Build Simul (2008) 1: 25 35

DOI 10.1007/s12273-008-8101-4
RESEARCH ARTICLE

Predictive Simulation-Based Lighting and Shading Systems

Control in Buildings
Ardeshir Mahdavi( )
Department of Building Physics and Building Ecology, Vienna University of Technology,
Karlsplatz 13, 1040 Vienna, Austria

Received: 5 November 2007 / Revised: 7 January 2008 / Accepted: 15 January 2008

Tsinghua Press and Springer-Verlag 2008

Abstract This paper presents a prototypically implemented daylight-responsive lighting and shading
systems control in buildings that makes use of real-time sensing and lighting simulation. This system
can control the position of window blinds and the status of the luminaires. It operates as follows: (1) at
regular time intervals, the system considers a set of candidate control states for the subsequent time
step; (2) these alternatives are then virtually enacted via a lighting simulation application that receives
input data from a self-updating model of sky (luminance distribution maps obtained via calibrated
digital photography), room, and occupancy; (3) the simulation results are compared and ranked
according to the preferences (objective function) specified by the occupants and/or facility manager to
identify the candidate control state with the most desirable performance.
Keywords simulation, lighting, shading, control, sky luminance mapping, calibration

Introduction

A prototypical implementation of a daylight-responsive

lighting and shading systems control in buildings is
presented that makes use of real-time sensing and lighting
simulation. This system can control the position of window
blinds and the status (on/off, dimming level) of the
luminaires (Mahdavi and Spasojevi 2007). It possesses an
internal digital representation involving models of the
room, the context (sky), and the occupancy. The room
model entails information about room geometry, furniture,
the location and size of windows, the physical properties
of room components (such as reflectance and transmittance),
as well as the position of virtual sensors that monitor
pertinent performance parameters (such as illuminance
levels or glare indices). The room model provides the
basis of systems internal representation and is updated
dynamically using an optically based location-sensing
system (Mahdavi et al. 2007). The sky model is generated
on a real-time basis using calibrated digital photography
E-mail: amahdavi@tuwien.ac.at

(Mahdavi et al. 2006; Spasojevi and Mahdavi 2007).

Toward this end, the buildings weather station is augmented
with a digital camera with a fish-eye converter. From
images, the sky luminance model is extracted in terms of
luminance values of (256) discrete sky patches.
In general, the need for a control action arises if a
change occurs in one or more of the following:
a) Room configuration (e.g., position of furniture and
partition walls)
b) Outdoor daylight (sky) conditions
c) Occupancy (presence) and/or occupant settings (e.g.,
preferred illuminance levels, weights in the prescribed
objective functions)
To provide and maintain the desired performance under
such dynamically changing internal and external conditions,
a model-based control approach is taken (Mahdavi 2001
and 2004). Specifically, the proposed control system
operates as follows:
) At regular time intervals, the system considers a set of
candidate control states (i.e., a set of alternative
combinations of the states of control devices such as
position of blinds, dimming levels of luminaires) for
the subsequent time step.

26

Build Simul (2008) 1: 25 35

) These alternatives are then virtually enacted via

lighting simulation. Thereby, the simulation application
uses the aforementioned digital representation of the
room, sky, and occupancy toward the prediction of the
implications of the alternative control actions,
resulting in values for corresponding performance
indicators such as task illuminance levels.
) These results are compared and ranked according to
the preferences (objective function) specified by the
occupants and/or facility manager to identify the
candidate control state with the most desirable
performance.
) The system either autonomously instructs the pertinent control device-actuator(s) or informs the user to
adjust the control state.
Note that the above regime merely reflects the working
of the system in the automated control model. Users can
override (or suspend) the automated control mode at any
time. Moreover, the system can be configured to allow for
multi-user operation scenarios. However, should multiple
users require changes in the state of the same zone,
appropriate mechanisms (such as well-known agent-based
negotiation routines or hierarchically organized access
privileges) must be implemented.
As mentioned earlier, the controller application in an
actual implementation of a simulation-assisted daylightresponsive illumination control system must be equipped
with dynamic models of room, context (sky), and occupancy.
Ideally, all these three models should be self-updating: a
manual updating scenario would involve a bottleneck,
limiting thus the practical applicability of simulation-based
building systems control (Mahdavi 2004). In our implementation, the room is equipped with a location-sensing
system (Mahdavi et al. 2007) which automatically tracks
changes in the position of moveable furniture elements
(including the aforementioned luminaires). Presence of the
people in the room (at the workstations) is monitored with
occupancy sensors. User preferences (e.g., desirable
illuminance levels, relative weights for objective function)
can be communicated to the lighting control application
via occupants and facility managers computers. The next
section of the paper describes the approach toward
generation of a self-updating sky luminance distribution
model via calibrated digital photography. This dynamically
updated model provides the specification of external
illumination (daylight) input data for the simulation
application in the controller.
2
2.1

Sky luminance model

The basic idea

The controller application in the implemented systems

makes use of regular simulation-based predictions of the

performance implications of multiple control options.
However, reliable prediction of daylight availability in
indoor environments via computational simulation requires
reasonably detailed and accurate sky luminance models: the
standard sky models, still used in most daylight simulation
applications are not sensitive to transient luminance
variations in different areas of the sky. Previous research
efforts have shown that accurate sky scanning data can
provide the basis for detailed sky models (Igawa et al.
1999). Moreover, sky-scanning measurements can be used
in simulation tools to facilitate the creation of sky
luminance distributions of real occurring skies (Mardaljevic
1999). However, high cost of sky scanners necessitates the
deployment of more affordable technologies for obtaining
accurate sky models for various locations and over
statistically representative periods of time. The past
research efforts (Roy et al. 1998; Spasojevi and Mahdavi
2007; Mahdavi et al. 2006) have demonstrated that sky
luminance mapping with digital photography can provide
an alternative to high-end research-level sky scanners. This
approach requires, however, calibration, as the camera is
not a photometric device.
In the present research effort, the use of a digital camera
with a fish-eye converter is further explored toward
provision of sky luminance maps of various real occurring
skies. Toward this end, an original calibration method is
developed. This method involves simultaneous generation
of digital images of the sky hemisphere and measurement
of global external horizontal illuminance. A digital camera
equipped with a fisheye lens makes it possible to capture
images of the entire sky dome. Such a camera is used for
the present research, mounted on top of the tower of the
Vienna University of Technology. Synchronous measurements of outdoor daylight levels are performed using a
precision illuminance meter (with an error range of
maximum 10%) integral to the Departments weather
station. For each image, the initial estimate of the
illuminance resulting from all sky patches on a horizontal
surface can be compared to the measured global illuminance.
The digitally derived luminance values of the sky patches
can be corrected to account for the difference between
measured and digitally estimated horizontal illuminance
levels. For overcast and intermediate sky conditions
(without direct sun), this correction may be applied
uniformly to all sky patches. The presence of direct sun,
however, necessitates the application of a differential
(non-uniform) correction to sky patches. Based on a set of
iterations, a simple method is devised for the distribution
of this difference across the sky model. Thereby, the
difference between measured and calculated global
illuminance can be assigned to a sky area associated with
the sun position (Mahdavi et al. 2006).

Build Simul (2008) 1: 25 35

2.2

A test of calibration options and performance

An algorithm proposed by Roy et al. (1998) allows deriving

the luminance values of the particular portions of the sky
dome using the RGB values of the images pixels and the
camera metadata such as shutter speed, f-stop number, and
ISO number. This algorithm is used as the basis of the
initial calculation of sky patch luminance values. A pattern
of 256 patches is used to map sky luminance distributions.
As mentioned before, the difference between the global
horizontal illuminance derived from this initial sky patch
luminance distribution and the measured global illuminance
provides a means for calibration of digital sky images. A
correction factor, defined as a ratio of the measured global
illuminance to the horizontal illuminance obtained from
initially calculated sky patch luminance values, can be
uniformly applied toward calibration of the sky luminance
maps of cloudy skies without visible sun (Spasojevi and
Mahdavi 2007). While experimental results suggest that
the uniform correction factor corresponding to the cloudy
skies without visible sun does not exceed 1.1, it may be
much larger in case of sunny skies. Thus, for sky conditions
involving visible sun, the application of a uniform
correction factor is not appropriate. In order to arrive at a
generally applicable calibration procedure to derive sky
luminance maps from digital images, the aforementioned
illuminance difference is assigned to a sky region
associated with the sun position. To identify the proper
extent of such a region experimentally, three alternatives
are compared here. The respective sky areas in these three
cases are given in Table 1.

27

To compare the options described in Table 1, three

distinct and independent experiments are conducted.
A. Illuminance measurements with sky monitoring device
A sky monitoring device, hosting twelve illuminance
sensors, is placed on the roof of building housing the test
space to measure the horizontal illuminance resulting
from twelve different sky sectors. The measurement error
range associated with this sensor type (used also in the
experiments B and C described below) is maximum 10%.
The sky images are collected using a digital camera. A
precision illuminance meter is used to obtain the global
horizontal illuminance levels. The data set included 2547
sky images collected in July 2003 and November 2005.
Figure 1 shows the projection of the twelve sky sectors
viewed by the sky-monitoring device onto the fish-eye
images obtained from the digital camera.
The sky luminance distribution maps derived from sky
images are calibrated using the three aforementioned
methods. The illuminance predictions resulting from
calibrated sky luminance maps are compared to the
respective photometric measurements.

Table 1 The three specification options for sky regions associated

with the sun position
Solar region

Description

) CIRCUMSOLAR Five patches, one corresponding to the sun

REGION
position and the four adjacent patches
represent sun and its circumsolar region.
The initial luminance values of the other
251 patches remain unaltered.
) SUN PATCH

A patch where the sun position has been

detected is modeled as a source of direct
illumination. The initial luminance values
of the other 255 patches remain unaltered.

) SOLAR DISC

A small solid angle with an opening cone

subtending 0.533 degrees (Duffie and
Beckman 1991) is modeled as a source of
direct illumination. Solar disc is separately
modeled and added to the sky luminance
map. The luminance of all 256 sky patches
within the initial sky luminance map
remains unaltered.

Fig. 1 Fish-eye projection of twelve sky sectors as seen by

illuminance sensors

B. Illuminance measurements in a scale model

In this experiment, a scale model (1:5) of a simple
rectangular room with one window opening (Fig. 2) is
placed on the roof of the test building. Illuminance
measurements in this model are performed over the course
of 10 days in November 2005. Outdoor illuminance levels
are measured and the luminous state of the sky dome is
captured with a digital camera. The orientation of the scale
model is changed a number of times so that the scale models
window opening faced both South and North directions.
Using the lighting simulation system RADIANCE
(Ward 1994; Ward Larson and Shakespeare 2003), indoor
illuminance levels are computed for the same three positions
in the scale model. Thereby, the sky luminance models are

28

constructed using the camera-based sky luminance values

as derived via the aforementioned three calibration methods.
Subsequently, simulation and measurement results are
compared.

Build Simul (2008) 1: 25 35

interval) are compared with daylight simulation results. In

this case too, the simulations are performed with the
RADIANCE simulation system using calibrated imagebased sky luminance data as the context model.
Given these experimental setups, the three previously
mentioned options (circumsolar region, sun patch, solar disc)
are applied to the sky luminance maps of sunny skies. All
calibration procedures involved uniform correction of sky
luminance maps for cloudy skies.
Comparison using the sky-monitoring device. The
following figures (see Figs. 4 6) illustrate comparisons of
the measured illuminance values resulting from the twelve
sky sectors for 2547 sky instances with the corresponding
illuminance levels derived from camera-based luminance
maps.

Fig. 2 Illustration of the scale model of a rectangular room

C. Measurements of indoor illuminance in a real space

A sixteen-day experiment (in May and June 2005) involves
illuminance measurements inside a test space at Vienna
University of Technology (see Fig. 3). This real office space
has two casement windows equipped with movable blinds.
The blinds are continuously moved through seven different
positions, and illuminance levels are measured for every
blinds position using six sensors located on the desks
(Fig. 3, E1 to E6).

Fig. 4 Comparison of measured external illuminance levels caused

by twelve sky sectors with the corresponding camera-based
values (CIRCUMSOLAR REGION)

Fig. 5 Comparison of measured external illuminance levels caused

by twelve sky sectors with the corresponding camera-based
values (SUN PATCH)

Fig. 3 Test room with six measurement points

Simultaneously, the camera with the fish-eye lens is

used to capture the sky images, and the measurements of
global illuminance are obtained from the weather station.
Illuminance measurements (collected in 15-minute time steps
and considering all seven blinds positions for each time

Fig. 6 Comparison of measured external illuminance levels caused

by twelve sky sectors with the corresponding camera-based
values (SOLAR DISC)

Build Simul (2008) 1: 25 35

29

Comparison using the scale model. The simulated

illuminance levels at three points on the floor of the
scale model are compared to the respective photometric
measurements (see Figs. 7 9).

Three sets of calibrated sky luminance maps within the

RADIANCE lighting simulation system are applied to
obtain respective illuminance predictions. The comparison
results are shown in Figs. 10 12.

Fig. 7 Simulated versus measured illuminance values in the

scale model (CIRCUMSOLAR REGION)

Fig. 10 Comparison of simulated and measured indoor illuminance

values in the test room (CIRCUMSOLAR REGION)

Fig. 8 Simulated versus measured illuminance values in the scale

model (SUN PATCH)

Fig. 11 Comparison of simulated and measured indoor illuminance

values in the test room (SUN PATCH)

Fig. 9 Simulated versus measured illuminance values in the

scale model (SOLAR DISC)

Fig. 12 Comparison of simulated and measured indoor illuminance

values in the test room (SOLAR DISC)

Comparison based on measurements in a real space.

The illuminance measurements during a sixteen-day
experiment for six measurement points on the working
planes in the test room are compared with the respective
simulation results. The measurements involve the sensor
readings in fifteen-minute intervals. For every time step
the illuminance measurements corresponding to the seven
different blinds positions are collected (one seven-statecycle is completed within four minutes). A sky image at
the beginning of each cycle is used to derive the respective
sky luminance map. In total, 620 skies are considered.

The experiments involving the sky-monitoring device

and the scale model have a more controlled character:
the geometry of the monitoring device and the scale model
are much simpler than that of the real space. Likewise, the
simulation of light inter-reflections in a scale model (with
a limited number of surfaces with well-defined reflection
properties) is less error-prone. This explains the generally
higher correlations between measured and camera-based
values in these cases as compared to the respective results
for the test room. The latter has a significantly more
complicated geometry, a higher number of surfaces, and

30

Build Simul (2008) 1: 25 35

multiple furniture elements, leading to a more challenging

modeling task. Nonetheless, the comparisons of measured
and camera-based illuminance values suggest that
independent of the underlying experiment the calibration
method involving solar disc (use of smaller source of
direct illuminance) performs best (see Table 2 with the
comparison of the correlation coefficients for all calibration
methods and experiments). Calibrated thus, real-time digital
photography can provide a feasible technical solution
toward provision of real-time sky models to the simulation
module within the control application of the proposed
lighting and shading control system.
Table 2 Correlation coefficients regarding measured and camerabased illuminance values for the three experiments and the three
calibration methods
Calibration method
CIRCUMSOLAR

SUN

SOLAR

REGION

PATCH

DISC

Sky monitoring device

0.95

0.95

0.98

Scale model

0.88

0.76

0.96

Test room

0.68

0.80

0.85

Experiment

Fig. 13 Simplified illustration of the test space ( L1 , L2 : luminaires;

B: blind; E1 to E6 : virtual illuminance sensors)

Since all these devices affect both Em1 and Em2 , a

central control instance (C) is required to coordinate the
three devices toward the most preferable control state
(Fig. 14).

Implementation of the model-based control system

prototype

3.1

Test space

As the test space, an office at the Vienna University of

Technology was selected (see Fig. 13 for a schematic
illustration). This offices two windows are equipped with
automatically controllable blinds. Artificial illumination is
provided by two freestanding luminaires.
3.2

Control devices and control state space

Three control devices are considered (see Fig. 13), namely

the two luminaires ( L1 , L2 ) and the window blinds (B ). In
the control scenarios considered in this paper, the two
blinds are controlled simultaneously. They can be moved
up and down, and the slats can be set into horizontal and
vertical positions. As a primary indicator of lighting
performance, we considered the two mean illuminance
levels Em1 and Em2 (see Eqs. (1) and (2)):

Fig. 14 Control system influence graph representing the flow of

information form controller to devices and influence (causal)
relationships from devices to sensors ( L1 , L2 : luminaires; B: blind;
C: central control instance; Em1 and Em2 : mean workstation
illuminance levels as per Eqs. (1) and (2))

To each device, we allocate a discrete number of possible

states. The luminaires can be in any of 10 possible
dimming positions (see Table 3). The blinds can be in one
of seven possible positions (Fig. 15).
Table 3 Dimming steps and power output

Em1 = ( E1 + E3 + E5 ) / 3,

(1)

Dimming step

10

Em2 = ( E2 + E4 + E6 ) / 3.

(2)

Power output (%)

20

30

40

50

60

70

80

90

100

Build Simul (2008) 1: 25 35

31

ranges of measurable indoor environmental parameters

must be formalized and implemented. Moreover, such
preference functions need not be static, but can be
dynamically changed by users to facilitate transient
changes in operational requirements.

Fig. 15 Blind states considered in the experiment

3.3

Control objective

Fig. 16 Preference function for task illuminance

To demonstrate the working of the control system, a simple

control scenario involving three objectives is considered:
) Minimize the deviation of the prevailing values
of Em1 and Em2 from the preferred (user-specified)
illuminance levels. The concept of useful daylight
illuminance (Nabil and Mardaljevic 2005) provided, in
this case, the basis for corresponding preference
function PE (see Fig. 16).
) Minimize electrical energy use. A possible formulation
of the corresponding preference function PL is shown
in Fig. 17. It is obtained by inverting the luminaires
dimming curve (luminous flux as a function of electrical
energy input).
) Minimize cooling load. A possible corresponding
preference function ( PC ) for the latter objective is
shown in Fig. 18. The reasoning behind this
formalization is to avoid unnecessary high illuminance
levels as they typically involve heat gain, which is
undesirable in a building in cooling mode.
The overall behavior of the control system is deter-
mined through a utility function (UF). The objective of the
control process is to maximize UF. Equation (3) provides
an example for such a utility function:
UF = wE1 PE1 + wE2 PE 2 + wC PC + wL PL . 

(3)

In this equation PE1 , PE 2 , PC and PL are the preferences

for illuminance levels ( Em1 and Em2 ), cooling load, and
electrical energy consumption. The corresponding weights
are represented by wE1 , wE2 , wC , and wL . Related
illustrative preference functions are shown in Figs. 16 18.
Note that these functions represent only highly simplified
examples and serve toward demonstration of the overall
system operation. In real use situations, more sophisticated
functional relationships between user preferences and

Fig. 17 Preference function for electrical power

Fig. 18 Preference function for cooling

3.4

Control process

Consider the following scenario involving a control cycle

that is repeated regularly (in this case, every 15 minutes).
At the time step ti , the controller application C goes over
a list of candidate states for each device (i.e., L1 , L2 , B )
the time step ti +1 . In the present case, four alternative
options are considered for each device. These options are
the devices current position, the two neighboring states,
and a fourth randomly chosen option from the rest of
the devices control state space. Thus, the resulting overall
option space encompasses a maximum of 64 distinctive
control states.

32

Build Simul (2008) 1: 25 35

This approach to the selection of the candidate control

state space as a sub-region of the entire theoretically
possible control state space may be characterized as a
combination of greedy search and stochastic jumps. It is
intended to demonstrate one of the many possibilities to
deal with very large search spaces in cases, which involve
numerous control devices and corresponding device states.
From the computational point of view, an exhaustive
coverage of such large search spaces may easily become
infeasible for realistic applications (involving a large
number of rooms and building services). In the present
case, an exhaustive search for the optimal control state at
each time interval would involve 700 options (7 possible
blinds positions and 2 luminaires with 10 dimming positions
each). An overview of a number of approaches toward
efficient operation of model-based control strategies
(including search space reduction and accelerated
computation via substitution of simulation engines with
neural networks) can be found in Mahdavi (2004) and
Chang and Mahdavi (2002).
To predict the illuminance levels Em1 and Em2 at
time ti +1 due to these options, the control application uses
numeric simulation. For this purpose, the lighting
simulation program RADIANCE is adopted. Illuminance
levels at Em1 and Em2 are computed at each time step for
the aforementioned 64 device state configurations. Note
that, currently, the sky model at time ti is used to predict
illuminance levels at time ti +1 . In future tests, trend
analysis algorithms will be applied to obtain a modified
version of the sky model at time ti for computations
pertaining to time step ti +1 .
Given the obtained values of Em1 and Em2 as well as
electrical energy use (derived based on the identified
dimming state of the luminaires), UF values can be
derived using Eq. (3). Thus, at each time step, the control
state with the maximum utility function can be identified
for the subsequent time step.
3.5

performance (based on data for a day in May 2005). Figure

27 provides information toward a more comprehensive
evaluation of the systems performance over the entire
15-day duration of the experiment.
Figure 19 shows the measured external global horizontal
illuminance. To allow for an objective evaluation of
controllers performance, throughout the experiment the
resulting illuminance levels are measured at each time step
not only for the systems recommended blind position, but
also for the remaining 6 positions. In other words,
measured Em1 and Em2 values are obtained for all possible
blind positions at every time step.
To derive UF (see Eq. (3)), the following weights are
used: wE1 = wE2 = 0.3; wC = wL = 0.2. The resulting recommendations of the control system for dimming positions of
the two luminaires and the blind position are shown in
Fig. 20 and 21 respectively. Figure 22 shows the control
systems predictions of the illuminance levels Em1 and
Em2 as a result of the control systems recommended
shading and luminaire states. Figures 23 and 24 include
comparisons of predicted and measured reference
illuminance levels Em1 and Em2 . Figure 25 shows the
electrical power requirement. Figure 26 shows the course
of UF.
Figures 23 and 24 display a relatively good match
between control systems predictions and actually achieved

Fig. 19 Prevailing external global horizontal illuminance levels

in the course of one day

System performance

To illustrate and benchmark the working of the abovedescribed control method, we documented the operation of
the system in the course of fifteen days (in May and June
2005). In this case, the control systems reassessment of the
desirable control state occurs regularly every 15 minutes.
Note that this is not a fix interval and can be changed at
any time, as long as the set interval is not shorter than the
time needed for the simulation-based assessment of the
control state space (less than 15 minutes in the present
scenario). Figures 19 26 illustrate the results of the test
operation in terms of systems recommendations and its

Fig. 20 Systems recommendations for the luminaires dimming

positions

Build Simul (2008) 1: 25 35

Fig. 21 Systems recommendations for the blind positions

Fig. 22 Task illuminance levels ( Em1 and Em2 ) as the result of

systems operation

Fig. 23 Predicted versus measured task illuminance level Em1

Fig. 24 Predicted versus measured task illuminance level Em2

33

Fig. 25 Electrical power requirement as the result of systems

operation

Fig. 26 Utility function (UF) values resulting from systems

operation

illuminance levels in the test room for the selected day.

A more comprehensive evaluation of the systems
performance over the entire duration of the experiment
requires, however, an in-depth analysis of the respective
data. As mentioned before, the controller application
considers at each time interval (i.e., every 15 minutes) 64
combinations of the control device states. An exhaustive
search for the optimal control state at each time interval
would involve, however, 700 options (7 possible blinds
positions and 2 luminaires with 10 dimming positions
each). Since measured illuminance levels ( Em1 , Em2 ) are
obtained for every interval and every possible blind
position over the test period, the performance of all 700
configurations at each time interval can be objectively
ranked based on corresponding utility function values.
This availability of measured data for the entire search
space at each time interval allows for an objective
evaluation of the performance of the control method.
What was, then, the objective rank of the control
systems recommended control state amongst all 700
possible control states? Figure 27 illustrates the results in
terms of a relative frequency graph (for 590 intervals over
the test period). It suggests that for approximately 74% of
all intervals, the control state recommended by the

34

Build Simul (2008) 1: 25 35

controller is amongst the top 5% of all possible options. In

more than 80% of the time, the controllers recommendation
is amongst the top 10% list of objectively ranked control
states. Only less than 6% of the controllers recommendations fall outside the top 25% control options. This level of
performance appears quite promising, given the large list
of potential sources of error in the processes involved (e.g.,
sky scanning, image calibration, lighting simulation, indoor
measurements).

Fig. 27 Ranking of the systems recommendation amongst all

possible control options over the experiment period of 15 days,
expressed in terms of relative frequency distribution

Conclusion

The architecture and a prototypical implementation of a

simulation-assisted lighting and shading systems control in
buildings are presented. To dynamically provide the context
model (sky luminance distribution map) to the simulation
engine of the lighting control system, an original method
is developed, whereby calibrated sky luminance maps are
derived based on digital photography. The results suggest
that calibrated sky luminance maps derived based on
digital photography can provide a reliable basis for locally
representative sky descriptions in daylight simulation
tools. Thus, the effectiveness of daylight simulation can
be enhanced to support model-based lighting control
applications.
A prototypical instance of such a model-based control
approach was developed, implemented, and tested. This
system considers at regular times a set of candidate control
states for the subsequent time step. These alternatives are
then assessed via lighting simulation, resulting in values for
pertinent performance indicators. These results are compared
and ranked according to the preferences specified by the
occupants and/or facility manager to identify the candidate
control state with the most desirable performance. The
system can either autonomously instructs the pertinent
control device-actuator(s) or informs the user to adjust the
control state. Experiences to date (specifically a 15-day
continuous operation experiment) with the performance of

the simulation-based controller prototype are promising.

Developed to its full potential, the proposed system could
offer a number of advantages:
) Given the role of virtual sensors, the reliance on physical
sensors for performance monitoring can be reduced,
resulting in a more efficient sensory infrastructure.
) Physical sensors can typically monitor only limited
kinds of performance indicators (e.g., illuminance in
case of lighting controls). Using virtual sensors, more
elaborate performance indicators (such as illuminance
distribution uniformity factors and various glare
indices) can be considered.
) In a model-based control system, changes in rooms (e.g.,
remodeling, retrofit) can be digitally reflected in the
building model, thus reducing the need for extensive
reconfigurations of physical sensory components.
) Given a model-based control strategy, integration of
multiple systems (heating, cooling, ventilation, lighting,
etc.) could be achieved in a more transparent fashion,
resulting in comprehensive performance specification
and monitoring.
Work is under way to extend the methodology towards
the integrated control of buildings lighting and thermal
systems. Moreover, the scalability of the system and its
self-updating capability is to be improved via further
implementation efforts involving larger objects and multiple
environmental systems.
Acknowledgment

The research presented in this paper was supported, in part,

by two grants from the Austrian Science Foundation (FWF),
project numbers P15998-N07 and L219-N07. The author
gratefully acknowledges the contributions of Bojana
Spasojevi, particularly with regard to data collection and
option testing as related to development of the calibration
procedure for sky luminance mapping.
References
Chang S, Mahdavi A (2002). A hybrid system for daylight responsive
lighting control. Journal of the Illuminating Engineering
Society, 31:147 157.
Duffie JA, Beckman WA (1991). Solar Engineering of Thermal
Processes, 2nd edn. New York: Wiley.
Igawa N, Nakamura H, Matsuura K (1999). Sky Luminance
Distribution Model for Simulation of Daylit Environment. In:
Proceedings of Sixth International IBPSA Conference, Kyoto,
Japan, Vol. II, pp. 969 975.
Mahdavi A (2004). Self-organizing models for sentient buildings. In:
Malkawi A, Augenbroe G (eds), Advanced Building Simulation
(pp. 159 188). New York: Spon Press.
Mahdavi A (2001). Simulation-based control of building systems
operation. Building and Environment, 36(6): 789 796.

Build Simul (2008) 1: 25 35

Mahdavi A, Spasojevi B (2007). Incorporating Simulation into

Building Systems Control Logic. In: Zhao B et al. (eds),
Proceedings of the 10th International Building Performance
Simulation Association Conference and Exhibition (BS2007),
Beijing, China, pp. 1175 1181.
Mahdavi A, Icoglu O, Camara S (2007). Vision-Based Location
Sensing and Self-Updating Information Models For SimulationBased Building Control Strategie. In: Zhao B et al. (eds),
Proceedings of the 10th International Building Performance
Simulation Association Conference and Exhibition (BS2007),
Beijing, China, pp. 1291 1298.
Mahdavi A, Tsiopoulou C, Spasojevi B (2006). Generation of
detailed sky luminance maps via calibrated digital imaging, In:
Koenigsdorff R, van Treeck C (eds), Energieeffizienz von
Gebuden und Behaglichkeit in Rumen, BauSIM2006 (IBPSA),
Eigenverlag TU Mnchen, Mnchen, pp. 135 137.
Mardaljevic J (1999). Daylight Simulation: Validation, Sky Models
and Daylight Coefficients. PhD Dissertation. Institute of Energy
and Sustainable Development, De Montfort University, Leicester,
UK.

35

Nabil A, Mardaljevic J (2005). Useful daylight illuminance: a new

paradigm for assessing daylight in buildings. Lighting Research
Technology, 37: (41 57).
Roy GG, Hayman S, Julian W (1998). Sky Modeling from Digital
Imagery. ARC Project A89530177, Final Report. The University
of Sydney, Murdoch University, Australia, 72 pages.
Spasojevi B, Mahdavi A (2007). Calibrated Sky Luminance Maps
For Advanced Daylight Simulation Applications. In: Zhao B et
al. (eds), Proceedings of the 10th International Building
Performance Simulation Association Conference and Exhibition
(BS2007), Beijing, China, pp.1205 1210.
Ward G (1994). The RADIANCE lighting simulation and rendering
system. In: Proceedings of the 21st Annual Conference on
Computer Graphics and Interactive Techniques (SIGGRAPH 94),
pp. 459 472.
Ward Larson G, Shakespeare R (2003). Rendering with Radiance:
The Art and Science of Lighting Visualization, Revised Edition.
Space and Davis, CA, USA.