Отправлено: 05.11.18 15:53. Заголовок: Automatic interpretation of geophysical well logs

Automatic interpretation of geophysical well logs
ANDRÉS ALCOLEA RODRÍGUEZ1*, PAUL MARSCHALL2,
CHRISTOPHE NUSSBAUM3 & JENS KARL BECKER2
1HydroGeoModels AG, Tösstalstrasse 23, 8400 Winterthur, Switzerland
2Nationale Genossenschaft für die Lagerung radioaktiver Abfälle (Nagra),
Hardstrasse 73, 5430 Wettingen, Switzerland
3Swisstopo, Route de la gare 63, 2882 St-Ursanne, Switzerland
*Correspondence: andres.alcolea@hydrogeomodels.ch


Abstract:
Stratigraphic sequences in boreholes are commonly estimated by interpreting combinations of well
logs. The interpretation is generally tedious and is made some time after log completion, which often leads to a
loss of valuable first-hand information gathered on-site. This may lead to delayed or potentially poor on-site
decisions. To make things worse, the standard interpretation of well logs is, at least to a certain degree, subjective
and based on the manipulation of data, which may be difficult to trace in the long term. Small changes in
lithology are often disregarded and alternating thin layers presenting different lithologies are often combined in
one single (notably thicker) stratigraphic unit. Therefore an automatic parameter-based and thus traceable and
objective quick look at the lithology immediately after log completion represents both a valuable tool to help with
on-site decisions and a solid, mathematically based starting point for further physically based interpretations
carried out by log analysts. We present a workflow for the interpretation of well logs defined as an optimization
problem. The workflow is applied to the characterization of metre- to decametre-scale stratigraphic units along
13 boreholes in northern Switzerland (one-dimensional resolution) and to millimetre-scale features over a wall
at the Mont Terri underground rock laboratory in Switzerland (two-dimensional resolution). The results show
that: (1) the workflow accurately maps lithological changes; (2) the interaction with the analyst is minimized,
which reduces the subjectivity of the interpretation; and (3) outputs are available for on-site decisions.