Intelligent Light Control using Sensor Networks
Vipul Singhvi1,3, Andreas Krause2,
Carlos Guestrin2,3, Jim Garrett1, Scott Matthews1
Carnegie Mellon University
1 Department of Civil and Environmental Engineering2 School of Computer Science
3 Center of Learning and Discovery
Motivation Current built infrastructure
Trillions of dollars investment Cost over the life cycle Research shows potential gains from
reducing operating cost and improving occupant performance $10 - $30 billion/yr from reduced
energy consumption $20 - $160 billion/yr gained from
improvement in comfort leading to better occupant performance
Reduction in energy cost related to reduced comfort & performance: Complex tradeoff optimization
Life cycle building cost
Salary cost overbuilding life cycle
Maintenance andoperation
Construction
Sensor networksSmart monitoring and actuationcan significantly reduce cost and improve occupant performance
10
Motivating Scenario
Lo
uve
rs
All lights0-10 levels
10
5
5
0
05
0
0
0
Operator
Controller
06
AndyBob
Louvers/ Blinds
Feedback
Coordinate lighting to make everybody happy
Strategy to exploit natural lighting
Predictive light control
Challenges
Knowing the current state Light levels and occupants location
Capturing occupant and operator preferences & happiness
Optimization of tradeoff Occupants happier OR save more
energy
Desk
Knowing the current state of the world Indoor Environment
Light levels Pervasive sensor
network Wireless or Wired
Tracking occupants Smart tags RFID tags Camera tracking
Utility Theory: Framework to compare choices based on preferences
Personal preference Attributes: Coolness, Horse Power, Mileage,
COST…. Representation complexity of utility function
Preferences & Happiness
Lamborghini, Second Hand 2003 model, $50,000
Toyota Corolla, New 2006 model,$30,000
Occupant preference:Comfort Light level
Utility Function Task dependent
Light levels Depends on lamp setting Use sensing to learn effect of
lamps on person i – Control lamp settings a to max.
occupant preferences, a=(a1,…,an), aj – level of lamp j
i
Building Operations: Occupants
argmax
1
!n
ii
Try tomakeeverybody happy
aa
iBob
Andy
Building Operations: Operator Operator preference: Cost
Operating cost Maintenance Cost
Decreases monotonically with the energy expended
Utility function
aj , jth lamp
1
( )j j
n
j
a
100 200 300
Operating Cost
0.00
0.20
0.40
0.60
0.80
1.00
No
rma
lize
d u
tili
ty
j
Cheaper the better
Utility Maximization : Tradeoff Maximize system utility: Make occupants and operator
happy!
a = (a1,…….an)
Scalarization technique
is the tradeoff parameter
1 1
( )) (,( )n n
j jj
ii
U F a
a a
OccupantsOperator
1
( ) (* )n
ii
U
a a a
* argmax ( )U
aa a
a6
a5
a4
a4
a2
a1
Utility Maximization: Complexity
Evaluating U(a) for combinations of all lamp setting for just 6 lamps the total number is 106
Evaluating argmax U(a) is also over that big space Exponential in number of lamps!
* argmax ( )Ua
a a 1
( ) (* )n
ii
U
a a a
10 levels
10 levels
10 levels
10 levels
10 levels
10 levels
a6
a5
a4
a4
a2
a1
Reducing Complexity
Exploit problem structure: Zoning
11
*
1
( ),...,g (ar )k jii
m
ji
n
ji
a amax a
a
a
1 2 3 2 3 4
3 4
1 2
3 4
6
4 5
*
15 6
, , , ,
, ,
( ) ( )
( ) ( )argmax *
,(
,)i
i
a a a a a a
a a a a aa
a
a
a
Distributed action selection approach (Guestrin ’03) Exact solution to the coordination problem
Open-loop controller: Coordinated Lighting
Control law using Occupant utility and Coordination Graph
approacha
Test Bed Control Schematics 10 table lamps 12 motes aka occupants Size: 146 * 30 in., 7 zones
146 in.
1234567
Coordinated Lighting: Results
• Comparison to greedy approach
•Each occupant comes and actuates the light
•Caveat: cannot reduce the level of a already ON light
• At = 0.4, reduction in comfort = 7% but reduction in energy cost = 30%
Greedy Heuristic
Energy Cost
Measured utility
30%
0.4
Coordinated Illumination
Coordinated Lighting
Performs significantly better than typical greedy approach
Solves the complex optimization using the structure of the problem (zoning)
Coordinated Lighting
Natural Lighting
Predictive light control
Closed-loop controller: Daylight
Harvesting
Control law
aOnline sensing using sensor network
Current Light Level
Sense natural light levels Actuate lamps to compensate for extra light
Variability using the real sunlight data from Pittsburgh
Day Light Harvesting: Sun Simulation Simulated sun using
overhead lamps
Real sun intensitiesMeasured intensities at center
Sun Lamps
Daylight Harvesting: Utility Redefined Represents the sunlight
intensity at time t and point in space x,
New utility definition
Maximization problem
:T X R
1 1, ( ,( , , ) ( , ) ( ))i
n m
ji j
i jit tU x a
xa x a
Sun Lamps
* argmax ( , , )( , )t U ta
ax xa
Running the Simulations
Day Light Harvesting: Evaluation
Gamma values (0.01, 0.4), same setup
Gamma = 0.01, 15% of energy savings Gamma = 0.4, 55% of energy savings Loss in occupant utility due to too much light
Shading, Louvers
Measured Utility
Energy Cost
Measured Utility
Energy Cost
Day light harvesting
Builds on the coordinated lighting approach Saves significant (~50%)energy cost during
sun time Long term sensor
deployment: battery life Sensor scheduling
Save battery life
Coordinated Lighting
Natural Lighting
Predictive light control
Spatial correlation in sunlight distribution Temporal correlation in sunlight intensity Use only a small number of sensor Estimate the light levels at other times and
locations
Active Sensing aka Sensor Scheduling
Desk?
???
When and Where to sense!
Active Sensing: Scheduling Use sunlight observation (samples) to estimate the
current sunlight intensity distribution
1 1 1,1 ,( , ) ,..., ( )|) ( ,,l r l rP
The utility formulation then changes to conditional expected utility
Choose a set of observations that yields best maximum expected utility values
,1 1
( | )( , , | ) ,, ( ) )( ) (i
n m
ji j
jiEU t xPO Ot a
xa x o ao
Sunlight Distribution Conditioned on observation
Active Sensing Calculating a set of observation that maximize
More observation: better accuracy but high battery cost
Constraint the observations to a budget Allocate strategically to max. EU
(1: ) (1: )( ) ( )( max ( , , | ))t t
t T
J O P EU tO O
a
o
a xo o
* argmax ( )O
O J O
Active Sensing: Single Sensor
Optimal solution for single sensor budget allocation in polynomial time (Krause & Guestrin ’05)
Xi where i is the time step, (5 times steps, Budget 2) For just 2 sensors: complexity is NP-hard
* argmax ( )O
O J O
X1 X2 X3 X4 X5X1 X2 X3 X4 X5
X1 X2 X3 X4 X5
Y1 Y2 Y3 Y4 Y5
X1 X2 X3 X4 X5
Y1 Y2 Y3 Y4 Y5
X
1
X
2
X
3
X
4
X
5
Y1 Y2 Y3 Y4 Y5
Heuristic for solving multiple sensor Coordinate ascent scheme (uses optimal solution for
single sensor)
Guaranteed to improve score on each iteration, guaranteed to not perform worse than independent scheduling
Can be used for more than 2 sensors
Active Sensing: Heuristic
X1 X2 X3 X4 X5
Y1 Y2 Y3 Y4 Y5
Optimize sensor 1X1 X2 X3 X4 X5
Y1 Y2 Y3 Y4 Y5
X1 X2 X3 X4 X5
Y1 Y2 Y3 Y4 Y5
Optimize sensor 2
Active Sensing: Results
3 sensors, upto 10 readings per sensor in a day Energy saving are close approximation compared to
sensing continuously Even a small number of readings (3) provides results
as good as continuous
Energy Cost
Measured Utility
No sensing
1 obs./sensor 10 obs./sensor
Active Sensing for Daylight Harvesting Exploit temporal correlation in sunlight
intensity to schedule sensing Significant reduction in sensing requirement
for comparable performance Can be integrated in the coordinated lighting
formulation
Coordinated Lighting
Natural Lighting
Predictive light control
Predictive light control
Probabilistic model on mobility People move independent of each other
Modeled using a random walk Stay in same position Move left, move right
Zone 1Zone 2Zone 2
Integrating mobility
Assuming full observability
Computing expected utility
( 1) 1( . | , )t t ti i iP x x x
1
1,
1
1
( ,( , , , ) ( ( ),( )) ( )ii
n mt t
ji j
ti iix x jx tP xEU t Ex a
a x x a
Probability of motion
Predictive Lighting: Results
20 step random walk Total utility increase of about 25% Low values of trade-off parameter, system prefers
occupants comforts
Occupant Utility
Energy CostTotal Utility
No
rmal
ized
Sca
le
Occupant Utility
Energy Cost
Using prediction
Without prediction
ConclusionCoordinated lighting strategy•Maximizes happiness using utility maximization •Solves complex coordination problem
Day light harvesting•Exploits natural light sources using sensors •50-70% reduction in energy consumption
Active sensing •Sensor scheduling using sunlight distribution• Substantial increase in network life time
Predictive Light control•Captures occupant mobility•Higher total utility for the system