FixtureFinder: Discovering the Existence of Electrical and Water Fixtures
Vijay Srinivasan*, John Stankovic, Kamin Whitehouse
University of Virginia*(Currently affiliated to Samsung)
Motivation For Fixture Monitoring
CookingToileting
Home Healthcare Applications
7 KW hours 400 liters
Resource conservation applications
Fixture Monitoring Using Smart meters
Whole house
power orwater flow
Time
Power meter
Water meter
Bathroom Kitchen
Bedroom Livingroom
2000 W
100 W
100 W100 W
100 litres/hour
100 litres/hour
• Poor accuracy for low power or low water flow fixtures• False positive noise• Identical fixtures
Existing Fixture Monitoring Techniques
Direct metering on each fixture Indirect sensing + smart meter
Single-Point Infrastructure sensing
Images courtesy: HydroSense and Viridiscope (Ubicomp 2009)
Requires users to:• Identify each fixture, and for each fixture:• Install a sensor, or• Provide training data
FixtureFinder
Power meter
Water meter
Bathroom Kitchen
Bedroom Livingroom
• Automatically:– Identify fixtures– Infer usage times– Infer resource
consumption2 PM5 PM…
Single-Point Infrastructure sensing
Training data
7 KW hours
400 liters
Home security or automation sensors
Light and motion
+Lights, sinks and toilets
FixtureFinder Insights
Bathroom Kitchen
Bedroom Livingroom
Fixtures identical in meter data
Unique in (meter, sensor) data
100 W100 W, 30 lux
100 W, 50 lux
Light sensor
Power meter
Water meter
FixtureFinder Insights
Bathroom Kitchen
Bedroom Livingroom
100 W, 30 lux
100 W, 50 lux
Light sensor
False positive noise in meter
and sensor data
1. Eliminate noise events in one stream when no activity in other stream
2. Eliminate unmatched noise
Power meter
Water meter
ON-OFF pattern
Bedroom light sensor data
Power meter data
Outline
• FixtureFinder algorithm• Case studies• Experimental setup• Evaluation results• Conclusions
FixtureFinder Algorithm Inputs
Stream 1
Stream 2
Power meter
Water meter
Light or motion sensors
or
• Four step algorithm
Step 1 – Event Detection
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
500 60
40
40
140
60200
Stream 1
Stream 2
False positives events:
True positive events:
40 lux100 Watts
For example:
Edge detection algorithms
Key challenge: Large number of false positives
100
40
100
Light sensor
Power meter
Step 2 – Data fusion
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
100 500 60
40
40
140
100 60200
40
Stream 1
Stream 2
40 lux100 Watts
For example:
Light sensor
Power meter
Fixture use creates events in multiple streams simultaneously
Compute event pairs
Eliminate temporally isolated false positives
Step 3 – Matching
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
100 500 60 40
140
100 60200
40
Stream 1
Stream 2
40 lux100 Watts
For example:
Light sensor
Power meter
Fixture use occurs in an ON-OFF pattern
Match ON event pairs to OFF event pairs
Eliminate unmatched false positives
High match probability
Step 3 – Matching
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
100 60
100 60
40
Stream 1
Stream 2
40 lux100 Watts
For example:
Light sensor
Power meter
High match probabilityTwo ON-OFF event pairs:(40,100) or (40,60) ?
True event pairs are more likely than noisy event pairs
High pair probability
Use both match and pair probabilities to compute ON-OFF event pairs
Soft clustering and Min Cost Bipartite matching (Described in paper)
Low pair probabilityAll false positives eliminated in this example!
Step 4 – Fixture Discovery
Stream 1 (Light)
intensity
Stream 2 (Power) intensity
ON Time OFF Time
41 102 5 PM 6 PM
62 103 5:30 PM 6:15 PM
43 99 8 PM 10 PM
60 101 7 PM 8 PM
61 100 9 PM 10 PM
Step 3: Matching
ON-OFF events
Clustering
Clustering based on:(stream 1 intensity, stream 2 intensity)
40 lux,100 watts
60 lux,100 watts
Fixtures discovered
Outline
• FixtureFinder algorithm• Case studies• Experimental setup• Evaluation results• Conclusions
Light Fixture Discovery
Power meter
Water meter
Bathroom Kitchen
Bedroom Livingroom
Apply FixtureFinder algorithm on every (light sensor, power meter)
40 lumens, 100 watts
40 lumens, 150 watts
Unique fixture usage defined by:Light sensor locationLight intensityPower consumption
Light Fixture Discovery
Bedroom light sensor data
Bedroom light fixture ON-OFF events
Power meter data
Large number of false positives after step 1
False positives eliminated after steps 2 and 3
Water Fixture Discovery
Power meter
Water meter
Bathroom Kitchen
Bedroom Livingroom
Fused motion sensor stream
Apply FixtureFinder algorithm on (fused motion sensor, power meter)
Unique fixture usage defined by:Motion sensor signatureFlow rate
100 litres/hour
100 litres/hour300 litres/hour
Water Fixture Discovery
Two toilets with the same flow signature but different
motion signatures
Water Fixture Discovery
Two toilets with the same motion signature but
different flow signatures
Use event pair probability to pair simultaneous toilet events with correct rooms
Outline
• FixtureFinder algorithm• Case studies• Experimental setup• Evaluation results• Conclusions
In-Situ Sensor Deployments in Homes
Power meter(TED 5000)
Water meter(Shenitech)
X10 motion Custom light sensing mote
One per room in a central location(Except in 3 large rooms wheretwo sensors were used)
One per home
In-Situ Sensor Deployments in Homes
Smart switch Smart plug
Contact switches on water fixtures
Ground truth for light fixtures
Ground truth for water fixtures
All sensors deployed in 4 homes for 10 days(Except water meter deployed in 2 homes for 7 days)
Outline
• FixtureFinder algorithm• Case studies• Experimental setup• Evaluation results• Conclusions
Fixture Discovery Results
Discovered all sinks and toilets across 2 homes
Discovered 37 out of 41 light fixtures across 4 homes
Undiscovered lights:- All in large kitchens- Task lighting or under-cabinet lighting- Used rarely (1-3 times)- Low energy consumption
One false positive light with negligible energy consumption
Fixture Usage Inference Results
Recall: % of ground truth fixture events detected by Fixture Finder
Precision: % of detected fixture events that are supported by ground truth
Results shown for light fixtures
99% precision64% recall
True positive ON-OFF events from fixtures
Single-Point Infrastructure
sensing
Training data
High precision usage data
Fixture Usage Inference Results
Recall: % of ground truth fixture events detected by Fixture Finder
Results shown for light fixtures
92% precision82% recall
Balanced precision and recall
Home Activity Monitoring applications
Precision: % of detected fixture events that are supported by ground truth
Analysis of FixtureFinder Steps• Step 1: Event
Detection– ME: Meter event
detection– SE: Sensor event
detection• Step 3: Matching
– MM: Meter event matching
– SM: Sensor event matching
• Step 2: Data Fusion– SMF: Sensor
meter data fusion• FixtureFinder
Small reduction in recall
Significant increase in precision with steps 2, 3, and FixtureFinder
Results shown for light fixtures
Light Fixture Energy Estimation
• 91% average energy accuracy for top 90% energy consuming fixtures
Water Consumption Estimation
• 81.5% accuracy in Home 3• 89.9% accuracy in Home 4
Home 3 Home 4
B – BathroomK – KitchenS – SinkF – Flush
Outline
• FixtureFinder algorithm• Case studies• Experimental setup• Evaluation results• Conclusions
Conclusions
• FixtureFinder combines smart meters with existing home security sensors to automatically:– Identify fixtures– Infer usage times– Infer resource consumption
• Demonstrated for light and water fixtures• Complements other fixture monitoring
techniques by providing training data without manual effort
Future Improvements
• Expand scope to include:– Additional electrical appliances and water fixtures– Additional sensing modalities such as routers,
smart switches, infrastructure sensors• Extend algorithm to multi-state appliances– Not just two-state ON-OFF
• Explore temporal co-occurrence over multiple timescales
ThanksQuestions?
FixtureFinder Approach
Power meter
Water meter
Home security or automation sensors
+
• Automatically discover low power or low water flow fixtures– Lights, sinks,
and toilets
Bathroom Kitchen
Bedroom Livingroom
Light and motion
Step 3 – Bayesian Matching• Two matches possible
– (40,100) or (40,60)• Assumption: Edge pairs
from true fixtures are more frequent than noisy edge pairs– P(40,100) >> P(40,60)
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
100 60
100 60
40
Hidden variables
Stream 1 cluster
Stream 1 edge Stream 2
edge
Stream 2 cluster
Observed variables
Step 3 – Bayesian Matching• Incorporate edge pair
probability into a match weight function
• Perform optimal bipartite matching based on match weight function
• Eliminate unlikely matches
Stream 1
Stream 2
Time
ON
OFF
ON
OFF
40
100 60
100 60
40