1 SpotLight : Focusing on Energy Consumption of Individuals Younghun Kim Zainul M. Charbiwala Akhilesh Singhania Mani B. Srivastava

2 What is SpotLight? An WSN application  Profile individual’s energy consumption  Provide real-time energy profile  Help users identify areas of inefficient power consumption A prototype implementation  May not be a thorough solution  May have many limitations  But can point out interesting points and observations

3 Overview Motivation Problem description and discussion Related work Prototype system implementation and design Evaluations Conclusion Future work

4 Motivation How much energy do I consume?  What about in the lab and home? I am unaware of my energy usage pattern  Hard for me to optimize energy consumption What about other resources?  Water and Gas

5 Levels of Electricity Monitoring

6 Fundamental Questions How to measure energy consumption of an appliance  Direct Measurement Easy and precise but may not be scalable  Indirect Measurement Hard and less precise but may be scalable How to know who is using it  Annotation : Users write down when and which  Inference From absolute position From relative position among users and appliances And on/off status of appliances

7 Intuitive Ways to Answer to the Questions How do we know someone is using appliances  An appliance is on  A is in front of the appliance and interacts with it How do we say someone is wasting energy  A turned on a light  A left it on and went somewhere Some ambiguities from socio-economical reasons  On behalf of B, A is using an appliance. Who is responsible for the energy consumption?  A cooks for others, who is responsible?

8 Objectives Give some useful information for users  Amount of wasted energy  Amount of useful energy per each appliance or each person Construct fair and comprehensive policy  Less arguable policy  Agreeable policy  Reliable policy Develop an easy deployable and usable system  Less human intervention  Off-the-shelf experience

9 Related Works Intelligent Powermeter, Der intelligente Stromzähler, Germany  Real-time Powermeter monitoring device From

10 Related Works Demand response system, Berkeley  Control HVAC/Lighting  Based upon occupancy or demand From Demand Response Enabling Technology Development Tech report, Berkeley

11 Related Works MEMS nonintrusive electrical monitoring device, Leland et. al., Berkeley  Electrical monitoring  Energy scavenging for motes From Energy scavenging power sources for household electrical monitoring, Eli S. Leland et al., Berkeley

12 Related Works At the Flick of a Switch, Shwetak N. Patel et al., Gatech  Detecting and Classifying Unique Electrical Events on the Residential Power Line  Less intrusive, real-time From At the Flick of a Switch, Shwetak N. Patel et. al, Gatech

13 Related Works Ambient Devices  Provide users with customized information From

14 Classes of Appliances Underlying Assumptions  Someone, using appliances, is benefiting from them  Indirect use, ambiguous use: not a technical question or even harder to be addressed Classification of appliances  Class I: Serviced inside physical vicinity. E.g. Television, coffee machine and most of home appliances. Focus of our work  Class II: Serviced with remote access. E.g. servers, networked printers and some office appliances  Class III: Serviced independent of access. E.g. Refrigerator, alarm clocks

15 Definition : Service Range Service range of an appliance is a user defined vicinity boundary Token : Abstraction that indicates current status of usage

16 Token Issue/Expire : Case I

17 Token Issue/Expire : Case II

18 Prototype System Implementation Users carry tag motes Appliances instrumented with a power meter and tag reader Server to store and process data

19 Hardware Prototype MicaZ ‘appliance’ motes  MicaZ+COTS Power Measurement device  Measuring power consumption and RSSI value MicaZ ‘tag’ motes  MicaZ+Accelerometer  Triggered by accelerometer and broadcast its ID MicaZ ‘base’ mote  MicaZ+Laptop  Processing and backup Sensorbase  Remote spatio-temporal DB

20 Service Range and Radio Signal Strength MicaZ Mote as Tags and Readers  RSSI values give rough range information RSSI based localization technique makes use of it RSSI gives rough range information between a tag and reader  Relative position could be inferred  Limitation : Service Range cannot be arbitrary Note : Home and office setting is static

21 Technical Challenges RSSI values vary depending upon Antenna characteristics, users movement, obstruction of LOS and others Simple threshold may not be good In some cases,  Calibration helps : Antenna Characteristics  Hysteresis helps : Noise along boundary Those hamper usability issues

22 Evaluation Setting : Ideal Service Range 4 ‘appliance’ Motes  Three SOS S/W modules : PM, tag reader, UART 2 ‘tag’ Motes  One SOS S/W module : tag 1 BaseNode Mote  One SOS module : Snooping

23 Measured Service Range : Coffee Maker

24 Measured Service Range : Bedroom Lamp

25 Measured Service Range : Livingroom Lamp

26 Measured Service Range : TV

27 Some Observations Service Range for RFID tag I  about -26 of the living-room lamp  about -20 of the TV  about -5 of the coffee machine Service Range for RFID tag II  about -22 of the living-room lamp  about -14 of the TV  about -8 of the coffee machine => Need to calibrate tag characteristics => Need to reject noise effect

28 Three Schemes Unified Token Scheme  Service range = Single RSSI value Calibrated Token Scheme  Service range = Tag specific RSSI value Hysteresis Token Scheme  Service range = Tag specific RSSI value  Expire tokens when RSSI overcomes Hysteresis

29 Ground Truth Using Campaignr with a local Sensorbase  Took Pictures and annotated users’ status  Sampling rate : 10s Ground Truth tells us  Who is/are using appliances  Who has/have left appliances on/off

30 Evaluation in terms of False Positive and Negative : TV User 200User 201 Schemes Correct (%) False Positive (%) False Negative (%) Correct (%) False Positive (%) False Negative (%) Unified Calibrated Hysteresis

31 Evaluation in terms of False Positive and Negative : Bedroom Lamp User 200User 201 Schemes Correct (%) False Positive (%) False Negative (%) Correct (%) False Positive (%) False Negative (%) Unified Calibrated Hysteresis

32 Evaluation in terms of False Positive and Negative : Livingroom Lamp User 200User 201 Schemes Correct (%) False Positive (%) False Negative (%) Correct (%) False Positive (%) False Negative (%) Unified Calibrated Hysteresis

33 Performance of Various RSSI, Vicinity, Token Mechanisms

34 Appliance Power and Energy Consumption per User

35 Useful/Wasted Power and Energy for TV

36 Useful and Wasted Consumption for all Appliances

37 Conclusion Proof-of-concept implementation  Presented a system that profiles energy consumption pattern in individual level  Detection scheme was reasonably successful under simple evaluation setting  Presented various information that encourages people optimize their energy consumption pattern But the system needs to  Be easy to deploy and use  Have comprehensive and fair policy

38 Future work More comprehensive view needed Deployment under more complex setting needed Monitor other resources : Water Cope with various RSSI uncertainties or try other localization technique? Improve usability and scalability Cope with various limitations