Metering, Monitoring and Making Sense of Energy Use in ‘Mixed-Use’ Buildings Rajesh K. Gupta Professor & Chair, Computer Science & Engineering Associate Director, California Institute for Telecommunications & Information Technology University of California, San Diego Yuvraj Agarwal, Rajesh Gupta Thomas Our Team Bharath Seemanta John Sathya Kaisen

Buildings are an important research focus All electricity in the US: 3,500 TWh  ~500 power Buildings: 2,500 TWh All electronics: 290 TWh Buildings consume significant energy >70% of total US electricity consumption >40% of total carbon emissions Buildings consume significant energy >70% of total US electricity consumption >40% of total carbon emissions Bruce Nordman, LBNL BuildSys 1 PC per 200 sq. foot 1 PC = $100 1W saved = ~2W less imported = 5W less produced. 1 PC per 200 sq. foot 1 PC = $100 1W saved = ~2W less imported = 5W less produced.

Energy Dashboard

Looking across 5 types of buildings From: Yuvraj Agarwal, et al, BuildSys 2009, Berkeley, CA. more IT

Modern Buildings Are IT Dominated: 50% of peak load, 80% of baseload

Two Steps to Improving Energy Efficiency 1. Reduce energy consumption by IT equipment  Servers and PCs left on to maintain network presence  Key Idea: “Duty-Cycle” computers aggressively  SleepSever: maintains seamless network presence 2. Reduce energy consumption by the HVAC system  Energy use is not proportional to number of occupants  Key Idea: Use real-time occupancy to drive HVAC  Synergy wireless occupancy node 6

Duty Cycling: Processors, HVAC Why not power-down machines that are not working?  Or power-down building HVAC systems Runs into several use model problems  “Always ON” abstraction of the internet Unlike light-bulb, ‘when not in room, turn off the light’  Use model for the user/application and the infrastructure are different Network, enterprise system maintenance: distributed control of duty-cycling has its own usability problems.

Collaborating Processors Somniloquy daemon Somniloquy daemon Host processor, RAM, peripherals, etc. Operating system, including networking stack Apps Network interface hardware Secondary processor Embedded CPU, RAM, flash Embedded OS, including networking stack wakeup filters Appln. stubs Host PC  Fundamental Problem: Our Notions of Power States  Hosts (PCs) are either Awake (Active) or Sleep (Inactive)  Power consumed when Awake = 100X power in Sleep!  Users want machines with the availability of active machine, power of a sleeping machine. Somniloquy SleepServers Somniloquy SleepServers Maintain availability across the entire protocol stack, e.g. ARP(layer 2), ICMP(layer 3), SSH (Application layer)

Somniloquy exploits heterogeneity to save power and maintain availability % less energy than using host PC. Increase battery life from 60 hrs Stateful applications: Web download “stub” on the gumstix 200MB flash, download when Desktop PC is asleep Wake up PC to upload data whenever needed

SleepServers for Enterprises: Architecture Respond: ARPs, ICMP, DHCP Wake-UP: SSH, RDP, VoIP call Proxy: Web/P2P downloads, IM

Average Power 26 Watts Average Power 96 Watts DE Total estimated Savings for CSE (>900PCs) : $60K/year Deployed SleepServers across 50 users Energy Savings: 27% - 85% (average 70%)

12 Scenario: CSE Energy Use Reductions Deploy Somniloquy / Sleepserver – Machine room : 142 kW  71 kW – PC Plug loads : 130 kW  70 kW Ventilation system: – New fans, chillers : 65 kW  52 kW Lighting: – Fluorescent lighting  LED – Motion-detector controlled hallway lighting evenings & weekends: 50 kW  11 kW 80 kBTU/ft 2 42 kBTU/ft 2

13 Could CSE become a ZNEB? Solar energy : 2700 m 2 roof 111 kBTU/ft 2 Solar PhotoVoltaic: 20% efficient 22 kBTU/ft 2 How do we achieve 42 kBTU/ft 2 ? – Tracking solar PV : add 30% irradiance 28 kBTU/ft2 – Increase PV efficiency : 29% efficient 42 kBTU/ft2 Dramatic improvements in energy efficiency and solar conversion efficiency needed for ZNEB

Wait for global warming or better solar cells? Is that it?

15 Buildings 2.0: Occupancy-Driven Smart Buildings Use occupancy and activity to drive energy efficiency in HVAC system usage. Increased HVAC when a room has more occupants. Reduced cooling when a room is empty. When there are less people in the room, reduce cooling. When there are more, increase cooling as required to maintain comfort. Occupancy Performability Adaptive Envelope Occupancy Performability Adaptive Envelope

16 HVAC: Central control and Static Schedules 16 HVAC ON 5:15AM 6:30PM Some people actually arrive 2 hours later! HVAC starts at this time Un-Occupied Periods HVAC stops at this time

17 Energy Consumption in a Mixed-Use Building 17 HVAC loads significant: Electrical ( >25%) and Thermal – Electrical (air handlers, fans, etc), thermal (chilled water loop) – HVAC load independent of the actual occupancy of building

18 Relating HVAC Energy Use and Occupancy Controlled experiment in CSE over 3 days: Fri, Sat, Sun – Friday: Operate HVAC system normally – Weekend: HVAC duty-cycled on a floor-by-floor basis – 1 floor (10am – 11am), 2 floors (11am – 12pm), ….., ….. Occupancy affects HVAC energy – Points to the benefits of fine-grained control 18

19 Occupancy Driven HVAC control 19 Key Design Requirements: Inexpensive (less than 10$) Battery powered – 4-5 year life Multiple sensors for accuracy Synergy Occupancy Node CC2530 based design 8051 uC radio Zigbee compliant stack PIR + Magnetic reed switch

20 Accuracy of Occupancy Detection 20 Over 96% occupancy accuracy with Synergy node

21

22 Deployment across 2 nd floor of CSE Offices, 20 Labs. - 8 Synergy Base Stations Control individual HVAC zones based on real-time occupancy information! Floormap: 2 nd Floor

23

24 Implementation: Interfacing with the EMS 24 NAE Windows Server with OPC Tunneller BACnet OPC DA Server HVAC Control Occupancy Data Analysis Server (ODAS) Database Sheeva Plug base stations Occupancy nodes Metasys ADX NAE … Database Occupancy Data Analysis Server Database to store mapping, MetaSys EMS – proprietary protocols OPC tunnel to communicate with EMS Actuation based on modifying status for individual thermal zones Use priorities levels -- co-exist with current campus policies. Occupancy data not visible externally

25 HVAC Energy Savings Estimated 40% savings if deployed across entire CSE! Detailed occupancy can be used to drive other systems. 25 HVAC Energy Consumption (Electrical and Thermal) during the baseline day. HVAC Energy Consumption (Electrical and Thermal) for a test day with a similar weather profile. HVAC energy savings are significant: over 13% (HVAC-Electrical) and 15.6% (HVAC-Thermal) for just the 2 nd floor

26 Summary HVAC energy not proportional to occupancy – Use of static schedules is common – Significant energy wasted Fine-grained occupancy driven HVAC control – Occupancy node: accurate, low cost, wireless – Interface with existing building SCADA systems Evaluation: Deployment in the CSE building/UCSD – 11.6% (electrical) and 12.4% (thermal) savings – Estimate over 40% savings across entire building 26

27 Some (Recent) Pointers “Evaluating the Effectiveness of Model-Based Power Characterization”, USENIX Advanced Technical Conference (ATC), "Duty-Cycling Buildings Aggressively: The Next Frontier in HVAC Control", ACM/IEEE IPSN/SPOTS, "Occupancy-Driven Energy Management for Smart Building Automation", ACM BuildSys "SleepServer: A Software-Only Approach for Reducing the Energy Consumption of PCs within Enterprise Environments", USENIX ATC, "Cyber-Physical Energy Systems: Focus on Smart Buildings", DAC "The Energy Dashboard: Improving the Visibility of Energy Consumption at a Campus-Wide Scale“, ACM BuildSys "Somniloquy: Augmenting Network Interfaces to Reduce PC Energy Usage", NSDI

An exciting time to be doing research in embedded systems with tremendous potential to solve society’s most pressing problems. Thank You Rajesh Gupta Rajesh Gupta