Printer Friendly

Development of empirical office occupancy profiles based on information technology sensor systems.

BACKGROUND

The ever-increasing demands in energy efficiency are quickly changing the way how buildings are designed and operated. In building design community, professionals are adopting building simulation as tools to inform both the envelope and mechanical system design throughout various design stages. But recent research has shown significant deviation between the estimated building performance through simulations and the actual performance (Newsham, Mancini et al. 2009; Scofield 2009). Though many factors are concluded to have contributed to the observed discrepancy, the higher-than-expected occupant loads (including the occupancy dependent plug loads), due to inaccurate estimate of occupancy related inputs made during the design process, are identified as the common significant factor from field studies conducted on six high performance buildings (Torcellini, Deru et al. 2006).

Occupant behavior is well known to be random and difficult to predict. Occupancy rate, also referred as occupancy diversity factor or occupancy schedule, is commonly used in building simulation practice to estimate time dependent occupancy related loads, such as internal gains from people, hot water usage, and etc.. Due to the random nature of occupant behavior, it has been technically challenging and economically expensive for researchers to conduct experiments to develop reliable and up-to-date occupancy rate profile for design practice.

There are a number of methods researchers and building operators use to measure occupancy for various purposes and at various resolutions. Depending on the mechanism used, these methods are often categorized into two general groups: direct method and implicit method.

The most common direct method in industry is through occupancy sensor, including passive infrared occupancy sensors, ultrasonic occupancy sensors, and etc.. These occupancy sensors are relatively low cost and have been successfully implemented in buildings for occupancy based lighting control. Unfortunately they are able to detect presence of occupants but not the number of occupants present in the space. Therefore they are less helpful in terms of informing or measuring quantitative occupancy rate in multi-occupancy spaces. The occupancy sensors have been used in a few field studies to help develop occupancy profiles, mainly in single occupancy space (Keith and Krarti 1999; Rubinstein, Colak et al. 2003; Mahdavi 2009; Duarte, Van Den Wymelenberg et al. 2013). The occupancy detection accuracy could be enhanced by combining occupancy sensors with other sensors in buildings, such as light switch sensors (Rubinstein, Colak et al. 2003). Besides occupancy sensors, other methods of directly measuring occupancy include security cameras, doorway counting sensors, badge log, manual collection through personal headcounts. Those devices may not be installed for occupancy detection purpose originally but could be used for occupancy estimate under certain circumstances. For instance, doorway counting sensor data can be used to estimate occupancy when the building has controlled access. Compared to occupancy sensors, these methods could provide a quantitative estimate of occupancy in buildings, including both single-occupancy and multi-occupancy spaces. But sometimes the estimation process may be more time consuming (such as in case of using security camera). And since those devices may have been readily available, it requires no additional front cost to conduct occupancy studies other than hiring personnel for data processing. Davis and Nutter (2010) took the advantage and developed occupancy diversity factors for common university building types by using a number of occupancy measurement methods, including security cameras, classroom scheduling data, doorway counting sensors, and manual headcounts. However, there is one thing that all the direct methods mentioned previously fail to offer. It is the spatial resolution in occupancy measurements, i.e. where occupants stay in the building. Additional measurements have to be taken in order to capture the spatial aspect, such as installing additional cameras in space of interests as seen in the study by Mahdavi (2009).

The implicit method does not measure occupancy rate directly but measures other parameters that are correlated with occupancy. Thus it requires a conversion between the implicit measures and occupancy. Abushakra and Claridge (2008) discover a linear relationship between occupancy rates and the lighting and equipment consumption in office buildings. Their results indicate that it is appropriate to use this implicit method to derive occupancy profiles whenever the hourly monitored data for lighting and equipment consumption is available. They also suggest that cautions should be taken when applying this method to perimeter zones where daylight control is implemented. Melfi et al. (2011) review and investigate the feasibility of using existing information technology (IT) infrastructure to measure real-time occupancy. Their research summarizes a few implicit measures that could be used to estimate building occupancy, such as Wi-Fi host count, telephone calls, keyboard and mouse use, existing microphone and webcam on PCs and laptops. Their experimental studies not only demonstrate the feasibility of IT based implicit occupancy sensing but also reveal that the IT based implicit methods are able to detect beyond simple occupant counts: they are also able to detect occupants' location and activities. Though no additional cost is needed, Melfi et al. emphasize the challenge of developing a reliable conversion factor between occupancy and the plausible implicit measures that are not well understood at this moment.

This study presents an implicit method of measuring real-time occupancy rate based on an IT platform-integrated sensor system. This IT platform-integrated sensor set is originally designed to provide detailed personalized sensor data on energy usage and local ambient condition for individual occupant (the "ambient" here refers the indoor environment relative to the occupants themselves). They are installed in individual work space, aiming to enhance occupants' comfort and promote energy conservation behavior in the work environment. More details on the integrated platform, the sensor framework, and the end user experience can be found in previous publications (Milenkovic, Hanebutte et al. 2011; Milenkovic, Dang et al. 2013; Milenkovic, Hanebutte et al. 2013). This study presents a creative way of exploiting the sensor set's data acquisition mechanism and using that information to estimate real-time building occupancy. The rest of the paper will introduce the calculation algorithm and the resulting occupancy profiles from a pilot study conducted in an office building.

METHOD

The installed IT platform-integrated sensor set contains two types of sensors: (1) physical sensors that measure local environmental conditions at individual work space (i.e. temperature, relative humidity, and light intensity); and (2) a custom software sensor that estimates computer power consumption levels by tracking computer power states defined by EnergyStar and ECMA-383 (details on the software sensors and their accuracy can be found in (Milenkovic, Dang et al. 2013)).

The physical sensors are attached to each work space at a desired location of choice (at initial prototyping stage) and connected to a desktop or a dock station in case of a laptop through a USB port. The local ambient conditions (temperature, relative humidity, and lighting levels) are only recorded to the database when somebody logs into the desktop or docks their laptop to the dock station. The software sensor is installed on each desktop or laptop and the power level of each desktop/laptop is recorded whenever the desktop/laptop is on. Therefore an occupant's presence is required whenever there are local ambient condition parameters recorded. The only exception is when someone remotely logs into his/her desktop which is considered as a small chance event and does not impose a significant impact to the building wide occupancy estimate (in fact this situation doesn't apply to the pilot study presented here). Computer power states, such as short idle (i.e. active use), long idle, and sleep modes, can be identified from software sensor measurements and then used as a secondary source to help confirm occupants' presence.

Given the mechanism how the two sets of sensor data are recorded, their associated timestamps are used to estimate real-time occupancy, further supported by the derived computer power states. Given that occupants may stay mobile within the work space during work hours and take their laptops with them (such as conferences, group meetings, etc.), local ambient condition measurements (which could be temperature, humidity, and light levels but we will refer to temperature measurements for simplicity for the rest of the papers) are not required to be continuous. But at least one single temperature measurement must show up in the database on any given day in order to confirm the concerned individual's presence in the work space.

This study proposes the following algorithm to confirm one's presence at any given time in the work space during typical work hours (from 8AM to 6PM): (1) at least one temperature measurement exists on that day; (2) PC power consumption measurement exists. In addition, if a continuous long idle event is observed across 6pm, the time lapsed from the beginning of that continuous long idle event will be overwritten as "absence" indicating that the user has gone home early without logging off from the desktop (or turning off their laptops). For hours outside of the typical work hours (8AM-6PM), the algorithm confirms one's presence only if both the temperature and power measurements exist and the measured power state indicates active mode. Figure 1 shows an example scenario for a typical weekday and the resulting occupancy status determined based on the algorithm explained above. Though the example shown in Figure 1 demonstrates the binary occupancy status at a hourly interval, the actual sensor set measures and records data more often (once every 3 seconds at the initial prototyping stage) and the measurement interval may be customized for individual organization. The office occupancy rate, also known as diversity factor, for each space is then calculated by adding up each individual' presence at a given time and then being divided by the total number of occupants. Essentially the occupancy rate indicates the percentage of people present in the building as compared to the maximum expected occupancy.

RESULTS

The IT platform-integrated sensor sets were deployed for performance testing in two pilot office buildings at the initial prototyping stage in 2012 (Milenkovic, Hanebutte et al. 2013). This paper only presents results from one of the sites due to the page limit. The selected pilot site also represents a more typical office environment, i.e. employees working in the office have typical work hours of 8AM-6PM.

The pilot experiment took place in a 5-story office building of 2422 [m.sup.2] (27,146 square feet), which is mainly open plan office space. A random sample of 23 occupants participated the pilot study, coming across from different floors, different business departments, and different professions. The sensor sets were deployed at each participant's desk and his/her computer. The experiment lasted approximately 3 months, from Mid-June 2012 to Mid-September 2012. Timestamps of the temperature and computer power consumption measurement data from the sensor set are then analyzed based on the algorithm presented in the method section. This study uses four parameters to characterize occupants' presence in the work space: starting time, leaving time, total number of hours per day an individual spends at the work place, and the average daily occupancy profile.

Figure 2 shows a histogram of starting time and leaving time of all participants in the pilot office building throughout the experiment. On average, people come to work around 9:30AM with a standard deviation of 1.5 hours. And they leave around 5:30PM with a standard deviation of 1.8 hours. And they spend an average of 8.14 hours in the work place (i.e. the office building) with a standard deviation of 2.3 hours (Figure 3). Since our algorithm is not able to detect lunch break or other short break in the middle of the day if any (unless people actually turn off their computer while out for lunch or field meeting in the middle of the day), the total occupancy hour at a given day might have been overestimated in this study.

Figure 4 shows the resulting average 24-hr occupancy profile. The average weekday occupancy profile leads to a classic bell curve with a peak occupancy rate of 37%, taking place right before lunch hours. As discussed before, the proposed algorithm is not able to detect small absence in the middle of the day unless computers are turned off. And yet the lunch break is captured in the average 24-hr occupancy profile. Closer examination of the raw data show that lots of experiment participants did turn off their computers while out for lunch. And this conscious mind of energy conservation could be the result of the sensor's intended functionality (Milenkovic, Hanebutte et al. 2013). The average 24-hr occupancy profile varies between weekdays, with a lower occupancy observed on Fridays as compared to other weekdays.

Figure 5 shows the average 24-hr occupancy profiles categorized by business departments within the participating organization and the profession of each participant. Results show that there is a significant difference in the occupancy diversity factor between different departments, with peak values varying as much as 69% between the management and global departments and 16.5% between the sales and service departments. The numeric number in the bracket behind each category legend in Figure 5 represents the total number of participants from that corresponding category (either by department or by profession) and doesn't represent the exact number of samples used in the average calculation. For instance, the MGMT department's 24-hr average occupancy profile is the average of 58 experimental daily occupancy rate profile that is estimated based on the occupancy status of 3 volunteer participants. Results also show slight variation in occupancy characteristic in terms of the starting and ending times among different professions and different business departments. For instance, the administrative personnel appears to have the best punctuality as compared to engineers and sales people, suggested by their nearly rectangular occupancy profile curve. And more engineers come early and leave late as indicated by the wider variation in the occupancy profile curve. Since each sensor set is deployed to a specific work space, the corresponding sensor data are also accompanied with a spatial tag, such as floors or thermal zones. Though not reported in this particular paper, this zone based occupancy information, together with local ambient environment conditions measured by the physical sensors, could be extremely valuable for any occupancy-centered zone based climate control.

Figure 6a shows the actual occupancy rate at 15 minute intervals for the office building throughout the entire experiment period. The value at each time interval can also be thought as the "real-time" occupancy, where the occupancy rate at each interval represents the percentage of people actually present in the work place (i.e. the office building). The sensor data show a clear distinction in occupancy between weekdays and weekends (and a holiday in the 3rd week of August, which has near zero occupancy). The observed peak occupancy is about 70%. And Figure 6a also shows a clear pattern of occupancy from one week to another, with exceptions to the four weeks in late July and early August. Likely due to the fact that families tend to take vacations in summer, there are four weeks from the late July to the 3rd week of August where a considerably lower occupancy is observed in the pilot office building. The specific reason why considerably lower-occupancy is observed during those four weeks is unknown and might even have something to do with company culture and management. Figure 6b shows a comparison of average 24-hr weekday occupancy profiles among the entire experiment period, "typical" weekdays, the transition period, and the vacation period. The peak occupancy shifts from 37% in overall average to 46% in "typical" weeks, and to 22% in "vacation" weeks. These measurable gaps indicate that current practice of simply classifying occupancy profiles (or occupancy schedules in building simulation) into weekdays and weekends might not be sufficient in terms of capturing occupancy variation throughout the year. A "holiday affected" occupancy profile might be needed for the extended weeks when a lower than normal occupancy takes place due to either vacations or holidays.

DISCUSSION

As briefly addressed before, the algorithm proposed in this study is not able to capture short absence that happens in the middle of the day, such as lunch leave, client visit, and etc., unless users/occupants always turn off their computers every time they leave the office. As a result of that, the daily occupancy this study presents might be slightly higher than the actual one.

In the meantime, this IT platform-integrated sensor set was originally developed to provide highly personalized occupant experience in built environment and was not intended for occupancy detection in the beginning. Thus information on the actual occupancy in both pilot offices was not collected in the original experiments. As a result of that, we are not able to validate and evaluate the accuracy of the proposed algorithm in estimating real-time occupancy at this time. But we have qualitatively compared results from this study to those reported from recent occupancy studies (General Services Administration 2009; Duarte, Van Den Wymelenberg et al. 2013) and found that the estimated occupancy rate from this study agrees well with what have been observed in other studies. A rigid validation shall be planned in the next round of deployment study.

CONCLUSION

This study presents an implicit method to estimate real-time occupancy in office buildings, based on a creative way of analyzing timestamps of data collected by IT platform-integrated sensor sets. Although the sensor itself is still at its prototyping stage and not intentionally developed for occupancy detection purpose, results from pilot studies demonstrate its promising functionality in capturing dynamic occupancy variation in work environment. This sensor could be easily deployed in both single occupancy space and open. Since it is already integrated with the existing IT platform, this sensor could provide a non-invasive way to collect occupancy data in an office environment without additional first costs. Pilot experiment results show that the average peak occupancy rate is less than 40% and the daily occupancy profile varies among different weekdays, across different (business) departments and different professions. The resulting occupancy rate varies considerably from what has been typically used in current design practice, which suggests that an up-to-date occupancy profile might be in need for building design community.

ACKNOWLEDGEMENTS

The authors would like to thank Intel Corporation for the financial support to this study and the Intel POEM development team for providing access to the pilot experiment data. Special thanks go to Catherine Huang, Ulf Hanebutte, and Milan Milenkovic from the Intel Labs for providing invaluable insights and inputs throughout the project.

REFERENCES

Abushakra, B. and D. E. Claridge (2008). "Modeling Office Building Occupancy in Hourly Data-Driven and Detailed Energy Simulation Programs." ASHRAE Transactions 114(2): 472.

Davis, J. A. and D. W. Nutter (2010). "Occupancy diversity factors for common university building types." Energy and buildings 42(9): 1543-1551.

Duarte, C., K. Van Den Wymelenberg and C. Rieger (2013). "Revealing occupancy patterns in an office building through the use of occupancy sensor data." Energy and buildings 67(0): 587-595.

General Services Administration (2009). The new federal workplace. Washington, D.C.

Keith, D. M. and M. Krarti (1999). "Simplified Prediction Tool for Peak Occupancy Rate in Office Buildings." Journal of the Illuminating Engineering Society 28(1): 43-56.

Mahdavi, A. (2009). "Patterns and Implications of User Control Actions in Buildings." Indoor and Built Environment 18(5): 440-446.

Melfi, R., B. Rosenblum, B. Nordman and K. Christensen (2011). Measuring building occupancy using existing network infrastructure. International Green Computing Conference and Workshops (IGCC), Orlando, FL, IEEE.

Milenkovic, M., T. Dang, U. Hanebutte and C. Huang (2013). Platform-integrated Sensors and Personalized Sensing in Smart Buildings. Proceedings of SensorNets 2013, Barcelona Spain.

Milenkovic, M., U. Hanebutte and U. Hanebutte (2011). Demo Abstract: POEM--A User-Centric Approach to Energy Efficiency in Office Buildings. BuildSys 2011: the 3rd ACM workshop on embeded sensing systems for energy efficiency in buildings. Seattle, WA.

Milenkovic, M., U. Hanebutte, Y. Huang, D. Prendergast and H. Pham (2013). Improving user comfort and office energy efficiency with POEM (personal office energy monitor). CHI '13 Extended Abstracts on Human Factors in Computing Systems. Paris, France, ACM, New York, USA: 1455-1460.

Newsham, G. R., S. Mancini and B. J. Birt (2009). "Do LEED-certified buildings save energy? Yes, but...." Energy and Buildings 41(8): 897-905.

Rubinstein, F. M., N. Colak, J. D. Jennings and D. Neils (2003). Analyzing Occupancy Profiles from a Lighting Controls Field Study. CIE Session 2003, San Diego, CA.

Scofield, J. H. (2009). "Do LEED-certified buildings save energy? Not really.." Energy and Buildings 41(12): 1386-1390.

Torcellini, P. A., M. Deru, B. Griffith, N. Long, S. Pless, R. Judkoff, et al. (2006). Lessons learned from field evaluation of six high-performance buildings, National Renewable Energy Laboratory.

Huafen Hu, PhD

Associate Member

Chad Miller

Huafen Hu is an assistant professor, and Chad Miller is a research assistant in the Department of Mechanical Engineering at Portland State University in Portland, OR.

Figure 1: An example scenario from 7AM to 7PM on a weekday with
the resulting binary occupancy status (1 for presence and 0 for
absence) estimated based on the proposed algorithm. This example
scenario could represent someone who has to come early for a
teleconference in the office and leaves early in the afternoon.

           Temperature     Computer      Occupancy
             Sensor        Activity

 7:00 AM        0              0             0
 7:30 AM        0         1 (Active)         1
 8:00 AM        0          1 (Sleep)         1
 9:00 AM        1         1 (Active)         1
10:00 AM        1         1 (Active)         1
11:00 AM        1         1 (Active)         1
12:00 AM        1        1 (Long Idle)       1
 1:00 PM        1        1 (Long Idle)       1
 2:00 PM        1         1 (Active)         1
 3:00 PM        1         1 (Active)         1
 4:00 PM        1         1 (Active)         1
 5:00 PM        1        1 (Long Idle)       0
 6:00 PM        1        1 (Long Idle)       0
 7:00 PM        0        1 (Long Idle)       0
COPYRIGHT 2014 American Society of Heating, Refrigerating, and Air-Conditioning Engineers, Inc. (ASHRAE)
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Hu, Huafen; Miller, Chad
Publication:ASHRAE Transactions
Article Type:Report
Date:Jul 1, 2014
Words:3604
Previous Article:A hand calculation method of smoke movement through a high-rise airtight single shaft.
Next Article:Wind-driven airflow through building openings: preliminary experimental results using particle image velocimetry.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters