Two-level expert system for well-log correlation
One limitation of conventional (either manual or machine) well-log correlation is that correlation loops do not close; that is, different correlations result depending on the order in which logs are matched. The authors have devised a two-level expert system for a well-log correlation which addresses this nonclosure problem. The first level of the expert system uses heuristic rules to determine zone attributes and degree of matching. The zones or segments of logs are predetermined either by eye or by any zonation algorithm. The second level of the system deals with machine learning and dynamic programming (an optimization technique that uses recursion to find best matches). A set of optimal weights is obtained through machine learning. These weights subsequently will be used in a dynamic programming technique. Training sets for machine learning are selected to reflect geologic settings of the region. Dynamic programming is then used to match the individual zones of logs, which may exhibit gaps, repetitions, and/or thickening-thinning. Five examples having the following features will be illustrated in the poster session: anticline, unconformity, reverse fault, normal fault with an anticline, and growth fault. The first four are made up of synthetic logs, and the fifth is a real-world example of a Gulf Coast growth fault having an extra layer on the downthrown side.
- Research Organization:
- Univ. of Alabama, Tuscaloosa (USA)
- OSTI ID:
- 6018286
- Report Number(s):
- CONF-890404-
- Journal Information:
- AAPG Bull.; (United States), Journal Name: AAPG Bull.; (United States) Vol. 73:3; ISSN AABUD
- Country of Publication:
- United States
- Language:
- English
Similar Records
Evidence for unconformities in the Cenozoic of the central and south Texas Continental Shelf from regional electric log and seismic cross sections
Aquifer Hydrogeologic Layer Zonation at the Hanford Site