Geology Reference
In-Depth Information
2001 ). Box et al.( 2004 ) and Box and Lawrey ( 2003 )
describe a process in which log editing is approached
via a multi-linear regression of data from a large
number of wells including all relevant logs and vari-
ables (sonic, density, gamma ray, resistivity, depth
and pressure). Whilst using several logs rather than
one to make a prediction can improve the results the
process needs a thorough sense check. There is usu-
ally a fairly obvious single transform relationship that
forms the basis of the prediction and the question is
whether additional logs will help. Figure 8.49 (third
column) shows an example where a simple transform
of a neutron porosity log has been used to predict a
sonic log in a sequence of sands and shales. The
addition of the gamma ray ( Fig. 8.49 , fourth column)
makes a slight improvement but the addition of the
density log does not ( Fig. 8.49 , fifth column). As with
all prediction methods blind testing and cross valid-
ation (e.g. Hampson et al., 2001 ) are useful in assess-
ing the accuracy of predictions.
Other techniques that are commonly employed
for log prediction are neural networks and fuzzy logic.
Shear velocity prediction using fuzzy logic and genetic
algorithms has been described by Cuddy and Glover
( 2002 ). Using neural net or fuzzy logic prediction
techniques without adequate QC, however, is likely
to be misleading.
An effective test of a log prediction is to perform a
well tie. Figure 8.50 illustrates an example where a
sonic log was predicted from a density and resistivity
log, together with a reasonable well tie based on the
predicted sonic.
A good example of log prediction forming a useful
quality control on measured data is found in V s pre-
diction. Comparing the measured Poisson
GR
NPhi / Rhob
DTP1
DTP2
DTP3
200ft
Figure 8.49 Results of multilinear regression to predict
compressional slowness (black ¼ original sonic log, purple ¼
prediction), using (1) neutron porosity, (2) neutron porosity and
gamma ray, and (3) neutron porosity, gamma ray and density. Note
how there is a small improvement in the prediction around the
sands with the inclusion of the gamma ray, but no additional
improvement with the addition of the density log. Also note how
the thin hard carbonate-rich layers (high density spikes) are not
predicted.
8.4.3 Log prediction
A variety of approaches can be taken to predicting
sonic and density logs with a number of log trans-
forms and rock physics models having been described
in Section 8.2. Log predictions can provide a useful
means of log quality control (i.e. identifying poor
sections of log) as well as being a tool for infilling
missing sections.
A key issue in the application of both transforms
and rock physics models is the need to establish local
calibration before applying them. This calibration
step clearly requires other well data to be available
with the same geology and a similar effective pressure
regime. It is important that the geological and rock
physics context of transforms and models is correctly
understood before application. In particular, rock
types and facies need to be discriminated adequately
before applying transforms or models. It may be
possible, for example, to address sands and shales in
the same model (e.g. Xu
'
s ratio log
with the predicted Poisson
s ratio log can identify
zones that may benefit from editing. Another import-
ant aspect of V s prediction is the application of empir-
ical trends to data from hydrocarbon bearing wells.
The effect of the hydrocarbon on the compressional
velocity has to be removed before brine-based empir-
ical transforms can be applied. To do this requires
some knowledge of an appropriate stiffness model as
described in Section 8.2.3. If shear velocity data are
available in nearby wells, dry rock trends can be
determined, and then the Gassmann methodology
method described in Section 8.2.3 can be employed
to derive the
'
White, 1995 ) but lithologies
such as salt, coal and carbonate layers will need to be
addressed separately.
A common approach to log prediction is to use
multi-linear regression methods (e.g. Hampson et al.,
-
compressional velocity in the wells.
If there are no wells with shear data then a dry rock
model needs to be invoked (see Section 8.2.3 ).
'
wet
'
186
 
Search WWH ::




Custom Search