Visible to the public File preview

Methods
 for
 Network
 Enabled
 Embedded
  Monitoring
 and
 Control
 for
 High-­‐Performance
  Buildings

Prabir
 Barooah
 (UF),
 Prashant
 G.
 Mehta
 &
 Sean
 Meyn
 (UIUC),
 Luca
 Carloni
  (Columbia
 U),
 Martin
 Krucinski
 (UTRC) Acknowledgement:
 NSF
 grant
 CNS-­‐0931885 (March
 2010)



Occupancy
 modeling
 and
 prediction,
 control
 of
 evacuation Platform-­‐based
 design
 space
 exploration
 for
 cyber
 systems
  relevant
 to
 building
 monitoring
 and
 control Energy
 efficiency
 improvement
 through
 estimation
 and
  control
 
  Learning
 and
 adaptation
 of
 models
 and
 networks
 across
  buildings
 and
 time-­‐scales









Occupancy
 modeling
 and
 prediction,
 control
 of
 evacuation Platform-­‐based
 design
 space
 exploration
 for
 cyber
 systems
  relevant
 to
 building
 monitoring
 and
 control Energy
 efficiency
 improvement
 through
 estimation
 and
  control
 
  Learning
 and
 adaptation
 of
 models
 and
 networks
 across
  buildings
 and
 time-­‐scales







Buildings and Energy (U.S.A.)
>
 70%
 of
 electricity*,
 40%
 of
 total
 energy,
 ... >
 commercial
 vs.
 residential
 :
 close
 to
 1
  Sources
 of
 Inefficiency:
  (1)
 over-­‐design,
 (2)
 poor
 operation
  HVAC
 ~
 50%
heating,
 ventilation,
 air
 conditioning 2050
 U.S.
 building
 energy
 use

BAU with savings

State of the art?
inefficiency:
 (1)
 over-­‐design
 (2)
 poor
 operation
  90 80 Temp
 (F) 70 60 50 0 hour-­‐-­‐-­‐> 5 10 15 noon 20 serve
 the
 occupants,
  not
 empty
 space! Air
 quality!!!
  (ASHRAE
 standards) Supply
 air Room Pugh
 Hall,
 Univ
 of
 Florida

Room
 249,
  Pugh
 Hall

High performance buildings
efficiency:
 (1)
 ...
 (2)
 better
 operation
 
“minimize
 energy
 use
 while
 maintaining
 comfort
 and
 IAQ”

weather
  model occupancy
  model

optimizer temp
 +
 air
 quality model min
 E
  subj.
 to
 ....

complexity
 and
 uncertainty
 makes
 modeling/prediction/design
 challenging *)
 Delay
 between
 action
 and
 its
 effect *)
 Air
 quality
 and
 temp
 vs.
 energy Occupancy
  prediction

none

some

perfect

Predictive control w/o occp. prediction
Set-­‐point
 based
 control
  (PID+logic) Floating-­‐point
 control: occupied T-­‐des

T-­‐des

*)
 allow
 temp
 within
 a
 band un-­‐occupied *)
 wider
 band
 when
 unoccupied *)
 maintain
 minimum
 airflow
 to
  remove
 VOC
 and
 CO2
  Formulated
 as
 a
 RHC
  *)
 if
 occupied
 now,
 assume
 will
 be
  (Receding
 horizon
 optimal
 control)
  occupied
 for
 the
 next
 T
 minutes problem *)
 control
 changes
 only
 every
 T
  minutes *)
 minimize
 energy
 over
 the
 next
 NT
  minutes
 

Example
76

Room
 Temperature

0.4

Power
 consumption PID
  RHC

74
0F

0.3

72

RHC PID

70

68

hour
 -­‐-­‐-­‐-­‐>
6 12 20 24

kW
0.2 0.1 0

4

4

6

12

20

24

occupancy

occupancy

30%
 savings
 over
 PID (ex
 2:
 occupied
 18
 of
 24
 hours:
 10%)

Example
76

Room
 Temperature

0.4

Power
 consumption PID
  RHC

74
0F

0.3

72

RHC PID

70

68

hour
 -­‐-­‐-­‐-­‐>
6 12 20 24

kW
0.2 0.1 0

4

4

6

12

20

24

occupancy

occupancy

30%
 savings
 over
 PID (ex
 2:
 occupied
 18
 of
 24
 hours:
 10%)


Large
 savings
 are
 possible
 with
 only
 occupancy
 measurement,
 w/o
 prediction bottleneck
 is
 not
 temperature
 control
 but
 air
 quality/ventilation
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 (health
 problems,
 lawsuits,
 ..)
 



Example
76

Room
 Temperature

0.4

Power
 consumption PID
  RHC

74
0F

0.3

72

RHC PID

70

68

hour
 -­‐-­‐-­‐-­‐>
6 12 20 24

kW
0.2 0.1 0

4

4

6

12

20

24

occupancy

occupancy

30%
 savings
 over
 PID (ex
 2:
 occupied
 18
 of
 24
 hours:
 10%)


Large
 savings
 are
 possible
 with
 only
 occupancy
 measurement,
 w/o
 prediction bottleneck
 is
 not
 temperature
 control
 but
 air
 quality/ventilation
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 (health
 problems,
 lawsuits,
 ..)
 



Measuring occupancy
Most
 sensors
 are
 “flow”
 sensors
 =>
 large
 drift! Occp(k)
 =
 flow
 in(1)
 +
 flow
 in(2)
 +
 .....
 +
 flow
 in(k)
 
200

data
 from
 Pugh
 Hall
 @UF
 

150

#
 of
 occupants

100

50

0

−50 0

6

hour
 of
 day

10 12

15

20

24

Measuring occupancy
Most
 sensors
 are
 “flow”
 sensors
 =>
 large
 drift! Occp(k)
 =
 flow
 in(1)
 +
 flow
 in(2)
 +
 .....
 +
 flow
 in(k)
 
200

12

data
 from
 Pugh
 Hall
 @UF
 
10

experiment
 in
 a
 6
 occupant
 office Software
  sensor
 (video) TRUE CPS
 

150

#
 of
 occupants

#
 of
 occupants
6 10 12 15 20 24

8 6 4 2 0 0

100

50

0

CO2

−50 0

hour
 of
 day

8

12

16

20

24

Measuring occupancy
Most
 sensors
 are
 “flow”
 sensors
 =>
 large
 drift! Occp(k)
 =
 flow
 in(1)
 +
 flow
 in(2)
 +
 .....
 +
 flow
 in(k)
 
200

12

data
 from
 Pugh
 Hall
 @UF
 
10

experiment
 in
 a
 6
 occupant
 office Software
  sensor
 (video) TRUE CPS
 

150

#
 of
 occupants

#
 of
 occupants
6 10 12 15 20 24

8 6 4 2 0 0

100

50

0

CO2

−50 0

hour
 of
 day

8

12

16

20

24

Cyber
 Physical
 Sensor: software
 to
 detect
 mouse/keyboard
 idle

Exploit inter-zone convection
Which
 rooms
 have
 convection
 interaction
 (airflow
 thermal
 exchange)? A
 network
 structure
 identification
 problem
  proposed
 solution
 based
 on
 estimating
 conditional
 probabilities

Room 243 78

76
Temperature(F)

74

72

70

68 0

10

20

Time (hr)

30

Without Convection With Convection Actual 40 50

Test-bed Development


BACnet
 stack
 server
 


Pugh
 Hall
 @UF
 campus LEED
 Silver,
 66
 zones

runs
 on
 a
 PC
 residing
 @Pugh
 
 
  communication
 through
 VLAN
 
UF Campus Map





Capable
 of
 high
 frequency
 control
  of
 all
 actuators
 (not
 just
 “set
  point”
 change) Control
 software
 being
 stress-­‐ tested

http://ca

Campus Map

University of Flo

Search
BUILDING LOCATOR MAP FEATURES



Art, Culture & History Campus Life Sports & Recreation Sustainable Campus Transportation Technology CALCULATE DISTANCE
Welcome to the UF Campus Map!

PPD

Since these maps are based on the Google maps and are draggable, you can use your mouse or the directional arrows to pan left, right, up and down to see areas that are hidden offscreen. You can also use the slider to zoom in and zoom out. Single-clicking on a map feature (such as a building

Team
UF Prabir
 Barooah Sean
 Meyn UIUC Prashant
 Mehta Columbia United
 Tech.
 Res.
 Center Luca
 Carloni Martin
 Krucinski

Siddharth
 Goyal Chenda
 Liao Yashen
 Lin

Kun
 Deng

Adam
 Tilton

Marcin
 Szczodrak

Collaborators: Sanjay
 Ranka,
  faculty,
 CISE@UF (cyber
 physical
 sensor)

Tim
 Middelkoop,
  faculty.,
 ISE@UF (Pugh
 hall
 DAC
 server)