- Email: [email protected]
Statistical simulation of user behaviour in low-energy office buildings
Solar Energy 81 (2007) 676–682 www.elsevier.com/locate/solener
Statistical simulation of user behaviour in low-energy office buildings J. Pfafferott *, S. Herkel Fraunhofer Institute for Solar Energy Systems, Heidenhofstr. 2, D-79110 Freiburg, Germany Received 23 August 2005; received in revised form 10 August 2006; accepted 15 August 2006 Available online 16 October 2006 Communicated by: Associate Editor Mattaios Santamouris
Abstract A large number of design guidelines and tools are available for the design of passive cooling systems. Using the underlying thermodynamic models, a certain input (e.g. air change rate, internal heat gains or sun control) results in a certain output (i.e. room temperature). However, in real buildings the room temperature at a given outdoor temperature is a distribution rather than a single value. Therefore, the building engineer should take uncertainties into account, since the actual use of the building, the building physical properties or the user behaviour are statistically distributed. One promising approach to include these uncertainties in the design procedure is the use of statistical models: the design parameter is defined by a mean value and its deviation. From a control theoretical point of view, the deterministic controlled system responds to random disturbance variables by a statistically distributed response function. Considering the institute building of Fraunhofer ISE as example, this study shows how statistical simulations can be applied to the design process of passive cooling in low-energy office buildings. 2006 Elsevier Ltd. All rights reserved.
1. Introduction In preparation for the statistical simulation, several investigations on detail phenomena concerning the thermal behaviour of the Fraunhofer ISE building (Fig. 1) were carried out to provide a reliable simulation model (Pfafferott, 2004). For this study, a validated building simulation model is used to investigate the impact of user behaviour on the thermal building performance. The influence of material properties and parameters in the air-flow network on the room temperature have been analysed by a sensitivity analysis in order to prepare and to evaluate the building simulation model. This model validation has been carried out using data sets from April and July 2002 (Pfafferott et al., 2003), while the statistical simulation with regard to the user behaviour uses a data set from the summer of *
Corresponding author. Tel.: +49 761 4588 5129. E-mail address: jens.pfaff[email protected] (J. Pfafferott). URL: www.ise.fraunhofer.de (J. Pfafferott).
0038-092X/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.solener.2006.08.011
2003. All simulations have been carried out using the simulation program ESP-r (Clarke, 2001). The functionality (and the programming) of Monte Carlo-simulations is briefly described by Macdonald and Strachan (2001). The statistical simulation is supposed to deal with uncertain input parameters and should show, whether typical occupancy patterns are effective for modelling the thermal building behaviour and how far statistical models can be used successfully to take uncertainties into account. The procedure can be described according to the control theory (Fig. 2). Excitation function. The input parameters are statistically analysed with regard to the use of the building (i.e. equipment and occupancy) and user behaviour (i.e. ventilation and sun control). Through this comprehensive data evaluation, statistical models for the simulation input are derived. Controlled system. Using these data series as input parameters, the thermal building performance is simulated: many parameter combinations are considered by the
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
677
Fig. 1. View of the Fraunhofer ISE building from west. The monitored offices are located in the middle building wing on the 2nd and 3rd floor with south orientation.
open control loop exication function user behaviour (probability function)
response function room temperature (statistical analysis)
controlled system building (validated simulation model)
from 25.9 C to 26.9 C, since the offices vary in the daily attendance, the office equipment (internal heat gains) and the user behaviour regarding sun protection (solar heat gains) and ventilation (heat losses). As the statistical models are deduced from a long-term monitoring, the following sections shortly describes the statistical data analysis for 16 offices, during the considered summer period.
Fig. 2. Methodological approach.
2.1. Variation in time of room temperatures
2. Data evaluation For the statistical data analysis and simulation, (1) the time period and (2) the sample of offices must be defined. For this data sample, the variation in time of room temperatures and the heat gains are statistically analysed by mean value and distribution. The user behaviour is analysed and modelled with special regard to the ventilation (Herkel et al., 2005). Time period. Starting from the hourly ambient air temperatures in 2003, a characteristic period can be defined. The period June 12 to July 23, 2003 is short enough to carry out many simulation runs in a row and long enough to cover different summer weather conditions. Similar ambient air temperatures and sun positions in this time period will produce likely consistent user behaviour. Office sample. Sixteen offices in the Fraunhofer ISE building are considered for the statistical data evaluation and simulation. Since all rooms are located on the first and second floor in the same part of the building, the climate impact from the outside is the same. However, the mean room temperatures (during the summer period) vary
The following procedure provides the variation in time of room temperatures and its statistical distribution: 1. Preparation of hourly room temperatures in each of the 16 offices. 2. Calculation of daily mean room temperature and daily temperature fluctuation in each office. 3. From this time series, mean room temperature, between 16% and 84% quantile are identified for each day. The minimum/maximum temperatures specify the limit of variation. Exemplarily for this procedure, Fig. 3 shows the time behaviour of the daily mean room temperature (in 16 offices) and its daily deviation/variation in all rooms: • The daily mean room temperature varies from 22.3 C to 28.8 C. • The mean deviation is 0.4 K and varies from 0.15 K to 0.9 K.
Mean room temperature [ºC]
simulation, and each is distributed around a true mean value. It is reasonable to assume that the parameters are normally distributed. As all input parameters are varied simultaneously, the room temperature is also distributed around a true mean value. Response function. The simulated room temperatures are statistically analysed and compared with the monitored room temperatures. For further information concerning the building, the HVAC system and its operation the reader is referred to the project documentation in SolarBau:Monitor (Internet platform, 2005) and to the cross section analysis on low-energy buildings by Voss et al. (2005).
30 28 26 24 22 20 0
7 14 21 28 35 Period of comparison [June 12 - July 23, 2003]
42
Fig. 3. Mean room temperatures with minimum/maximum (grey lines) and 1 r deviation (black dashes).
678
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
As the climatic and the building physical boundary conditions are (almost) identical in all rooms, the temperature variations result mostly from variations in use of the offices. 2.2. User behaviour regarding ventilation The following procedure provides the user behaviour regarding ventilation and its statistical distribution: 1. Preparation of the hourly status (open or closed) of door and louver (indoor) and the two upper and downer part of a two-segment window (outdoor) in each of the 16 offices. If a room has more than one window these openings will be combined by an OR-relation: if one of the window segments is open, the whole window is considered to be open. 2. The time series differentiates working days and weekends. 3. All data lines are sorted by the time of day. 4. From this information, relative frequencies of opening are calculated for each ventilation component and for each hour of the day. 5. Using the hourly data regarding the status of all openings in each of the 16 offices, the local distribution of hydraulic resistances (mean value and statistical distribution) is calculated using an air-flow network with resistances, parallel (16 offices) and in series (office-corridor). Fig. 4 shows the average user behaviour in 16 offices, during 30 working days: as expected, doors are closed outside business hours. More or less than 50% of all doors are opened, during the working hours. In general, windows are opened at arrival and are closed bit by bit, when the ambient air temperature is increasing. In the morning, more windows are turned open, while in the afternoon most open windows are tilted. At least 50% of all windows are open, during night. At least 80%/90% of all small windows above the window and louvers above the door are opened, during
Relative frequency of opening
100% 80% 60% 40% 20% 0% 00:00
large window small window louver, indoor door
04:00
08:00
12:00
16:00
20:00
00:00
Fig. 4. Time dependent user behaviour with regard to ventilation, during the working days. (Manual opening of ventilation components.) ‘‘Large window’’ refers to the downer and ‘‘small window’’ to the upper part of the two-segment windows. The small window and the louver above the door provide for a basic ventilation in conjunction with the exhaust air ventilator.
the whole day. These two ventilation flaps are manipulated very rarely and provide for a basic ventilation in the rooms. The exhaust fan delivers an air change rate of 1 h 1 (480 m3/h for 8 offices), during the working hours and 5 h 1 (2400 m3/h for 8 offices), during night ventilation in each floor. If the hydraulic resistance was the same in each of the offices, each office would get the same fraction of the air-volume flow, which is 1/8 or 12% of the total air-volume flow. Due to the changing distribution of open and closed flaps, doors and windows, the air-flow network (hydraulic resistances in series) constantly changes. Some offices get less than 4% and other rooms up to 18% of the air flow. 2.3. Variation in time of heat gains The following procedure provides occupancy patterns, the time variation of electricity consumption, the use/control of sun-shading and their statistical distributions: 1. The hourly heat gains due to persons are calculated from the signal of an ultrasonic sensor and correspond to 80 W/person. Missing or incorrect data are substituted by an average time profile, which is calculated from all available sensors. 2. The total electric power consumption [Wtotal] is hourly measured. Furthermore, the electricity consumption [kW hoffice/(m2 d)] in an office depends on the electric connection power and the estimated operation time of the equipment. Using the connection power and the current attendance for each office, the total electricity consumption can be divided into the electricity consumption in each office [Woffice]. 3. The hourly solar heat gains are calculated from the solar radiation on the window, which is calculated from meteorological data, taking into consideration shading from adjacent buildings and the facade as well as the position of the venetian blinds. The data acquisition is realised by a web camera. The error-prone data evaluation by image processing results in some data gaps, which are substituted depending on their length of time: if the status of the venetian blind is not recognised, the last valid value will be extrapolated. If the data gap covers several hours, the missed value will be substituted by the mean value of all known positions. 4. Through this process, hourly data for internal (persons and lighting/equipment) and solar heat gains are available for each office. The simulation deals with 16 different time profiles for heat gains. Fig. 5 outlines the daily heat gains in all of the 16 offices for 30 working days and its variation. The heat gains vary from one room to the next and can vary from one day to the next: • Due to the different application of computers, the daily heat gains from equipment [grand average: 137 W h/ (m2 d)] varies from 58 W h/(m2 d) to 295 W h/(m2 d).
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
equipment persons solar
300 200 100 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Office room (number)
Fig. 5. Daily heat gains and their 1 r deviations in 16 office rooms, of which 8 are in the 1st and 8 are in the 2nd floor.
• Since the rooms can be occupied by as few as one and as many as four persons and the mean attendance in the office is variably long (e.g. laboratory workers), the mean daily heat gains from persons [grand average: 54 W h/(m2 d)] varies greatly from 26 W h/(m2 d) to 103 W h/(m2 d). • In comparison, the variation in solar heat gains [grand average: 158 W h/(m2 d)] is relatively small: The room, in which the occupants close the venetian blinds most frequently, gains meanly 106, the room with rarely closed blinds 210 W h/(m2 d). • The mean heat gains are slightly higher in the 8 offices in the 2nd floor (362 W h/(m2 d)) than in the 1st floor (333 W h/m2 d). • Heat gains can vary heavily from one day to the next: In some offices, the standard deviation reaches 70% (equipment), 93% (persons) and 43% (solar) of the mean value. The 8 offices on the 2nd floor are 0.4 K cooler than the 8 offices on the 3rd floor. As the impacts (i.e. thermal stratification, sun-shading by adjacent building parts and the higher heat gains) on the temperature difference cannot be distinguished from each other, the 16 offices are evaluated together. Confirmatory, a sensitivity analysis shows that the internal heat gains and the user behaviour are the dominant parameters in the energy balance compared to the thermal stratification or the sun-shading by adjacent building parts. Based on this statistical analysis and the validated simulation model (building physics and ventilation model), the heat flows to calculate the room temperature can be modelled statistically, taking into account user behaviour with respect to attendance, use of electric equipment, sun-shading (manual control of Venetian blinds) and use of windows (outdoor) and doors/louvers (indoor). 3. Statistical simulation for summer 2003 There are approximately 1018 possible combinations from the 16 heat gain series and the status of ventilation components within the period of 1008 h, but the number of simulation runs can be reduced to 1000 by the use of statistically distributed input parameters. The Monte-Carlo
simulation deals with statistical input parameters, which are defined by a true average and a realistic deviation. All input parameters are varied at the same time. The validated simulation model is applied to the statistical simulation, which requires that all internal surfaces have adiabatic boundary conditions. The following procedure describes the program flow of the statistical simulation according to the Monte–Carlo simulation concept: 1. Calculation of user behaviour concerning the window opening. For each hour of the day, the status of the two-segment window and the door/louver is calculated with the Gaussian distribution. Mean value and standard deviation are known from the data analysis. The time series is mathematically conservative: the opening status increases or decreases over the time, but does not oscillate from one hour to the next. 2. Random determination of mechanical air change according to the status of openings. 3. The hourly time series of the internal and solar heat gains for one room is taken from the data analysis. At the next time step, the next room is chosen. After 16 simulation runs, the procedure starts with the first room again. (For each simulation run, the hourly room air temperatures are saved.) 4. The statistical analysis corresponds to the data analysis of the monitored room temperatures according to Section 2.1. Since the input variables are statistically distributed, the calculated room temperatures are statistically distributed as well. The simulated temperatures are compared with the monitored room temperatures, in order to evaluate the statistical input models. Fig. 6 shows that the mean temperature variation in the simulation closely approximates the variation in measurements: • In this 42-day period, the mean measured and simulated room temperatures differ from each other by only 0.2 K. • Every day, the simulated room temperatures meet the monitored temperatures within the standard deviation.
Mean room temperature [ºC]
Heat gain [Wh/(m2 d)]
400
679
30 28 26 24 22 measurement simulation ambient air
20 18 16 0
7 14 21 28 35 42 Period of comparison [June 12 - July 23, 2003]
Fig. 6. Daily mean room temperatures and their 1 r deviations: the measured room temperature is 26.4 C and the simulated 26.2 C. (The mean ambient air temperature is 22.9 C.)
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
• As expected, the deviation (of 16 offices) is higher in the simulation than in the monitored data at each day, since the balancing heat transfer between adjacent rooms (monitoring) is disconnected by the adiabatic boundary conditions (simulation model). While Fig. 6 shows a good agreement in the mean temperatures, Fig. 7 evaluates the dynamic temperature behaviour:
Daily room temperature [ºC]
680
30 measurement simulation ambient air
28 26 24 22 20 0
• In this 42-day period, the mean measured and simulated daily fluctuations differ from each other by 0.5 K. Since temporary fluctuations of heat gains or losses cannot be balanced by heat conduction from/to adjacent rooms due to the adiabatic boundaries, the simulation shows generally a higher daily fluctuation. • The variation in time is very similar. The simulated daily temperature fluctuations match those measured, during 39 of the 42 days. • Noteworthy, the deviation (of 16 offices) in the simulation is smaller than in the monitored room temperatures on some days. High variations in measurements may be explained by the measuring system and the local position of temperature sensors.
ΔT room temperature [K]
Conclusion: Both the daily energy balance (mean room temperature) and the dynamic thermal building behaviour (temperature fluctuation) are simulated realistically using the user model. The passive cooling system is designed in order to hold the room temperature in the comfort range. During this extreme weather situation, the room temperatures did not meet the commonly used comfort criteria according to DIN 1946-2 or ASHRAE 55. Taking a daily mean temperature of 26 C as a reference temperature, the duration curve can be used in order to estimate the thermal comfort taking pattern from the standard (DIN 4108-2). Fig. 8 shows clearly that the simulation and the monitoring result are almost in the same benchmark: the ambient air temperature exceeds 26 C for 6, the measured room temperature (mean value from 16 office
20
measurement simulation ambient air
16 12 8 4
7 14 21 28 35 Number of days [June 12 - July 23, 2003]
42
Fig. 8. Duration curve for daily mean temperatures and their 1 r deviations. (Due to the heat gains, the daily mean room temperature is always higher than the daily mean ambient air temperature. However, the maximum hourly room temperature is below the maximum hourly ambient air temperature due to the heat storage capacity of the room.)
rooms) for 27 and the simulated room temperature (mean value from 1000 simulation runs) for 25 of 42 summer days. The results of this comprehensive data evaluation reveal, that the thermal building model combined with the statistical user model simulates the thermal building performance accurately and, hence, yields the same evaluation of the room temperature as the monitoring. 4. Statistical simulation for design In the design process, the user behaviour with regard to the ventilation and the manual control of the sun-shading and the internal heat i.e. gains (use of office equipment, electric lighting and occupancy patterns) had to be estimated and was taken from the available user models at the year of design (1998), cf. Herkel’s report on the design of the Fraunhofer ISE building (Herkel, 2001). With the statistical simulation, these design assumptions can be checked against the real building operation – assuming that users behave in the design summer (i.e. test reference year TRY 7 (Jahn, 1986)) similar to the summer 2003. Fig. 9 shows that the test reference year TRY 7 takes neither high ambient air temperatures nor long summer periods with warm weather into account. The design data set differs greatly from the measurements in 2002 (typical summer) and 2003 (hot summer). In the following, the predictions from the design simulation are reviewed using the statistical simulation with the weather data set TRY 7 and the user models, which were derived from the data evaluation. For comparison studies, the design and the statistical simulation has to be adjusted in order to operate with the same time schedules of heat gains:
0 0
7 14 21 28 35 42 Period of comparison [June 12 - July 23, 2003]
Fig. 7. Daily fluctuation of room temperatures and their 1 r deviations: the measured fluctuation is 3.5 K and the simulated 4.0 K. (The fluctuation of the ambient air temperature is 11.8 K.)
• As the heat gains are imported from a time series, the design model must be simulated with the calendar of the year 2003 and the weather data from the TRY 7 to take the working days and weekends accurately into account.
Ambient air temperature [ºC]
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682 32 Freiburg - 2003 Freiburg - 2002 Freiburg - TRY 7
28 24 20 16 12 0
14 28 42 56 70 84 Number of summer days [June 1 - August 31]
Fig. 9. Comparison of ambient air temperature for different years: in the design simulation the test reference year TRY 7 was used. The summer 2002 was a typical summer. Obviously, the test reference year does not consider high summer ambient air temperatures. The summer 2003 was 5.2 K warmer than the test reference year and 3.7 K warmer than the summer 2002. For design studies, planners do not only use the test reference year, but also ‘‘design days’’ in order to check the passive cooling concept against extreme weather conditions.
• The solar heat gains in 2003 differ from TRY 7 due to different solar radiation data. Therefore, the solar heat gains in the statistical model are adapted to the test reference year by multiplication with the ratio of the solar radiation in 2003 to the solar radiation in the TRY 7.
28
1. The design simulation operates with constant air change rates, which depend on the room air temperature and the time of the day. Though the statistical simulation is based on a much more complex air-flow network, which takes user behaviour regarding the ventilation into account, the air change rates are similar in the design and the statistical simulation. This is an insignificant difference. 2. While the night ventilation is interrupted in the design simulation at low ambient air temperatures, this cooling potential is used in the ordinary operation (statistical model), during the summer nights. Due to this different operation management, the night ventilation potential is better utilised by the statistical than by the design simulation at low ambient air temperatures. In spite of the higher nocturnal heat dissipation, the mean room temperature calculated by the statistical simulation with the real user behaviour is 1 K higher than the room temperature predicted by the design simulation. This is due to the higher heat gains. 3. The monitored internal heat gains are higher than their estimates in the design simulation, since the work stations are located in the offices and not – according to the assumptions from the design phase – in separated IT rooms, cf. Fig. 11. 4. Fig. 11 outlines that the monitored solar heat gains, during the working days are higher than their estimates in the design simulation. As the Venetian blinds are mainly used to protect from glare and not to prevent overheating, they are more rarely closed in summer time than estimated with a setpoint for blind control of 200 W/m2 solar radiation on the facade. (Noteworthy, the actually realised g-value of the facade is lower than the design value: during the weekends, when the blinds are closed, the monitored solar heat gains are lower than the design value.) These are two significant differences concerning the internal and solar heat gains between the estimated (design simulation) and the monitored data (statistical simulation). Table 1 shows the mean room temperature and the mean heat gains: in the statistical simulation, the significantly higher heat gain results in a higher (mean) room temperature, since the additional heat gain cannot be counterbalanced by the heat dissipation due to the (night) ventilation.
ambient air design simulation statistical simulation
26
400 Heat gain [Wh/(m2 d)]
Daily room temperature [ºC]
The duration curve is an important criterion in the design process. Fig. 10 shows that the assumptions from the design procedure were too confident: taking 24 C (mean daily temperature) as a standard of comparison, the design simulation forecasts only 3 days with a higher daily room temperature, but the daily room temperature exceeds 24 C for 20 days in the statistical simulation based on monitored data. Both the design and the statistical model are simulated with the same meteorological boundary conditions (here: test reference year TRY 7), but the simulated room temperatures differ greatly. As all other boundary conditions are (almost) identical, the higher room temperatures must be directly associated with the use of the building. What are the reasons, that the room temperature is higher in the operation of the building than predicted in the design process?
24 22 20 18 0
7 14 21 28 35 42 Number of days [TRY 7 : June 12 - July 23]
Fig. 10. Duration curve of daily mean room temperatures. During a long period the actual room temperature is at least 1 K above the room temperature predicted by the design simulation.
681
300
internal equipment
persons solar
200 100 0 design
monitoring
weekend
design
monitoring
working days
Fig. 11. Daily heat gains from design and statistical simulation.
682
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
Table 1 Mean room temperature and heat gains from design and statistical simulation for the period from June 12 to July 23
Ambient air (TRY 7) Design simulation (room) Statistical simulation (room)
Mean temperature (with standard deviation)
Heat gains (working days)
Heat gains (weekend)
18.47 C
–
–
22.61 C
214 W h/(m2 d)
75 W h/(m2 d)
Monitoring, data acquisition and data evaluation is a time-consuming task. We want to thank A. Gerber, S. Oexle and A. Adlhoch (data acquisition), H. Seitz, J. Matsche and H. Qiu (air change measurements and air flow modelling), E. Selg, A. Gorzelanny and U. Knapp (user behaviour regarding electric internal heat gains, control of venetian blinds and window opening), who contributed to this study. References
23.87 C (±1.01 K)
2
380 W h/(m d)
2
121 W h/(m d)
5. Conclusions Three lessons can be learned from the statistical simulation: 1. A statistical data analysis provides adequate user models for application in building simulation. 2. The Monte–Carlo simulation is an appropriate tool to calculate the thermal building performance with a true mean value and its statistically relevant deviation. 3. Statistical simulations can be advantageously applied to the design process and enhance the significance and clarity of simulation results. The design of passive cooling concepts should consider the user behaviour realistically according to this statistical data analysis (Voss et al., 2005).
Acknowledgements This work was funded by the German Ministry of Economics and Labour BMWA under reference 03350006O (MessISE) and 16IN0248 (InnoNet QUALIPASS).
ASHRAE 55, 2004. Thermal Environmental Conditions for Human Occupancy, ASHRAE, Atlanta, USA. Clarke, J., 2001. Energy Simulation in Building Design. ButterworthHeinemann, Oxford. DIN 1946-2, 1994. Raumlufttechnik – Gesundheitstechnische Anforderungen, Beuth-Verlag, Berlin (in German). DIN 4108-2, 2003. Thermal Protection and Energy Economy in Buildings: Part 2 – Minimum Requirements to Thermal Insulation, Beuth-Verlag, Berlin. Herkel, S. et al., 2001. Neubau Eines Institutsgeba¨udes mit Gesamtenergetischer Optimierung unter Hoher Ausnutzung der Solarenergie. Fraunhofer Institute for Solar Energy Systems ISE, Freiburg (in German). Herkel, S., Pfafferott, J., Knapp, U., 2005. A Preliminary Model of User Behaviour Regarding the Manual Control of Windows in Office Buildings, Building Simulation 2005, Montreal. Jahn, A. 1986. Entwicklung von Testreferenzjahren TRY fu¨r Klimaregionen der Bundesrepublik. Bundesministerium fu¨r Forschung und Technologie (Forschungsbericht T 86-051), 1986 (in German). Macdonald, I., Strachan, P., 2001. Practical application of uncertainty analysis. Energ. Buildings 33, 219–227. Pfafferott, J., 2004. Enhancing the Design and the Operation of Passive Cooling Concepts. Fraunhofer IRB-Verlag, Stuttgart. Pfafferott, J., Herkel, S., Ja¨schke, M., 2003. Design of passive cooling by night ventilation: evaluation of a parametric model and building simulation with measurements. Energ. Buildings 35, 1129–1143. SolarBau:Monitor: Internet Platform, Project Documentation Fraunhofer ISE, July 31, 2005.
Statistical simulation of user behaviour in low-energy office buildings J. Pfafferott *, S. Herkel Fraunhofer Institute for Solar Energy Systems, Heidenhofstr. 2, D-79110 Freiburg, Germany Received 23 August 2005; received in revised form 10 August 2006; accepted 15 August 2006 Available online 16 October 2006 Communicated by: Associate Editor Mattaios Santamouris
Abstract A large number of design guidelines and tools are available for the design of passive cooling systems. Using the underlying thermodynamic models, a certain input (e.g. air change rate, internal heat gains or sun control) results in a certain output (i.e. room temperature). However, in real buildings the room temperature at a given outdoor temperature is a distribution rather than a single value. Therefore, the building engineer should take uncertainties into account, since the actual use of the building, the building physical properties or the user behaviour are statistically distributed. One promising approach to include these uncertainties in the design procedure is the use of statistical models: the design parameter is defined by a mean value and its deviation. From a control theoretical point of view, the deterministic controlled system responds to random disturbance variables by a statistically distributed response function. Considering the institute building of Fraunhofer ISE as example, this study shows how statistical simulations can be applied to the design process of passive cooling in low-energy office buildings. 2006 Elsevier Ltd. All rights reserved.
1. Introduction In preparation for the statistical simulation, several investigations on detail phenomena concerning the thermal behaviour of the Fraunhofer ISE building (Fig. 1) were carried out to provide a reliable simulation model (Pfafferott, 2004). For this study, a validated building simulation model is used to investigate the impact of user behaviour on the thermal building performance. The influence of material properties and parameters in the air-flow network on the room temperature have been analysed by a sensitivity analysis in order to prepare and to evaluate the building simulation model. This model validation has been carried out using data sets from April and July 2002 (Pfafferott et al., 2003), while the statistical simulation with regard to the user behaviour uses a data set from the summer of *
Corresponding author. Tel.: +49 761 4588 5129. E-mail address: jens.pfaff[email protected] (J. Pfafferott). URL: www.ise.fraunhofer.de (J. Pfafferott).
0038-092X/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.solener.2006.08.011
2003. All simulations have been carried out using the simulation program ESP-r (Clarke, 2001). The functionality (and the programming) of Monte Carlo-simulations is briefly described by Macdonald and Strachan (2001). The statistical simulation is supposed to deal with uncertain input parameters and should show, whether typical occupancy patterns are effective for modelling the thermal building behaviour and how far statistical models can be used successfully to take uncertainties into account. The procedure can be described according to the control theory (Fig. 2). Excitation function. The input parameters are statistically analysed with regard to the use of the building (i.e. equipment and occupancy) and user behaviour (i.e. ventilation and sun control). Through this comprehensive data evaluation, statistical models for the simulation input are derived. Controlled system. Using these data series as input parameters, the thermal building performance is simulated: many parameter combinations are considered by the
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
677
Fig. 1. View of the Fraunhofer ISE building from west. The monitored offices are located in the middle building wing on the 2nd and 3rd floor with south orientation.
open control loop exication function user behaviour (probability function)
response function room temperature (statistical analysis)
controlled system building (validated simulation model)
from 25.9 C to 26.9 C, since the offices vary in the daily attendance, the office equipment (internal heat gains) and the user behaviour regarding sun protection (solar heat gains) and ventilation (heat losses). As the statistical models are deduced from a long-term monitoring, the following sections shortly describes the statistical data analysis for 16 offices, during the considered summer period.
Fig. 2. Methodological approach.
2.1. Variation in time of room temperatures
2. Data evaluation For the statistical data analysis and simulation, (1) the time period and (2) the sample of offices must be defined. For this data sample, the variation in time of room temperatures and the heat gains are statistically analysed by mean value and distribution. The user behaviour is analysed and modelled with special regard to the ventilation (Herkel et al., 2005). Time period. Starting from the hourly ambient air temperatures in 2003, a characteristic period can be defined. The period June 12 to July 23, 2003 is short enough to carry out many simulation runs in a row and long enough to cover different summer weather conditions. Similar ambient air temperatures and sun positions in this time period will produce likely consistent user behaviour. Office sample. Sixteen offices in the Fraunhofer ISE building are considered for the statistical data evaluation and simulation. Since all rooms are located on the first and second floor in the same part of the building, the climate impact from the outside is the same. However, the mean room temperatures (during the summer period) vary
The following procedure provides the variation in time of room temperatures and its statistical distribution: 1. Preparation of hourly room temperatures in each of the 16 offices. 2. Calculation of daily mean room temperature and daily temperature fluctuation in each office. 3. From this time series, mean room temperature, between 16% and 84% quantile are identified for each day. The minimum/maximum temperatures specify the limit of variation. Exemplarily for this procedure, Fig. 3 shows the time behaviour of the daily mean room temperature (in 16 offices) and its daily deviation/variation in all rooms: • The daily mean room temperature varies from 22.3 C to 28.8 C. • The mean deviation is 0.4 K and varies from 0.15 K to 0.9 K.
Mean room temperature [ºC]
simulation, and each is distributed around a true mean value. It is reasonable to assume that the parameters are normally distributed. As all input parameters are varied simultaneously, the room temperature is also distributed around a true mean value. Response function. The simulated room temperatures are statistically analysed and compared with the monitored room temperatures. For further information concerning the building, the HVAC system and its operation the reader is referred to the project documentation in SolarBau:Monitor (Internet platform, 2005) and to the cross section analysis on low-energy buildings by Voss et al. (2005).
30 28 26 24 22 20 0
7 14 21 28 35 Period of comparison [June 12 - July 23, 2003]
42
Fig. 3. Mean room temperatures with minimum/maximum (grey lines) and 1 r deviation (black dashes).
678
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
As the climatic and the building physical boundary conditions are (almost) identical in all rooms, the temperature variations result mostly from variations in use of the offices. 2.2. User behaviour regarding ventilation The following procedure provides the user behaviour regarding ventilation and its statistical distribution: 1. Preparation of the hourly status (open or closed) of door and louver (indoor) and the two upper and downer part of a two-segment window (outdoor) in each of the 16 offices. If a room has more than one window these openings will be combined by an OR-relation: if one of the window segments is open, the whole window is considered to be open. 2. The time series differentiates working days and weekends. 3. All data lines are sorted by the time of day. 4. From this information, relative frequencies of opening are calculated for each ventilation component and for each hour of the day. 5. Using the hourly data regarding the status of all openings in each of the 16 offices, the local distribution of hydraulic resistances (mean value and statistical distribution) is calculated using an air-flow network with resistances, parallel (16 offices) and in series (office-corridor). Fig. 4 shows the average user behaviour in 16 offices, during 30 working days: as expected, doors are closed outside business hours. More or less than 50% of all doors are opened, during the working hours. In general, windows are opened at arrival and are closed bit by bit, when the ambient air temperature is increasing. In the morning, more windows are turned open, while in the afternoon most open windows are tilted. At least 50% of all windows are open, during night. At least 80%/90% of all small windows above the window and louvers above the door are opened, during
Relative frequency of opening
100% 80% 60% 40% 20% 0% 00:00
large window small window louver, indoor door
04:00
08:00
12:00
16:00
20:00
00:00
Fig. 4. Time dependent user behaviour with regard to ventilation, during the working days. (Manual opening of ventilation components.) ‘‘Large window’’ refers to the downer and ‘‘small window’’ to the upper part of the two-segment windows. The small window and the louver above the door provide for a basic ventilation in conjunction with the exhaust air ventilator.
the whole day. These two ventilation flaps are manipulated very rarely and provide for a basic ventilation in the rooms. The exhaust fan delivers an air change rate of 1 h 1 (480 m3/h for 8 offices), during the working hours and 5 h 1 (2400 m3/h for 8 offices), during night ventilation in each floor. If the hydraulic resistance was the same in each of the offices, each office would get the same fraction of the air-volume flow, which is 1/8 or 12% of the total air-volume flow. Due to the changing distribution of open and closed flaps, doors and windows, the air-flow network (hydraulic resistances in series) constantly changes. Some offices get less than 4% and other rooms up to 18% of the air flow. 2.3. Variation in time of heat gains The following procedure provides occupancy patterns, the time variation of electricity consumption, the use/control of sun-shading and their statistical distributions: 1. The hourly heat gains due to persons are calculated from the signal of an ultrasonic sensor and correspond to 80 W/person. Missing or incorrect data are substituted by an average time profile, which is calculated from all available sensors. 2. The total electric power consumption [Wtotal] is hourly measured. Furthermore, the electricity consumption [kW hoffice/(m2 d)] in an office depends on the electric connection power and the estimated operation time of the equipment. Using the connection power and the current attendance for each office, the total electricity consumption can be divided into the electricity consumption in each office [Woffice]. 3. The hourly solar heat gains are calculated from the solar radiation on the window, which is calculated from meteorological data, taking into consideration shading from adjacent buildings and the facade as well as the position of the venetian blinds. The data acquisition is realised by a web camera. The error-prone data evaluation by image processing results in some data gaps, which are substituted depending on their length of time: if the status of the venetian blind is not recognised, the last valid value will be extrapolated. If the data gap covers several hours, the missed value will be substituted by the mean value of all known positions. 4. Through this process, hourly data for internal (persons and lighting/equipment) and solar heat gains are available for each office. The simulation deals with 16 different time profiles for heat gains. Fig. 5 outlines the daily heat gains in all of the 16 offices for 30 working days and its variation. The heat gains vary from one room to the next and can vary from one day to the next: • Due to the different application of computers, the daily heat gains from equipment [grand average: 137 W h/ (m2 d)] varies from 58 W h/(m2 d) to 295 W h/(m2 d).
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
equipment persons solar
300 200 100 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Office room (number)
Fig. 5. Daily heat gains and their 1 r deviations in 16 office rooms, of which 8 are in the 1st and 8 are in the 2nd floor.
• Since the rooms can be occupied by as few as one and as many as four persons and the mean attendance in the office is variably long (e.g. laboratory workers), the mean daily heat gains from persons [grand average: 54 W h/(m2 d)] varies greatly from 26 W h/(m2 d) to 103 W h/(m2 d). • In comparison, the variation in solar heat gains [grand average: 158 W h/(m2 d)] is relatively small: The room, in which the occupants close the venetian blinds most frequently, gains meanly 106, the room with rarely closed blinds 210 W h/(m2 d). • The mean heat gains are slightly higher in the 8 offices in the 2nd floor (362 W h/(m2 d)) than in the 1st floor (333 W h/m2 d). • Heat gains can vary heavily from one day to the next: In some offices, the standard deviation reaches 70% (equipment), 93% (persons) and 43% (solar) of the mean value. The 8 offices on the 2nd floor are 0.4 K cooler than the 8 offices on the 3rd floor. As the impacts (i.e. thermal stratification, sun-shading by adjacent building parts and the higher heat gains) on the temperature difference cannot be distinguished from each other, the 16 offices are evaluated together. Confirmatory, a sensitivity analysis shows that the internal heat gains and the user behaviour are the dominant parameters in the energy balance compared to the thermal stratification or the sun-shading by adjacent building parts. Based on this statistical analysis and the validated simulation model (building physics and ventilation model), the heat flows to calculate the room temperature can be modelled statistically, taking into account user behaviour with respect to attendance, use of electric equipment, sun-shading (manual control of Venetian blinds) and use of windows (outdoor) and doors/louvers (indoor). 3. Statistical simulation for summer 2003 There are approximately 1018 possible combinations from the 16 heat gain series and the status of ventilation components within the period of 1008 h, but the number of simulation runs can be reduced to 1000 by the use of statistically distributed input parameters. The Monte-Carlo
simulation deals with statistical input parameters, which are defined by a true average and a realistic deviation. All input parameters are varied at the same time. The validated simulation model is applied to the statistical simulation, which requires that all internal surfaces have adiabatic boundary conditions. The following procedure describes the program flow of the statistical simulation according to the Monte–Carlo simulation concept: 1. Calculation of user behaviour concerning the window opening. For each hour of the day, the status of the two-segment window and the door/louver is calculated with the Gaussian distribution. Mean value and standard deviation are known from the data analysis. The time series is mathematically conservative: the opening status increases or decreases over the time, but does not oscillate from one hour to the next. 2. Random determination of mechanical air change according to the status of openings. 3. The hourly time series of the internal and solar heat gains for one room is taken from the data analysis. At the next time step, the next room is chosen. After 16 simulation runs, the procedure starts with the first room again. (For each simulation run, the hourly room air temperatures are saved.) 4. The statistical analysis corresponds to the data analysis of the monitored room temperatures according to Section 2.1. Since the input variables are statistically distributed, the calculated room temperatures are statistically distributed as well. The simulated temperatures are compared with the monitored room temperatures, in order to evaluate the statistical input models. Fig. 6 shows that the mean temperature variation in the simulation closely approximates the variation in measurements: • In this 42-day period, the mean measured and simulated room temperatures differ from each other by only 0.2 K. • Every day, the simulated room temperatures meet the monitored temperatures within the standard deviation.
Mean room temperature [ºC]
Heat gain [Wh/(m2 d)]
400
679
30 28 26 24 22 measurement simulation ambient air
20 18 16 0
7 14 21 28 35 42 Period of comparison [June 12 - July 23, 2003]
Fig. 6. Daily mean room temperatures and their 1 r deviations: the measured room temperature is 26.4 C and the simulated 26.2 C. (The mean ambient air temperature is 22.9 C.)
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
• As expected, the deviation (of 16 offices) is higher in the simulation than in the monitored data at each day, since the balancing heat transfer between adjacent rooms (monitoring) is disconnected by the adiabatic boundary conditions (simulation model). While Fig. 6 shows a good agreement in the mean temperatures, Fig. 7 evaluates the dynamic temperature behaviour:
Daily room temperature [ºC]
680
30 measurement simulation ambient air
28 26 24 22 20 0
• In this 42-day period, the mean measured and simulated daily fluctuations differ from each other by 0.5 K. Since temporary fluctuations of heat gains or losses cannot be balanced by heat conduction from/to adjacent rooms due to the adiabatic boundaries, the simulation shows generally a higher daily fluctuation. • The variation in time is very similar. The simulated daily temperature fluctuations match those measured, during 39 of the 42 days. • Noteworthy, the deviation (of 16 offices) in the simulation is smaller than in the monitored room temperatures on some days. High variations in measurements may be explained by the measuring system and the local position of temperature sensors.
ΔT room temperature [K]
Conclusion: Both the daily energy balance (mean room temperature) and the dynamic thermal building behaviour (temperature fluctuation) are simulated realistically using the user model. The passive cooling system is designed in order to hold the room temperature in the comfort range. During this extreme weather situation, the room temperatures did not meet the commonly used comfort criteria according to DIN 1946-2 or ASHRAE 55. Taking a daily mean temperature of 26 C as a reference temperature, the duration curve can be used in order to estimate the thermal comfort taking pattern from the standard (DIN 4108-2). Fig. 8 shows clearly that the simulation and the monitoring result are almost in the same benchmark: the ambient air temperature exceeds 26 C for 6, the measured room temperature (mean value from 16 office
20
measurement simulation ambient air
16 12 8 4
7 14 21 28 35 Number of days [June 12 - July 23, 2003]
42
Fig. 8. Duration curve for daily mean temperatures and their 1 r deviations. (Due to the heat gains, the daily mean room temperature is always higher than the daily mean ambient air temperature. However, the maximum hourly room temperature is below the maximum hourly ambient air temperature due to the heat storage capacity of the room.)
rooms) for 27 and the simulated room temperature (mean value from 1000 simulation runs) for 25 of 42 summer days. The results of this comprehensive data evaluation reveal, that the thermal building model combined with the statistical user model simulates the thermal building performance accurately and, hence, yields the same evaluation of the room temperature as the monitoring. 4. Statistical simulation for design In the design process, the user behaviour with regard to the ventilation and the manual control of the sun-shading and the internal heat i.e. gains (use of office equipment, electric lighting and occupancy patterns) had to be estimated and was taken from the available user models at the year of design (1998), cf. Herkel’s report on the design of the Fraunhofer ISE building (Herkel, 2001). With the statistical simulation, these design assumptions can be checked against the real building operation – assuming that users behave in the design summer (i.e. test reference year TRY 7 (Jahn, 1986)) similar to the summer 2003. Fig. 9 shows that the test reference year TRY 7 takes neither high ambient air temperatures nor long summer periods with warm weather into account. The design data set differs greatly from the measurements in 2002 (typical summer) and 2003 (hot summer). In the following, the predictions from the design simulation are reviewed using the statistical simulation with the weather data set TRY 7 and the user models, which were derived from the data evaluation. For comparison studies, the design and the statistical simulation has to be adjusted in order to operate with the same time schedules of heat gains:
0 0
7 14 21 28 35 42 Period of comparison [June 12 - July 23, 2003]
Fig. 7. Daily fluctuation of room temperatures and their 1 r deviations: the measured fluctuation is 3.5 K and the simulated 4.0 K. (The fluctuation of the ambient air temperature is 11.8 K.)
• As the heat gains are imported from a time series, the design model must be simulated with the calendar of the year 2003 and the weather data from the TRY 7 to take the working days and weekends accurately into account.
Ambient air temperature [ºC]
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682 32 Freiburg - 2003 Freiburg - 2002 Freiburg - TRY 7
28 24 20 16 12 0
14 28 42 56 70 84 Number of summer days [June 1 - August 31]
Fig. 9. Comparison of ambient air temperature for different years: in the design simulation the test reference year TRY 7 was used. The summer 2002 was a typical summer. Obviously, the test reference year does not consider high summer ambient air temperatures. The summer 2003 was 5.2 K warmer than the test reference year and 3.7 K warmer than the summer 2002. For design studies, planners do not only use the test reference year, but also ‘‘design days’’ in order to check the passive cooling concept against extreme weather conditions.
• The solar heat gains in 2003 differ from TRY 7 due to different solar radiation data. Therefore, the solar heat gains in the statistical model are adapted to the test reference year by multiplication with the ratio of the solar radiation in 2003 to the solar radiation in the TRY 7.
28
1. The design simulation operates with constant air change rates, which depend on the room air temperature and the time of the day. Though the statistical simulation is based on a much more complex air-flow network, which takes user behaviour regarding the ventilation into account, the air change rates are similar in the design and the statistical simulation. This is an insignificant difference. 2. While the night ventilation is interrupted in the design simulation at low ambient air temperatures, this cooling potential is used in the ordinary operation (statistical model), during the summer nights. Due to this different operation management, the night ventilation potential is better utilised by the statistical than by the design simulation at low ambient air temperatures. In spite of the higher nocturnal heat dissipation, the mean room temperature calculated by the statistical simulation with the real user behaviour is 1 K higher than the room temperature predicted by the design simulation. This is due to the higher heat gains. 3. The monitored internal heat gains are higher than their estimates in the design simulation, since the work stations are located in the offices and not – according to the assumptions from the design phase – in separated IT rooms, cf. Fig. 11. 4. Fig. 11 outlines that the monitored solar heat gains, during the working days are higher than their estimates in the design simulation. As the Venetian blinds are mainly used to protect from glare and not to prevent overheating, they are more rarely closed in summer time than estimated with a setpoint for blind control of 200 W/m2 solar radiation on the facade. (Noteworthy, the actually realised g-value of the facade is lower than the design value: during the weekends, when the blinds are closed, the monitored solar heat gains are lower than the design value.) These are two significant differences concerning the internal and solar heat gains between the estimated (design simulation) and the monitored data (statistical simulation). Table 1 shows the mean room temperature and the mean heat gains: in the statistical simulation, the significantly higher heat gain results in a higher (mean) room temperature, since the additional heat gain cannot be counterbalanced by the heat dissipation due to the (night) ventilation.
ambient air design simulation statistical simulation
26
400 Heat gain [Wh/(m2 d)]
Daily room temperature [ºC]
The duration curve is an important criterion in the design process. Fig. 10 shows that the assumptions from the design procedure were too confident: taking 24 C (mean daily temperature) as a standard of comparison, the design simulation forecasts only 3 days with a higher daily room temperature, but the daily room temperature exceeds 24 C for 20 days in the statistical simulation based on monitored data. Both the design and the statistical model are simulated with the same meteorological boundary conditions (here: test reference year TRY 7), but the simulated room temperatures differ greatly. As all other boundary conditions are (almost) identical, the higher room temperatures must be directly associated with the use of the building. What are the reasons, that the room temperature is higher in the operation of the building than predicted in the design process?
24 22 20 18 0
7 14 21 28 35 42 Number of days [TRY 7 : June 12 - July 23]
Fig. 10. Duration curve of daily mean room temperatures. During a long period the actual room temperature is at least 1 K above the room temperature predicted by the design simulation.
681
300
internal equipment
persons solar
200 100 0 design
monitoring
weekend
design
monitoring
working days
Fig. 11. Daily heat gains from design and statistical simulation.
682
J. Pfafferott, S. Herkel / Solar Energy 81 (2007) 676–682
Table 1 Mean room temperature and heat gains from design and statistical simulation for the period from June 12 to July 23
Ambient air (TRY 7) Design simulation (room) Statistical simulation (room)
Mean temperature (with standard deviation)
Heat gains (working days)
Heat gains (weekend)
18.47 C
–
–
22.61 C
214 W h/(m2 d)
75 W h/(m2 d)
Monitoring, data acquisition and data evaluation is a time-consuming task. We want to thank A. Gerber, S. Oexle and A. Adlhoch (data acquisition), H. Seitz, J. Matsche and H. Qiu (air change measurements and air flow modelling), E. Selg, A. Gorzelanny and U. Knapp (user behaviour regarding electric internal heat gains, control of venetian blinds and window opening), who contributed to this study. References
23.87 C (±1.01 K)
2
380 W h/(m d)
2
121 W h/(m d)
5. Conclusions Three lessons can be learned from the statistical simulation: 1. A statistical data analysis provides adequate user models for application in building simulation. 2. The Monte–Carlo simulation is an appropriate tool to calculate the thermal building performance with a true mean value and its statistically relevant deviation. 3. Statistical simulations can be advantageously applied to the design process and enhance the significance and clarity of simulation results. The design of passive cooling concepts should consider the user behaviour realistically according to this statistical data analysis (Voss et al., 2005).
Acknowledgements This work was funded by the German Ministry of Economics and Labour BMWA under reference 03350006O (MessISE) and 16IN0248 (InnoNet QUALIPASS).
ASHRAE 55, 2004. Thermal Environmental Conditions for Human Occupancy, ASHRAE, Atlanta, USA. Clarke, J., 2001. Energy Simulation in Building Design. ButterworthHeinemann, Oxford. DIN 1946-2, 1994. Raumlufttechnik – Gesundheitstechnische Anforderungen, Beuth-Verlag, Berlin (in German). DIN 4108-2, 2003. Thermal Protection and Energy Economy in Buildings: Part 2 – Minimum Requirements to Thermal Insulation, Beuth-Verlag, Berlin. Herkel, S. et al., 2001. Neubau Eines Institutsgeba¨udes mit Gesamtenergetischer Optimierung unter Hoher Ausnutzung der Solarenergie. Fraunhofer Institute for Solar Energy Systems ISE, Freiburg (in German). Herkel, S., Pfafferott, J., Knapp, U., 2005. A Preliminary Model of User Behaviour Regarding the Manual Control of Windows in Office Buildings, Building Simulation 2005, Montreal. Jahn, A. 1986. Entwicklung von Testreferenzjahren TRY fu¨r Klimaregionen der Bundesrepublik. Bundesministerium fu¨r Forschung und Technologie (Forschungsbericht T 86-051), 1986 (in German). Macdonald, I., Strachan, P., 2001. Practical application of uncertainty analysis. Energ. Buildings 33, 219–227. Pfafferott, J., 2004. Enhancing the Design and the Operation of Passive Cooling Concepts. Fraunhofer IRB-Verlag, Stuttgart. Pfafferott, J., Herkel, S., Ja¨schke, M., 2003. Design of passive cooling by night ventilation: evaluation of a parametric model and building simulation with measurements. Energ. Buildings 35, 1129–1143. SolarBau:Monitor: Internet Platform, Project Documentation Fraunhofer ISE, July 31, 2005.