|
|
2.2. PROSPECTOR
Duda et al. 1978 SRI International
The work behind PROSPECTOR was very much inspired by the encouraging results reported from the MYCIN effort (Shortliffe 76). The medical type of diagnosis that MYCIN was able to demonstrate lent itself very well to that of analyzing rocks and minerals. PROSPECTOR was patterned according to this type of schemata, but pioneered some important new aspects on its own.
PROSPECTOR was built to help geologists in exploring for hard-rock mineral deposits (Duda 78; Gaschnig 81; Duda 81). To achieve that, the developers created a knowledge model along the same lines as that of MYCIN. With the aid of this model, the system should be able to emulate the reasoning process of an experienced exploration geologist in asserting the likelihood that a given prospect area could contain the type of mineral that the geologist would desire. The user of PROSPECTOR would obtain some promising field data and then consult the system for assessing these data. Based on the input given, the system would infer the presence of minerals such as sulfides, uranium, lead, zinc, copper, nickel, and others.
A novel feature of PROSPECTOR at the time was its ability to accept rock data and information on minerals found together with other observations that the user volunteered. In fact, part of the input came from a digitized map. After the first match on this input, the system would possibly request additional information in a standard question-driven manner.
Part of the output could be given in the form of color-coded graphical displays that showed the favorbility of each cell with respect to prospect drilling.
PROSPECTOR also demonstrated capabilities of explaining its conclusions and actions. When prompted for data, the user could enter WHY, thus initiating a process whereby the system produced a simple geological rationale for its query. In order to achieve this, the system would trace its rule network and produce an answer. In this way it reflected the causal structure between the rules themselves.
The system was built up around the backward driven rule schemata that became very popular during the early 1980s (Duda 81). An example rule would typically be of the form:
TABLE 1 |
A List of Principal Systems on Geology and Mineralogy |
|
Year reported |
Name of system |
Domain knowledge |
Technology applied |
Reference |
|
1978 |
PROSPECTOR |
Minerals, prospecting |
Production rules, Bayesian inferencing |
(Gaschnig 81) |
1981 |
DIPMETER ADVISOR |
Log interpretation |
Production rules, goal-directed programming, and more |
(Davis 81) |
1982 |
ELAS |
Log interpretation |
Production rules |
1982 |
LITHO |
Interpretation of oil-well drilling logs |
Production rules/EMYCIN |
(Bonnet 83) |
1982 |
PHOENIX |
Oil well log interpretation/knowledge engineering |
Models and production rules |
(Barstow 82) |
1982 |
Amoco/X-ray |
Mineralogy |
Production rules, various inferencing techniques |
(Ennis 82) |
1983 |
ANALOG |
Petroleum geology |
Production rules |
(Hawkins 83) |
1984 |
IKBM |
Evaluate petrophysical formations |
Production rules, explicit geological models |
(ES 84) |
1987 |
MuPETROL |
Basin classification |
Production rules, explicit geological models |
1987 |
SPECTRUM |
Remote sensing/geology |
Mixed-initiative, agents |
(Borchardt 87) |
1987 |
META/LOG |
Log interpretation and resource estimation |
Hybrid approach, production rules, blackboard |
(SPECTRUM 87) |
1988 |
EXPERTEST |
Reservoir and formation modeling |
Production-rules |
1988 |
XX |
Identification, modeling, analysis of hydrocarbon plays and prospects |
Suite of many knowledge modules, production rules, approximate reasoning |
(Kendall 88) |
1990 |
Contouring Assistant |
Intelligent front-end and advisor to a complex contouring and gridding system for geological data |
Object-oriented technology, production rules, hybrid system, interface to numerical models |
(Sines 96) |
1990 |
Baldwin et al. |
Log interpretation/dentification of minerals and lithofacies |
Neural networks/Hyper Cube |
(Baldwin 90) |
1991 |
GeoX |
Analysis of hydrocarbon plays, resource estimation |
Decision support, object-oriented technology, production rules, structured justification |
(Stabell 90) |
1992 |
I2SAdvisor |
Remote sensing for geological exploration |
Planning, decision-support, case-based planning |
(Bremdal 95) |
1992 |
PLAYMAKER |
Characterization of hydrocarbon plays |
Production rules, knowledge discovery techniques |
1993 |
Sismonaute |
Detecting and interpreting wave fronts in seismic simulations |
Model-based reasoning |
(Junker 95) |
1993 |
GeCoS 3D |
Building 3D models from geoloogical cross-sections |
Assumption-based truth maintenance, model-based reasoning, constraints |
(Hamburger 95) |
1993 |
Kemme's E&P |
Subsurface modelling |
Business models, object-oriented methods, constraints |
(Kemme 95) |
1993 |
Urwongse |
Prediciting hydrocarbon potential |
DBMS, fuzzy analogs, neural networks |
(Urwongse 95) |
1993 |
MatchMod |
Interpretation of X-ray diffraction patterns |
Genetic algorithms |
(Schuette 95) |
1993 |
Maceral analysis |
Automatic analysis of coal macerals |
Production rules, hybdrids, image analysis |
(Catalina 94) |
1995 |
Baygun et al. |
Geological modeling of a mature hydrocarbon reservoir |
Neural networks, fuzzy classification |
(Baygun 96) |
1995 |
Hatton et al. |
Crude oil fingerprinting |
Self-organizing neural networks |
(Hatton 96) |
1995 |
Abel et al. |
Petrographic analysis |
Case-based reasoning |
(Abel 96) |
|
IF: There is hornblende pervasively altered to biolite THEN: There is strong evidence (320,0.001) for potassic zone alteration
|