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