Exploiting Home Automation Protocols for Load Monitoring in Smart - PowerPoint PPT Presentation

Exploiting Home Automation Protocols for Load Monitoring in Smart Buildings David Irwin, Anthony Wu†, Sean Barker , Aditya Mishra, Prashant Shenoy, Jeannie Albrecht‡ University of Massachusetts Amherst Amherst College† Williams College‡ Department of Computer Science

Why Smart Buildings?  ~70% of grid power usage  Smart buildings for grid efficiency  Economic benefits  Environmental benefits University of Massachusetts Amherst - Department of Computer Science 2

Demand-Side Energy Management  Managing energy usage • Shifting loads • Reducing loads Peak Usage Off-Peak Usage Shiftable Load  DSEM components: • Continuous energy monitoring • Load control off on University of Massachusetts Amherst - Department of Computer Science 3

Energy Monitoring Systems  Primary goals • Inexpensive • Non-invasive  Examples • ViridiScope [Ubicomp 09] • ACme [SenSys 09] • Single-point sensing • ElectriSense [Ubicomp 10] • Flick of a Switch [Ubicomp 07] • Many others  Monitoring ≠ Control University of Massachusetts Amherst - Department of Computer Science 4

Exerting Control on Electrical Loads  Home automation (HA) products  Inexpensive and mature • X10 (1975) • Insteon (2001) • Z-Wave (2005)  Already deployed in smart grid trials www.insteonsmartgrid.com University of Massachusetts Amherst - Department of Computer Science 5

Combining Monitoring and Control  Load control requires adding HA-like on / off hardware to devices events  Augment HA with monitoring  Challenges: (1) Very low bandwidth and primitive HA protocols (2) Coarse-grained events rather than fine-grained data streams (3) Mapping events to power data power data University of Massachusetts Amherst - Department of Computer Science 6

Our System: AutoMeter  AutoMeter : our HA system for building-wide monitoring  Low-cost, off-the-shelf components  Wall switches • on/off/dim event notifications  Power meters • queries for outlet-level data  Prototype deployment in a home University of Massachusetts Amherst - Department of Computer Science 7

AutoMeter Architecture Building/Circuit Power AutoMeter Panel Load Controller Disaggregation Switch Events Plug Power University of Massachusetts Amherst - Department of Computer Science 8

System Components  Using the Insteon HA protocol  Why Insteon? • Low cost (~$40 per device) • More reliable than X10 • Non-proprietary • Not solely reliant on wireless  Complications • Reverse engineering meter protocol • No notifications from meters level changed? • Usable bandwidth <180 bps ... University of Massachusetts Amherst - Department of Computer Science 9

Insteon Protocol Overview  Increasing message reliability: • Message propagation, ACKs, retransmissions  Bandwidth limits: • Theoretical: 2880 bps, practical: <180 bps Plug h 3 o p s = 1 Switch hops = 1 2 PowerLine hops = 2 hops = 2 Plug hops = 2 2 hops = 3 Plug 1 hops = 2 3 = s p o h Switch PLM USB Controller 1 hops = 3 "Query Plug 3" University of Massachusetts Amherst - Department of Computer Science 10

Current AutoMeter Deployment  3 bedroom, 2 bath house, 34 wall switches • 20 Insteon SwitchLinc relays • 10 Insteon SwitchLinc dimmers • 30 Insteon iMeter Solos • TED 5000 for aggregate readings • GuruPlug control server connected to Insteon PowerLine Modem (PLM)  Entire equipment budget: $3025  96.7% of total TED energy use accounted for in a two-week period University of Massachusetts Amherst - Department of Computer Science 11

Issues Encountered  1. Low bandwidth and message losses • Trading off query rate and reliability  2. Learning switch power usage • Proactive and reactive strategies  3. Tagging aggregate power variations • Remapping power changes back to events University of Massachusetts Amherst - Department of Computer Science 12

Problem 1: Low Data Rates  <180 bps over power line  No collision avoidance • Serial meter queries • Asynchronous switch events ‘switch off’  Approach: insert delay X ‘query meter’ between subsequent queries  Delays for single meter multiplied by # of meters! University of Massachusetts Amherst - Department of Computer Science 13

Meter Query Losses  Approximate query duration: 1.0333 sec  Reliability vs. global query rate  Much lower reliability with lower interarrival times % Queries Received 100 80 60 40 Home Deployment 20 Isolation Model (no retransmissions) 0 0 1 2 3 4 5 6 7 8 9 10 Interarrival Time (sec) 1.0333s University of Massachusetts Amherst - Department of Computer Science 14

Wall Switch Event Losses  Cannot control when event messages occur  Increase interarrival time to reduce collisions  Round-robin queries every 10s (5 min / device) • <5% switch loss probability 100 Home Deployment 90 Model (no retransmissions) % Events Lost 80 70 60 50 40 30 20 10 0 0 1 2 3 4 5 6 7 8 9 10 Interarrival Time (sec) University of Massachusetts Amherst - Department of Computer Science 15

Smart Polling  Idea: How much energy could we miss between queries to a device?  Cap amount of unaccounted energy  Per-device query rate • power usage and typical duration on or off? on or off? vs. slow queries fast queries University of Massachusetts Amherst - Department of Computer Science 16

Problem 2: Learning Switch Power  Switches only report on / off / dim  Goal : learn switch power  Use aggregate TED data ‘switch usage: 100W’ h o n ’ ‘ s w i t c  Simple proactive approach • Programmatically disable all loads ‘power: 100W’ • Turn device on, record delta • Repeat for each device • 93% accurate, but requires cycling University of Massachusetts Amherst - Department of Computer Science 17

Reactive Approach: Learning on-the-fly  Learning power values on-the-fly  Problems encountered • Delayed data points • Simultaneous events ‘delta: 61W, record in 55-65W bin’ • Bad readings h o n ’ ‘ s w i t c  Record deltas around events ‘last power: 432W, • ‘Bin’ them based on delta size new power: 493W’ • Avg most common bin (e.g., 55-65W deltas) as energy value  Intuition: over many events, bins will reveal true value University of Massachusetts Amherst - Department of Computer Science 18

Reactive Approach: Binning  Wide range of energy deltas around events  Bins usually identify true delta • But...need enough data points • And highly correlated events are bad 40 guestbath:overheadlight guestbath:sinklight masterbath:sinklight 35 30 Number Events 25 20 15 10 5 0 15- 25- 35- 45- 55- 65- 75- 85- 95- 105- 115- 125- 135- 145- 155- 165- 175- 25W 35W 45W 55W 65W 75W 85W 95W 105W 115W 125W 135W 145W 155W 165W 175W 185W Watt Bins University of Massachusetts Amherst - Department of Computer Science 19

Problem 3: Tagging Power Variations  Objective: tag aggregate data with specific events  Problem: errors in aggregate data • Reading errors (2% TED error) • Timing errors (missed readings, 5-minute frequencies) • Rampup errors (TED readings change gradually)  Many events missed even with high deltas 100 80 Percentage 60 40 20 Individual Events:Building Events 0 0 100 200 300 400 500 600 700 800 900 Threshold (W) Figure 6. We use AutoMeter’s switch, plug, and circuit University of Massachusetts Amherst - Department of Computer Science 20

Conclusions  HA protocols show promise for providing monitoring capabilities • Smart polling • Accurate building data • Other types of data – circuits, topologies, ...  Issues encountered • Switch power: learn proactively or reactively over time • Outlet power: cope with limitations with intelligent polling  Time and cost not a significant barrier for complete HA instrumentation University of Massachusetts Amherst - Department of Computer Science 21

Questions? Department of Computer Science