Page 97 - Well Logging and Formation Evaluation

Advanced Log Interpretation Techniques 87

overlap much in the N dimensional space they occupy. Also, the method
does not work well with parameters that vary gradually with depth.
The advantage of the method over other approaches such as neural net-
works is that one is able to see, through plotting the membership func-
tions with respect to a certain variable, whether or not it is applicable to
include a certain variable or not. Also, the method can generate a confi-
dence level for the output classification, as well as a “second choice.” The
use of fuzzy logic has been mainly in acoustic and elastic impedance
modeling, where one can investigate whether or not, for instance, there is
any acoustic impedance contrast between oil- and gas-filled sandstones.
If there is, the membership functions may be used as input to a seismic
cube for allocating facies types to parts of the seismic volume, thereby
showing up potential hydrocarbon zones.
Fuzzy logic may also be useful to allocate certain facies types to the
logs, as for instance a basis of applying a different poroperm model. In
my experience with using fuzzy logic, I have often found that one starts
out with too many facies, which then are found to overlap each other.
Also, the effect of adding more log types as variables, which may be only
loosely related to the properties one is interested in, is generally detri-
mental. In many respects, fuzzy logic is similar to the statistical analyses
packages described earlier. In common with these, it has the advantage
over deterministic techniques in that it can handle a lot of variables impar-
tially and simultaneously. However, also in common with those packages,
it can easily generate rubbish unless great care is taken with the input.

Exercise 5.2. Fuzzy Logic

1. Set up a fuzzy logic model to distinguish between net and non-net on
the basis of GR using the data from the core as a learning set.
2. Apply the model to the lower half of the entire logged interval.
Compare the average net/gross with that derived using the conven-
tional analyses.

5.6 THIN BEDS

Conventional petrophysics relies on the logs being able to resolve the
individual beds in order to determine such properties as R t , rho b , and GR.
While examination of any core will reveal many features that are far below
the resolution of all but imaging logs, this commonly does not pose any