Machine Reasoning about Anomalous Sensor Data
Matt Calder, Francesco Peri, Bob Morris
Center for Coastal Environmental Sensoring Networks CESNUniversity of Massachusetts Boston
Goal
Provide scientists with software to explore domain hypotheses about their data
Outline1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
UMB CESN
• Interdisciplinary Research effort• Oceanography
• Biology
• Computer Science
• Policy / Law
• Cyber-infrastructure – Smart Sensor Networks
Outline1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
Algal Bloom ?
Benthic Resuspension ?
Aha!
Outline1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
Knowledge Representation• An ontology is a model of the relationships between concepts (ideas) of a particular domain. • OWL Web Ontology Language from the W3C
• Classes, Properties, Instances
Semantic Reasoners• Validation
• Checks that the constraints made in the ontology are not violated
• For example, a temperature sensor should not have taken any measurements other than temperature measurements.
• Inference and Rules• An inference is a conclusion drawn from the the truth
value of previously known facts
• antecedent -> consequence
• A ∧ B ∧ C -> D
Rule Example in Jena RL
[winter rule: (?x measurementOf Temperature)
(?x type Average),(?x value ?v),lessThan(?v, 0) →
(Season isWinter true) ]
In English:If x is a temperature and is an
average and has value v and v is less than 0 then it is winter.
Outline1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
Knowledge System
PhysicalPropertyPhysicalProperty
Measurement
Sensor
hasTakencanMeasure
real number dateTime
value timestamp
CESN Sensor Ontology: Core Components
Domain Knowledge Ontology: Ocean Events
OceanEvent
AlgalBloom BenthicResuspension
subClass subClass
dateTime
occurredAtTime
occurredAtLocationInfluencedBy
cesn:Locationcesn:PhysicalProperty
By the way…
Was it an Algal Bloom? ….No. It was winter!
Was it bethic diatom resuspension? Maybe – That is consistent with data and knowledge
Outline
1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
Sensor Data Reasoning System
Outline
1. Outline2. Motivation3. Knowledge Representation4. Our Knowledge System5. Software Architecture6. What’s missing (future work)
To Be Done• Distributed Sensor Reasoning Systems• Integrate with a stronger observations
ontology such as OBOE Ontology from SEEK
• User Interfaces for Rules • Investigate scalability and performance of
large sensor data sets.• Integrate with our existing SOS server• Collaborate with others
Summary
• Software System to test domain knowledge hypothesis about Sensor Data•
Thanks. Any Questions?
Key Components
Ontology
Rules
Software – Jena framework
Top Related