S MART C AP : Flattening Peak Electricity Demand in Smart Homes - PowerPoint PPT Presentation

S MART C AP : Flattening Peak Electricity Demand in Smart Homes Sean Barker , Aditya Mishra, David Irwin, Prashant Shenoy, and Jeannie Albrecht† University of Massachusetts Amherst Williams College† Department of Computer Science

Pervasive Computing in Smart Homes ! Smart homes : efficiency, automation, convenience ! Enabled by pervasive computing • Smart meters, energy sensors, load controllers • Appliance integration ! Greening smart homes • Why? 73% of U.S. electricity ! Economic benefits: • infrastructure, energy costs ! Environmental benefits: • carbon footprint, renewables Sean Barker (sbarker@cs.umass.edu) 2

Demand-Side Energy Management ! Control consumer-side energy demand • Respond to energy availability Dryer • Reduce peak usage, fluctuations Heater Power ! Components of DSEM Light • Monitoring (data collection) • Control (e.g., load shifting) Time Peak Usage Off-Peak Shiftable Load ! We focus on performing peak load reduction Sean Barker (sbarker@cs.umass.edu) 3

Benefits of Peak Load Reduction ! For utilities: • Lowered peak grid demand • Infrastructure savings • Transmission & distribution ( loss ∝ current 2 ) 47%! ! For consumers: • Variable pricing cost savings • Battery efficiency • Assist with capping Sean Barker (sbarker@cs.umass.edu) 4

Challenges of Peak Load Reduction ! Change user behavior • Users don’t want to! • Maintain household routines Lights off until 9 pm! ! Inflexible loads • Unacceptable: lights, TV • Inconvenient: dishwasher ! Goal: transparent peak reduction • No user cooperation • No negative impact Sean Barker (sbarker@cs.umass.edu) 5

Outline ! Motivation ! Home measurement study • What can we schedule transparently? ! SmartCap scheduler • How can we perform transparent peak reduction? ! Evaluation on home data • How effectively does SmartCap flatten demand? ! Conclusions Sean Barker (sbarker@cs.umass.edu) 6

Types of Loads Not OK to interfere! ! Interactive loads • Controlled by users • TV, lights, microwave • Little scheduling freedom! ! Background loads • Not controlled by users • A/C, refrigerator, heater • Don’t care how objective is met • Significant scheduling freedom! OK to change on/off times! Sean Barker (sbarker@cs.umass.edu) 7

SmartCap Measurement Study ! Home monitoring deployment • 3 occupants, one year (so far) ! Instrumented all outlets and switches • Lights, TV, dishwasher, freezer, A/C, etc. renewables Outlet/Switch power Meters readings SmartCap Gateway commands Appliances grid power power readings Programmable Panel Switches Meter Sean Barker (sbarker@cs.umass.edu) 8

1. Background Loads are Significant ! Few major background loads • heat recovery ventilator • refrigerator Load Peak Average Quantity • freezer Refrigerator 456W 74W 1 • dehumidifier Freezer 437W 82W 1 • air conditioner (x3) HRV 1129W 24W 1 Dehumidifier 505W 371W 1 Main A/C 1046W 305W 1 ! 8% of loads but Bedroom A/C 1 571W 280W 1 Bedroom A/C 2 571W 141W 1 59% of energy use Background 4715W 1277W 7 Interactive 9963W 887W 85 ! Background loads are few but major energy users Sean Barker (sbarker@cs.umass.edu) 9

2. Background Loads are Periodic ! Periodicity: regular on/off intervals ! Mostly (but not fully) independent of user behavior unusual 500 450 event 450 freezer refrigerator 400 400 350 350 regular 300 irregular 300 intervals 250 intervals 250 200 200 150 150 100 100 50 50 0 0 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am 140 160 HRV dehumidifier 140 120 120 100 100 80 80 60 60 40 40 20 20 0 0 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am ! Background loads vary but have useful periodicity Sean Barker (sbarker@cs.umass.edu) 10

3. Interactive Loads are Unpredictable mealtime 3000 ! “Peaky” total load interactive peaks 2500 Power (watts) • Brief, high-power devices 2000 1500 ! Many unpredictable 1000 individual loads 500 • Human usage patterns 0 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am • May change over time Time 200 1200 television coffeepot 1000 150 800 100 600 400 ! Peak reduction must 50 200 0 0 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am compensate for 100 18 90 lamp 16 80 14 interactive loads 70 12 60 10 50 8 40 6 30 4 20 entertainment center 2 10 0 0 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am 5 am 9 am 1 pm 5 pm 9 pm 1 am 5 am Sean Barker (sbarker@cs.umass.edu) 11

Flattening in SmartCap ! Background loads cycle on and off ! Exact on/off times (mostly) don’t matter ! Schedule cycling of background loads (a) no scheduling (b) with scheduling peak = 3000W power A/C power 3 A/C 2 peak = 1000W A/C A/C A/C A/C 1 1 2 3 one hour period one hour period ! Interleave background loads to flatten peaks Sean Barker (sbarker@cs.umass.edu) 12

Scheduling Loads: Slack ! Periodic background loads exhibit ‘slack’ • Measure of how long background load can remain off • Based on guardband (e.g., fridge temperature range) Active cooling: 160 40 Power accumulating slack Temperature 140 39.5 Temperature (F) Power (watts) 120 39 100 80 38.5 60 38 40 37.5 20 Passive warming: 0 37 consuming slack Time (6 hours) ! Control slack by modifying device duty cycle Sean Barker (sbarker@cs.umass.edu) 13

SmartCap Scheduler ! Schedule based on remaining slack ! Least Slack First (LSF) • Operate loads in order of ascending slack ! Online scheduler • Respond to foreground (interactive) loads (c) offline scheduling (d) online scheduling interactive loads power power peak = 2000W interactive loads peak = 1000W A/C A/C A/C 1 2 3 one hour period one hour period ! Preempt background loads by interactive loads Sean Barker (sbarker@cs.umass.edu) 14

SmartCap Evaluation ! Evaluate LSF on home data 160 40 Power Temperature 140 • Computed per-period slack 39.5 Temperature (F) Power (watts) 120 39 • Linear slack model 100 80 38.5 60 38 40 ! Flattening metric 37.5 20 0 37 Time (6 hours) • Average deviation from mean ! Flattening period (a) One day (b) Four hours Daytime Nighttime ! High deviation period (mealtimes) or low (nights) Sean Barker (sbarker@cs.umass.edu) 15

Evaluation: Smart Home ! Day-long periods % Deviation Decrease 70 % LSF Improvement 60 No Improvement • Flattening on 91% of days Up to 50% 50 40 improvement 30 • 16% average flattening 20 10 0 -10 Untimely 0 10 20 30 40 50 60 70 80 Days ! Four-hour periods interactive loads (a) Each Day • High variance (31%) % Deviation Decrease 25 >1kW % LSF Improvement No Improvement 20 • >20% flattening Variance 15 10 • Low variance (69%) 5 0 • <3% flattening -5 0 5 10 15 20 25 30 35 40 4-hour Periods ! LSF especially good at flattening high peak periods Sean Barker (sbarker@cs.umass.edu) 16

Evaluation: Electric Vehicle ! Electric Vehicle (EV) • Peaky typical usage • Grid unreadiness at scale • EV load added to home data 7000 No Scheduling LSF (3.0kW) 6000 Power (watts) 5000 Peak power 4000 periods reduced 3000 2000 1000 0 0 10 20 30 40 50 60 70 80 90 100 Percentage of Time ! EV is good candidate for LSF: 22% flattening Sean Barker (sbarker@cs.umass.edu) 17

Evaluation: Lab Testbed ! Live testbed for active control, repeatability • ‘Smart appliances’ via programmable Insteon switches ! Active background scheduling in LSF 3500 Background No Scheduling loads pushed LSF Aggregate Power (watts) 3000 forward 2500 Background 2000 1500 1000 500 0 Time (4 hours) ! 23% flattening (background + interactive) on testbed Sean Barker (sbarker@cs.umass.edu) 18

Related Work ! Demand-side energy management • Load shifting [Keshav, GreenNet 10] • Prediction [Schülke, SmartGridComm 10] • Batteries [Zhu, BuildSys 11], [Bar-Noy, WEA 08] ! Background schedulers • Optimizing for renewables [Taneja, SmartGridComm 10] • Offline scheduling [Bakker, SmartGridComm 10] Sean Barker (sbarker@cs.umass.edu) 19

Conclusions ! Demand-side energy management for peak reduction ! SmartCap flattens transparently • Modifies only background loads • Interactive loads unaffected ! 20-30% flattening using Least Slack First ! Additional savings possible with modest user changes • Subject of ongoing work Sean Barker (sbarker@cs.umass.edu) 20

Questions? Sean Barker sbarker@cs.umass.edu Department of Computer Science

Least Slack First Threshold ! Threshold power to preempt loads • Start scheduling when threshold is reached No Scheduling 5000 LSF (2.2kW) Power (watts) 4000 A/C's 1, 2, 3 activate 3000 2000 Stacking 1000 triggers scheduling 0 6 am 7 am 8 am 9 am Time (Hours) ! Adaptive threshold (moving average of past use) Sean Barker (sbarker@cs.umass.edu) 22