People, Systems, Environment: Exploring the patterns and impact of control-oriented occupant actions in buildings

Ardeshir Mahdavi
Department of Building Physics and Building Ecology Vienna University of Technology, Vienna, Austria

ABSTRACT: In most buildings, people operate control devices such as windows, shades, luminaires, radiators, and fans to bring about desirable indoor environmental conditions. These control actions have a significant yet often neglected impact on buildings' performance and sustainability (energy use, indoor climate). A better understanding of the logic and patterns of such control-oriented user behavior can not only facilitate better predictions of buildings' performance, but also support the effective operation of buildings' service systems. This paper describes an extensive empirical study of control-oriented user behavior (as related to systems for lighting, shading, ventilation, and heating) in a number of office buildings in Austria. The results specifically shed light on the relationships between control actions and environmental conditions inside and outside buildings. Keywords: building systems, controls, energy, indoor climate, people, behavior

1. INTRODUCTION
Multiple studies have been (and are being) conducted internationally to collect data on building users interactions with building control systems and devices (e.g., [1], [2], [3], and [4]). Such data can bring about a better understanding of the nature, type and frequency of control-oriented user behavior in buildings and thus support the development of corresponding behavioral models for integration in building performance simulation applications. Moreover, such data could support the effective (and proactive) operation of building service systems for indoor environmental control. The present contribution describes an effort to observe control-oriented occupant behavior in 42 offices in two office buildings over a period of one year and 6 offices in the third office building over a period of nine months. Specifically, states and events pertaining to occupancy, systems, indoor environment, and external environment were monitored. Weather stations, a number of indoor data loggers, and digital cameras were used to continuously monitor and record every five minutes such events and states (occupancy, indoor and outdoor temperature and relative humidity, internal illuminance, external air velocity and global irradiance, status of electrical light fixtures, position of shades). The results reveal distinct patterns in the collected data. Specifically, control lighting and shading behavior tendencies show dependencies both on indoor and outdoor environmental parameter. A summary of these tendencies are presented and their principal potential as the basis of empirically grounded user action models are explored.

2. METHOD
2.1 Objects Data collection was conducted in three office buildings in Austria. One of these is an educational (university) building. We refer to this building henceforth as "FH" (see Figure 1). The second building is a large high-rise office complex, referred to, in this paper, as "VC" (see Figure 2). An important feature of VC is its use as one of the major seats of international organizations, resulting in a very diverse occupancy profile in cultural terms. The third office building is used by a governmental organization. It is referred in present paper as "HB" (see Figure 3). We selected 13 scientific staff offices in FH, 29 singleoccupancy offices in VC, and 6 offices in HB. All th selected offices in FH face east, situated on the 4 , th th 5 and 6 floors. Ten offices are single-occupancy, two are double-occupancy, and one is tripleoccupancy. In case of VC, 15 offices face north (code: "VC_NO") and 14 face south-west (code: th "VC_SW"). The offices are located on the 12 and th 13 floor of the building. In HB, two offices are singleoccupancy and four offices are double occupancy. st Three offices are located on the 1 floor and three on nd the 2 floor. All selected offices in HB face northeast. To exemplify the layout of the offices in these buildings, Figures 4 to 6 provide corresponding schematic plans (two single-occupancy and one th double-occupancy offices on the 5 floor of FH, three th single occupancy offices on the 12 floor of VC_SW, and one single-occupancy office and two doublend occupancy offices on the 2 floor of HB). The workstations are mostly equipped with desktop computers and in some cases task lights. Both VDTbased and paper-based tasks are performed.

Figure 4: Schematic plan of sample offices in FH Figure 1: External view of FH

Figure 5: Schematic plan of sample offices in VC_SW

Figure 2: External view of VC

Figure 6: Schematic plan of sample offices in HB

Figure 3: External view of HB

In case of HB, the offices are equipped with two rows of luminaires with 4 or 6 (58 W) fluorescent lamps divided into two circuits and manually controlled by two switches near the entrance door, external manually operated shading elements and internal curtains, two or three operable windows, and two or three radiator units positioned underneath windows. 2.2 MONITORED PARAMETERS The intention was to observe user control actions pertaining to lighting and shading systems while considering the indoor and outdoor environmental conditions under which those actions occurred. Occupancy and the change in the status of ambient light fixtures were captured using a dedicated sensor. Shading was monitored via time-lapse digital photography: The degree of shade deployment for each office was derived based on regularly taken digital photographs of the faade. Shade deployment degree was expressed in percentage terms (0% denotes no shades deployed, whereas 100% denotes full shading).

The offices in FH are equipped with the followings environmental control systems: Three/four luminaires (58W each), divided into two circuits manually controlled via switches near the office door; External motorized screen shades operated by a switch mounted on a panel under the window; Fan coil under the window for fine adjustment of temperature. The systems installed in the offices of VC are as follows: Three rows of luminaires with 9 or 12 fluorescent lamps (36 W) divided into two circuits and manually controlled by two switches near the entrance door, internal manually operated shading, and three to four fan coil units (located below windows) for fine adjustment of temperature.

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

Mean occupancy [%]

The external weather conditions were monitored using two three weather stations, mounted on the top of the building in case of VC and HB, and in case of FH, on the rooftop of a close-by university building. Internal climate conditions (temperature, relative humidity, illuminance) were measured with autarkic loggers distributed across the workstations. To obtain information regarding user presence and absence intervals, occupancy sensors were applied, which simultaneously monitored the state of the luminaires in the offices. All of the above parameters were logged regularly every 5 minutes. Monitored indoor parameters o included room air temperature (in C), room air relative humidity (in %), ambient illuminance level at the workstation (in lx), luminaires status (on/off), and occupancy (present/absent). Monitored outdoor environmental parameters included air temperature, -1 relative humidity, wind speed (in m.s ) and wind direction, as well as horizontal global illuminance and -2 horizontal global irradiance (in W.m ). Vertical global irradiance incident on the faade was computationally derived based on measured horizontal global irradiance (cp. [5]). Collected data were stored and processed in a data base (see the structure in Table 1) in terms of events (E) and states (S) for further analysis. Events are either system-related (Es) or occupancy-related (Eo). States can refer to systems (Ss), indoor environment (Si), outdoor environment (Se), and occupancy (So). For the purposes of the present analysis the range of data considered was limited to working days between the hours 8:00 to 20:00 in case of FH and VC and between the hours 6:00 to 18:00 in case of HB. The collected data was mainly analyzed to explore hypothesized relationships between the nature and frequency of the control actions on one side and the magnitude and dynamism of indoor and outdoor environmental changes on the other side.

3. RESULTS
3.1 Occupancy Figure 7 shows the mean occupancy level in FH, VC, and HB over the course of a reference day (averaged over the entire observation period). Note that these values represent the presence in/at the users offices/workstations, not merely the presence in the building. Moreover, as Figure 8 demonstrates, the occupancy patterns can vary considerably from office to office.
80 70 60

Mean occupancy [%]

50 40 30 20 10 0 06:00

08:00

10:00

12:00

14:00

16:00

18:00

20:00

Time

FH

VC

HB

Figure 7: Mean occupancy level for a reference day in FH, VC, and HB.
80 70 60 50 40 30 20 10

Table 1: The structure of the collected data DATA Events (E) TYPE Systemrelated (Es) OccupancyRelated (Eo) States (S) Systemrelated (Ss) Indoor environ. (Si) Outdoor environ. (Se) OccupancyRelated (So) INSTANCES Switch lights on/off Pull shades up/down Entering into (or leaving) an office Lights on /off Position of shades/windows Air temperature Illuminance level Outdoor temperature Global irradiance Office/workstation occupied/vacant

0 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00

Time

Figure 8: Observed occupancy levels in 7 different offices in FH for a reference day. 3.2 Lighting Figure 9 shows the observed effective lighting operation in the course of a reference day expressed in terms of effective electrical power. The information in this Figure concerns the general light usage in all observed offices. Figure 10 shows the probability that an occupant would switch the lights on upon arrival in his/her office as a function of the prevailing task illuminance level immediately before arrival. Figure 11 shows the normalized relative frequency of (intermediate) actions "switching the lights on" (by occupants who have been in their office for about 15 minutes before and after the occurrence of the action) as a function of the prevailing task illuminance level immediately prior to the action's occurrence. Normalization denotes in this context that the actions

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

are related to both occupancy and the duration of the time in which the relevant illuminance ranges (bins) applied. Figure 12 shows the normalized relative frequency of all "switching the lights on" actions (upon arrival and intermediate) as a function of the time of the day. In this case too, actions are normalized with regard to occupancy. Note that Figure 12 includes also the corresponding mean global horizontal irradiance levels. Figure 13 shows the probability that an occupant would switch off the lights upon leaving his/her office as a function of the time that passes before he/she returns to the office. Figure 14 shows the normalized frequency of the (intermediate) "switching the lights off" actions as a function of the prevailing illuminance level immediately prior to the action's occurrence. Normalization denotes in this case the consideration of occupancy and the applicable durations of the respective illuminance bins while deriving the actions' frequency.

25

20
Relative frequency [%]

15

10

0
49 9 099 99 9 90 029 9 59 9 79 9 69 9 19 9 39 9 89 9 10 020 050 030 040 060 070 080 010 00
400 300 200 100 0

E [lx]

FH

VC

Figure 11: Normalized relative frequency of intermediate switching the lights on actions in FH and VC as a function of the prevailing task illuminance level.

60

Relative frequency [%]

8 7 6

40

30

Mean load [W.m -2]

5 4 3 2 1 0 06:00

20

10

0
08:00- 09:00- 10:00- 11:00- 12:00- 13:00- 14:00- 15:00- 16:00- 17:00- 18:0008:59 09:59 10:59 11:59 12:59 13:59 14:59 15:59 16:59 17:59 19:00

Time
08:00 10:00 12:00 14:00 16:00 18:00 20:00

FH

VC

BH

Mean global irradiance

Time

FH

VC

HB

Figure 9: Lighting operation in FH, VC, and HB offices over the course of a reference day.

Figure 12: Normalized relative frequency of switching the lights on actions in FH, VC, and HB together with mean horizontal global irradiance over the course of a reference day.

100
35 30

90 80 70
Probability [%]

25

60 50 40 30 20 10

Probability [%]

20 15 10 5 0 0-99 100199 200299

FH

0
0-14 1529
300399 400499 500599
VC

3044

4559

6074

7589

90- 105- 120- 135- 150- 165- 180 104 119 134 149 164 179
FH VC HB

600699

700799

800899
HB

900999

1000

Duration of absence [min]

E [lx]

Figure 10: Probability of switching the lights on upon arrival in the office in FH, VC, and HB as a function of the prevailing task illuminance level prior to the action's occurrence.

Figure 13: Probability of switching the lights off as a function of the duration of absence (in minutes) from the offices in FH, VC, and HB.

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

-2

50

Mean global irradiance [W.m ]

500

45 40

Relative frequency [%]

35 30 25 20 15 10 5 0

occupancy and the duration of times in which the prevailing irradiance was within a certain range (bin). Note that in this case the definition of opening/closing actions is not limited to actions resulting in fully opening/closing the shades. Rather, it denotes a relative occupant-driven change in the position of the shades. This means that even an incremental change (e.g. changing from 80% to 40% or changing from 20% to 40%) is considered an opening/closing action.
90 80
69 9 39 9 49 9 59 9 79 9 89 9 99 9

Mean shade deployment [%]

60 0-

30 0-

40 0-

50 0-

70 0-

80 0-

90 0-

70 60 50 40 30 20 10 0 JAN FEB MAR APR MAY

FH

E [lx]

FH

VC

Figure 14: Normalized frequency of intermediate switching the lights off actions in FH and VC offices as a function of the prevailing task illuminance level. 3.3 Shades Figures 15 and 16 show the mean monthly shade deployment degree for FH, HB, VC_NO, and VC_SW.
60

JUN

JUL

AUG SEP OCT NOV DEC

VC_SW

VC_NO

Mean shade deployment [%] .

50

Figure 17: Mean monthly shade deployment degree in FH, VC_NO, and VC_SW.
25

40

30

20

Relative frequency [%]

20

15

10

10

0
25 0 15 0 40 0 10 0 35 0 20 0 30 0 45 0 50 0 50

Global horizontal irradiance [W.m-2]

FH

HB

Figure 15: Mean shade deployment degree in FH and HB as a function of global horizontal irradiance.
80 70

49

14 9

19 9

24 9

49 9 49 9

34 9

29 9

0-

50

10 0-

15 0-

20 0-

39 9

30 0-

Global vertical irradiance [W.m ]

Mean shade deployment [%]
60 50 40 30 20 10 0
49 50 -9 9 49 99 99 49 035 03 40 04 10 01 15 01 20 02 25 02 30 03 45 04 50 0 49 49 99 99

-2

FH

40 0-

35 0-

45 0-

25 0-

44 9

Figure 18: Normalized relative frequency of opening shades in FH as a function of the global vertical irradiance and VC.
25

Relative frequency [%]

20

15

Global vertical irradiance [W.m ]

-2

VC_NO

VC_SW

10

Figure 16: Mean shade deployment degree as function of global vertical irradiance in VC_SW and VC_NO. Figure 17 shows, for FH, VC_NO, and VC_SW the mean monthly shade deployment degree over a year. Figures 18 and 19 show the normalized relative frequency of the actions "opening shades" and "closing shades" as a function of global vertical irradiance incident on the faade for FH and VC. Normalization means that the frequency of actions (opening and closing shades) is related here to both

49

14 9

0-

50

29 9

34 9

35 0-

40 0-

10 0-

15 0-

39 9

20 0-

Global vertical irradiance [W.m ]

-2

30 0-

25 0-

FH

45 0-

44 9

19 9

24 9

Figure 19: Normalized relative frequency of closing shades actions in FH and VC as a function of the global vertical irradiance.

10

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

50 0

-9

50 0

-9

VC

VC

4. DISCUSSION
4.1 General observations The monitored occupancy in FH and VC (Figure 7) and the obviously related lighting loads (see Figure 9) reveal a similar pattern for FH and VC (also known from other office buildings). However, the monitored occupancy in HB (Figure 7) and the corresponding people and lighting loads (Figure 9) reveal patterns that deviate from typical schedule assumptions for office buildings. Moreover, the maximum occupancy levels are noticeably lower in FH. This may be due to the circumstance that FH houses offices for teaching and research staff, who spend a considerable amount of time in classrooms and laboratories. The relationship between "switching on the lights" actions and the prevailing illuminance levels in the monitored buildings (see Figures 10 and 11) suggests that only illuminance levels below 100 lx were likely to trigger actions at a non-random rate. As to the action "switching the lights off", a clear relationship to the subsequent duration of absence is evident (see Figure 13). Occupants do switch off the lights more frequently if they are going to be away from the offices for longer periods. On the other hand, lights are not necessarily switched off by the occupants if the illuminance level in the office is already sufficient (or more than necessary) for performing typical tasks. In fact, such intermediate switching off actions appear to occur at a noticeably higher rate only once the illuminance level in the office rises above 1000 lx (see Figure 14). The mean shade deployment levels differ from building to building and faade to faade (see Figures 15 to 17). In case of FH, where we studied the eastfacing faade, a difference in the level of shade deployment can be seen between the high-radiation summer months and the low-radiation winter months (Figure 17). Moreover, an evident relationship between shade deployment and the magnitude of solar radiation is observable in FH and HB data (Figure 15). This relationship provides a very effective basis for modeling the state of shades (see, as an example, Figure 24). In case of VC_SW and VC_NO the shade deployment level does not vary much in the course of the year or in terms of vertical irradiance classes (see Figure 16), but there is a significant difference in the overall shade deployment level between these two facades (approximately 75% in case of the southwest-facing faade, 10% in case of the north-facing faade). The relative small variation range in the monthly shade deployment levels in VC_SW and VC_NO may be partly due to the fact that the manual shade operation mechanism is, in this case, much more difficult to handle than the mechanically supported shade operation system in FH. Our observations did not reveal a clear relationship between "opening shades" actions and the incident radiation on the faade (see Figure 18). However, the corresponding analysis of the "closing shades" actions shows for FH a higher action frequency once the incident radiation rises above 200 -2 W.m (see Figure 19).

4.2 Preliminary Models Our observations underscore the need for typologically differentiated occupancy models for different buildings. Patterns obtained from one building cannot be transposed to other buildings without extensive calibration measures considering differences in buildings' use (function), size, context (physical, climatic, cultural), orientation, envelope, systems, etc. Nonetheless, efforts are justified to apply the collected data to date toward the generation of preliminary models of user presence and behavior. As the data presented in this paper result from actual long-term observations and high-resolution measurements in typical office buildings, they can be justifiably assumed to be much more representative than (and thus superior to) most currently applied simulation input assumptions. Figures 20 to 25 provide examples of such preliminary models for office buildings in Austria: Figure 20 illustrates a simulation input data model of the mean hourly occupancy levels (based on data from FH and VC). Figure 21 contains a simulation input data model of mean hourly people (sensible) load (based on data from FH and VC).

60

50

Mean occupancy [%]

40

30

20

10

0 06:00

08:00

10:00

12:00

14:00

16:00

18:00

20:00

Time

FH+VC

Figure 20: Illustrative simulation input data model of mean hourly occupancy levels in office buildings in Austria (based on data from FH and VC).
4

Mean occupancy load [W.m -2]

0 06:00

08:00

10:00

12:00

14:00

16:00

18:00
FH+VC

20:00

Time

Figure 21: Illustrative simulation input data model of the mean hourly people load (sensible) for office buildings in Austria (based on data from FH and VC).

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

11

Figure 22 shows an Illustrative simulation input data model for mean hourly lighting load (based on FH and VC data). Figure 23 provides a probability model for switching the lights off as a function of the duration of absence from the offices (based on data collected in FH, VC, and HB). Figure 24 includes a shade deployment model as a function of incident irradiance on a faade (based on FH data).
5

Finally, Figure 25 illustrates a simulation input data model for normalized relative frequency of closing shades actions as a function of the global vertical irradiance (based on data collected in FH and VC). More representative people presence and actions models can improve the accuracy of simulation studies (for example, to explore the impact of thermal improvement measures on the building's energy use).

20

Relative frequency [%]

15

Mean load [W.m -2]

10 y = 1,9042x - 1,0491 R2 = 0,9496 5

1
0

49

34 9

39 9

19 9

24 9

29 9

50 -9

0-

20 0-

25 0-

15 0-

30 0-

06:00

08:00

10:00

12:00

14:00

16:00

18:00
FH+VC

20:00

10 0-

Time

Global vertical irradiance [W.m ]

-2

Figure 22: Illustrative simulation input data for mean hourly lighting load (based on FH and VC data).
80 70 60 50 40 30 20 10 0
30 45 60 75 15 90 10 5 13 5 12 0 15 0 18 0 16 5

Figure 25: Illustrative simulation input data model for normalized relative frequency of closing shades actions as a function of the global vertical irradiance (based on data collected in FH and VC).

y = 4,8856x + 6,4315 R2 = 0,9499

Duration of absence [min]

Figure 23: Probability model for switching the lights off as a function of the duration of absence from the offices (based on data collected in FH, VC, and HB).
70 60 50 40 30 20 10
49 9 9 9 9 9 9 9 9 50 -9 9 29 39 34 44 19 24 49 00000000050 14 0

y = 2,8803x + 27,531 2 R = 0,9559

Global vertical irradiance [W/m ]

Figure 24: Illustrative model shade deployment models as a function of incident irradiance on a faade (based on FH data).

4.3 Energy issues On a general level, our observations regarding occupancy in the buildings we studied suggest that the environmental systems in a considerable number of office buildings may in fact be "over-designed", in a sense that they are dimensioned for occupancy levels that seldom occur. Figure 26 and Table 2 specifically illustrate the potential for reduction of electrical energy use for lighting in the sampled offices. Thereby, two energy saving scenarios are considered. The first scenario requires that the lights are automatically switched off after 10 minutes if the office is not occupied. The second scenario implies an automated dimming regime, whereby luminaires are dimmed down so as to maintain an illuminance level of 500 lx at the workstation while minimizing electrical energy use for lighting. The estimated saving potential in electrical energy use for lighting of the sampled offices is significant. The cumulative energy saving potential for all sampled offices is 69% for FH and 71% for VC and 66% in HB (Table 2). This translates (for VC) into a cumulative annual energy saving potential of 17 -2 kWh.m . Note that a lighting system retrofit and the resulting electrical energy use reduction would increase the heating loads and decrease the cooling loads. Given the magnitude of required cooling loads in office buildings, the overall thermal implications of a lighting retrofit are positive both in energetic and monetary terms.

Mean shade deployment [%]

Probability [%]

25

30

35

12

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

45

40

20

10

15

45 0-

35 0-

40 0-

49 9

14 9

44 9

80 70 FH VC HB

REFERENCES
[1] Hunt D. 1979. "The Use of Artificial Lighting in Relation to Daylight Levels and Occupancy", Bldg. Envir.14, 2133, 1979. [2] Mahdavi A, Lambeva L, Prglhf C, et. al. 2006. "Integration of control-oriented user behavior models in building information systems", Proceedings of the 6th European Conference on Product and Process Modelling (13-15 September 2006, Valencia, Spain): eWork and eBusiness in Architecture, Engineering and Construction. Taylor & Francis/Balkema. ISBN 10: 0-415-41622-1.pp.101107. [3] Newsham GR. 1994. "Manual Control of Window Blinds and Electric Lighting: Implications for Comfort and Energy Consumption", Indoor Environment 1994, 3: 13544. [4] Reinhart C. 2002. "LIGHTSWITCH-2002: A Model for Manual Control of Electric Lighting and Blinds", Solar Energy 2002, v.77 no. 1, 15-28. [5] Mahdavi A, Dervishi S, and Spasojevic B. 2006. "Computational derivation of incident irradiance on building facades based on measured global horizontal irradiance data", Proceedings of the Erste deutschsterreichische IBPSA-Konferenz - Munich, Germany (2006),123-125.

Saving potential [%]

60 50 40 30 20 10 0 1 2 1+2

Scenarios

Figure 26: Saving potential for two control scenarios (cp. Text) in view of electrical energy use for lighting in FH, VC, and HB Table 2: Saving potential (electrical energy for lighting) for various scenarios and buildings Energy saving scenarios Building Saving potential in HB FH VC % % % kWh.m . a
-2 -1

1 15 26 28 6.8

2 51 43 43 10.2

1+2 66 69 71 17.0

5. CONCLUSION
A study was presented concerning user control actions in three office buildings in Austria. The results suggest that it is possible to identify general patterns of user control behavior in buildings in terms of their relationship to indoor and outdoor environmental parameters such as illuminance and irradiance. The compound results of the ongoing case studies are expected to lead to the development of robust occupant behavior models that can improve the reliability of building performance simulation applications and enrich the control logic in building automation systems. Moreover, the obtained information will support the assessment of energy saving potential due to consideration of occupancy and behavioral patterns in office buildings.

ACKNOWLEDGEMENT
The research presented in this paper was supported in part by a grant from the program "Energiesysteme der Zukunft", "Bundesministerium fr Verkehr, Innovation und Technologie (BMVIT)"; project number: 808563-8846. The author would also like to gratefully acknowledge the contributions of his research team members (G. Suter, A. Mohammadi, L. Lambeva, E. Kabir, C. Prglhf, J. Lechleitner, and S. Dervishi) toward the collection and analysis of the data presented in this paper.

PLEA2007 The 24th Conference on Passive and Low Energy Architecture

13