lable at ScienceDirect

Building and Environment 89 (2015) 229e243

Contents lists avai

Building and Environment

journal homepage: www.elsevier .com/locate/bui ldenv

Development of computational algorithm for prediction ofphotosensor signals in daylight conditions

Younju Yoon a, 1, Ji-Hyun Lee b, 2, Sooyoung Kim c, *

a Samsung C&T Corporation, Construction Technology Center, Seoul, South Koreab Graduate School of Culture Technology, Korea Advanced Institute of Science and Technology, Daejeon, South Koreac Department of Interior Architecture & Built Environment, Yonsei University, Seoul, South Korea

a r t i c l e i n f o

Article history:Received 7 January 2015Received in revised form22 February 2015Accepted 24 February 2015Available online 5 March 2015

Keywords:Photosensor signalLighting controlDaylight conditionsShielding

* Corresponding author. Tel.: þ82 2 2123 3142; faxE-mail addresses: younju.yoon@samsung.com (Y.

(J.-H. Lee), sooyoung@yonsei.ac.kr (S. Kim).1 Tel.: þ82 10 9379 3371.2 Tel.: þ82 42 350 2919.

http://dx.doi.org/10.1016/j.buildenv.2015.02.0300360-1323/© 2015 Elsevier Ltd. All rights reserved.

a b s t r a c t

This study aims to develop and validate an annual photosensor performance simulation method (APPSM)to compute the photosensor signals for a lighting control system under various daylight conditions. Aseries of computer simulations using PSENS, which is a simulation program within Radiance softwarewere conducted and field measurements were performed under various daylight conditions in order tovalidate the simulation results of APPSM.

Results indicate that the photosensor signals predicted by PSENS and APPSM showed a strong linearcorrelation. Prediction results by APPSM generally consisted with the results field measurements,although slight differences between them existed under particular daylight conditions. The differences inphotosensor signals between the prediction by APPSM and measurement effectively decreased asshielding conditions were applied to photosensors.

A strong linear relationship existed between the photosensor signals obtained from prediction byAPPSM and the field measurements. The prediction models for the photosensor shielding conditionswere acceptable with a significance level of 0.01. The majority of percent differences between themeasured and simulated photosensor signals were within 10% under clear and partly cloudy skyconditions.

© 2015 Elsevier Ltd. All rights reserved.

1. Introduction

A daylight responsive dimming control system,whichmaintainsa design level of illuminance at the task surface with the use ofelectric lighting, can significantly reduce the electric lighting use inspaces where daylight is a useful source of illumination [1e7].Daylight associated with electric light which is controlled by adaylight dimming system contributes to occupants' visual comfort,psychological satisfaction and work productivity [8e15]. Despitethe energy saving potentials and occupants' satisfaction obtainedfrom the daylight dimming control system, it has not been widelyused in commercial buildings due to the difficulty in the calibration

: þ82 2 313 3139.Yoon), jihyunlee@kaist.ac.kr

and optimization of the system performance and malfunction ofcontrol system devices [16e19].

For the daylight dimming system, the ideal location for a pho-tosensor would be on the workplane but this position is inappro-priate because the photosensor would likely be disturbed or shadedby activities in the room. Thus, the photosensor is mounted on theceiling rather than the workplane to minimize interference fromactivities in the room. Controlling workplane illuminance with asensor located on the ceiling complicates photosensor control[20e25]. The correlation of the illuminance levels between theworkplane and ceiling is strongly influenced by the position andspatial response of the photosensors [4,20,23,26e28].

Therefore, a proper calibration of a photosensor is critical toensure reliable operation of lighting control system and lightingenergy savings. In order to achieve a target light level at theworkplane, the calibration of the photosensor for a particularsensor position and orientation for optimum operation adjusts thecontrol algorithm, which is based on sensor signal to ballast outputrelationship [29e31].

Nomenclature

usun solid angle of the sunusky,patch solid angle of the sky patchEref,sun reflected illuminance contribution from the sunDCref,sky i reflected component of daylight coefficient from

sky patch iLsun luminance of the sunWi weighting factor for sky patch i

Y. Yoon et al. / Building and Environment 89 (2015) 229e243230

Knowing the annual performance of the photosensor is essentialto determine an optimum sensor position and spatial response thatinfluence lighting energy savings without undershooting or over-shooting a target illuminance in space. Several computer simula-tion tools, such as SPOT, PSENS and DAYSIM, have been used toanalyze photosensor system performance and the behavior oflighting control systems.

SPOT is used to simulate photosensor performance and deter-mine optimum photosensor positioning using annual daylightperformance in a space [32]. The annual daylight simulation isbased on bihourly daylight factors for three representative daysunder clear and overcast sky conditions. The hourly computation ofdaylight performance is computed by interpolating pre-computeddaylight factors. Accordingly, annual daylight performance isinaccurately estimated, especially for clear and partly cloudy skyconditions. In addition, the spatial distributions of the photosensorsthat control the influence of the photosensors on the dimmingcontrol system were not specifically included.

DAYSIM, which is based on Radiance, allows users to modelannual variations of daylight in space and to analyze complexelectric lighting systems controlled by a photosensor dimmingsystem. Annual daylight simulation by DAYSIM is based on daylightcoefficients approach and the Perez sky model [33e36]. Consid-ering the spatial response of the photosensors and lighting controlalgorithms for the dimming and switching of the electric lightingsystem, the DAYSIM incorporates a separatemodule, which permitsthe analysis of electric lighting and the modeling of integratedphotosensor lighting controls [37].

Also, it was known that the simulation error of DAYSIMexpressed in mean bias errors (MBEs) ranged from 6 to 20% ascompared to measured data [35,38]. However, the performance ofthe photosensor analysis module for calculations needs furthervalidation against measured data under real daylight conditions.

PSENS, which is a simulation procedure within the Radiancesoftware, is able to simulate the behavior of photosensors inresponse to various daylight and electric lighting conditions. Pho-tosensor response is a function of the luminance distributions fromall surfaces which are detected by the sensor. Therefore, it com-putes a photosensor signal by multiplying the luminance distri-bution of a space captured as a fisheye view with the angularsensitivity data of the photosensor.

The accuracy of the PSENS program in Radiance was validatedwith the percent error range of ±6.7% using measured data [39].However, examining the annual performance of photosensors us-ing PSENS is impractical since PSENS requires hourly fish-eye im-ages of the space for an entire year, which is a very computationallyextensive procedure. Accordingly, the performance of the photo-electric dimming control system simulated by PSENSwas examinedfor limited photosensor conditions under various daylight condi-tions such as cloudiness levels.

Under these circumstances, an accurate annual photosensorperformance simulation method, which computes photosensor

signals considering its spatial sensitivity in response to variousdaylight conditions, are necessary in order to predict overall systemperformance prior to installation in the field and actual commis-sioning of photosensor-based lighting control system.

Therefore, this study develops an annual photosensor perfor-mance simulation method (APPSM) and validates its accuracyagainst an existing computation software and field measurementdata. The photosensor signals computedwith APPSM can be used toassist in the commissioning of photosensor-based lighting controlsystem.

In this study, the theoretical foundation for the APPSM is pro-posed, and validations of computational results from the APPSMwere performed in order to prove the accuracy of the computa-tional results. The validation was performed by comparing thephotosensor signals computed by APPSM with those predicted bythe PSENS program in Radiance for typical classrooms, since PSENSis a validated program using field measurement data and proven tobe accurate [38]. Also, the variations in illuminance computed bythe APPSMwere compared with actual data that were acquired in afull-scale mock-up space under a variety of sky conditions.

2. Development of computational algorithm

2.1. Computational algorithm for annual photosensor performancesimulation method

The primary computational method developed in this study wasthe APPSM. The theoretical foundations that form the fundamentalgrounds for this method are briefly discussed in this section. TheAPPSM adopts the daylight coefficient theory and considers the sunand sky as separate light sources [40]. As shown in Fig.1, the APPSMfor the sky uses the daylight coefficients for the 145 sky patchescovering the sky hemisphere to compute the illuminance contri-bution from each sky patch to an illuminance calculation pointincluding interreflections [41].

This fixed numerical relationship between the luminance fromthe surface of each sky patch and the resulting illuminance at thecalculation point is called a daylight coefficient. Once the daylightcoefficients have been computed, the illuminance contributionfrom the sky with an arbitrary luminance distribution can be easilycalculated by summing the multiplication of the luminance valuesof the 145 sky patches with their respective daylight coefficients. Inthis study, the daylight coefficients are computed using the rtcon-trib program in Radiance, which traces rays from each analysispoint and records the hits of individual sky patches until they reachthe requested bounce numbers.

In the APPSM, the contribution from the sun is divided intodirect and reflected components to achieve higher accuracy in theprediction of illuminance from the sun. The direct illuminance iscalculated using rtrace, which is one of the computation programsin Radiance and takes only a fewminutes to compute all hourly sunconditions occurring for an entire year. The reflected illuminance ofthe sun can be computed using the reflected daylight coefficientsfrom the 145 sky patches, which have been computed to derive theilluminance contribution from the sky.

Two different approaches were considered to compute the re-flected component of the daylight coefficient for a specific sunposition [40,42,43]. One is to use the single-sky-patch condition. Inthis case, the reflected component of the daylight coefficient for thesky patch that contains the sun is used. For example, sky patch 132shown in Fig. 2 has the sun at 11 a.m. on June 17 within the patch.Therefore, the reflected daylight coefficients from sky patch 132 areused to compute the reflected sun illuminance for the sun at 11 a.m.on June 17.

Fig. 1. Tregenza Sky Patches (a) Sky patches [40] (b) Contribution of sky patch to illuminance [33].

Fig. 2. Visualization of indirect sky daylight coefficients calculation. (a) Single sky patch (b) Four sky patches.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 231

The other approach is to use the four-sky-patches condition,where the reflected components of daylight coefficients for the fourneighboring sky patches are interpolated according to differentweighting factors. The weighting factors are determined by therelative distances from the centers of the four sky patches to thecenter of the sun. The distance fraction of each sky patch to the sumof all four distances is applied as the interpolation factor for eachsky patch.

For example, the four sky patches 131, 132, 141, and 142 sur-round the sun at 11 a.m. on June 17, as shown in Fig. 2. The distanceratios from the center of the sun to the centers of sky patches 131,132, 141, and 142 are 0.54:0.41:0.75:1.0. Accordingly, the indirectluminances assigned to sky patches 131, 132, 141, and 142 are 28%,37%, 20%, and 15%, respectively.

By borrowing the daylight coefficients of the sky patch tocompute the reflected sun illuminance, the sun is modeled as aninfinitely distant light source, approximately 724 times its actualsize. Therefore, the sun's luminance needs to be reduced by a factorof 724, which accounts for the difference in solid angle of a skypatch to the sun. The procedure to compute the solid angle of thesun is shown in Equation (1).

Solid angle of the sun ¼ p� 0:5� p

180=2

2

¼ 0:0000598115

� �

(1)

The reflected illuminance contribution from the sun iscomputed using Equation (2). Direct sunlight is calculated for thesun positions for every hour of the year. The total illuminance fromthe sun is the sum of the direct illuminance and the reflectedilluminance caused by the sun.

1Þ Single� sky� patch condition

Eref ;sun ¼ DCref ;sky � Lsun � usun

usky patch

2Þ Four� sky� patches condition

Eref ;sun ¼�DCref ;sky1 �W1 þ DCref ;sky2 �W2 þ DCref ;sky3

�W3 þ DCref ;sky4 �W4

�� Lsun � usun

usky patch

(2)

The computation algorithm of APPSM, which was developed inthis study, employs the four-sky-patches condition to compute the

Y. Yoon et al. / Building and Environment 89 (2015) 229e243232

reflected illuminance by the sun, since it has been confirmed thatinterpolations using four daylight coefficients performed betterthan the single-sky-patch condition [40,42,43].

2.2. Modeling of photosensors

In order to model the performance of a photosensor, this studymodifies the initial ray distribution at the photosensor location tofollow the spatial sensitivity distribution of a photosensor. Threesteps were employed to model the photosensor.

First, the spatial sensitivity distribution of a photosensor ismodeled in the APPSM by placing a sphere with a diameter of2.54 cm at the photosensor location. The spatial transmission of thesphere follows the spatial sensitivity distribution of the photo-sensor. The transmission function is applied using a “trans”materialin Radiance.

Second, photosensor signals due to the sky are derived from thedaylight coefficients, which are computed using rtcontrib based onan octree file including the geometrical description of the simula-tion space and the sphere. When Radiance runs the rtcontrib pro-gram, it assumes that the illuminance sensor already has a cosinespatial sensitivity distribution, which needs to be removed tocompute the photosensor signals. Therefore, the actual spatialsensitivity distribution of the photosensor is divided by the cosinespatial sensitivity distribution.

Finally, photosensor signals due to the reflected contribution ofthe sun are computed by interpolating the reflected daylight co-efficients of four neighboring sky patches based on the relativedistance from each sky patch to the sun. Direct sun contributionsare computed using the rtrace program in Radiance.

3. Validation procedure for computation algorithm

3.1. Simulation conditions for validation of APPSM against PSENS

The accuracy of APPSM was validated by comparing its calcu-lation results with those of PSENS, since PSENS effectively predictsthe behavior of photosensors in response to various daylight andelectric lighting conditions, and is a validated program using fieldmeasured data [39]. PSENS computes photosensor signals by pixel-by-pixel multiplication of two fisheye images; one generated fromangular acceptance curve of the photosensor and the other from a180 or 360� fisheye simulation of the space as seen by thephotosensor.

The space that was modeled for PSNES and APPSM in this studywas a typical classroom. The space was assumed to be located inBoulder, Colorado, USA (latitude: 40� N, longitude: 105.2� W). Thedimensions of the space were 9.8 m � 9.1 m � 3.7 m, as shown inFig. 3.

The façade of the classroom included double-pane clear win-dows separated by mullions into upper and lower windows. Theratio of window to wall on the façade was 59.45%. The trans-mittance of thewindowswas assumed to be 0.67. The reflectance ofthe ceiling, walls, and floor were assumed to be 0.75, 0.5 and 0.25,respectively, which are typically used values for classrooms. Theground reflectance was 0.2. No desktop or electric lighting wasassumed to be installed in the classroom.

A Venetian blind was assumed to be installed on the inside ofthe windows for a particular simulation case. The reflectance of theblind slats was assumed to be 0.71. The width of the blind slats andthe distance between themwas 2.54 cm. The building façades withthe windows faced south and north in each simulation.

As shown in Fig. 3, it was assumed that two photosensors werepositioned on the ceiling aiming downward without being tilted.The photosensors were assumed to have a 45� cosine sensitivity

distribution pattern, as shown in Fig. 3. Accordingly, the photo-sensors detected signals from the area within 45� from the nadir,and no signals were assumed to be detected from the area beyond45� from the nadir.

The TMY2 weather file for Boulder was used to generate Perezweather sky files for every hour of the year [44,45]. The sky con-ditions used in the simulations were clear, intermediate, andovercast sky according to the opaque sky cover field in the TMY2file. Opaque sky cover is a data element of TMY2 weather file. It isdefined as the amount of sky dome in tenths covered by clouds orobscuring phenomena that prevent observing the sky or highercloud layers at the hour indicated [46]. The sky conditions of 0 and10 indicate a completely clear sky and an overcast sky, respectively.

The simulations were performed for several days inMarch on anhourly basis from 08:00 to 17:00. A summary of the daylight con-ditions used for each simulation case is shown in Table 1. Theanalysis results discussed in this study are based on selected datafor time periods that effectively represented sky conditions on eachday.

3.2. Experimental conditions for validation of APPSM against fieldmeasurement

Field measurements were performed in a full-scale mock-upmodel to monitor the variations in daylight illuminance under avariety of sky conditions and to comparemeasurement results withthe simulation results from APPSM. The full-scale mock-up spacewas constructed for the examination of lighting control perfor-mance and visual environment under real daylight conditions. Themodel space was located in Seoul, South Korea (latitude: 37.10� N,longitude: 126.58� E). The detailed layout of the full-scale mock-upspace is shown in Fig. 4.

The long axis of the model was tilted toward east by 22� fromthe northesouth axis, and double skin envelopes were installed atthe façade of the model space. Both external and internal envelopeswere covered with equal double pane glazing. The cavity spacebetween the envelopes was 3.6 m wide, 0.9 m deep, and 2.65 mhigh.

The side surfaces of the cavity space were also covered with thesame glazing used for the internal and external envelopes. Thetransmittance of light and solar irradiance of the glass used for theglazing was 0.621 and 0.348, respectively. Venetian blinds with2.54 cm blind slats were installed on the inside of the internalenvelope to control the penetration of daylight under diverse skyconditions.

The space was 3.6 m wide, 4.2 m deep, and 2.65 m high. Themodel space was furnished as a small private office. The wall sur-face was finished with white wallpaper and the floor was finishedwith light beige linoleum. No specular reflection was generatedfrom the wall and floor surfaces. An array of 0.6 m by 0.6 m sus-pended grids covered with white acoustic panel boards wasinstalled on the ceiling. Six fluorescent lighting fixtures wereinstalled on the ceiling, but they were not operated during theentire period of the experiment to monitor the variations indaylight illuminance in the model space.

To monitor the photosensor illuminance according to theshielding conditions, three types of photosensors (full-shielding,partial-shielding, and no-shielding) were installed at one point onthe ceiling aiming normal to the floor. The sensors were positioned3.3 m away from the internal envelope, as shown in Fig. 4.

The shielding conditions of the photosensors are described inFig. 5. The full-shielding condition provided a 76.8� aperture with a38.4� half angle of field view. For the partial-shielding conditions,the sensor was shielded 90� toward clockwise and counterclock-wise from facing directly toward the double skin envelope. The rest

Fig. 3. Space layout and photosensor condition. (a) Plan (b) Elevation of front façade (c) Photosensor spatial sensitivity distribution (45� cosine).

Table 1Simulation conditions for photosensor illuminance by PSNES and APPSM.

Window orientation Blind condition Photosensor sensitivity Surface reflectance Calculation method Number of sky patch

South No 45� cut off - Ceiling: 0.75- Wall: 0.50- Floor: 0.25- Outdoor ground: 0.20

Four sky patch 14545� 45� cut off

North No 45� cut off

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 233

of the sensor was open to the rear area of the space to receive thereflected component of light from the rear area. The surface color ofthe shielding material for the sensor was dark brown.

Data monitoring was conducted from December 2009 to August2011 on a daily basis to examine the variations in daylight illumi-nance according to the shielding configurations under a variety ofsky conditions. Daylight illuminance at the photosensors wasmonitored from 07:00 to 18:00, and the monitoring interval wasone minute. The sensor signal was detected in current and wasconverted into a standard unit of illuminance using a calibrationconstant assigned to each sensor. The sensitivity deviations of thesensor were within 1% and have been used effectively in otherstudies [47e49]. An automatic data logging system that contains 35differential channels for the detection of signals was used to recordthe illuminance and irradiance data. The accuracy range of datalogger is from �2.5 mV to 2.5 mV when temperature is between0 8 �C and 40 8 �C [50].

The conditions of the full-scale mock-up model space weremodeled for the simulation by APPSM. The Perez sky models weregenerated using the measured global horizontal irradiance anddirect normal irradiance derived from the direct horizontal irradi-ance after dividing by the solar zenith angle for a given time. Thesimulation results from APPSM were compared with the results ofthe field measurements.

4. Results

4.1. Comparison of simulation results by PSENS and APPSM undermodeled sky

Simulation results from PSENS and APPSM were compared un-der various daylight conditions to validate the computational re-sults from APPSM. Daylight illuminance levels at the twophotosensor positions shown in Fig. 3 were computed for opaquesky cover, which is widely used to determine sky conditions in atheoretical model.

The simulation results from the two computational methods areshown in Figs. 6e8. In the figures, an opaque sky cover value of0 indicates a clear sky and 10 indicates a overcast sky. Overall, thedaylight illuminance levels of the two photosensors were higher forthe south-facing conditions under lower opaque sky cover. Fig. 6shows the comparison of photosensor signals computed byPSENS and APPSM for the south-facing clear window simulationcase. The prediction result by the two computational methodsshowed narrow ranges of difference between them, when overcastsky conditions were formed.

Overall, the differences in estimating photosensor signals usingAPPSM became larger when the sky conditions were clear andallowed direct sunlight patterns inside the space. For the

Fig. 4. Space layout for field measurement. (a) Plan (b) Section.

Fig. 5. Shielding conditions for photosensors. (a) Full shielding (b) Partial shielding (c) No shielding.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243234

photosensor position #1, the maximum difference between APPSMand PSENS was 179.2 lx, when the opaque sky cover value was 2.However, most of the difference fell into within 102.5 lx. For pho-tosensor #2 under lower values of opaque sky cover, the majority ofdifference between them was within 78.5 lx.

For the rest of opaque sky cover, the PSENS and APPSM gener-ated close daylight illuminance at the two photosensors. Themaximum difference between themwas 48.3 lx for the opaque skycover value of 6. The majority of prediction results by PSENS andAPPSM showed close deviation between them.

To examine the impact of direct illumination from the sun onthe prediction of photosensor signals, simulation cases of a south-facing window with a 45� tilted blinds and a north-facing window

were analyzed. As shown in Fig. 7, the photosensor signalscomputed with APPSM agree well with the signals computed byPSENS. Overall, the illuminance by PSENS was slightly greater thanthat of APPSM, but the difference range was very narrow. Themaximum difference in photosensor signals between APPSM andPSENS was 16.4 lx for the sky with an opaque sky cover of 2.

Compared to the case of noblind condition under lower opaquesky cover value, the difference range of illuminance generated byPSENS and APPSM became very narrow due to effective control ofdirect sunlight. Also, the difference between prediction results bythe two computational methods became narrower as opaque skycover value increased. This tendency also occurred under noblindconditions when the influence of direct sun was strong. This result

0

500

1000

1500

2000

2500

3000

-1 0 1 2 3 4 5 6 7 8 9 10 1Opaque Sky Cover

Phot

osen

sor

Illum

inan

ce [l

x]

1

PSENSAPPSM

0

500

1000

1500

2000

2500

3000

-1 0 1Opaq

Pho

tose

nsor

Illu

min

ance

[lx]

2 3 4 5 6 7 8 9 10 11ue Sky Cover

PSENS

APPSM

(a)

(b)

Fig. 6. Photosensor Illuminance by PSENS and APPSM (South-facing, Noblind). (a)Photosensor position #1 (b) Photosensor position #2.

Fig. 7. Photosensor Illuminance by PSENS and APPSM (South-facing, 45� blind). (a)Photosensor position #1 (b) Photosensor position #2.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 235

indicates that high accuracy in the prediction of photosensor sig-nals can be achieved with APPSM when direct sunlight is not pre-sent in an interior space.

Fig. 8 compares the photosensor signals computed by APPSMand PSENS for a simulation case with a north-facing clear window.Under all sky conditions, the prediction results from PSENS andAPPSM were close. All differences were within 3.5 lx except for onecase for a particular sky condition. As sky conditions transitionedfrom clear to overcast, the difference became narrower. It appearsthat the accuracy in the prediction of the photosensor signals wasvery high due to the absence of direct sunlight patterns on theinterior surfaces.

In order to examine the results computed by PSENS and APPSM,Bivariate analysis method which generates Pearson Correlation,and linear regression analysis with analysis of variance (ANOVA)were employed in this study. The results from PSENS and APPSMwere selected as variables for the analysis. Scatter plots showingthe relationships between the computational results are shown inFig. 9. The ANOVA test and Pearson Correlation results for theirrelationship are summarized in Tables 2e4.

Overall, the test results for their relationship were acceptablewith a significance level of 0.01. The Pearson Correlation co-efficients ranged from 0.9737 to 0.9990. This implies that the linearequations describe the correlation between the results from PSENSand APPSM is appropriate. Also, the coefficient of determination(r2) ranged from 0.9825 to 0.9993. The models imply that very

strong linear correlations exist between the computational resultsof PSENS and APPSM. In summary, the computational resultsshowed a strong foundation for further analysis to validate theresults computed by APPSM against the results of PSENSsimulations.

4.2. Variations in outdoor daylight illuminance and sky conditions

Field measurements were performed to monitor the variationsin daylight illuminance and sky conditions during the entire datamonitoring period. Data that represent specific sky conditions wereselected among the entire set of data collected during the moni-toring period in order to examine the variations in outdoor daylightilluminance that affect photosensor signals. Fig. 10 show the vari-ations in outdoor global illuminance and vertical illuminance to anexternal envelope under selected sky conditions.

As shown in Fig. 10(a), the clear sky conditions provided stablevariations in global and vertical illuminance for the entire timeperiod. Statistics of outdoor daylight iluminance for clear sky con-dition are summarized in Table 5. Global illuminance showed cleardifferences according to the months, which have different solaraltitudes. The mean illuminance in December and April was65,850 lx and 31,200 lx, respectively with standard deviation of18,400 lx and 15,620 lx. The maximum illuminances in April andDecember were 87,700 lx and 49,950 lx, respectively. Mode, whichappears most frequently in the monitored data, was 85,700 lx and47,750 lx for April and December, respectively.

Fig. 8. Photosensor Illuminance by PSENS and APPSM (North-facing, Noblind). (a)Photosensor position #1 (b) Photosensor position #2.

Fig. 9. Relationship between photosensor illuminance by PSENS and APPSM. (a)South-facing condition (b) North-facing condition.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243236

The vertical illuminance monitored at the external envelope ofthe full-scale mock-upmodel also showed stable patterns similar tothat of the global illuminance. Unlike the case of global illuminance,the vertical illuminance in April was lower than that in Decemberdue to the direct sunlight that reaches the sensor with a lowerincident angle. The mode of vertical illuminance was 70,900 lx, andmean values in April and December were 46,290 lx and 53,720 witha standard deviation of 22,540 lx and 26,230 lx, respectively. Themaximum illuminances in April and December were 72,700 lx and88,200 lx, respectively.

The variations in outdoor daylight illuminance under a partlycloudy sky are shown in Fig. 10(b) and statistics are summarized inTable 5. Overall, illuminance varied with unstable patterns andshowed much fluctuation. Global illuminance in June was higherthan that in December, but vertical illuminance in June was lowerthan that in December. This pattern was similar to that observedunder clear sky conditions.

Mode of global illuminance in June and December was 72,100 lxand 1360 lx, respectively. Mean illuminance in June and Decemberwas 58,520 lx and 22,080 lx with a standard deviation of 12,870 lxand 12,710 lx, respectively. Mode of vertical illuminance was22,340 lx and 2460 lx in June and December, respectively. Mean inJune and December was 22,440 lx and 31,290 lx with standarddeviations of 8606 lx and 20.542 lx. This implies that the data set ofilluminance has a narrower distribution from the mean comparedwith clear sky conditions.

Fig. 10(c) shows the variations in outdoor daylight illuminanceunder overcast sky conditions and statistics are summarized inTable 5. Global and vertical illuminance levels showed stable pat-terns of change without significant fluctuation. The monthlychanges in illuminance were not significant compared to thosefound in the cases of clear and partly cloudy sky conditions.

The maximum global illuminance levels in July and Decemberwere 24,530 lx and 12,470 lx, respectively. Mean was 6670 lx and7020 lx with a standard deviation of 5197 lx and 2817 lx. This resultimplies that the illuminance data set had the narrowest distribu-tion, among three daylight conditions. Mean of vertical illuminancein July and December was 3010 lx and 3820 lx with a standarddeviation of 2353 lx and 1690 lx, respectively.

In this study, sky ratios were used in order to clarify sky con-ditions. The sky ratio is defined as a ratio of outdoor diffused irra-diance to global irradiance under daylight conditions. According tothe sky ratio, sky condition is classified into clear (SR � 0.3), partlycloudy (0.3 < SR < 0.8), and overcast (SR � 0.8) sky [51,52]. The skyratios of some selected days among the data monitoring period areshown in Fig. 11. The sky ratios were calculated based on the irra-diance data monitored in the field measurements. Statistics of skyratio for each sky condition are summarized in Table 6.

The majority of sky ratios in April was stable and did not exceed0.31. This indicates a clear sky condition under which direct sun-light was strong and the diffused component from the sky wasweak. Mean and mode were 0.16 and 0.12. This implies that the skycondition was clear during monitoring period.

Table 2Relationship between calculated photosensor illuminance by PSENS and APPSM.

Window orientation Photosensor position Statistics ANOVA

Variable Unstandardized coefficients t Sig.

B Std. error

South # 1 (Constant) 13.2613 4.2387 3.13 0.00 F (83.1) ¼ 40455.4r2 ¼ 0.9980, Sig. ¼ 0.00PSENS1 0.9458 0.0047 201.14 0.00

# 2 (Constant) �7.3871 6.5448 �1.13 0.26 F (83.1) ¼ 4665.1r2 ¼ 0.9825, Sig. ¼ 0.00PSENS2 1.1225 0.0164 68.30 0.00

North # 1 (Constant) �0.8054 0.6441 �1.25 0.22 F (27.1) ¼ 61130.1r2 ¼ 0.9996, Sig. ¼ 0.00PSENS1 1.0056 0.0041 247.24 0.00

# 2 (Constant) �0.7576 0.6114 �1.24 0.23 F (27.1) ¼ 37322.5r2 ¼ 0.9993, Sig. ¼ 0.00PSENS2 1.0079 0.0052 193.19 0.00

Table 3Relationship between calculated photosensor illuminance by PSENS and APPSM (South-facing condition).

Statistics PSENS (ps #1) PSENS (ps #2) APPSM (ps #1) APPSM (ps #2)

PSENS (ps #1) P. C 1.00Sig.N 85

PSENS (ps #2) P. C 0.9926 1.00Sig. 0.00N 85 85

APPSM (ps #1) P. C 0.9990 0.9899 1.00Sig. 0.00 0.00N 85 85 85

APPSM (ps #2) P. C 0.9877 0.9912 0.9890 1.00Sig. 0.00 0.00 0.00N 85 85 85 85

Where, ps: photosensor, P. C: Pearson correlation, N: number of data, Sig.: significance level.

Table 4Relationship between calculated photosensor illuminance by PSENS and APPSM (North-facing condition).

Statistics PSENS (ps #1) PSENS (ps #2) APPSM (ps #1) APPSM (ps #2)

PSENS (ps #1) P. C 1.00Sig.N 29

PSENS (ps #2) P. C 0.9748 1.00Sig. 0.00N 29 29

APPSM (ps #1) P. C 0.9998 0.9737 1.00Sig. 0.00 0.00N 29 29 29

APPSM (ps #2) P. C 0.9763 0.9996 0.9757 1.00Sig. 0.00 0.00 0.00N 29 29 29 29

Where, ps: photosensor, P. C: Pearson correlation, N: number of data, Sig.: significance level.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 237

In contrast, the majority of sky ratios in December ranged from0.8 to 1.0. This indicates that the sky condition was overcast andthat the diffused component dominated the illuminance. Both clearand overcast sky conditions showed very stable patterns of change.As shown in Fig. 11 and Table 6, the ratio indicated that the sky wasan overcast condition (Mean: 0.92, Standard Deviation: 0.036).

The sky ratios in June showed noticeable fluctuation, and themajority of values ranged from 0.3 to 0.8. This indicates a partlycloudy sky, where the influences of the direct and diffused com-ponents of daylight varied within a short time period (Mean: 0.52,Standard Deviation: 0.181). Mode, which appears most frequentlyin the monitored data, was 0.39.

4.3. Validation of APPSM against results of field measurement

The daylight illuminance of photosensors according to theshielding conditions shown in Fig. 5 was analyzed to examine the

differences between the results from field measurements andsimulations by APPSM. The variations in daylight illuminance ofphotosensors under three representative sky conditions are shownin Figs. 12e14. Overall, the computational results from APPSMweregenerally consistent with the results of the field measurements,although differences between them varied according to day, solaraltitude, sky conditions, and shielding configurations.

Fig. 12 shows the trend in daylight illuminance of photosensorsduring selected days in April and December under clear sky con-ditions. The illuminance varied stably over the entire time periodand showed differences according to the shielding conditions.Overall, the difference range between measurements and compu-tational results was narrower for April than for December. Detailedstatistics for the difference between measured and simulatedphotosensor illuminance are summarized in Table 7.

The illuminance of the fully-shielded photosensor showed thelowest values due to the effective blocking of direct sunlight by the

Fig. 10. Variation of outdoor illuminance. (a) Clear sky (b) Partly cloudy sky (c)Overcast sky.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243238

shielding configuration. The lower solar altitude in winter causedthe deep penetration of daylight into the indoor space andincreased daylight illuminance at the photosensors, because noshading device was installed in the model space.

For unshielded photosensor conditions, the difference wasgreater than those of the two shielded photosensors. Themaximumdifference between measured and simulated illuminance occurredaround 11:00, when the maximum direct solar component reachedthe envelopes. Shielding conditions effectively reduced the differ-ence between the measured and simulated illuminances. Partialshielding conditions that allowed the photosensors to detect

reflected components from the rear wall surfaces generated slightlybigger differences in the early morning, when direct sunlightpenetrated deeply into the inside of the space with a lower incidentangle.

This implies that APPSM considered the reflected componentsfrom real wall surfaces more effectively. When direct sunlight wasnot available in the space, the differences were not significant. Thistendency was stronger when full shielding conditions were usedfor the photosensor.

Fig. 13 shows variations in photosensor illuminance by daylightunder partly cloudy sky conditions. Detailed statistics for the dif-ference betweenmeasured and simulated photosensor illuminanceare summarized in Table 7. Photosensor illuminance varied unsta-bly over the entire period due to the change in sky ratio. Themeasured and simulated illuminances of the photosensors showedno significant differences in June. More difference between themoccurred for the unshielded photosensor conditions. The majorityof difference ranges were within 100 lx. Like the case of clear skyconditions, the shielding conditions contributed to reduce the dif-ferences between the measured and simulated illuminance levels.

In December, when direct sunlight was strong with a lowerincident angle, the difference range was wider than that in June. Inparticular, unshielded conditions showed significant differences.Fully-shielded conditions reduced the difference effectively. Thedifference range of photosensor illuminance was the greatest forunshielded conditions and smallest for fully shielded conditions.Also, the simulated illuminance levels predicted by APPSM werelower than the measured levels.

These results imply that the difference between measured andsimulated illuminance was greater when the incident angle ofsunlight was lower in the winter season, and that the reflectedcomponents of incident direct sunlight were efficiently detected bythe photosensors.

Fig. 14 shows the variations in photosensor illuminance underovercast sky conditions, where the sky ratio was greater than 0.8.Overall, the variation range of illuminance was not greatercompared to the cases of clear and partly cloudy sky conditions. Thepatterns of change in illuminance from computations and fieldmeasurements were similar to one another. Detailed statistics forthe difference between measured and simulated photosensor illu-minance are summarized in Table 7.

For unshielded conditions, the majority of differences werewithin 50 lx in July. The differences did not exceed 53.2 lx and32.1 lx under the partial and full shielding conditions, respectively.The illuminance calculated by APPSM showed slightly lower valuesthan those of the field measurements.

The differences between the simulated and measured illumi-nances in December were greater than those in July. In particular,the difference was noticeable around 11:00, when the incidentangle of direct sunlight was approximately normal to the envelopesof the mock-up model space under clear sky conditions. When thesun moved out of the viewing angles of the windows, the differ-ences reduced noticeably. Also, the shielding conditions effectivelyreduced the differences.

Under overcast sky conditions, the sun is invisible due to cloudsin sky, but the sun exists beyond the cloud layers. This situationgenerates different sky brightness according to the position of thesun in the sky, although direct sunlight is not available. It appearsthat the difference between measured and simulated illuminanceoccurred due to the different luminous distributions from the skysurface, which was influenced by the position of the sun.

In order to analyze the relationship between computationalresults from APPSM and field measurements under various skyconditions, linear regression analysis with analysis of variance(ANOVA) was used in this study. For the analysis, the measured and

Table 5Statistics of outdoor daylight illuminance.

Sky Day Outdoor illuminance Mean [Klx] Min. [Klx] Max. [Klx] Range [Klx] Mode [Klx] Std. deviation [Klx]

Clear Apr/3 Global 65.85 25.67 87.70 62.03 85.70 18.404Vertical 46.29 7.49 72.70 65.21 70.90 22.545

Dec/9 Global 31.20 1.75 49.95 48.20 47.75 15.627Vertical 53.72 1.42 88.20 86.78 70.90 26.235

Partly cloudy June/10 Global 58.52 31.56 105.7 74.14 72.1 12.870Vertical 24.44 9.62 45.29 35.67 22.34 8.606

Dec/6 Global 22.08 0.87 49.84 48.97 1.36 12.712Vertical 31.29 0.53 80.7 80.17 2.46 20.542

Overcast July/2 Global 6.67 0.23 25.53 25.30 3.37 5.197Vertical 3.01 0.11 12.78 12.67 1.52 2.353

Dec/5 Global 7.02 1.31 12.47 11.16 8.65 2.817Vertical 3.82 0.71 10.22 9.51 1.78 1.696

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 239

simulated signals predicted by APPSM were chosen as variables.Scatter plots implying the relationships between the measured andsimulated results are shown in Fig. 15. Statistical ANOVA test resultsfor the regression are summarized in Table 8.

Overall, the regression models were acceptable with a signifi-cance level of 0.01. Themodels also guaranteed an acceptable linearrelationship between the results from APPSM and the field mea-surements. The coefficient of determination (r2) ranged 0.4865 to0.9820. The regression models imply that very strong linear cor-relations exist between the measured and simulated results.

Except for the full shielding condition under an overcast sky, allcoefficients were greater than 0.6299. For instance, all coefficientswere greater than 0.9436 under clear sky conditions. In summary,this result indicates that there is a strong linear relationship be-tween the photosensor illuminances from field measurements andthe computer simulations by APPSM.

Frequency analysis was performed for the final step of validationof the results computed by APPSM under various daylight condi-tions. The percent differences between measured and simulatedphotosensor illuminances were calculated to examine the de-viations between them. The analysis results are summarized inTable 9.

0.0

0.2

0.4

0.6

0.8

1.0

1.2

8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00ime

Sky

Rat

io

T

Apr/3 June/10 Dec/5

Fig. 11. Variation of sky ratio.

Table 6Statistics of sky ratio.

Day Sky Mean Min. Max. Range Mode Std. deviation

Apr/3 Clear 0.16 0.12 0.30 0.18 0.12 0.042June/10 Partly cloudy 0.52 0.24 0.84 0.60 0.39 0.181Dec/5 Overcast 0.92 0.61 1.00 0.39 0.93 0.036

For clear sky conditions, 97.5% of the percent differences werewithin 10% of the difference range. This implies that the potentialdeviation between measured and simulated photosensor illumi-nances was less than 10%. For partly cloudy sky conditions, thepercent difference diminished slightly, but 91.74% of cases fellwithin 10% of the percent difference range. However, the percentdifferences under overcast sky conditions showed wide rangescompared to those under clear and partly cloudy sky conditions.

This result implies that the deviation ranges between measuredand simulated photosensor illuminance levels were most narrow

Fig. 12. Photosensor illuminance (Clear sky, Noblind). (a) Photosensor illuminance(April/3) (b) Photosensor illuminance (Dec/9).

Fig. 13. Photosensor illuminance (Partly cloudy sky, Noblind). (a) Photosensor illumi-nance (June/10) (b) Photosensor illuminance (Dec./6).

Fig. 14. Photosensor illuminance (Overcast sky, Noblind). (a) Photosensor illuminance(July/2) (b) Photosensor illuminance (Dec./5).

Y. Yoon et al. / Building and Environment 89 (2015) 229e243240

for clear sky and partly cloudy sky conditions. This directly signifiesthat the simulated illuminance predicted by APPSM was effectivelyconsistent with the results of the field measurements. The simu-lated photosensor illuminance levels were not fully consistent withmeasured illuminance under all sky conditions, but still providedreliable results.

5. Conclusions

This study was performed in order to examine the accuracy ofthe computational algorithms of APPSM for a photosensor-basedlighting control system. A series of computer simulations wereconducted and field measurements were carried out under a vari-ety of daylight conditions to validate simulation results effectively.The summary of findings is as follows:

1. The photosensor signals predicted by PSENS and APPSM showeda very strong linear correlation under all sky conditions. Underclear sky conditions, the difference range between them showedthe greatest value for the no-blind case and decreased effec-tively when direct sunlight was controlled by shading devices.In particular, the difference was effectively reduced in the caseof north-facing conditions and overcast sky conditions, underwhich the direct component of daylight from the sun was notstrong in the model space.

2. The prediction results computed by APPSM were generallyconsistent with the results of field measurements, althoughslight differences between them existed according to daylightconditions. The differences between measured and simulatedilluminances were greater when the incident angle of the sun-light was lower in winter, and the reflected components ofdirect sunlight were efficiently detected by photosensors inspace.

3. The difference in photosensor signals between prediction andmeasurement effectively decreased as shielding conditionswere applied to the photosensors. The difference was not sig-nificant when direct sunlight was not available in the modelspace. Under overcast sky conditions, the differences in thepredicted and measured photosensor signals were not greatercompared to the cases of clear and partly cloudy sky conditions.

4. Linear regression analysis indicates that a strong relationshipexists between the photosensor signals obtained from predic-tion by APPSM and the field measurements. The predictionmodels between them for three photosensor shielding condi-tions were acceptable with a significance level of 0.01. Also, themajority of percent differences between measured and simu-lated photosensor signals were within 10% under clear andpartly cloudy sky conditions. This implies that the predictionresults of APPSM are generally consistent with measurementresults and that the prediction still provides reliable results.

In this study, the prediction results by APPSM were comparedwith simulation and measurement results for limited daylight

Table 7Statistics of difference between measured and simulated photosensor illuminance (Measured illuminance e Simulated illuminance by APPSM).

Sky Day Photo sensor shielding Mean [lx] Min. [lx] Max. [lx] Range [lx] Mode [lx] Std. deviation [lx]

Clear Apr/3 No 51.8 �160.1 344.0 504.1 157.7 134.2Partial �31.7 �303.1 103.1 406.2 �23.3 102.7Full �19.2 �141.9 55.1 197.0 �27.4 41.7

Dec/9 No 672.3 �406.4 2307.8 2714.2 �406.4 592.5Partial 115.3 �666.6 578.4 1245.0 �666.6 307.5Full �102.6 �584.3 122.4 706.7 �584.3 150.5

Partly cloudy June/10 No �67.6 �190.3 102.5 292.8 �108.8 77.6Partial �32.2 �145.9 70.3 216.2 �29.4 55.4Full �29.6 �82.7 31.4 114.1 �24.1 25.0

Dec/6 No 258.9 �459.5 1161.2 1620.7 �101.3 404.0Partial 49.5 �434.1 452.6 886.7 333.1 215.4Full �71.7 �244.2 91.4 335.6 56.3 95.2

Over-cast July/2 No 21.8 �49.0 125.1 174.1 20.5 29.3Partial 5.9 �39.5 53.2 92.7 �1.4 17.1Full 2.2 �32.1 21.9 54.0 �11.3 12.4

Dec/5 No �180.1 �519.9 16.1 536.0 �250.5 139.2Partial �134.2 �354.4 9.5 363.9 �354.4 104.7Full �84.0 �252.3 15.1 267.4 �41.9 65.5

Fig. 15. Relationship between measured and simulated illuminance. (a) Outdoor global illuminance (b) Unshielded photosensor (c) Partly-shielded photosensor (d) Fully-shieldedphotosensor.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 241

conditions due to research limitations. There is an increase in erroras the solar altitude angle lowers in December, especially for theovercast sky condition in this study. Hence, it appears that theovercast sky model applying the Perez sky model based onmeasured direct and global irradiance values might not ideallyreplicate the actual overcast sky luminance distribution.

Field measurement was not conducted at the same location asthe simulation location, due to some limitations in research period,budget, and administration supports. Different sites may have

impact on the test results. However, the validation results of thisstudy are still effective, since the photosensor signals due todaylight are influenced by sun position and the amount of clouds aswell as their distribution across the sky.

In future studies, equal conditions for simulations and experi-ments would be useful. Also, further analysis of diverse daylightand building conditions under a variety of sky conditions would beuseful to examine the variances between field measurements andsimulation results. Although the prediction results were similar to

Table 8Relationship between illuminance by APPSM and measurement.

Condition Sky Statistics ANOVA

Variable Unstandardized coefficients t Sig.

B Std. error

Outdoor global Clear (Constant) 2840.68 454.84 6.25 0.00 F (372.1) ¼ 12012.87r2 ¼ 0.9969, Sig. ¼ 0.00Outdoor-Measured 0.904 0.01 109.64 0.00

Partly cloudy (Constant) 2238.17 211.39 10.59 0.00 F (372.1) ¼ 41149.76r2 ¼ 0.9922, Sig. ¼ 0.00Outdoor-Measured 0.980 0.00 202.85 0.00

Overcast (Constant) �106.73 67.08 �1.59 0.11 F (372.1) ¼ 20394.01r2 ¼ 0.9820, Sig. ¼ 0.00Outdoor-Measured 0.899 0.01 142.81 0.00

Noshield Clear (Constant) 242.54 21.08 11.51 0.00 F (372.1) ¼ 7448.94r2 ¼ 0.9523, Sig. ¼ 0.00Noshield-Measured 0.739 0.01 86.31 0.00

Partly cloudy (Constant) 325.96 20.27 16.08 0.00 F (372.1) ¼ 3955.22r2 ¼ 0.9613, Sig. ¼ 0.00Noshield-Measured 0.738 0.01 62.89 0.00

Overcast (Constant) 16.74 11.07 1.51 0.13 F (372.1) ¼ 990.11r2 ¼ 0.9263, Sig. ¼ 0.00Noshield-Measured 1.318 0.04 31.47 0.00

Partial shield Clear (Constant) 43.22 15.68 2.76 0.01 F (372.1) ¼ 6340.04r2 ¼ 0.9436, Sig. ¼ 0.00Partial shield-Measured 0.921 0.01 78.99 0.00

Partly cloudy (Constant) 142.16 14.96 9.50 0.00 F (372.1) ¼ 3032.56r2 ¼ 0.9504, Sig. ¼ 0.00Partial shield-Measured 0.863 0.02 55.07 0.00

Overcast (Constant) 25.81 7.84 3.29 0.00 F (372.1) ¼ 634.99r2 ¼ 0.6299, Sig. ¼ 0.00Partial shield-Measured 1.361 0.05 25.20 0.00

Full shield Clear (Constant) �33.16 6.73 �4.93 0.00 F (372.1) ¼ 11442.43r2 ¼ 0.9684, Sig. ¼ 0.00Full shield-Measured 1.126 0.01 106.97 0.00

Partly cloudy (Constant) 30.09 6.37 4.73 0.00 F (372.1) ¼ 5564.40r2 ¼ 0.9720, Sig. ¼ 0.00Full shield-Measured 1.097 0.01 74.59 0.00

Overcast (Constant) 18.05 4.88 3.70 0.00 F (372.1) ¼ 353.46r2 ¼ 4865, Sig. ¼ 0.00Full shield-Measured 1.477 0.08 18.80 0.00

Table 9Percent difference between illuminance by APPSM and measurement (unit: %).

Percent differencerange [%]

Clear sky Partly cloudy Overcast sky

No shield Partial shield Full shield Out door No shield Partial shield Full shield Out door No shield Partial shield Full shield Out door

0e5 66.06 75.84 75.54 95.72 67.58 67.58 58.41 99.39 38.84 36.39 15.60 68.5010 31.50 21.71 21.10 3.36 24.16 22.63 27.22 0.61 22.63 21.41 19.27 22.3215 1.83 2.14 2.14 0.00 3.67 3.36 8.26 16.82 16.82 20.49 8.8720 0.61 0.31 0.61 0.92 2.45 2.45 1.53 7.65 7.65 15.29 0.3125 0.61 1.22 1.53 1.83 7.95 5.50 8.5630 0.61 1.53 1.53 5.50 4.28 8.26>30 0.31 0.92 1.22 0.61 7.95 12.54Total 100 100 100 100 100 100 100 100 100 100 100 100

Y. Yoon et al. / Building and Environment 89 (2015) 229e243242

the field measurements, further study would be necessary todevelop more reliable prediction tools that reflect actualconditions.

Acknowledgments

This research was supported by the Basic Science ResearchProgram through the National Research Foundation of Korea (NRF)funded by the Ministry of Science, ICT and Future Planning (NRF-2014R1A2A1A11051162).

References

[1] Li DHW, Lam TNT, Wong SL. Lighting and energy performance for an officeusing high frequency dimming controls. Energy Convers Manag 2006;47:1133e45.

[2] Lee E, Selkowitz S. The New York Times Headquarters daylighting mockup:monitored performance of the daylighting control system. Energy Build2006;38:914e29.

[3] Onaygıl S, Güler €O. Determination of the energy saving by daylight responsivelighting control systems with an example from Istanbul. Build Environ2003;38:973e7.

[4] Kim S, Al-zoubi H, Ihm P. Determining photosensor settings for optimumenergy saving of suspended lighting systems in a small double-skinned office.Int J Energy Res 2009;33:618e30.

[5] Roche L. Summertime performance of an automated lighting and blindscontrol system. Light Res 2002;34:11e27.

[6] Yang I, Nam E. Economic analysis of the daylight-linked lighting control sys-tem in office buildings. Sol Energy 2010;84:1513e25.

[7] Li D, Lam J. Evaluation of lighting performance in office buildings withdaylighting control. Energy Build 2001;33:793e803.

[8] Escuyer S, Fontoynont M. A field study of office workers' reactions. Light ResTechnol 2001;33:77e99.

[9] Boyce P, Veitch J, Newsham G, Jones C, Heerwagen J, Myer M, et al. Occupantuse of switching and dimming controls in offices. Light Res Technol 2006;38:358e78.

[10] Vine E, Lee E, Clear R, DiBartolomer D, Selkowitz S. Office worker responses toan automated Venetian blind and electric lighting system: a pilot study. En-ergy Build 1998;28:205e18.

[11] Foster M, Oreszczyn T. Occupant control of passive systems: the use ofVenetian blinds. Build Environ 2002;36:149e55.

[12] Begemann SHA, van den Bled G, Tenner A. Daylight, artificial light and peoplein an office environment, overview of visual and biological responses. Int J IndErgonom 1997;20:231e9.

[13] Boubekri M, Hulliv R, Boyer R. Impact of window size and sunlight penetra-tion of office workers' mood and satisfaction e a novel way of assessingsunlight. Environ Behav 1991;23:474e93.

[14] Inoue T, Kawase T, Ibamoto T, Matsuo Y. The development of an optimalcontrol system for window shading devices based on investigations in officebuildings. ASHRAE Trans 1998;94:1034e49.

[15] Boyce P, Eklund N, Simpson S. Individual lighting control: task performance,mood, and illuminance. J Illum Eng Soc N Am 2000;29:131e42.

[16] Bierman A, Conway KM. Characterizing daylight photosensor system perfor-mance to help overcome market barriers. J Illum Eng Soc 2000;29(1):101e15.

Y. Yoon et al. / Building and Environment 89 (2015) 229e243 243

[17] Newsham GR, Mancini S, Marchand RG. Detection and acceptance of demandresponsive lighting in offices with and without daylight. Leukos 2010;4(3):130e56.

[18] Rude D. Why do daylight harvesting projects succeed or fail? Con-structionSpecif 2006;59(9):108.

[19] Love J. Field performance of daylighting system with photoelectric controls.In: Proceedings of the IESNA annual conference, Houston, TX; 1993.

[20] Choi A, Song K, Kim Y. The characteristics of photosensors and electronicdimming ballasts in daylight responsive dimming systems. Build Environ2005;40:39e50.

[21] Doulos L, Tsangrassoulis A, Topalis FV. Reviewing the role of photosensors inlighting control system. In: Proceedings of the 6th WSEAS internationalconference on application of electric engineering, Istanbul, Turkey; 2007.

[22] Doulos L, Tsangrassoulis A, Topalis FV. Optimizing the position and the field ofview of photosensors in daylight responsive systems, Lux Europa. In: The 11thEuropean lighting conference, Istanbul; 2009.

[23] Mistrick R, Chen C, Bierman A, Felts D. A comparison of photosensor-controlled electronic dimming system performance in a small office. J IllumEng Soc 2000;29:66e80.

[24] Rubinstein F, Ward G, Verderber R. Improving the performance of photo-electrically controlled lighting systems. J Illum Eng Soc 1989;18(1):70e94.

[25] Ward G, Mistrick R, Lee E, McNeil A, Jonsson J. Simulating the daylight per-formance of complex fenestration systems using bidirectional scattering dis-tribution functions within Radiance. Leukos 2011;7(4):241e61.

[26] Kim S, Song K. Determining photosensor conditions of a daylight dimmingcontrol system using different double-skin envelope configuration. IndoorBuilt Environ 2007;16:411e25.

[27] Choi A, Sung M. Development of a daylight responsive dimming system andpreliminary evaluation of system performance. Build Environ 2000;35:663e76.

[28] Kim S, Kim J. The impact of daylight fluctuation on a daylight dimming controlsystem in a small office. Energy Build 2007;39:935e44.

[29] RPI. Photosensors. LRC Specifier Report. Troy, NY: Rensselaer PolytechnicInstitute; 2007.

[30] Kim S, Mistrick R. Recommended daylight conditions for photosensor systemcalibration in a small office. J Illum Eng Soc N Am 2001;30:176e88.

[31] Chen C. An analysis of automated daylight dimming system performanceunder optimal calibration conditions. Master thesis. University Park, PA, USA:The Pennsylvania State University; 1999.

[32] Clevenger C, Rogers Z. Project 6.2 program technology and product designtools. SPOT Development Report. Boulder: Architectural Energy Corporation;2004.

[33] Mardaljevic J. Simulation of annual daylight profiles for internal illuminance.Light Res Technol 2000;32(2):111e8.

[34] Reinhart C, Andersen M. Dynamic RADIANCE-based daylight simulations for afull-scale test office with outer Venetian blinds. Energy Build 2001;33(7):683e97.

[35] Reinhart C, Walkenhorst O. Development and validation of a radiance modelfor a translucent panel. Energy Build 2006;38(7):890e904.

[36] Bourgeois D, Reinhart C, Ward G. A standard daylight coefficient model fordynamic daylighting simulations. Build Res Inform 2008;36(1):68e82.

[37] Mistrick R, Casey C. Performance modeling of daylight integrated photosensorcontrolled lighting systems. In: Proceedings of the 2011 winter simulation,Phoenix; 2011. p. 903e14.

[38] Reinhart C, Andersen M. Dynamic RADIANCE-based daylight simulations for afull-scale test office with outer Venetian blinds. Energy Build 2001;33(7):683e97.

[39] Enrich Charles E, Papamichale K, Lai J, Revzan K. A method for simulating theperformance of photosensor-based lighting controls. Energy Build2002;34(9):883e9.

[40] Yoon Y. Development of a fast and accurate annual daylight approach forcomplex window systems. Ph.D. Dissertation. University Park, PA, USA: ThePennsylvania State University; 2006.

[41] Tregenza P. Subdivision of the sky hemisphere for luminance. Light ResTechnol 1987;19:13e4.

[42] Reinhart C, Walkenhorst O. Validation of dynamic RADIANCE-based daylightsimulations for a test office with external blinds. Energy Build 2001;32:683e97.

[43] Reinhart C, Herkel S. The simulation of annual daylight illuminance distri-butions e a state-of-the-art comparison of six RADIANCE-based methods.Energy Build 2000;32(2):161e87.

[44] Perez R, Seals R, Michalsky J. All-weather model for sky luminancedistribution-preliminary configuration and validation. Sol Energy 1993;50:235e45.

[45] Perez R, Ineichen P, Seals R, Michalsky J, Stewart R. Modeling daylight avail-ability and irradiance components from direct and global irradiance. Sol En-ergy 1990;44(5):271e89.

[46] Marion W, Urban K. User's manual for TMY2s. NREL; 1995. p. 20 (Table 3-2).[47] LI-COR, Inc. LI-COR sensor instruction manual. 1991.[48] Kim Y-M, Lee J-H, Kim S-M, Kim S. Effects of double skin envelopes on natural

ventilation and heating loads in office buildings. Energy Build 2011;43:2118e26.

[49] Kim S, Kim J-J. Influence of light fluctuation on occupants' visual perception.Build Environ 2007;42:2888e99.

[50] Campbell Scientific, Inc. CR23X micro Logger operator's manual. 2000.[51] Rea M. The lighting handbook. 10th ed. New York: Illuminating Engineering

Society of North America; 2011.[52] Pattanasethanon S, Lertsatitthanakorn C, Atthajariyakul S, Soponronnarit S. All

sky modeling daylight availability and illuminance/irradiance on orizontalplane for Mahasarakham, Thailand. Energy Convers Manag 2007;48:1601e14.