such
other
level
tiple
s for
used
d as
, the
et of
r to
hole
| be
ring
ered
cies
ra
tity
ned
ent
cts.
ster
for
y to
on.
the
ing
the
at
of
ter
the
ing
ing
(2)Imperatives for human managers: Their definition is
similar to one of imperatives for automated
managers.
3. IMPLEMENTATION
Our approach to solve the above problems of
environmental disaster management: information
distribution strategy and automatic/adaptive
management is to incorporate learning and inference
abilities into management system of environmental
disaster to automate the process of global view
construction, measurement interpretation, problem
forecasting, problem diagnosis, and decision making
related to environmental disaster management.
3.1 Global View
To build the information infrastructure about
environmental disaster management, a set of global
view is constructed with the help of the above policy
hierarchy. A global view is a virtual object class
defined from MIB via logical rules. From the
relationship between making policies, these global
views serve as windows through which management
applications can access physical entities about
environmental disaster.
ED Control DKB
Triggering
Logging
HD PKB
Leammg
Fig 2 Decision Process Model
3.2 Decision Process
To. implement the above automatic and adaptive
abilities, the disaster patterns related to policy attributes
must be learned from a historical database which
contains a chronological measurement trace. These
discovered patterns have been represented and describe
the correlation between disaster objects. Based on these
disaster patterns and prespecified domain knowledge,
forward and backward inference can be triggered to
access global views, predict disaster status, fire control
actions, and reported problems. Fig. 2 is an abstract
decision process model of management system of
environmental disaster. The extensional database(ED)
is used to represent the basic facts about distribution,
severity and performance measurement alarms, and
events of disaster. Each disaster has an associated ED
which is its view about the disaster. The historical
database(HD) is the temporal historical database which
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
encodes time in the disaster trace. Disaster patterns are
learned from HD and stored in Pattern knowledge
Base(PKB). The domain knowledge base(DKB) is for
prespecified problem solving and general relationship
knowledge.
3.3 Reasoning Process
Each disaster pattern is represented as a logical rule in
PKB, and describes a correlation between the attributes
of disaster objects and related policies. These
correlations come from HD, where selected attributes
are logged according to the specific management
policy. If the status of disaster objects satisfies the body
part in the rule, the pattern from past experience
represents that it is very likely that the status of the
disaster object also satisfies head port with some
probability. A logical rule has the generic form: IF X
THEN Y, where X is its body part and Y is its head
part. If some undesired status of a disaster object is
foreseen to occur, it can further fire some logical rules
in DKB and then trigger preventive control actions for
environmental disaster.
4. DISASTER MANAGEMENT
From the policy hierarchy, the major components of
environmental disaster management include fault
management, distribution management, and influence
management. As its name implies, fault management is
responsible for detection, isolation, and recovery from
component failures and inflicted damages related to
environmental disaster. Distribution management is
related to determination of disaster influence and
accommodation of an disaster distribution changes
including those requested by fault management and
influence management. Influence management is
responsible for reducing the disaster influence by
adjusting the disaster control decisions, and critical for
efficient control of large-scale disaster that occurs in a
dynamic environment.
4.1 Basic Model
Modeling management information of environmental
disaster is to map disaster distribution, characteristics,
and events to objects in ED. An inheritance hierarchy
can represent a simple classification of disaster object
classes, where the elements class has three subclasses:
distributions, characteristics, and events. Physical
entities class has two subclasses: affected entities and
geographic positions.
4.2 Disaster data
Disaster management data in the policy hierarchy of
environmental disaster management can be broadly
classified into the followings.
775
