2. June 26-27, 2002 2
Outline
Motivating tasks
Background information
The OntoWEDSS decision-support
system with the WaWO ontology
Results
Conclusions and perspectives
3. June 26-27, 2002 3
Motivating tasks
Improvement of the modeling of the
information about the wastewater
treatment process and of wastewater
management
Solution of complex problems related
to wastewater using ontologies
Integration of ontologies in the
reasoning of decision support systems
4. June 26-27, 2002 4
Outline
Motivating tasks
Background information
The OntoWEDSS decision-support
system with the WaWO ontology
Results
Conclusions and perspectives
5. June 26-27, 2002 5
Ontologies: definition
An ontology is a formal and explicit
specification of a shared conceptualization,
which is readable by a computer.
An ontology describes the shared model of
a domain. Everybody following a particular
ontology understands all the categories and
the relations comprised in that ontology and
behave accordingly.
6. June 26-27, 2002 6
PLANNING / PREDICTION/SUPERVISION
AI
MODELS
STATISTICAL
MODELS
NUMERICAL
MODELS
GIS
(SPATIAL DATA)
DATA BASE
(TEMPORAL DATA)
USER INTERFACE
Background
/ Subjective
Knowledge
ECONOMIC
COSTS
USER
Decision
/ Actuation
ENVIRONMENTAL
/ HEALTH
REGULATIONS
Spatial /
Geographical
data
On-line
data
Off-line data
DATA MINING
KNOWLEDGE ACQUISITION/LEARNING
EXPLANATION ALTERNATIVES
EVAL.
REASONING / MODELS’ INTEGRATION
BIOLOGICAL
/ CHEMICAL
/ PHYSICAL
ANALYSES
SENSORS
ON-LINE /
OFF-LINE
ACTUATORS
Feedback
ENVIRONMENTAL SYSTEM / PROCESS
DECISIONSUPPORTDATAINTERPRETATION
DIAGNOSIS
Environmental decision-support systems
7. June 26-27, 2002 7
Outline
Motivating tasks
Background information
The OntoWEDSS decision-support
system with the WaWO ontology
Results
Conclusions and perspectives
8. June 26-27, 2002 8
OntoWEDSS: profile (1)
Use of ontologies in domain modeling and
clarification of existing terminological
confusion in wastewater domain
Automatic, reliable discovery and
management of problematic states in real-
world domains
Composition, interoperation and reuse of
different reasoning systems (rule-based,
case-based and ontology-based)
9. June 26-27, 2002 9
Environmental process supervision and
management distributed in 3 layers:
perception, diagnosis and decision support
Incorporation of wastewater microbiological
knowledge into the reasoning process and
representation of cause-effect relations
Resolution of existing reasoning-impasses
OntoWEDSS: profile (2)
11. June 26-27, 2002 11
WaWO
- Frame-based representation
- Hierarchy used for:
Queries
Language analysis
Reasoning
- Standard but specialized:
Storm is an
Operational-Problem
Bacterium is a
Wastewater-Biological-
-Living–Object
- Metazoan represented:
Nematode
Rotifer
12. June 26-27, 2002 12
Reasoning
with
ontologies
Role or
Phenomenon
categories
Occurrents
Relations
13. June 26-27, 2002 13
SupervisionSupervision
modulemodule
RBES
Does
RBES’s
diagnostics
exist?
CBRS
CBRS’s
inference
RBES’s
inference
No
Yes
No
No
Yes
No
Does
CBRS’s
diagnostics
exist?
RBES’s
Diagnostics
=
CBRS’s
Diagnostics?
Yes
CBRS’s >
constant β ?
Yes
Does
CBRS’s
diagnostics
exist?
No
CBRS’s
Diagnostics
Yes
RBES’s
Diagnostics
CBRS’s Diagnostics
RBES’s Diagnostics
CBRS’s
Diagnostics
RBES’s
Diagnostics
WaWO’s
Diagnostics
WaWO
Reasoning
integration
14. June 26-27, 2002 14
Functionalities
Input (modeling and execution)
List of descriptors to use
Weight of descriptors (optional)
New-problem’s descriptors values
Output (execution)
Diagnosis of the current state of the WWTP
(with reliability factor)
Trace of the reasoning
List of actions to take according to the current
situation
16. June 26-27, 2002 16
Action suggestion
Change Sludge-Recirculation-External to 120
Destruction of filaments via chlorine addition
Addition of inorganic coagulant
Check out Food-To-Micro-Organism-Ratio
Remove aeration-tank and clarifier foam
Reduce waste-activated-sludge flow rate
(FlowRate-WAS)
17. June 26-27, 2002 17
Outline
Motivating tasks
Background information
The OntoWEDSS decision-support
system with the WaWO ontology
Results
Conclusions and perspectives
18. June 26-27, 2002 18
Database description
Initial set: 790 days with 21 quantitative and
qualitative descriptors (out of 170)
Filters: missing values, labels
Final set for CBRS training: 186 days
Bulking-Sludge labeled: 29 days (16%)
Lack of benchmarks
High number of descriptors
Multiple labels
Problems
19. June 26-27, 2002 19
Evaluation results: CBRS and RBES
Focus on the
most
representative
problematic
situation:
bulking
sludge
20. June 26-27, 2002 20
OntoWEDSS evaluation
Average
successful
outcomes:
65%
Average
successful
outcomes:
88%
21. June 26-27, 2002 21
Outline
Motivating tasks
Background information
The OntoWEDSS decision-support
system with the WaWO ontology
Results
Conclusions and perspectives
22. June 26-27, 2002 22
Conclusions
Research tool to explore the possibilities and
the potential of introducing ontologies into
decision support systems, using an
environmental domain as case study
Creation of an ontology for the domain of
wastewater treatment process
Ontological representation of two kinds of
cause-effect relations:
micro-organisms ↔ problematic situations
state of the plant ↔ suggested actions
23. June 26-27, 2002 23
Perspectives
Further refinement and update of
current AI modules
Simulation and prediction of the
evolution of a treatment plant’s state
Integration of the ontology with some
temporal reasoning
Reasoning with variations/transitions
of descriptors’ values
27. June 26-27, 2002 27
Axioms
Example of causality axiom:
Physical entities may causally affect other
physical entities
Different views of the same entity may be
described with different words, definitions
and axioms.
Each category in the hierarchy inherits all
the properties and axioms of every category
above it.
28. June 26-27, 2002 28
Ontologies: languages
KIF: meta-format for knowledge interchange
Ontolingua: KIF-based; object-oriented using a
Frame Ontology; Web interface (on-line collaboration);
translation to various languages; large repository
RDFS: resources as Web addresses; primitives for
classes and properties
OIL: RDFS-based; entirely Web-driven; combination
of frame-based modeling and description logic
DAML+OIL: designed for Web-agents; richer
modeling primitives (e.g., properties with cardinality)
29. June 26-27, 2002 29
Decision-support systems
User friendliness
Assistance in problem formulation
Framework for information capture
Specific KBs
Integration of different AI systems
(RBES and CBRS, generally)
Generation of different strategies
30. June 26-27, 2002 30
Rule-based expert system
These systems express regularities as
rules. They typically follow a situation-
action paradigm: the set of rules let
them directly suggest what action to
take in a given situation.
The domain is so complex that causes
other than the given action may also
contribute to a resulting situation.
31. June 26-27, 2002 31
Case-based reasoning system
These systems express regularities
and singularities as cases, each of
which encodes some effects of an
action under a specific situation. They
also follow a situation-action paradigm:
the adaptation of the actions taken in
previous similar situations let them
suggest about the current actions to
take.
32. June 26-27, 2002 32
The chicken-and-egg paradox
in modeling and diagnosis
The situations (set of descriptors’ values)
cannot be defined without first knowing what
diagnostics they correspond to.
And most diagnostics can be hard to define
as such, until the corresponding situations
have been identified.
Expert often have to use trial-and-error
methods.
Set of
descriptor values
Diagnostics
DIAGNOSIS
Situation
modeling
33. June 26-27, 2002 33
Functional parameters
Activation cycle
1 hour (5 min in case of detected emergency)
Accuracy (based on focused evaluation)
Cost
Allegro Common LISP
Experiment Number
of data
Correct
classification
G-1
G-2
G-3
8
10
11
100%
90%
70%