Lighting Res. Technol.

2011; 43: 235–248

Control of wireless-networked lighting in

open-plan offices
Y-J Wen PhD and AM Agogino PhD
Department of Mechanical Engineering, University of California at Berkeley, Berkeley CA, USA

Received 14 June 2010; Revised 2 August 2010; Accepted 8 August 2010

This paper presents an energy-efficient lighting control system for open-plan

offices with two novel features: A versatile ‘plug-and-play’ wireless-networked
sensing and actuation system, and a control method incorporating multiple
management strategies to provide occupant-specific lighting. Workstation-based
wireless photosensors are employed to measure task illuminance, and individual
addressability of wireless-enabled dimmable luminaires increases the freedom of
the lighting system to fulfil various lighting requirements. Daylight harvesting,
occupancy control and light level tuning strategies are formulated into an
optimisation problem that generates light outputs for each luminaire to produce
the desired lighting at each workstation with minimal energy usage. A pilot study
of the wireless-networked lighting system implemented in a small open-plan
office has shown more than 60% energy savings potential compared to
conventional practice without any control.

1. Introduction including unit price, installation, rewiring

and commissioning, are the primary hurdles,
Grid-based electric lighting consumes 19% of especially for legacy buildings.4–6
total electricity production worldwide, and Unsatisfactory performance due to improper
offices account for 19% of the lighting energy commissioning and neglected retro-commis-
consumed by all types of commercial build- sioning has often led to occupant dissatisfac-
ings.1 Forty to fifty percent of energy savings tion, eventually rendering the systems
can potentially be generated with a combina- obsolete.7 As a result, most buildings have
tion of energy-efficient lighting management been missing the opportunity to generate
strategies.2–4 In order to realise the goal of energy savings from efficient lighting control.
net-zero energy buildings in the next 15- to Wireless technology has been identified as
20-year timeframe as set out by many a promising wiring and communication alter-
countries, it is crucial for lighting systems to native for building systems.8 Reduced cost
achieve or even exceed the expected energy and complexity as well as increased flexibility
savings. on wiring are the primary advantages of
Advanced lighting components and sys- wireless-enabled systems over their wired
tems that are capable of implementing counterpart. The emerging wireless sensor
energy-efficient lighting control strategies and actuator network (WSAN) technology
have existed for more than a decade but further enhances the versatility of wireless
have not been able to make significant market sensing and actuation entities, which can
penetration. Exorbitant initial costs, self-configure into a fully connected commu-
nication network. When integrated with
Address for correspondence: Yao-Jung Wen, 411 Cornell
Avenue, Apartment 9, Albany, CA 94706, USA. lighting control systems, the WSAN technol-
E-mail: riowen@berkeley.edu ogy can potentially overcome many practical
ß The Chartered Institution of Building Services Engineers 2010 10.1177/1477153510382954
236 Y-J Wen and AM Agogino

challenges encountered by wired lighting sys- the commercial sector.3 In the US, for exam-
tems and bring energy savings into reality. ple, lighting alone accounts for 27–43% of the
This paper describes a lighting system for energy usage in office buildings depending on
open-plan offices with two major novelties: the climate and often contributes the largest
(1) demonstration of a ‘plug-and-play’ system proportion of energy consumption among all
of wireless lighting sensors and actuators, an electrical systems.9 Being the single largest
economical solution for implementing light- energy consumer in the commercial sector,
ing controls in retrofitting projects; (2) a office buildings therefore have a major role to
control framework capable of working with play in increasing energy efficiency and
any lighting configuration to provide occu- reducing carbon footprint.
pant-specific lighting in the most efficient Various lighting control technologies have
manner. Wireless photosensors are deployed been developed for increased energy effi-
as workstation-based light sensing devices, ciency, including daylight harvesting, occu-
and individually addressable and dimmable pancy sensing, light level tuning, etc. Daylight
wireless luminaires are utilised to deliver task harvesting is the most energy-efficient control
lighting to the office space. For a given strategy for offices in perimeter zones where
occupancy status and occupants’ lighting significant daylight is available. Research has
preferences, the control technique actuates also supported this strategy by showing that
each luminaire at a different level to comple- daylighting can indeed displace electric light-
ment available daylight and results in lighting ing from the occupant’s perspective.10
that satisfies each occupant’s need while A lighting zone typically uses at most two
consuming the least amount of energy. reference points for responding to daylight,
An implementation of the wireless-networked and electric lights in the same zone are actu-
lighting control system in a small open-plan ated uniformly in accordance. Occupancy
office has demonstrated more than 60% sensing dims or turns off the lights in a zone
energy savings and high levels of occupant after the space becomes vacant for a certain
satisfaction. amount of time and is the most pervasive
The rest of this paper is organised as fol- control strategy implemented in offices now-
lows. Sections 2 and 3 discuss energy-efficient adays. Light level tuning generates energy
lighting control strategies and WSAN tech- savings by reducing electric light level away
nologies, respectively. Formulation of the
from the recommended standard according to
lighting control problem is presented in
occupants’ lighting preferences. This tech-
Section 4. The modelling approach for office
nique is particularly effective in private offices
lighting is presented in Section 5 followed by
but not in open-plan offices since no single
the controller design in Section 6. Section 7
light level will satisfy all individuals’
shows an implementation and the results of
testing the lighting control system. needs.11,12 The potential lighting energy sav-
ings is estimated to be up to 50% when a
proper combination of control strategies is
2. Energy-efficient lighting control effectively implemented.2–4
strategies As mentioned above, each lighting control
strategy has certain limitations, and most of
Lighting consumes more than 2000 terawatt- them are implemented at the zone level in
hours of electricity globally, which corre- current practice. Our research incorporates
sponds to approximately 1800 million metric the aforementioned lighting control strategies
tons of carbon dioxide emissions annually; at workstation level, utilising wireless-
and 48% of lighting electricity is attributed to networked sensors and actuators to create a
Lighting Res. Technol. 2011; 43: 235–248
 Wireless-networked lighting 237

more personalised lighting environment in but can also be energy efficient.14,15 In certain
open-plan offices and further boost energy leading-edge practices, one or more over-
savings. head luminaires is purposely aligned and
associated with each workstation as the
designated light source for the occupant.
3. Wireless sensor and actuator Luminaire-embedded photosensors as well
networks as workstation-specific occupancy sensors
are implemented for harvesting daylight and
Wireless sensor network (WSN) technology to dim or turn off the lights when a worksta-
emerged about a decade ago as a futuristic tion becomes unoccupied.16 This approach,
mesh network comprising miniature sensing however, requires precise spacing and
units with computational and wireless com- arrangement of both luminaires and worksta-
munication capabilities.13 Wireless sensor tions, which may be practical in new con-
nodes can be massively deployed to environ- structions and major renovations but not in
ments, where communication over physical legacy buildings, which will still dominate the
wires could be expensive or even prohibited, built space in the foreseeable future. The work
and self-organise into a communication net- presented in this paper targets the challenging
work for monitoring purposes. WSN tech- open-plan office configuration, where there
nologies have been identified as a promising is no luminaire directly above each worksta-
solution for advanced building operation
tion or cubicle serving as a dedicated light
systems to circumvent costly installation and
source. This is the most common arrangement
wiring, particularly in legacy buildings.8
in most offices, especially those that have
Recently researchers have integrated actu-
undergone reconfigurations due to change
ators with the same wireless technology to
of tenants.
realise WSANs. In addition to passively
Exploiting the wireless capability, photo-
monitoring the environment, a WSAN
sensors can be flexibly embedded into each
actively interacts with the physical world
where the actuators perform actions based workstation instead of being hardwired in the
on the data collected from the sensors. Our ceilings or luminaires. Similarly, workstation-
research is a realisation of such a WSAN, based occupancy sensors can also be deployed
where wireless photosensors measure task in the same manner, although not explicitly
illuminance and wireless ballast actuators tackled in this research.
dim the lights based on the decision made The goal of this lighting control system is
by the controller. Specifically, the illuminance to make maximal use of available daylight
on each workstation in an open-plan office and deliver specified task lighting to each
is determined by more than one luminaire occupied workstation with the least amount
as well as daylight, and thus coordination of energy consumption. The block diagram in
among ballast actuators is critical to respond Figure 1 illustrates the concept of the system,
to available daylight and deliver desired where C is the lighting controller, A is the
lighting. wireless dimmable luminaire network, G is
the illuminated space and S represents wire-
less-networked sensors used to measure task
4. Lighting control problem description illuminances at each workstation. While the
explicit objective is to bring the task illumi-
Recent studies have pointed out that allowing nances close to the specified level, the implicit,
occupants to work under their preferred yet equally important, objective is to minimise
lighting may not only increase productivity overall energy usage. The electric light model
Lighting Res. Technol. 2011; 43: 235–248
238 Y-J Wen and AM Agogino

Daylight
Specified
lighting for each +
workstation Luminaire +
+ C A G Actual
output levels
– lighting in the
office space

Task illuminance at
S
each workstation

Figure 1 Block diagram of the lighting control system

200

150
Illuminance (lux)

100

50

0
0
1
2
3
4
5 5
6 4
7 3
8 2
Y dimension (m) 1
9 0
X dimension (m)

Figure 2 Example of an illuminance model

for control purposes is introduced in the next Basic knowledge specific to a room configu-
section followed by the controller design. ration has to be specified, including room
dimensions, surface reflectances, luminaire
locations and photometrics. The space at the
5. Lighting model workplane level is first geographically dis-
cretised into a grid of squares with prescribed
In this research, a model of illuminance at the resolution, and the illuminance at the centre
workplane level is derived for each luminaire of each square is than calculated. The gener-
using RADIANCE,17 an image rendering ated illuminance model is represented in a
engine based on backward ray tracing. matrix for easy manipulation where each of
The procedure for generating the illuminance the elements is the workplane level illumi-
models is partly inspired by the Sensor nance corresponding to the centre of each
Placement Optimization Tool (SPOTTM).18 square. Figure 2 graphically illustrates one
Lighting Res. Technol. 2011; 43: 235–248
 Wireless-networked lighting 239

such model for a single luminaire, where the fractions of the full light outputs, that is, the
x- and y-axes are the room dimensions. dimming levels, for each luminaire.
Notice that a model can still be arranged
into a matrix even if the space is not rectan- 2 3
gular, but care has to be taken to correctly li11  li1n
X
K X
K
6
associate each element with its geographical E¼ di  l i ¼ di  4 ... .. .. 7 ð1Þ
. . 5
location. i¼1 i¼1
Every model is derived assuming the dim- lim1  i
lmn
mable light is at full output. The relationship
between a dimming level and the actual light
output from a luminaire is assumed to be
linear, which is true for fluorescent lights with
analogue dimming ballasts, such as the ones 6. Control algorithm design
adopted in this research. A slightly more
complicated mapping for lights with logarith- As pointed out in Section 4, the goal of
mic dimming, such as DALI (Digital intelligent lighting control is twofold: to
Addressable Lighting Interface), can easily minimise the overall lighting energy usage
be derived. Despite different dimming level and to deliver satisfactory task lighting that
and light output relationships, power con- meets each occupant’s preference. Wireless-
sumption of a luminaire is always propor- networked photosensors have enabled light
tional to the actual light output.19 Also note sensing at the workstation level, but the fact
that the models are generated assuming an that multiple workstations may share light
empty space and will deviate from reality due from the same luminaire has made tuning for
to the unaccounted light reflection from personalised lighting a non-trivial problem.
furniture, lamp depreciation, etc., which con- With the setup of wireless individually con-
tribute to modelling uncertainties. While the trollable ballasts, one seemingly straightfor-
furniture size, type and arrangement in an ward control strategy is to link one or more
open-plan office have been recognised to have luminaires to the closest wireless sensor and
significant effect on the resulting task illumi- implement multiple closed control loops for
nance, the impact will be minimised since the occupant-specified lighting. This approach,
control loop is closed by workstation-specific however, will involve more efforts and judge-
wireless photosensors. The photosensors ment calls on pairing luminaires with sensors
ensure that the task lighting is regulated at during the commissioning process, thereby
the specified illuminances, as will be discussed defeating the flexibility of wireless lighting
in the next section. systems. Fighting between control loops may
The illuminance at any point in a space is also occur and cause unstable oscillations
the summation of the light from each lumi- since a photosensor is likely to detect and
naire. Suppose an office space with K lumi- respond to the light level changes from
naires is discretised into a grid of m-by-n adjacent control loops. The main concept of
squares, the generated models are K m-by-n the proposed lighting control algorithm is to
matrices, l1, l2, . . . , lK, associated with each of fulfil the two control objectives with an
the K luminaires indicated by the superscript optimal set of light outputs for the luminaires
number. The illuminance of the room at the exploiting the individual addressability of
workplane level, E, can then be represented as each wireless-enabled luminaire. The devel-
the linear combination of each model as oped control algorithm builds on our earlier
shown in Equation (1), where di’s are the research.20
Lighting Res. Technol. 2011; 43: 235–248
240 Y-J Wen and AM Agogino

The overall lighting in an office is consid- It is possible that the optimisation problem
ered as a linear combination of the light may not have a solution due to overly
contributions from each of the luminaires as stringent constraints. Infeasible problems are
explained in the previous section. Using the most likely to be caused by conflicting pref-
matrix representation of the illuminance erences specified for adjacent workstations.
models in Equation (1), the objective is to In the case of infeasibility, the equality
find the optimal set of di’s so that each constraints in Equation (3) are relieved to
workstation is properly illuminated as speci- inequality constraints, thus allowing some
fied. The elements in each of the K matrices tolerances. The physical meaning of this is
corresponding to the location where a pre- to allow the light at the workstations to be
ferred lighting is specified are picked out as regulated in certain tolerable ranges ("tol)
the elements in each column of L in Equation instead of demanding the exact amount of
(2). T0 in the matrix operation is a column light. In summary, iterations of lighting opti-
vector comprising the target illuminances at misation starts with the original linear pro-
each workstation, and d is a column vector gramming problem with the equality
whose elements are fractions of the full light constraints in Equation (3) and gradually
output of each luminaire. expands the tolerable range by relaxing the
equality constraints into inequality con-
2 32 3 2 3
l1pq l2pq lK
pq d1 epq straints as shown in Equation (4) until a
6 l1rs l2rs lK 7
6 d 7 6 ers 7 solution is reached. Although there are more
6 rs 7 2
7¼6 7
Ld ¼ 6 . . . 76 . 6 . 7  T0 sophisticated ways of relaxing the constraints
4 .. ..    .. 54 .. 5 4 .. 5 considering each occupant’s tolerance and
l1xy l2xy lK
xy
dK exy sensitivity, Equation (4) represents the most
straightforward approach.
ð2Þ

Lighting energy and preference control is minkdk1 , subject to

formulated as a linear programming problem     
Ld  T0 þ etol , L T0 þ etol
in Equation (3), which makes sense both d
physically and mathematically. By minimising Ld  T0  etol , L T0 þ etol
the 1-norm of the vector d, the summation of DimLevelmin  d  DimLevelmax
the light output levels from each luminaire is
minimised, which translates into minimising ð4Þ
the energy usage of the resulting light settings.
The specified lighting for each occupied
workstation, Equation (2), is posed as the Modelling uncertainties and daylight
constraints to the linear programming prob- harvesting are accounted for in the feedback
lem. The physical dimming capabilities of loop. Suppose the actual illuminances on
each luminaire, DimLevelmin and DimLevelmax the workstations measured by the photosen-
are also considered as constraints. sors are Sk ¼ [spq(k) srs(k) . . . sxy(k)]T, where
the subscript of the bold-face capital s and
the number in the parentheses denotes the
number of the iteration, then the constraints
minkdk1 , subject to are modified to Equation (5) for the next
Ld ¼ T0 ð3Þ iteration of optimisation, as shown in
Equation (6). Each iteration brings the deliv-
DimLevelmin  d  DimLevelmax ered lighting closer to the desired lighting, and
Lighting Res. Technol. 2011; 43: 235–248
 Wireless-networked lighting 241

2 3

N
1 2 3 4
1

6 7
5 6 7 8

9 10 11 12
5 4

Figure 3 Floor plan of the open-plan office implementation. Shaded rectangles are luminaires

the algorithm terminates when the difference workstation-based photosensors, and imple-
between the desired and the measured illumi- mented with the research lighting control
nances falls into a small prescribed tolerance algorithm.
range.
Ld ¼ Tk1 þ ðTk1  Sk Þ ¼ 7.1 System implementation and setup
2 3 2 3 The test bed office (9.1  5.8  4.3 m3)
epq epq  spq ðkÞ
contains 10 personal workstations and some
6 e 7 6 ers  srs ðkÞ 7 shared work surfaces as shown in the floor
6 rs 7 6 7 ð5Þ
6 . 7þ6 .. 7  Tk plan (Figure 3). The workplanes are 0.74 m
6 . 7 6 7
4 . 5 4 . 5 above the floor. There are 12 2-lamp fluores-
exy exy  sxy ðkÞ cent light fixtures hanging 1.2 m from the
ceiling evenly mounted in the office as the
minkdk1 , subject to 12 equally spaced rectangles represent in
Figure 3. The numbers marked on the
Ld ¼ Tk ð6Þ corner of the luminaire indicates the order
DimLevelmin  d  DimLevelmax in which the controlled light output for the
luminaires will be presented later in the
discussion. The reflectances of the floor,
walls and ceiling are measured at 10%, 50%
7. Implementation and performance and 30%, respectively.
evaluation Originally the twelve luminaires were phys-
ically wired together and could only be turned
A pilot study was conducted in a shared- on/off by 3-way switches located at both ends
space office, which was retrofitted with of the office. The operation of this original
wireless-networked dimmable lights and lighting configuration will be referred to as
Lighting Res. Technol. 2011; 43: 235–248
242 Y-J Wen and AM Agogino

the ‘base case’ in the following discussion. computer, which also coordinated nodes of
Each luminaire was retrofitted with a dim- ballast actuators and photosensors that form
ming ballast and a prototype wireless ballast a wireless mesh network. The wireless actua-
actuation module developed at the early stage tor and sensor nodes were programmed so as
of this research as shown in Figure 4.21 to form a self-configuring, multi-hop network
Prototypes of the workstation-based photo- upon powering up. The network ensured that
sensor, shown in Figure 5, were deployed the actuation commands containing the opti-
directly to each workstation. The research mal light settings could be propagated from
lighting controller was implemented on a the controller to each luminaire and sensor

Figure 4 Prototype wireless ballast actuation module

Figure 5 Prototype workstation-based wireless photosensor

Lighting Res. Technol. 2011; 43: 235–248

 Wireless-networked lighting 243

readings could be relayed from each photo- of ten seconds was used to drive the daylight
sensor back to the controller. Issues specific simulating structure. The original time series
to wireless networks were not the focus and of daylight, shown in Figure 6(a), was sam-
were not tackled in this particular research. pled every ten minutes and scaled to fit the
It is worth mentioning that like most office operation range of the daylight simulating
buildings in California, USA, the original structure. The data points in Figure 6(a)
lights in this office were not designed to indicate the sampled data points for this test,
provide the task illuminance recommended and the light output from the daylight simu-
by the Illuminating Engineering Society of lating structure was adjusted accordingly
North America but could at most illuminate every minute instead of every ten minutes.
the workstations to a little less than 350 lux. As a result, the simulated daylight followed
The office implemented with the research the trajectory in Figure 6(b). The daylight
system was located in the core of the build- structure, however, provided only limited
ing and had no windows for testing day- light output and penetration through the
light harvesting. Thus, a separate structure room. Although not as strong as daylight
containing dimmable full-spectrum fluores- from a clear sky, the simulated daylight was
cent lights was constructed to simulate still sufficient for testing the system perfor-
a window, from which daylight was admitted. mance under the situation where part of the
The structure was mounted on the north room receives daylight and part of the room
side of the office and 2 m above the floor. receives almost no daylight.
A set of real daylight data measured with The office was virtually divided into a grid
a photosensor 1.5 m away from a north-facing of 0.1 m  0.1 m squares for generating the
window on an ordinary day at a sensing rate illuminance models. Other practical factors

(a) 2000
Illuminance (lux)

1500

1000

500

0
0 100 200 300 400 500 600 700
Time (min)
(b)
Percentage lighting output (%)

100

80

60

40

20

0
0 10 20 30 40 50 60
Time (min)

Figure 6 (a) Measured daylight illuminance sequence; (b) simulated daylight sequence

Lighting Res. Technol. 2011; 43: 235–248

244 Y-J Wen and AM Agogino

were also taken into account: (1) the resulting range of a typical dimming ballast was
optimal light settings were represented as 5–100%. The light would be turned off if
the percentage of the maximum output of the optimised light setting was zero.
the luminaires and rounded to integers as the
dimming resolution of the wireless ballast 7.2 Control performance
actuator was limited to 256 distinct levels; (2) This test considered seven occupied work-
the light level was set to 5% whenever the stations with a wide range of specified lighting
optimised output was less than 5% but requirements. The solid lines in each of the
greater than 0 since the effective output subplots in Figure 7 show the sensor readings
Sensor reading (lux) Sensor reading (lux) Sensor reading (lux) Sensor reading (lux) Sensor reading (lux) Sensor reading (lux) Sensor reading (lux)

Occupant 7
280
260
240
220

220 Occupant 5

200
180
160
140
220 Occupant 4
180
160
140
120

320 Occupant 3

300
280
260
240
320 Occupant 6

300
280
260
240
340 Occupant 1
320
300
280
260
320 Occupant 2
300
280
260
240
0 10 20 30 40 50 60

Time (min)

Figure 7 Lighting control performance at seven workstations in the office

Lighting Res. Technol. 2011; 43: 235–248

 Wireless-networked lighting 245

from the seven workstation-based photosen- lights was actually very subtle and nearly
sors, which are ordered by the level of unnoticeable to occupants.
daylight reception. The occupant numbers Figure 8 shows the percentage of energy
associate each occupant’s workstation to the usage of the system during the test compared
circled digits in Figure 3. The dash-dotted to the base case where all the lights were wired
horizontal lines (through the centre of the to be turned on/off together. The baseline
data) and dashed lines (outer higher and assumed that the lights were not dimmable
lower boundaries) in the same plots represent and consumed the same amount of power as
the lighting preference specified by each the dimmable lights at full output. It was
occupant and the  5% boundary, respec- found that 39.2% of the energy was used to
tively. In this pilot study, occupants’ lighting deliver the specified lighting to each worksta-
preferences were obtained at the early phase tion when the experiment started with no
of the networked lighting system implemen- extraneous light. In other words, for this set
tation, which allows each individual to con- of occupants, offering individual control
figure the lights to their satisfaction.21 The alone would have generated 60.8% savings
resulting task illuminances were measured as under this optimisation framework with non-
their preferences. All seven sensor readings, workstation-specific lights. The energy usage
especially the ones close to the daylight decreased after harvesting daylight, and could
simulating structure did fluctuate with the be as low as 34.8% when the simulated
change of simulated daylight, but were main- daylight was at full output. As an example,
tained within 5% of specified lighting. the output levels of the 12 luminaires at the
In order to ensure there was no unnecessary lowest energy consumption are 49%, 100%,
lighting fluctuation that may annoy the 27%, 0%, 73%, 15%, 0%, 0%, 11%, 53%,
occupants, new iterations of lighting optimi- 21% and 0% ordered by the number at the
sation would only be triggered to bring task corner of each luminaire in Figure 3. Each
illuminances back to the specified lighting if step in Figure 8 represents a new energy usage
any sensor reading exceeded the 5% bound- level after a new set of lighting optimisation
ary. From the observation made in the office iterations revised the light settings to com-
during the test, the change of the electric pensate for available daylight. The savings

40
Percent of energy consumption (%)

39

38

37

36

35

34
0 10 20 30 40 50 60
Time (min)

Figure 8 Lighting energy use with daylight harvesting

Lighting Res. Technol. 2011; 43: 235–248

246 Y-J Wen and AM Agogino

introduced by extraneous light will potentially constraints under the same lighting optimisa-
be more significant in practice when the tion framework.
available daylight is stronger. If all lights Deploying wireless photosensors at the
were controlled by a single photosensor, for workstation level may raise reliability con-
conventional daylight-responsive dimming to cerns as the battery-powered sensors have
maintain a uniform task illuminance of limited lifespan and are more prone to
350 lux, the savings could only be as high as disturbances from the occupants. Leveraging
23% assuming the photosensor was mounted their miniature size, multiple sensors can be
in the centre of the room. This signifies the embedded into the workstations and inte-
importance of the proposed control algorithm grated into sensor networks where sensor
incorporating multiple lighting control strat- validation and fusion techniques can be used
egies for deep energy savings. to extract pertinent sensor information for
control decisions.22,23
Occupancy sensing, which was not explic-
8. Conclusion and discussion
itly addressed in this research, is also essential
for the system to effectively optimise the
This research presents a promising approach
lighting condition based on the presence of
to implementing multiple modern lighting
control strategies, including daylight harvest- occupants. As the immediate next step of this
ing, light level tuning and occupancy control, research, occupancy sensing will be imple-
in open-plan offices even with less-than-ideal mented either bundled with the workstation-
configurations. The lighting control method based photosensors or as stand-alone sensors
dynamically tunes the lights to each occu- leveraging the same WSN technology.
pant’s requirement complementing available In daylit spaces, shading systems are impor-
daylight while minimising energy usage. The tant lighting control elements for preventing
implementation of the lighting control system glare, and more and more of those systems
has demonstrated a more than 60% potential have become motorised and automated in
for energy savings. Although actual savings in leading-edge practices. Advanced strategies
practice will vary significantly according to are under development to integrate electric
the lighting specification for each worksta- lighting systems and shading systems for
tion, occupancy status and available daylight, optimal savings and comfort.24 It is also a
minimal energy usage is always guaranteed. natural extension of the presented research to
The lighting control system realisation in look into the interaction and integration with
the open-plan office also shows the feasibility automatic shading controls.
of wireless-networked lighting systems for While the lighting preferences in the pilot
economical retrofitting and fast implementa- study of the system implementation were
tion, especially in legacy buildings. However, predetermined, the information should be
the proposed control technique may also be obtained in a non-intrusive manner in prac-
seamlessly applied to wired lighting systems tice. This can be achieved by providing the
with individual addressability, such as DALI. occupants with a graphical interface to spec-
While the theory and implementation pre- ify and change their lighting requirements and
sented herein focused only on task lighting, preferences as our previous research demon-
the same concept may also incorporate sur- strated.21 A future research direction is to
rounding lighting to account for proper incorporate machine learning into the lighting
ambient/task contrast in practice. The desired system in order to automatically learn occu-
surrounding lighting or the ambient/task light pants’ preferences through daily usage pat-
ratio can be described as an additional set of tern, overriding events, etc.
Lighting Res. Technol. 2011; 43: 235–248
 Wireless-networked lighting 247

Human perception is another important 3 Mills E. Why we’re here: The $230-billion
aspect affecting lighting control performance global lighting energy bill: Proceedings of the
and requires further investigation before the Right Light 5 Conference, Nice, France, 2002.
proposed control technique can be deployed 4 Architectural Energy Corporation, Lawrence
at a large scale. Frequent changes or notice- Berkeley National Laboratory. Retrofit
Fluorescent Dimming with Integrated Lighting
able jumps of light levels could be considered Control – Economic and Market
annoying by the occupants even though fast Considerations. Sacramento, CA: California
response is usually favoured from a control or Energy Commission, PIER Lighting Research
energy conservation standpoint. Therefore, a Program, 2003.
smooth transition between any two light 5 DiLouie C. Lighting automation becomes the
levels has to be enforced and factored into norm. Lighting strategies - online exclusive,
the control problem to retain occupants’ 2004. Retrieved 2 May 2010, from http://
satisfaction at all times. www.buildings.com/articles/
detail.aspx?contentID¼1979.
6 Littlefair P. Daylight Linked Lighting Control
in the Building Regulations. Watford, UK:
Acknowledgement Building Research Establishment, 1999.
7 Rubinstein F, Avery D, Jennings J, Blanc S.
This work was partially funded by the On the calibration and commissioning of lighting
Chancellor’s Advisory Committee on controls: Proceedings of the Right Light 4
Sustainability (CACS) of University of Conference, Copenhagen, Denmark, 1997.
California at Berkeley. The authors thank 8 Kintner-Meyer MCW, Conant R.
Raul Abesamis, Elaine Ito, Medardo Largoza Opportunities of wireless sensors and controls
and Sal Castros of the Physical Plant – for building operation: Proceedings of the 2004
Campus Service of UC Berkeley for providing ACEEE Summer Study on Energy Efficiency in
Buildings, Pacific Grove, CA, 2004.
retrofitting and consulting service in support 9 Platts. Managing Energy Costs in Office
of green energy projects. We also wish to Buildings. New York: Platts, 2002.
thank Jessica Granderson, James Bunnell and 10 Newsham GR, Aries MBC, Mancini S, Faye
Alireza Lahijanian for their research support G. Individual control of electric lighting in a
on the implementation. The ‘smart lighting’ daylit space. Lighting Research and Technology
project, from which this paper resulted, was 2008; 40: 25–41.
initiated by partial funding from UCOP 11 Tregenza PR, Romaya SM, Dawe SP, Heap
MICRO, GE Global Research, UC Energy LJ, Tuck B. Consistency and variation in
Institute and the Public Interest Energy preferences for office lighting. Lighting
Research (PIER) Energy Innovations Small Research and Technology 1974; 6: 205–211.
Grant Program. 12 Veitch JA, Newsham GR. Preferred luminous
conditions in open-plan offices: Research and
practice recommendations. Lighting Research
and Technology 2000; 32: 199–212.
References 13 Pister KSJ, Kahn JM, Boser BE. Smart Dust:
Wireless Networks of Millimeter-Scale Sensor
1 International Energy Agency. Light’s Labour’s Nodes. Berkeley: Electronics Research
Lost – Policies for Energy-Efficient Lighting. Laboratory Research Summary, 1999.
Paris: OECD/IEA, 2006. 14 DiLouie C. Lighting and productivity: Missing
2 Rubinstein F, Jennings J, Avery D, Blanc S. link found? (Source: Architectural Lighting, 03
Preliminary results from an advanced lighting October 2003). Retrieved 2 May 2010, from
controls testbed. Journal of the Illuminating http://www.archlighting.com/industry-news.
Engineering Society 1999; 28: 130–141. asp?articleID¼453031&sectionID¼0.

Lighting Res. Technol. 2011; 43: 235–248

248 Y-J Wen and AM Agogino

15 Newsham GR, Veitch JA. Individual control 21 Wen Y-J, Bonnell J, Agogino AM. Energy
over office lighting: Perceptions, choices and conservation utilizing wireless dimmable lighting
energy savings. Construction Technology control in a shared-space office: Proceedings of
Updates, 1998. the 2008 Annual Conference of the Illuminating
16 Galasiu AD, Newsham GR, Suvagau C, Engineering Society of North America,
Sander DM. Energy saving lighting control Savannah, GA. New York: IESNA, 2008.
systems for open-plan office: A field study. 22 Wen Y-J, Agogino AM, Goebel K. Fuzzy
Leukos 2007; 4: 7–20. validation and fusion for wireless sensor
17 Lawrence Berkeley National Laboratoy. networks: Proceedings of the ASME
RADIANCE Synthetic Imaging System. International Mechanical Engineering
Berkeley, CA: LBNL, 2006. Congress, Anaheim, CA, 2004.
18 Architectural Energy Corporation. SPOT: 23 Goebel KF, Agogino AM. Sensor validation
Sensor Placement þ Optimization Tool. and fusion for automated vehicle control using
Boulder, CO: AEC, 2006. fuzzy techniques. Journal of Dynamic Systems,
19 O’Rourke C. Dimming Electronic Ballasts. Measurement, and Control 2001; 123: 145–146.
Troy NY: Lighting Research Center, 1999. 24 Kurian C, Aithal R, Bhat J, George V. Robust
20 Wen Y-J, Agogino AM. Wireless networked control and optimisation of energy consump-
lighting systems for optimizing energy savings tion in daylight-artificial light integrated
and user satisfaction: Proceedings of the schemes. Lighting Research and Technology
Wireless Hive Networks Conference, Austin, 2008; 40: 7–24.
TX, 2008.

Lighting Res. Technol. 2011; 43: 235–248