[PPT] - Smart An Open Data Set and Tools for Enabling Research in PowerPoint Presentation

SLIDE 1

Department of Computer Science

An Open Data Set and Tools for Enabling Research in Sustainable Homes

Sean Barker, Aditya Mishra, David Irwin, Emmanuel Cecchet, Prashant Shenoy, and Jeannie Albrecht†

University of Massachusetts Amherst Williams College†

Smart

SLIDE 2

University of Massachusetts Amherst - Department of Computer Science

Smart Buildings for Sustainability

! 73% of U.S. grid power ! Efficiency, sustainability through smart homes

! Environmental benefits:

carbon footprint, renewables

! Economic benefits:

infrastructure, energy costs

2

SLIDE 3

! Algorithms, policies, ...

Flattening demand
Shifting demand using

stored energy

Optimizing renewables

! Data collection

Building a collection testbed
Scaling the testbed
Maintaining the testbed

Sean Barker (sbarker@cs.umass.edu)

Challenges of Smart Home Design

3

ne hour period

power power peak = 3000W peak = 1000W A/C 3 A/C 2 A/C 1 (b) with scheduling A/C 2 A/C 3 A/C 1

ne hour period

power peak = 2000W A/C 1 A/C 2 A/C 3

ne hour period

interactive loads power peak = 1000W (d) online scheduling

ne hour period

interactive loads (c) offline scheduling (a) no scheduling

SLIDE 4

! “I need [X] data to try [Y]”

Build a customized sensing system
Temporarily deploy in a home
Collect planned data and move on

Sean Barker (sbarker@cs.umass.edu)

The Conventional Approach

4

! Potential Drawbacks

Narrow data scope
Scalability of custom deployments
Verification may require broader

data (e.g., NILM)

SLIDE 5

Sean Barker (sbarker@cs.umass.edu)

The Smart* Approach

! Data collection in the Smart* project

Breadth of data from many sources
Scalability using off-the-shelf components
Continuity of data collection over long periods

! Key Statistics

Deployments in three homes
100+ distinct sensor streams
Energy usage, generation, weather, motion, doors, GPS...
House-level, circuit-level, device-level
Time granularities as fine as 1 second
Up to 2 years of history (and counting)

5

Smart

SLIDE 6

! Releasing two data sets today ! UMass Smart* Home Data Set

Significant subset of our home data

! UMass Smart* Microgrid Data Set

Energy dataset from 400+ homes

! Also some sensing utilities used in

ur infrastructure

! Goals:

Facilitating validation of techniques in sustainability
Identification of new research avenues in sustainable homes

Sean Barker (sbarker@cs.umass.edu)

The Smart* Open Data Sets

6

SLIDE 7

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation and overview ! Smart* Open Data Sets ! Potential Uses and Applications ! Smart* Software Tools ! Summary

7

SLIDE 8

! House-level data via commercial meter (e.g., eGauge, TED)

Real & apparent power, <1% error
One second granularity

! Voltage, frequency on both phases ! Circuit-level data via multiple current transducers (CTs)

Real & apparent, one second intervals
Many single-device circuits

Sean Barker (sbarker@cs.umass.edu)

Data Types: Aggregate Electrical Usage

8

SLIDE 9

! Two types of outlet-level energy meters ! Insteon iMeter Solo

Powerline protocol, descendant of X10

! Z-Wave Smart Energy Switch

Wireless protocol

! Almost all (>90%) loads monitored ! Scalability challenges within homes

E.g., low bandwidth, interference
[Hnat, SenSys 2011], [Irwin, BuildSys 2011]

Sean Barker (sbarker@cs.umass.edu)

Data Types: Outlet-Level Electrical Usage

9

SLIDE 10

! Multiple data granularities enable measurements

f accuracy

! Current deployment: >99% of second-level circuit readings within 4% of aggregate

Much better than our previous meter

Sean Barker (sbarker@cs.umass.edu)

Accuracy of Sub-House Readings

10

40 50 60 70 80 90 100 2 4 6 8 10

% Readings < Error Error (%)

Grid vs. ΣCircuits

SLIDE 11

! Renewable deployment at

ne home
Three solar panels
Two wind turbines
Micro-inverters feed back into

electric grid (net metering)

! Record current and attached battery voltage ! 5 second (average) data granularity

Sean Barker (sbarker@cs.umass.edu)

Data Types: Electrical Generation

11

SLIDE 12

! Wall switch events provided by Insteon-enabled switches ! Drop-in replacements for mechanical wall switches ! On/off/dim(%) events

Switch energy use derived

from events and known max wattages

Provides another level of

energy redundancy

Sean Barker (sbarker@cs.umass.edu)

Data Types: Switch Events

12 40 80 120 160 200 240 280 10 20 30 40 50 60 70 80 90 100

Dim Level (%) Wattage (W)

kitchen:lights:dim f(x)=(-13/5)x

SLIDE 13

! Motion data via Insteon (binary room-level readings) ! Door activity via Insteon (open/close) ! Heating activity via Insteon-enabled thermostats

Furnace on/off, temperature setpoint

Sean Barker (sbarker@cs.umass.edu)

Data Types: Other Events

13

SLIDE 14

! Data from deployed weather stations

One minute granularity

! Indoor readings

Temperature & humidity
Rooms and appliances

(e.g., fridge interior)

! Outdoor readings

Temperature & humidity
Rain and wind

Sean Barker (sbarker@cs.umass.edu)

Data Types: Environmental Data

14

SLIDE 15

! 443 unique homes ! Per-home electrical usage data ! Single 24-hour period ! One minute granularity ! Homes located in US

Sean Barker (sbarker@cs.umass.edu)

Microgrid Data Set

15

SLIDE 16

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation and overview ! Smart* Open Data Sets ! Potential Uses and Applications ! Smart* Software Tools ! Summary

16

SLIDE 17

Sean Barker (sbarker@cs.umass.edu)

Example Applications

! Cost Optimization: Energy storage to cut energy bills

[e-Energy 2012]

! Load Monitoring using home automation infrastructure

[BuildSys 2011]

! Renewable Prediction using weather forecasts

[SmartGridComm 2011]

! Demand Flattening: Load shifting to cut peak demand

[PerCom 2012]

! Privacy and commercial smart meters

[BuildSys 2010]
Closely related to Nonintrusive Load Monitoring (NILM)

17

SLIDE 18

! Many devices operate within a guardband range

Guardband provides ‘slack’ that can be used to timeshift

! E.g., power & environmental data reveals guardband

Sean Barker (sbarker@cs.umass.edu)

Demand Flattening: Load Shifting

18

20 40 60 80 100 120 140 160 37 37.5 38 38.5 39 39.5 40

Power (watts) Temperature (F) Time (6 hours)

Power Temperature

SLIDE 19

! Event collisions complicate disaggregation

What can we do to reduce them?

! Higher fidelity meters (smaller event ‘width’)

Sean Barker (sbarker@cs.umass.edu)

NILM: Simultaneous Events

19

2000 4000 6000 8000 10000 12000 14000 1 2 3 4

Time Intervals (1 Hz) Events During Interval

Coarser meters (7 second events) Finer meters (3 second events)

SLIDE 20

! Event collisions complicate disaggregation

What can we do to reduce them?

! Higher fidelity meters (smaller event ‘width’) ! Dedicated device meters (remove devices from trace)

Sean Barker (sbarker@cs.umass.edu)

NILM: Simultaneous Events

19

10000 20000 30000 40000 50000 60000 70000 1 2 3 4

Time Intervals (1 Hz) Events During Interval

0 dedicated meters (all devices) 1 dedicated meter 3 dedicated meters

SLIDE 21

Sean Barker (sbarker@cs.umass.edu)

Outline

! Motivation and overview ! Smart* Open Data Sets ! Potential Uses and Applications ! Smart* Software Tools ! Summary

20

SLIDE 22

! ‘Off the shelf’ sensors are easy to use!

...in theory, anyways

! Still have difficulties to deal with

Proprietary or difficult-to-script software
Immature open-source options
Not designed for continuous monitoring at scale

! Releasing utilities for Insteon and Z-Wave meters

Hides protocol details and simplifies configuration

! May release higher-level components of our sensing infrastructure in the future

Sean Barker (sbarker@cs.umass.edu)

Sensing Software Tools

21

SLIDE 23

Sean Barker (sbarker@cs.umass.edu)

Summary

! Data sets with both breadth and depth are important for research in sustainability ! Releasing two data sets (and related utilities) today

UMass Smart* Home Data Set
UMass Smart* Microgrid Data Set
Periodic updates to come

! Go download them!

22

http://smart.cs.umass.edu

Questions? sbarker@cs.umass.edu