Information Technology Reference
In-Depth Information
information for problem solving.
(3) Constraint representation is easy to be implemented in parallel, because
information propagation in constraint network can be regarded running at the
same time.
(4) Constraint representation is suitable for system of gradual increase.
Constraint can be added to constraint network in gradual increase way.
(5) Constraint representation is easy to connect with problem solving models of
field. Mathematical planning, equation solving and other techniques can be
embedded into constraint system naturally.
Many constraint reasoning methods has been proposed in many years ' study.
According to data type of node in constraint network, constraint reasoning can be
classified into following categories.
(1) Relation reasoning: New constraint relation reasoned out in reasoning process
will be added into constraint network. Kuiper's ENV system, Simmon's
Quantity Lattice system and Brooks' CMS system, are all belong to relation
reasoning.
(2) Mark reasoning: Mark each node with a set of possible values, which will be
limited by constraint in constraint propagation.
(3) Value reasoning: The node is marked with constant value. Constraints obtain
values of unmarked nodes by using already marked nodes values.
SKETCHPAD and THINGLAB both use value reasoning.
(4) Expression reasoning: It is a extension of value reasoning and the node might
be marked with other node expressions. The different expressions of marked
node
should
be
equal
and
form
result
equations
to
be
solved.
CONSTRAINTS uses this kind of reasoning.
The value of constraint variable may be the number value, or not the number
value, i.e. symbolic value. In general, the values of symbolic variable form a
limited set. Therefore after constraint propagation stops, exhaust search can be
used to decide the consistency of symbolic variable. While the number variable
usually has limitless value domain, it is impossible to carry on the exhaustive
search.
Above four kinds of constraint reasoning all have some shortcomings. For
example, value reasoning can be not used in inequality constraint but in equation
constraint. Relation reasoning and mark reasoning are difficult to control and
Search WWH ::




Custom Search