344
A.M.J. Meijerink and C.M.M. Mannaerts
Knowledge representation & processing. There are generally two types of know 1edge contained in an expert system. A-priori knowledge represents the facts and the
rules one knows about a specific domain prior to any knowledge processing within
the system. Inferred knowledge consists of new facts or conclusions derived during
information processing by the expert system. A major concern is how to represent
facts and rules within the knowledge base. This involves maintaining a close correspondence between the computer and the real world facts and rules, and establishing
a representation that can be easily addressed, retrieved, modified, updated and
processed. Knowledge can be represented in the following ways i.e., object attribute
value triplets, semantic networks, frames and rules. Knowledge processing is basically
performed using a prioritized list of hypothesis, contained in an agenda. The verification of hypothesis is done using inference mechanisms, which let the expert expand
the search space for relevant conclusions without exhaustively evaluating all of the
rules in the knowledge-base. For example, the Nexpert Object contains a number of
search mechanisms for testing an agenda with priorities. Backward chaining is
provoked when a condition containing an unknown Boolean slot ( in fact, a hypothesis) is encountered. All rules pointing to that hypothesis are evaluated immediately or
with highest priority. Suggesting a hypothesis from, e.g., the user interface puts it on
the agenda for evaluation. Suggested hypotheses have priority over others except
those generated by backward chaining. When a hypothesis is used as data in a rule's
left hand side condition, then the hypothesis of that rule (using the hypothesis as data)
is put on the agenda immediately after evaluation of the first hypothesis. This is a
forward propagation inference mechanism for evaluating hypothesis. Semantic gates
are the basic mechanisms for automated goal generation or opportunistic reasoning.
Furthermore, we can distinguish among right-hand side actions, volunteer and context
or weak links.
We refer to Nexpert object (1991) for more details on these knowledge inference
mechanisms. Figure 15.6 shows a simplified version of knowledge processing for
watershed flood runoff simulation in an object-oriented data environment (Baten,
1994). The software environment is composed of Ilwis (lTC, 1994) as Remote
Sensing and Geographic Information System, HEC-1 (ASCE, 1987) as flood hydrology simulation model, dBase (© Borland), as data base and the Nexpert Object as
expert shell surrounding the RS/GIS, data bases and model. From Fig.l5.6, the rule
evaluation sequences and interaction of the rules with the object network and other
system components can be seen. The remote sensing input in this application pertains
to the land cover and associated curve numbers (CN), and the soil groups. Integration
of knowledge processing mechanisms with hydrology and water management is
usually done using an object-oriented approach, to capture the physical system e.g.,
a river basin or channel network system (Kim, 1990; Coad and Y ourdan, 1991). This
represents the more conventional application domain of knowledge engineering, i.e.,
object orientation, in hydrologic data base design and management. Expert systems
and knowledge-based engineering should have also extensive application in conceptual model building for solving hydrologic and water contamination problems (IGWC,
1992). The solution of many hydrologic and pollution problems in the aquatic
environment must be based on a broad or detailed preliminary analysis of envi-
A.M.J. Meijerink and C.M.M. Mannaerts
Knowledge representation & processing. There are generally two types of know 1edge contained in an expert system. A-priori knowledge represents the facts and the
rules one knows about a specific domain prior to any knowledge processing within
the system. Inferred knowledge consists of new facts or conclusions derived during
information processing by the expert system. A major concern is how to represent
facts and rules within the knowledge base. This involves maintaining a close correspondence between the computer and the real world facts and rules, and establishing
a representation that can be easily addressed, retrieved, modified, updated and
processed. Knowledge can be represented in the following ways i.e., object attribute
value triplets, semantic networks, frames and rules. Knowledge processing is basically
performed using a prioritized list of hypothesis, contained in an agenda. The verification of hypothesis is done using inference mechanisms, which let the expert expand
the search space for relevant conclusions without exhaustively evaluating all of the
rules in the knowledge-base. For example, the Nexpert Object contains a number of
search mechanisms for testing an agenda with priorities. Backward chaining is
provoked when a condition containing an unknown Boolean slot ( in fact, a hypothesis) is encountered. All rules pointing to that hypothesis are evaluated immediately or
with highest priority. Suggesting a hypothesis from, e.g., the user interface puts it on
the agenda for evaluation. Suggested hypotheses have priority over others except
those generated by backward chaining. When a hypothesis is used as data in a rule's
left hand side condition, then the hypothesis of that rule (using the hypothesis as data)
is put on the agenda immediately after evaluation of the first hypothesis. This is a
forward propagation inference mechanism for evaluating hypothesis. Semantic gates
are the basic mechanisms for automated goal generation or opportunistic reasoning.
Furthermore, we can distinguish among right-hand side actions, volunteer and context
or weak links.
We refer to Nexpert object (1991) for more details on these knowledge inference
mechanisms. Figure 15.6 shows a simplified version of knowledge processing for
watershed flood runoff simulation in an object-oriented data environment (Baten,
1994). The software environment is composed of Ilwis (lTC, 1994) as Remote
Sensing and Geographic Information System, HEC-1 (ASCE, 1987) as flood hydrology simulation model, dBase (© Borland), as data base and the Nexpert Object as
expert shell surrounding the RS/GIS, data bases and model. From Fig.l5.6, the rule
evaluation sequences and interaction of the rules with the object network and other
system components can be seen. The remote sensing input in this application pertains
to the land cover and associated curve numbers (CN), and the soil groups. Integration
of knowledge processing mechanisms with hydrology and water management is
usually done using an object-oriented approach, to capture the physical system e.g.,
a river basin or channel network system (Kim, 1990; Coad and Y ourdan, 1991). This
represents the more conventional application domain of knowledge engineering, i.e.,
object orientation, in hydrologic data base design and management. Expert systems
and knowledge-based engineering should have also extensive application in conceptual model building for solving hydrologic and water contamination problems (IGWC,
1992). The solution of many hydrologic and pollution problems in the aquatic
environment must be based on a broad or detailed preliminary analysis of envi-