Yexin Liu* A Self-Adaptive Optimization Solution to Petrophysical Properties Inversion from Well Logs
Calgary, AB
yexinliu@gmail.com
Summary
During the well logging processing (inversion) and interpretation, it is important for us to select the
fluid type, clay type and mineral components. Usually the clay and fluid types can be referred to field
experience and other measurements, but the mineral components could be variable from borehole
top to bottom (depth-variable). One component model maybe is appropriate for one geological
section, but the other section may need to use another mineral component model. Also it is difficult
for us to know the mineral components at the different sections!
Here we present a self-adaptive approach to add/remove the mineral components according to the
rules pre-defined by users and the best mineral components will be selected automatically during
the inversion. Also a multi-objective optimization mechanism has been used to inverse the porosity,
volumes of clay and mineral components from well logs using the selected mineral components
combined with both fluid and clay types.
During the inversion, according to field experience there are differing logging response equations
available from the same well log tool, which can be selected interactively. For example, the water
saturation can be calculated from dual-water model, and also be calculated from Archie or
Indonesian equations. Also the inversion procedure is a section/zone-based mechanism to meet the
petrophysical model depth-variable’s needs. The method has demonstrated a reasonable and
robust result after processing well logging dataset from both conventional sand-shale formation and
complex carbonate formation.
=================================================================
Because of ill-posed issues, solutions might not be unique and/or might not depend continuously on
the data. Hence their mathematical analysis is subtle. Usually when the logging reading is a minor
change, the output will have no solution or the output is not robust. Here a multi-objective
optimization solution is presented to solve the ill-posed issues. All well logging response equations
can be normalized using their standard deviation to remove measurement unit affection and then
the penalty factors will be applied to some objective functions when the target property cannot meet
the constrained conditions, and also a small random variable with mean zero to add the response
equation and target properties. The Broyden-Fletcher-Goldfarb-Shanno (BFSG) method is selected
to optimize and inverse the property and it performs extremely well.
===============================================
http://www.cspg.org/conventions/abstracts/2008abstracts/070.pdf Интересно как он работает?
Неужто с разностными производными?
Тогда как шаг подбирает? ;-)