST - ET: Analysis of Dynamic Data in Shale Gas Reservoirs – Part 2
p 18/18
All this is consistent with what we found in Part 1 when evaluating the different techniques
available today. Overall, they all show that some of the most significant aspects occur to be:
-
Nonlinearities: because of the very low permeability, the pressure gradients in the vicinity
of the fractures are so important that the usual PTA analysis assumption of constant
compressibility and viscosity does not hold anymore.
-
Specific geometry: even if the permeability is very low, effects such as interferences
between fractures can become of upmost importance after a few months or years. This
must be accounted for.
-
Transient flow: the use of material balance related techniques for the determination of the
SRV may introduce the belief that pseudosteady-state flow can be reached in shale gas
formations. We must forget about this: fluid flow in a typical shale gas play is bound to
remain transient for tens or even hundreds of years. The SRV depletion regime is merely a
transition towards radial flow in the matrix, it can only in the best cases behave “like a”
pseudosteady-state regime without ever reaching it.
Not surprisingly NL numerical modeling is the option that accounts for a maximum of those
specificities, even if some work is required (e.g., downscaling) to adapt the model. But looking
at real data showed that simpler methods are still valuable as long as the analyst keeps the
various limitations in mind: results from straight line analysis can feed a model fitting with an
analytical model, which in turn can be used as a starting point for a fit with a more complex NL
numerical model. This analysis path is suggesting a workflow that goes through the different
methods step by step, with increasing complexity
However forecasts should always be made with the most complex available model. In a
simulation forecast, we cannot waive out with complete certainty the impact of a given effect
until it is accounted for in the model…
The next question is then: is the best model we have good enough? We suspect that there
remain many ruling processes (secondary fracture networks, stress dependency, multiphase
flow, etc…) that we have to incorporate in our toolkit, even though the problem of unicity will
remain acute as long as we do not have any source of information to constrain the results.