Working together to put CPS Technology on the Sustainability Table David E. Culler University of California, Berkeley August 2, 2011 What can we put on the table? CPS-PI-11 8/2/2011 2 Sustainability? “Sustainable development should meet the needs of the present without compromising the ability of future generations to meet their own needs” Our Common Future, World Commission on Environment and Development, United Nations, 1987 CPS-PI-11 8/2/2011 3 Quantifying it - California Law AB 32 Reduce GHG emissions to 1990 levels by 2020 Governor’s executive order S-3-05 (2005) 80% reduction below 1990 levels by 2050 Renewable Portfolio Standard 33% renewables by 2020, 20% biopower procurement 480 => 80 mmT CO2e in 40 years Population expected to grow from 37 => 55 million Economic growth CPS-PI-11 8/2/2011 4 CPS-PI-11 8/2/2011 5 What it amounts to … Historical and BAU Emissions Energy emissions Nonenergy emissions 2020 Target CPS-PI-11 14.62 12.93 5 2050 1.56 0 8/2/2011 CA 2050 Target BAU 23.4 9.88 10 CA 2020 Target Historical 2020 15 CA 2005 2005 20 CA 1990 1990 Metric tons CO2 per capita per year 25 Per Capita CO2 Emissions 6 US 2003 GHG Emissions (MtCO2e/yr) 1,000 900 800 700 600 500 400 300 200 100 0 Can we do this? CPS-PI-11 8/2/2011 7 But CPS-PI-11 8/2/2011 8 My Take … We can’t build our way to sustainability The discussion is focused on the Physical New Windows, Materials, Motors, Biofuels, … Need to make the best of what we have, or will have, and how we use it… This takes observation, intelligence and control Demonstrable CPS technology on the table Efficiency & Supply-Following Loads 10 years to innovate, 30 years to scale CPS-PI-11 8/2/2011 9 What’s on the Table? – buildings CPS-PI-11 8/2/2011 10 4 Part GHG Reduction Plan Efficiency CPS-PI-11 8/2/2011 11 4 Part GHG Reduction Plan Efficiency Electrify CPS-PI-11 8/2/2011 12 4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel CPS-PI-11 8/2/2011 13 4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel CPS-PI-11 8/2/2011 14 All required for even 60% reduction but still fall short of 80% CPS-PI-11 8/2/2011 15 Efficiency CPS-PI-11 8/2/2011 16 Electrification CPS-PI-11 8/2/2011 17 Low Carbon Electricity Options CPS-PI-11 8/2/2011 18 Build Rate CPS-PI-11 8/2/2011 19 The Renewables are there CPS-PI-11 43% agriculture, 3.4% urbanized 8/2/2011 20 The Problem: Supply-Demand Match Baseline + Dispatchable Tiers Generation CPS-PI-11 Transmission Oblivious Loads Distribution Demand 8/2/2011 21 Load-following Supply (?) Growing proportion of renewables leads to higher price volatility. October 2008 to March 2010: >90 hours with negative prices; highest price reached: +€500/MWh, lowest -€500/MWh 494.26 330 300 270 240 210 180 150 120 90 60 30 01.03.2010 01.02.2010 01.01.2010 01.12.2009 01.11.2009 01.10.2009 01.09.2009 01.08.2009 01.07.2009 01.06.2009 01.05.2009 01.04.2009 01.03.2009 01.02.2009 01.01.2009 01.12.2008 -60 01.11.2008 €/MWh -30 01.10.2008 0 -90 -120 -500.02 -150 Daily maximum price Source: EEX spot prices. CPS-PI-11 Daily minimum price (indicated in red when negative) 8/2/2011 22 … and in CA CPS-PI-11 8/2/2011 23 Zero Emissions Load Balancing (ZELB) Just the emissions from the natural gas to firm the 33% renewables exceeds GHG target Even with 50% with natural gas & 50% with some yet-to-exist storage CPS-PI-11 8/2/2011 24 We Understand Supply CPS-PI-11 8/2/2011 25 Not demand CPS-PI-11 8/2/2011 26 Limits to Renewable Penetration Variability, Intermittency of Supply Visibility into Availability of Supply Ability of Loads to Adapt Algorithms and Techniques for Reactive Load Adaptation Capability of the Infrastructure to maintain the match CPS-PI-11 8/2/2011 27 CPS and the 4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel CPS-PI-11 Monitoring, Analysis, Modeling, Waste Elimination, Power Proportionality, Optimal Control Intelligence, Communication, adaptation in Everything ZELB. Supply-Following Loads, Energy SLA, Cooperative Grid 8/2/2011 28 CPS Technology … National-scale Physical Information Service Data collection, streaming, archiving, querying Modeling, Analysis, Control Representation, metadata, life-cycle Use others’, contribute yours CPS-PI-11 8/2/2011 29 Some starts CPS-PI-11 8/2/2011 30 Applications sMAP: Uniform Access to Diverse Physical Information Personal Feedback Modeling Visualization Control Storage Location Continuous Commissioning Actuation Debugging JSON Objects Authentication Physical Information sMAP CPS-PI-11 Water Structural Electrical Weather REST API Geographical Environmental Occupancy HTTP/TCP … Actuator 8/2/2011 31 sMAP ecosystem Applications sMAP Resources sMAP Gateway California ISO Database sMAP sMAP sMAP Google PowerMeter Weather AC plug meter Edge Router EBHTTP Translation EBHTTP / IPv6 / 6LowPAN Wireless Mesh Network Proxy Server Vibration / Humidity Internet sMAP sMAP Every Building Temperature/PAR/TSR Dent circuit meter Light switch CPS-PI-11 sMAP sMAP RS-485 Modbus sMAP Gateway sMAP Gateway Cell phone 8/2/2011 32 Cory CEC B2G Testbed DOP HVAC Whole Bldg MCL equip MCL infra Central vent office HVAC MCL vac Lighting servers Plug loads CPS-PI-11 inst Lab 199 HVAC Parking Lot 8/2/2011 33 DOE MELS => Appliance Energy LBNL Bldg 90 611 of 1200 loads CPS-PI-11 8/2/2011 34 DOE/UCB/Siemens Auto Demand Response Sutardja Dai Hall 400 Lights & Plugs 350 Power (kW) 300 Servers & Cooling 250 Nano Fab 200 150 HVAC 100 Other 50 0 www.openbms.org CPS-PI-11 8/2/2011 35 BACNet => sMAP Sensors, Actuators Controllers Sutardja Dai Hall Technical Specification Sheet Rev. 9, August 2003 Internet Modular Building Controller Siemens P2 over RS-485 safety, security, and lighting). Up to 100 modular field panels communicate on a peer-to-peer netw ork. Remote Login sMAP Features M odular hardw are components to match equipment to initial control requirements w hile providing for future expansion M odular, snap-in design simplifies installation and servicing Transparent view ing panels on the enclosure door to view the status indicator LEDs and override sw itch positions Integration platform for communications and interoperability w ith other systems and devices Proven program sequences to match equipment control applications Figure 1. M odular Building Controller. Description The M odular Building Controller (M BC) is an integral part of the APOGEE® Building Automation System. It is a high performance, modular Direct Digital Control (DDC) supervisory field panel. The field panel operates stand-alone or netw orked to perform complex control, monitoring and energy management functions w ithout relying on a higher level processor. The M BC provides central monitoring and control for distributed Floor Level Netw ork (FLN) devices and other building systems (e.g., chiller, boiler, fire/life CPS-PI-11 Advanced Proportional Integral Derivative (PID) loop tuning algorithm for HVAC control to minimize oscillations and guarantee precise control Built-in energy management applications and DDC programs for complete facility management Apogee BACnet/IP Server Gateway Modem Comprehensive alarm management, historical data trend collection, operator control and monitoring functions Support for peer-to-peer communications over Industry standard 10/100 Base-T TCP/IP netw orks. 8/2/2011 36 Data from 1 Modern Building 1358 control settings Set points, Relays (lights, pumps, etc), Schedules 2291 meters/sensors Power (building, floor, lights, chiller, pumps, etc) Current, voltage, apparent, real, reactive, peak Temp (rooms, chilled water, hot water) Air volume 4+ million Commercial, Alarms, Errors 2165 control outputs 110+ million Residential Dampers, valves, min/max flow, fan speed, PID parameters 72 other CPS-PI-11 8/2/2011 37 Current sMAP demography Name Sensor Type Physical Layer Cory Hall Submetering Dent Three-Phase Modbus + Ethernet 79 3318 Cory Hall Metering ION and pQube Meters HTTP/Ethernet 3 150 Cory Lab Temperature TelosB 802.15.4 + Ethernet 4 8 Cory Lab Machines ACme 804.15.4 + Ethernet 8 16 Cory Chilled Water HeatX Modbus + Ethernet 1 11 Cory Weather Vaisala WXT520 SDI-12 + Ethernet 1 11 Soda Hall Sun Blackbox Fan speed; environmental Http/Ethernet 10 84 Soda Lab Machines ACme 802.15.4 + Ethernet 40 80 Soda Lab Panel Veris E30 Modbus + Ethernet 1 42 Soda SCADA Data Barrington controls RS-485/various “1” 1670 LBNL Building 90 Acme 802.15.4 + Ethernet 550 1650 Campus Power Data Obvius Aquisuite; various XML/HTTP/Ethernet 4 100 Citris SDH BACnet Siemens Apogee BACnet/IP + RS-485 “1” 600+ Brower Buildinbg Johnsons Control BACnet “1” 1500 CPS-PI-11 Sense Points 8/2/2011 Channels 38 m e t e r resample aggregate query Time-series Interface Bucketing RPC Compression Storage mapper Key-Value Store Page Cache CPS-PI-11 Lock Manager readingdb D e n t c i r c u i t ~300k pts/sec streaming pipeline insert sMAP sMAP A Stream Engine SQL Storage Alloc. MySQL 8/2/2011 39 Scaling Out in the Cloud Stream Management System Data Source Data Processing Data Sink Query Translator SQL MR Hadoop HBase Hive Map-Reduce Columnar Data Storage Analytics Workloads HDFS Real Time Monitoring CPS-PI-11 Interactive Workloads 8/2/2011 40 15,000 pts and growing… CPS-PI-11 8/2/2011 41 CPS Technology National-Scale Physical Info Service Software foundations, platforms and solutions for Energy Efficiency and Agility CPS-PI-11 8/2/2011 42 Stages of Energy Effectiveness Waste Not Do Nothing Well !!! Power Proportionality Peak Performance : Power => Safety Optimize Partial Load - from nothing to peakl Sculpting Identify the energy slack and utilize it Negotiated Grid / Load / Human Interaction Plan, Forecast, Negotiate, Manage CPS-PI-11 8/2/2011 43 Our Buildings Lighting HVAC IT and Plug Load PDUs, CRACs Servers 8/2/2011 CPS-PI-11 44 PowerProportional Buildings ? 950 KW 1150 KW Cory Hall: Office + Semiconductor + IT Min = 82% of Max CPS-PI-11 8/2/2011 45 PowerProportional Buildings ? 1.45 MW 2.02 MW Stanley Hall: Office + BioScience - 13 NMRs Min = 72% of Max CPS-PI-11 8/2/2011 46 PowerProportional Buildings ? 202 KW LeConte Hall: Office 62 KW Min = 31% of Max CPS-PI-11 8/2/2011 47 Re-Flash the HVAC … Whole Bldg LoCal + ActionWebs inst Lab 199 HVAC CPS-PI-11 8/2/2011 48 Learning-based Model Predictive Control Mathematical model from Newton’s law of cooling time constant of room heating from occupants and equipment weather dT/dt = -krT - kcu(t) + kww(t)) + q(t) change in temperature over time AC cooling Model identified using semi-parametric regression (°C) 23 Temperature: Experimental (blue) Simulated (red) 22 21 11AM 12PM 1PM 2PM 3PM 4PM 5PM 6PM 7PM 8PM 7PM 8PM Heating from occupants and equipment (°C/s) Time 0.029 0.0285 11AM 12PM 1PM 2PM 3PM 4PM 5PM 6PM Time (Aswani, Master, Taneja, Culler, Tomlin, Proc. of IEEE, 2011) CPS-PI-11 8/2/2011 49 Living Lab… LBMPC saved 30% energy when hotter outside 70% energy when cooler outside Standard MPC was inconsistent Two-position control was inefficient Estimated Energy Tracking Error Temperature Variation Average External Load LBMPC 23.6kWh 0.75°C 0.13°C 11.0°C Linear MPC 30.5kWh 0.62°C 0.30°C 11.0°C Method Two-Position Control Experiment LBMPC Experiment CPS-PI-11 Measured Energy Two-Position 32.6kWh 35.1kWh 0.61°C 0.20°C 11.0 °C LBMPC 11.8kWh 13.3kWh 0.86°C 0.17°C 7.2°C Linear MPC 8.6kWh 0.93°C 0.21°C 7.2°C Two-Position 34.5kWh 0.55°C 0.19°C 7.2°C 8/2/2011 50 LBMPC to Minimize Energy… Estimates heating load using model Best control that considers estimated occupancy Cost Temperature Dynamics Temperature Constraints AC Constraints CPS-PI-11 Master, Taneja, Culler, Tomlin, 2011, submitted) 8/2/2011 (Aswani, 51 … or optimize to follow the supply Generation Transmission VPG CPS-PI-11 Distribution Load 8/2/2011 52 at Building-Scale, and beyond Temperature Blue – Measured; Red - Simulated 1 3 2 1.4 0.9 1.8 0.85 1.75 0.8 1.7 1.3 0.75 1.65 1.25 1.6 0.7 20 20 Tu W Th F Sa Su M Tu W Th F Sa Su 4 20 Tu W Th F Sa Su M Tu W Th F Sa Su 5 24 Tu W Th F Sa Su M Tu W Th F Sa Su Tu W Th F Sa Su M Tu W Th F Sa Su 6 24 4 7 1.6 8 24 1.7 1.45 1.6 1.4 Tu W Th F Sa Su M Tu W Th F Sa Su 9 24 7 Tu W Th F Sa Su M Tu W Th F Sa Su 8 9 1.75 1.55 1.7 22 1.5 1.65 1.55 °C 22 °C °C °C 1.4 1.35 24 1.6 °C 1.5 1.45 Tu W Th F Sa Su M Tu W Th F Sa Su 1.65 22 1.55 1.5 1.65 Tu W Th F Sa Su M Tu W Th F Sa Su 1.6 1.55 °C 22 20 Tu W Th F Sa Su M Tu W Th F Sa Su °C °C °C °C 20 Tu W Th F Sa Su M Tu W Th F Sa Su 6 1.65 1.8 20 Tu W Th F Sa Su M Tu W Th F Sa Su 5 24 22 1.2 Tu W Th F Sa Su M Tu W Th F Sa Su 1.75 22 1.35 °C 22 1.45 1.85 °C 22 °C °C °C °C 0.95 22 3 1.9 24 °C 2 24 1.5 °C 1 24 Heating from occupants, equipment, etc. 1.45 1.6 1.4 1.55 1.35 1.45 1.3 1.5 20 Tu W Th F Sa Su M Tu W Th F Sa Su CPS-PI-11 20 Tu W Th F Sa Su M Tu W Th F Sa Su 20 Tu W Th F Sa Su M Tu W Th F Sa Su 1.4 Tu W Th F Sa Su M Tu W Th F Sa Su Tu W Th F Sa Su M Tu W Th F Sa Su 8/2/2011 Tu W Th F Sa Su M Tu W Th F Sa Su 53 Making Sense out of Sensors ° ° ° ° ° ° ° ° ° CPS-PI-11 8/2/2011 54 1 67 133 199 265 331 397 463 529 595 661 727 793 859 925 991 1057 1123 1189 1255 1321 1387 1453 1519 1585 1651 1717 1783 1849 1915 1981 Supply Air Fan VFD 90 85 Input 80 Output 75 OAT 70 65 60 55 50 CPS-PI-11 8/2/2011 55 What’s really going on? Bring comfortable air from outside Cool it down till its really cold Push it out everywhere thru VAVs that are at minimum opening Reheat it to set point So the empty rooms will be comfortable This is going on everywhere! And we supply perfect, precious energy to do it! CPS-PI-11 8/2/2011 56 And IT is no better … Atom 333 Westmere Server Power Consumption 350 48 Active 87 300 Idle Watts 250 15 13 200 13 14 19 287 150 230 248 190 100 190 200 161 Compaq DL360 57 HP Integrity rx2600 4/12/11 SunFire x2100 Cyber Switching SunFire V60x Dell PowerEdge 1950 Core i7 CPS 2011 PowerEdge 1850 0 SunFire X2200 50 Supply-Following Computational Loads Background Processing (shiftable) Requests Availability IPS Forecasts QoS (fidelity & latency) Power Controllable Storage CPS-PI-11 8/2/2011 58 Energy-Availability Driven Scheduling Non-dispatchable, variable supply 14 x 10 4 12 Power (W) 10 8 6 4 2 0 0 200 400 Time (hrs) 600 800 Pacheco wind farm x 10 x 10 4 greedy pacheco 2.0x 4 5 5 4 Power (W) Power (W) 4 3 2 1 0 0 200 400 Time (hrs) 600 800 Power proportional, grid-aware loads 3 2 1 0 200 250 Time (hrs) 300 Scientific computing cluster CPS-PI-11 NREL Western Wind and Solar Integration Study Dataset 59 8/2/2011 http://wind.nrel.gov/Web_nrel/ Scheduling & Energy Storage CPS-PI-11 8/2/2011 60 CPS Technology National Physical Info Service Software foundations for Energy Efficiency and Agility Infrastructure for Energy Innovation CPS-PI-11 8/2/2011 61 Personalized Automated Lighting Control HTTP Python Control Process MySQL Python Django sMAP BACnet Gateway CPS-PI-11 Three controllable ballasts per fixture ~5 zones per floor 8/2/2011 62 Real Energy Savings SDH 4th Floor Lighting Energy Usage 4 68% 69% 3.5 KW 65% 63% 68% 71% 70% 61% 58% 3 60% 53% 2.5 50% 2 40% 1.5 30% Colab Savings 1 Floor kW 0.5 kW Collab 0 0 CPS-PI-11 80% 75% 20% 10% 0 0% 8/2/2011 63 From the agony of load-following … CPS-PI-11 8/2/2011 64 to Co-operative Energy Mgmt in a Cyber-Physical Grid • Availability • Pricing • Planning • Forecasting Source IPS • Tracking • Market energy subnet Intelligent Power Switch Load IPS • Monitor, Model, Mitigate • Deep instrumentation • Power proportional Design • Energy-Agile Control • Shifting, Scheduling, Adaptation CPS-PI-11 8/2/2011 65 Thanks CPS-PI-11 8/2/2011 66