Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus
(USC Thesis Other)
Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
i Exploring Participatory Sensing and the Internet of Things to Evaluate Temperature Setpoint Policy and Potential of Overheating/Overcooling of Spaces on the USC Campus by Zhengao Dong Presented to the FACULTY OF THE SCHOOL OF ARCHITECTURE UNIVERSITY OF SOUTHERN CALIFORNIA In partial fulfillment of the Requirements of degree MASTER OF BUILDING SCIENCE AUGUST 2019 ii ACKNOWLEDGEMENTS At first, I would like to express my sincere thanks to professor Kyle Konis, who spent lots of time on revising my thesis and giving me both technical and mental support. I would also thank professor Douglas Noble and Joon-Ho Choi for constructive suggestions and literary instruction. Moreover, the TrojanSense project team provided the basic data for the analysis during my thesis process. Many thanks to all team members: Kyle Konis (Faculty Lead), Naman Kedia (Developer and Designer), Simon Blessenohl (Advisor), Shuhui He (Software Developer), Danyang Zhang (Building Energy Modeler), Vishal Rahane (Software Developer), Ahsan Zaman (Software Developer). At the end, many thanks to my parents and friends who always support me during my 2-years study in a foreign land. I would not finish this thesis without their emotional support. Thanks to all the people who helped me during the thesis. iii COMMITTEE MEMBERS Kyle Konis, Ph.D., AIA Assistant Professor USC School of Architecture kkonis@usc.edu Douglas E. Noble, PH.D., FAIA Professor USC School of Architecture dnoble@usc.edu Joon-Ho Choi, Ph.D. Assistant Professor USC School of Architecture joonhoch@usc.edu iv ABSTRACT University of Southern California (USC) has more than 450 buildings, which consume a lot of energy. Among the energy consumption categories, HVAC system accounts for the largest portion. In order to meet USC Sustainability 2020 Goals, people need to find ways to save energy for university buildings. However, there is no feedback mechanism to collect real-time thermal information of every study space and occupants’ preferences. Nowadays, the school tends to set uniform HVAC standards for all rooms on campus. But every space has different operational characteristics, systems control, and architectural features, which leads to uneven thermal distribution and thermal discomfort. Additionally, no effective procedures and advanced technology for occupants (mostly are students) to inform HVAC management. Most schools use “work order” system to fix thermal issues, but this system needs several days to respond and not accessible for most students. Moreover, bad thermal conditions drive students to get out of rooms, and some classrooms may keep at a very low temperature even though there are no occupants. As a result, more energy can be wasted. Many previous studies have suggested that personalized (room-level or person-level) comfort models combined with occupants’ feedback can optimal thermal comfort satisfaction and energy efficiency. However, these studies were limited to single rooms or individuals. There were few kinds of research employing at a campus level. This study developed wireless temperature & humidity sensor nodes based on Internet of Things (IoT) technique. Powered by batteries, these IoT sensor nodes are portable so they could be installed in different spaces. By installing these sensors in 18 representative study spaces in the Watt Hall on the USC campus, people can get the real-time thermal condition around the building. Moreover, occupants in each space can submit their preferences through the TrojanSense App. The research collected the indoor temperature and humidity for 18 spaces in the Watt Hall with the IoT sensors and collected about 177 valid occupants’ feedback with the TrojanSense APP. After analysis, it verified that many spaces were overcooled in November, and most occupants in these spaces wanted to be 1-3 degrees warmer than the current HVAC set points. The results could be used to guide the future decision-making and improve the current temperature policies. The occupants in these spaces would have a better thermal comfort with a more rational decision, and the overall energy use efficiency would be improved as well. KEY WORDS: Internet of Things (IoT), Thermal Comfort, Temperature Setpoint Policy, Overcooling v TABLE OF CONTENTS ACKNOWLEDGEMENTS ......................................................................................................... ii COMMITTEE MEMBERS ........................................................................................................ iii ABSTRACT .................................................................................................................................. iv 1. INTRODUCTION..................................................................................................................... 1 1.1 Problems ............................................................................................................................................. 1 1.2 Goals ................................................................................................................................................... 2 1.3 Thermal Comfort ................................................................................................................................ 3 1.3.1 Temperature Policies.................................................................................................................... 4 1.3.2 Air Temperature Setpoints ........................................................................................................... 5 1.3.3 HOBO Test .................................................................................................................................. 6 1.3.4 Temperature Policy vs Observed Measurements ......................................................................... 8 1.4 TrojanSense......................................................................................................................................... 8 1.4.1 Internet of Things (IoT) ............................................................................................................... 9 1.4.2 IoT Sensors .................................................................................................................................. 9 1.5 Objectives ......................................................................................................................................... 10 1.6 Summary ........................................................................................................................................... 10 2. BACKGROUND AND LITERATURE REVIEW .............................................................. 11 2.1 Comfort Evaluation ........................................................................................................................... 11 2.2 Living Lab ......................................................................................................................................... 11 2.2.1 Example 1: UCLA ..................................................................................................................... 11 2.2.2 Example 2: UC DAVIS.............................................................................................................. 12 2.3 Personalized Comfort Model ............................................................................................................ 12 2.3.1 Definition ................................................................................................................................... 12 2.3.2 Current Studies ........................................................................................................................... 12 2.4 Modeling Frame ................................................................................................................................ 13 2.4.1 Data Collection .......................................................................................................................... 13 2.4.2 Data Selection ............................................................................................................................ 14 2.4.3 Data Visualization ...................................................................................................................... 14 2.4.4 Model Selection ......................................................................................................................... 14 2.4.5 Model Evaluation ....................................................................................................................... 14 2.4.6 Continuous Learning .................................................................................................................. 14 2.4.7 Case Study ................................................................................................................................. 15 2.5 Summary ........................................................................................................................................... 15 3. METHODOLOGY ................................................................................................................. 16 3.1 Methodology Diagram ...................................................................................................................... 16 3.2 IoT Sensors ....................................................................................................................................... 17 3.2.1 ESP32 Board .............................................................................................................................. 18 3.2.2 DHT22 Sensor ........................................................................................................................... 18 3.2.3 Arduino ...................................................................................................................................... 18 3.2.4 MQTT ........................................................................................................................................ 19 3.2.5 Adafruit IO ................................................................................................................................. 19 vi 3.3 Installation......................................................................................................................................... 21 3.4 Collection .......................................................................................................................................... 24 3.5 Feedback ........................................................................................................................................... 24 3.6 Data Analysis .................................................................................................................................... 24 3.7 Assessment ........................................................................................................................................ 25 3.8 Summary ........................................................................................................................................... 25 4. RESULTS ................................................................................................................................ 26 4.1 Research Questions ........................................................................................................................... 26 4.2 Basic Analysis ................................................................................................................................... 26 4.2.1 Central AC System Spaces ........................................................................................................ 27 4.2.2 Independent AC Unit Spaces ..................................................................................................... 48 4.2.3 Outdoor Spaces .......................................................................................................................... 51 4.3 Comprehensive Analysis .................................................................................................................. 54 4.3.1 Occupants’ Feedback ................................................................................................................. 55 4.3.2 MBS Corner ............................................................................................................................... 55 4.4 Outdoor Temperature ........................................................................................................................ 58 4.5 Summary ........................................................................................................................................... 82 5. DISCUSSION .......................................................................................................................... 84 5.1 Significance of the Thesis ................................................................................................................. 84 5.2 Evaluation of the Workflow .............................................................................................................. 85 5.3 Validation .......................................................................................................................................... 85 5.4 Limitations ........................................................................................................................................ 85 5.4.1 Insufficient Participation ............................................................................................................ 86 5.4.2 Limited Battery Life................................................................................................................... 86 5.4.3 WiFi Accident ............................................................................................................................ 86 5.4.4 Contradictory Feedback ............................................................................................................. 86 5.5 Summary ........................................................................................................................................... 87 6. CONCLUSION ....................................................................................................................... 88 6.1 Improvement to the current workflow .............................................................................................. 88 6.2 Future work ....................................................................................................................................... 88 6.2.1 Consider More Factors ............................................................................................................... 88 6.2.2 Energy Impact ............................................................................................................................ 88 6.2.3 Man-Machine Interaction ........................................................................................................... 88 6.2.4 Machine Learning ...................................................................................................................... 88 6.3 Summary ........................................................................................................................................... 88 REFERENCES ............................................................................................................................ 90 APPENDIX .................................................................................................................................. 92 1 1. INTRODUCTION As more and more environmental challenges have emerged, like climate change, exploding population growth, and the loss of biodiversity, environmental sustainability has become one of the most critical challenges for humanity in the 21 st century. Sustainability was first presented in the 1987 U.N. Brundtland Commission Report, Our Common Future (Barnaby, 1987), as “development that meets the needs of the present without compromising the ability of future generations to meet their own needs”. Nowadays, sustainability has a more comprehensive definition, which is “a process of maintaining energy and sources in a balanced range within a specific area, any change to this kind of balance is in harmony and make sure both current and future human needs and requirements” (“What is Sustainability”, n.d.). The National Environmental Policy Act of 1969 declared sustainability is a national policy. As a densely populated location, campus of a university should take sustainability into account at the local, regional, and global levels during the development process. Today, most universities, like Harvard, MIT, Stanford, and so on, have set Sustainability Office to investigate innovative solutions or high-end tools to save energy. Moreover, each institution invests lots of sustainability innovation fund to support relevant research study every year. All of these encouraged some significant achievement. For example, Harvard’s students developed NASA Flight instruments to study ozone depletion (Adam, 2017); UCLA’s students designed a new way by using wireless technology to speed the transition to electric vehicles (Judy, 2011). Similarly, the project is eventually desired to support the development of campus sustainability. 1.1 Problems The University of Southern California (USC) is a private research university in Los Angeles, California. It was founded in 1880 and kept the longest history of independent research in California (USC, 2013). USC has about 43,000 students in total, including 19,000 undergraduate students and 24,000 graduate students (USC, 2008). The University Park is the main campus of USC. It is in the University Park district, 2 miles southwest of Downtown Los Angeles. The campus' boundaries are Jefferson Boulevard on the north and northeast, Figueroa Street on the southeast, Exposition Boulevard on the south, and Vermont Avenue on the west. The whole campus has roughly 450 buildings. Most buildings are in the Romanesque Revival style, which is mainly red-brick building. As a school with more than 1000 classrooms in 450 buildings, USC consumes a lot of energy every year. Carbon emissions result from electricity consumption of campus heating impair the development of sustainability. According to USC Greenhouse Gas Emissions Inventory Summary (The Unlimited Carbon Assistance Network [UCAN], 2018), USC emitted 324,566 metric tons of carbon dioxide equivalent in its baseline year 2014. 10.47% of those emissions were “Scope 1” emissions, which means the greenhouse gas directly generated from burning fuels or using chemicals on the USC campus. Another 41.70% were “scope 2” emissions; these results from generating the electricity used on campus. The largest category of emissions—47.83% of the USC footprint—are the “Scope 3” emissions that are the indirect emission of campus operations; for example, the emissions that caused by student, staff, and faculty commuting, and business travel. 2 Table 1. USC Greenhouse Gas Emissions Inventory Summary Greenhouse Gas (GHG) performance is measured in emissions per square foot of built space. There is a gap between USC and other peer institutions on addressing environmental sustainability. When only considering about Scope 1 and Scope 2 emissions, which is used to gauge the combination of campus density, energy efficiency, and carbon intensiveness of campus fuel supple, USC performed well with a measure of 6.82 metric tons carbon dioxide equivalent (MTCDE) per gross square foot (GSF) compared to the national average of 13 MTCDE/GSF. But when incorporating Scope 3 into consideration, USC has 7.28 MTCDE per “full-time equivalent” student (FTE), a little bit higher than the national average value of 7.1 MTCDE/FTE. Figure 1. USC Greenhouse Gas Performance 1.2 Goals USC’s Mission— “the development of human beings and society as a whole through the cultivation and enrichment of the human mind and spirit” (“Mission Statement”, n.d.)—encourages all staff, faculty, students, and other stakeholders to develop campus-wide sustainable strategies. Therefore, the Provost has identified “security and sustainability” as one of four key initiatives of USC. The Sustainability Steering Committee (SSC), consisting of faculty, staff, and students from across the university, declared a Sustainability 2020 Plan in late 2015, which is USC’s first campus-wide sustainability plan (USC, 2015). The Sustainability 2020 Plan launched by the SSC set a series of goals in seven core categories of sustainability, including Education and Research, Engagement, Energy Conservation & Greenhouse Gas Mitigation, Transportation, Procurement, Waste Diversion, and Water Conservation. These goals are based on a comprehensive evaluation of USC’s capabilities and the STARS system. Besides, committees also evaluated baseline metrics and case studies, as well as excellent practices from comparable peer institutions. Setting a meaningful goal and ensuring it is achievable for USC in the near future is the most complicated part for committee members. 3 In terms of Energy Conservation & Greenhouse Gas Mitigation. The main goal is to reduce 20% of greenhouse gas emissions per square foot compared to 2014 by 2020. In order to achieve this goal, USC has encouraged several projects to reduce costs and improve energy efficiency since 2015. For example, the thermal energy storage system composed of an underground tank holding 3 million gallons of chilled water under the Cromwell Field (“Thermal Energy Storage”, n.d.). Chillers run at night to cool the water at night when electricity is cheapest, and chilled water from the tank could be used during daytime hours when electricity is more expensive. Several sustainable projects like this help USC to achieve the 20% reduction goal ahead of original schedule, and 47% greenhouse gas emission is cut down compared to 2014 level. Moreover, there is an opportunity to reduce more considering about other projected and ongoing projects. Although significant progress has been made, USC needs to do more to keep moving forward on the road to a sustainable future. Figure 2. USC Sustainability 2020 Goals (“USC Sustainability Report”, n.d.) 1.3 Thermal Comfort According to ANSI/ASHRAE Standard 55 (2013), thermal comfort is “the condition of mind that expresses satisfaction with the thermal environment and is assessed by subjective evaluation.” As we all know, human’s body will absorb heat from input food, and release heat during exercise. In general, this kind of heat transferring will stay in a stable range to maintain a body’s normal vital state. However, people may get or lose more heat since the surrounding environment changes. Human bodies tend to lose more heat in cold weather and can gain too much heat in hot weather. Both the too hot and too cold environments may lead to discomfort, so the thermal environment around people affect their satisfaction. Nowadays, it is generally accepted that a range of temperatures around 68 to 72 degrees Fahrenheit, but this may vary greatly depending on different factors. 4 In addition to air temperature, there are many factors that affect thermal comfort, including relative humidity, air speed, metabolic rate, clothing insulation, and activity level. The Predicted Mean Vote (PMV) models is generally accepted to evaluate the level of thermal comfort (Ole, 1970). It was developed by collecting data in a controlled climate chamber and it is an empirical model according to a large group of people’s sensations of thermal comfort. This model assumes people under the constant conditions, like metabolic rate and clothing. However, since the actual conditions are slightly different, standards give a small range of PMV for recommended comfort zone. Therefore, even though the USC consumes a lot of energy, it may not always transfer to a high level of occupants’ comfort in the buildings. Thermal comfort is a critical goal for the built environment since it affects people’s thermal satisfaction, health, and work productivity. In order to better understand the thermal comfort at USC, people need to know the temperature policies of the school. There are a variety of heating and cooling systems on campus, but they are all regulated by a central control system in the Facilities Management Services (FMS). Basically, most schools make their own rule about room temperature policies, including the setpoints of the HVAC systems in summer and winter separately. 1.3.1 Temperature Policies Temperature policies are rules that provide a framework to assist building managers and occupants for the sake of building a healthy, productive, and safe working environment. Usually, they will be divided into summer policies and winter policies. Temperature policies decide the running time, setpoints, operation and maintenance of the HAVC system, etc. They were developed and adopted by the stakeholders and facilities leaders representing all of the schools and administrative departments. The FMS in school should periodically evaluate and discuss their temperature policies based on improvements to building systems, advances in occupant comfort and productivity research, and feedback from occupants. In most universities, students and faculties are also welcomed to take part in the adjustment of the temperature policies. The graph below shows the comparison of HVAC setpoints between several universities. Table 2. Temperature Policies Between Universities University Heating Season (°F) Cooling Season (°F) Harvard University 68-71 74-76 Columbia University 66-72 74-78 Bentley University 68.5-75 75-80 Tufts University 68 78 Rice University 68-72 74-78 USC 68 76 As can be seen from the graph, most universities have the setpoint around 68-degree Fahrenheit in the heating season, and 76-degree Fahrenheit in the cooling season. However, most schools have a small range of temperature for people to adjust except the Tufts University and the USC, who have stable setpoints in both seasons. Taking Harvard University as an example, their space temperature ranges are based on established standards for human comfort, productivity, and safety. Actual space temperatures may vary in the range across Harvard’s buildings due to the wide range of space types and building control systems on campus. Maximum comfort and efficiency are the only goals when determining the specific targets in each building. For some locations that require specific temperature ranges, like research laboratories or healthcare rooms, building managers even can decide the setpoints below or over the range of recommended temperature (Harvard University Sustainability, n.d.). On the contrary, the temperature in many campus buildings is regulated by a central control system in the USC. The goal for all centrally controlled facilities is to heat spaces to 68 degrees in the heating seasons and cool spaces to 76 degrees in the cooling season when occupied (USC Facilities Management Services [FMS], n.d.). The temperature policies do not consider the different space types and buildings’ condition. Since different buildings have the different built year, material, structure, orientation and so on, the same HVAC setpoints may cause overcooling or overheating in some spaces. In fact, some students have complained about the overcooled temperature in the Doheny Memorial 5 Library, which has a three-stories height space. Although the FMS in the USC has the hotline for students and faculties to report an overheating or overcooling problem, the hotline is not easy to get in since a large volume of requests and calls. Even though occupants successfully reported their adjustable suggestion, the HVAC system may need more time to adjust the actual room temperature. In some large spaces like the Doheny Library, this kind of temperature delay would be longer. 1.3.2 Air Temperature Setpoints The HVAC setpoint is the temperature that the machine desired to achieve, and it could be accurate to single digit before the decimal point. Although one- or two-degrees’ change on the setpoint will not be noticeable for people, it can affect energy consumption much. According to Hoyt, Arens, and Zhang (2014), if the setpoint range is among the desired temperature range, then there is no need to heat or cool the space anymore, which means saving more energy. They also quantified the energy consumption based on different setpoint ranges in seven ASHRAE climate zones separately. Table 3. Energy Saving with Cooling Setpoint Table 4. Energy Saving with Heating Setpoint Hoyt and his colleagues found the average cooling HVAC energy savings can achieve 29% when increasing the cooling setpoint from 72 degrees to 77 degrees and reducing the heating setpoint from 70 degrees to 68 degrees can save an average of 34% of heating energy. Based on their study, the Center of the Built Environment in Berkeley developed a Setpoint Savings Calculator to calculate HVAC consumption between different setpoint ranges. Basically, 6 on the premise of occupants’ thermal comfort, expanding the modified temperature setpoint range as big as much can save more HVAC energy consumption. Figure 3. HVAC Savings Calculation 1.3.3 HOBO Test In order to better understand the thermal comfort in different spaces, three temperature data-logging studies using HOBOs on the USC campus were performed. A HOBO is a small portable temperature & humidity sensor, which runs with built-in batteries and controlled by a computer. The tiny volume allows it to be installed in corners around the campus and does not attract people’s attention. A HOBO can detect the temperature and humidity around it and record the data in its memory. The detection time could be adjusted, and each launch could last for months. Figure 4. HOBO 9 main libraries were chosen on campus and did the HOBO test in April, June, and August separately. In the test, the detection interval was set to 5 mins, and all HOBOs were running the same time. Basically, HOBOs were placed at the people-height level in the middle area of detecting spaces, and somewhere people will not aware of would be better. Since most libraries open from 9 am to 5 pm, detected data was filtered to get the final graphs below. 7 Figure 5. Temperature Comparison in April Figure 6. Temperature Comparison in June 8 Figure 7. Temperature Comparison in August In the box and whisker plots, the ends of the box mean the upper and lower quartiles, so the box covers the interquartile range. The whiskers are lines on the top or bottom outside of the box that extend to the highest lowest value. The median is also marked by a line inside the box. While the points outside of the extreme values mean abnormal values, which may be filtrated in technical analysis. Box and whisker plot is very useful for analyzing a large number of data and when several data sets are being compared. As we can see from the graphs, different libraries have different temperature ranges. Even though most of them have a temperature from 70 degrees to 74 degrees, there are a few exceptions people should aware. For example, Doheny Library is cooler than other libraries both in April and June; while the Gerontology Library is much hotter in August. Besides, many abnormal values, which do not obey the box and whisker rule, may affect FMS’s judgment in real life. 1.3.4 Temperature Policy vs Observed Measurements Comparing the USC temperature policies and actual room temperature, it could be found that the measured actual air temperatures do not match with stated policies. Although the FMS encourages students and faculties to give feedback when they feel uncomfortable, unpredictable abnormal values may affect the actual effect. The HVAC controller need not only a real-time monitoring system in each room, but also an intact feedback system to build data-driven thermal comfort models, which allow people to adjust the room temperature in advance. IoT sensors and TrojanSense are two tools that can achieve those goals. 1.4 TrojanSense The TrojanSense is a such kind of APP that is trying to collect people’s feedback and use them to improve energy use efficiency. Through the same-name app and associated website, occupants can choose the way they want to submit feedback either by IOS or PC. For most popular study spaces, the app can detect the room through a preset Bluetooth transmitter. But other spaces on campus can manually type the building’s name and room number and submit feedback as well. It is also possible to improve the data type by adding the Internet of Things (IoT) devices which were used to collect real-time indoor temperature and humidity. 9 This group project involves eight people so far. Professor Kyle Konis is the Faculty Leader, who lead the whole group and give professional advices; Simon Blessenohl is the Advisor, and he helped to design the business process and basic function; Naman Kedia, Shuhui He, Vishal Rahane, and Ahsan Zaman work as the Software Developer, and they all in charge for coding and back-end compiling; Danyang Zhang, who built three-dimension building model to verify research result, is the Building Energy Modeler; as for the author, he acted as the IoT Sensor Developer in the group. Figure 8. TrojanSense Website 1.4.1 Internet of Things (IoT) IoT is the network of physical devices, vehicles, home appliance, and other items embedded with electronics, software, sensors, and connectivity which enables physical items to connect, collect and exchange data. IoT creates opportunities for more direct integration of the physical world into computer-based systems and results in efficiency improvements, economic benefits, and reduced human exertions (Gershenfeld, Krikorian, & Cohen, 2004). 1.4.2 IoT Sensors IoT sensor is a temperature and humidity sensor based on IoT techniques. It is composed of a small electric board, a temperature sensor, and a battery box. The electric board has many welded circuits and ports, which can connect other sensors, lights, motors, or actuators. Basically, the board supports USB access to receive a signal from computers, then it will transfer and control other components according to the signal. The whole IoT sensor is as big as a palm, so it can be placed in every corner around the campus while people will not aware. In addition, the board used in this paper has Wi-Fi function, so it can send data to the cloud through the campus-wide free Wi-Fi. As a result, people can check the real-time temperature and humidity by accessing the cloud via any internet-access devices. With TrojanSense and IoT sensors, people can provide feedback indicating whether they want the temperature to be adjusted. In total 13 grades including warmer, cooler, or stay could be chose. Their subjective feedback will be paired with physical measurements from IoT sensors and used to develop data-driven comfort models. These models will be shared with the FMS to help inform adjustments to temperature setpoint ranges that reflect the preferences of USC’s diverse population. In addition to the potential to improve thermal comfort outcomes, adjustments that bring indoor temperatures closer to outdoor temperatures will help reduce the carbon emissions generated by our campus Heating, Ventilation, and Air Conditioning (HAVC) systems. 10 1.5 Objectives The research tried to develop a series of IoT sensors that can remotely detect real-time temperature and humidity. At the same time, the TrojanSense APP can collect occupants’ feedback about the indoor thermal condition. After comparing and analyzing the physical thermal data and people’s first-hand feelings, it can be used to evaluate the current temperature policies and be decision-making reference for FMS staff when they are trying to adjust the HVAC system. As a result, this personalized detecting not only make sure thermal satisfaction in different types of spaces but also improve the energy efficiency by avoiding overheating or overcooling. In the end, this project can promote the development of USC sustainability and help to build a more sustainable campus. 1.6 Summary The USC campus is a moderately densely built area. Hundreds of buildings in USC consume a large amount of energy. However, high energy consumption does not lead to high thermal comfort in many rooms. Because of a lack of appropriate data and means to analyze and interpret, the FMS staff cannot dynamically adjust HVAC system. So, they preset constant setpoints for the whole campus, which was proved to cause overheating or overcooling in some spaces and could waste a lot of energy. The research developed some IoT sensors to collect real-time room temperature, paired with occupants’ feedback from the TrojanSense APP, it can help the FMS personnel to improve their decision making as well as the school’s sustainability. 11 2. BACKGROUND AND LITERATURE REVIEW The goal of this research is to improve the thermal comfort in different campus spaces and energy efficiency. In order to understand how people can evaluate thermal comfort, several relative literature review would provide some ideas. As for the location of experiment, a building could work as a “living lab” to provide live data. Moreover, a series of steps are needed for the sake of building personalized comfort models, including data collection, data selection, data visualization, model selection, model evaluation, and continuous learning. Key findings from literature review include: • All thermal comfort evaluations need detailed data of the thermal environment; • When introducing rational indices into the thermal comfort analysis, the thermal comfort zone would be wider; • Personalized comfort models combined with occupants’ feedback can optimal thermal comfort satisfaction and energy efficiency; • Getting access to the building information with the Graphic User Interface can help building managers, occupants, and stakeholders make decision or improve energy-saving awareness; • Personalized comfort model should adapt to different weather conditions. 2.1 Comfort Evaluation Many previous researchers have proposed methods to evaluate thermal comfort. Olesen (2004) has given an overview: all thermal comfort evaluations need detailed data of the thermal environment. The data includes temperature (air, radiant, surface), relative humidity, air velocity, clothing, and activity type. Numeric data can be detected by installed sensors, but clothing and activity type can only be gotten through survey. Except field data, researchers also need to gather thermal response of occupants. This is usually measured by asking a choice between the ASHRAE thermal satisfaction scales, which use -3 to 3 to represent thermal satisfaction. The number is smaller, it means the person feel cooler, while bigger number indicates warmer. Then, choosing a proper statistical method to link thermal environment data with the responses together. That is the thermal comfort model of the investigative space. A guiding thermal comfort zone could be concluded around the states that are considered as neutral in the model. Finally, the thermal comfort model and can be used to predict the thermal status in the future and be introduced to other similar situations. However, the thermal environment is not constant. So, it is hard to explain some variables in the field data. What is worse, possible errors along with the chosen statistical method may cause severe inaccuracy on the final results. ISO 7730 (1994) and ASHRAE 55 are the most common standards in USA right now. Both of them are based on the occupants’ responses measured in a stable environment. When researchers introduce rational indices into the analysis, the thermal comfort zone would be wider. As a result, Nicol and Humphreys (1973) suggested that this conclusion could be used in the internal adaptation of occupants and the influence of their behavior towards the thermal conditions. Nicol and Humphreys also explored the relationship between PMV and subjective response. The conclusion is that the prediction is not affected much by the thermal environment on the field. Thermal comfort model based on one variable without detecting the environment is similar to using the PMV with all thermal data. They also suggest that using less sensors to predict the thermal comfort can save the cost and achieve the same goal (Altherr & Gay, 2002). 2.2 Living Lab Building Living Laboratories is an effective method to study thermal comfort. According to Almirall and Wareham (2011), “a Living Laboratory is a user-centered, open-innovation ecosystem, often operating in a territorial context (e.g. campus, city, region), integrating concurrent research and innovation processes within a public-private-people partnership”. This approach allows all involved stakeholders to test the performance of a product or service and its potential impact on users in a specific area. University campus is a representative example of Living Laboratory where all staff, faculty, and students do coursework and research using the physical campus as a testing ground for new sustainable concepts and technologies. Any problems during this process would timely appear and in turn help stakeholders to improve. 2.2.1 Example 1: UCLA UCLA is a representative university that investigates living laboratory for sustainability research. The Institute of the Environment and Sustainability Action Research Course in the school brings about 70 undergraduate students each 12 year to take part in sustainability research. Students would be paired with campus stakeholders to do hands on study on campus on topics from reducing emissions to saving water (“Living Laboratory Projects”, n.d.). For example, UCLA researchers found environmental concerns can motivate people to save energy instead of money (Hewltt, 2015). In order to find out the motivation that encourage people cut down electricity, the UCLA project, called Engage, installed smart-metering systems for 118 apartments at the campus’s University Village, which provide housing for graduate students and their families. The researchers divided participants into two groups. Except the basic weekly emails talking about how much energy they used, one group also was told how much more they paid for energy that their neighbor, while the other group was informed how many more pounds of air pollution, they were creating than the neighbor. The trial period ran from October 2011 to March 2012. During the six-month trial, the team also created a website that enable participants to track historical and real-time electricity use. The result shows that environment-focused messages caused people to cut an average of 8 percent of energy use. On the other hand, the cost-focused messages were less effective on electricity reduction of only 2 percent. Although that in part because electricity is relatively inexpensive, and occupants are overall well-educated, people are still worried more about environmental data since it involves air pollution and the risk for diseases. 2.2.2 Example 2: UC DAVIS UC Davis is another typical university on sustainability which was listed as “Gold” in STARS system. The Energy Conservation Office (ECO) in UC Davis has investigated campus thermal comfort since 2014. For the sake of avoiding an overheated or overcooled space, ECO developed a tool, called TherMOOstat, to collect occupants’ feedback in the buildings on campus. By using TherMOOstat, faculty and students can submit their perspectives about temperature, and it can help ECO to save energy and improve occupant comfort (Contreras, 2017). After reading each submission, the ECO would work with the Facilities Management Department to adjust the temperature remotely. The office also tried to look for any energy saving opportunities by aggregating and trending people’s feedback over a longer period of time. For instance, based on the TherMOOstat feedback, researchers found that the comfort band for UC Davis classroom spaces will be between 71 Fahrenheit to 76 Fahrenheit when it’s warmer outside, and between 69 Fahrenheit to 74 Fahrenheit when it’s cooler outside. This is very helpful to adjust setpoint of HVAC system according to outside temperature, which would improve thermal comfort and avoid unnecessary energy waste. 2.3 Personalized Comfort Model USC is a big campus with more than 450 buildings and 43,000 people, and it could be used as an existing living laboratory to work on campus-level thermal comfort study. Developing a proper comfort model is a critical method to understand and predict the thermal comfort in a specific area or space (Allen et al., 2015). However, a uniform thermal model cannot represent the whole campus which contains diverse spaces. If trying to simulate thousands of rooms with a single model, the result would be inaccurate and meaningless. Therefore, personalized comfort models associated with each type of spaces are needed to study the campus-level thermal comfort. 2.3.1 Definition A personalized comfort model predicts a single space’s thermal comfort instead of the average of a large group of rooms. It has several characteristics: (1) take a single room as the unit of analysis; (2) use only the feedback that occupants in the specific space to train a model; (3) has the capacity to adopt new data; (4) employ a data-driven approach (Talon, Goldstein, 2015). Personalized comfort model focus on a specific room. All analyzed data come from this space, so the final model is close to the actual building. Thus, it can help people to better understand specific comfort needs in the simulated room and provide more accurate adjustment suggestion. 2.3.2 Current Studies In recent years, there are more and more researchers pay attention to the personalized comfort model. Some of them were not studied thermal comfort, but they were adopting personalized models to simulate in other academic fields, which means this method can also be used in a wide range of disciplines. Some of them did study thermal comfort, but their models’ unit is individual occupant. In order to better understand the recent researches and development of personalized comfort models, people should review the relevant publications in the past decades. Li, Menassa, and Kamat (2017) collected occupants’ feedback from 7 subjects, including 3-point thermal preference (warmer/neutral/cooler), clothing insulation, heart rate, skin temperature, and human activity. Then he measured physical data including indoor air temperature, relative humidity, carbon dioxide, window state (open/close), outdoor 13 temperature, and outdoor humidity. He used Random Forest as the modeling method to simulate and predict occupants’ thermal preference, the accuracy of results could reach 80% after 50 samples. Jiang and Yao (2016) gathered ASHRAE 7-point thermal sensation scale, clothing insulation, metabolic rate, indoor air temperature, mean radiant temperature, and air velocity. Their modeling method is C-Support Vector that used to predict thermal comfort based on ASHRAE 7-point sensation scale. At the end, the mean accuracy of the proposed model achieves 89.8%. Q. Zhao et al. (2013) investigated personalized comfort model based on filed data from 9 subjects, including continuous thermal preference scale with 5 labels (hot/warm/neutral/cold/extremely freezing), indoor air temperature, mean radian temperature, relative humidity, and air velocity. Least square estimation was chosen as the modeling method to simulate the indoor thermal comfort. He calculated the mean square error of proposed model is 0.53, PMV model is 1.16. He also proposed a continuous learning method which is placing more emphasis on new data and gradually remove historical data. After the review, although some past studies achieved a considerable accuracy of the modeling results, the research found several gaps that exist in the current researches: (1) Lack of a unified modeling framework. Researchers focus on the accuracy of the model instead of developing a systematic method to evaluate the model; (2) Lack a connection to thermal comfort fundamentals. Some researchers provide their assumption in their models that are not associated with the existing thermal comfort study; (3) Lack the real-world integration. Some researchers did not describe how to integrate the comfort model to the real-world systems and how to transfer the modeling result to guide actual application; (4) Lack industry standards. Industry standards can help to evaluate comfort models’ performance in actual buildings, but there were no such standards applied to guide the evaluation of personalized comfort models. 2.4 Modeling Frame According to Kim, Schiavon, and Brager (2017), an intact personalized comfort modeling process includes the following steps: (1) Data preparation. Determine what kind of date will be used in the model and how to collect it; (2) Data collection. Collect raw data into preset forms that used for modeling; (3) Data selection. Confirm the algorithms that can eliminate extreme data and invalid data; (4) Model selection. Confirm an appropriate model for the given data and goals; (5) Model evaluation. Validate the performance of the chosen model with data and consider how to apply it in the actual application; (6) Continuous learning. Improve the model with incoming new data to make sure accuracy and efficiency. Figure 9. Modeling Frame 2.4.1 Data Collection Preparation and Collection phrases are work together to get the data that used for modeling personalized thermal model. Basically, the data should be: (1) expresses occupants’ perception of thermal condition; (2) describes the given conditions affecting their perception. The first type of data can be assessed by survey questionnaires that ask questions about thermal sensation, satisfaction, acceptability, preference, etc. (M. Schweiker et al., 2017). One assumption is made here that people’s “acceptability” of environment equals to “comfort”. But people may still feel “acceptable” even when the thermal condition is not in their comfort zone, just because they can tolerate it. So, thermal preference is a closer measurement of a thermal model but not 100% accurate. Although not perfect, thermal satisfaction still the best metric in the personalized thermal model since it is very straightforward, and it associates with the PMV model. When participants answer the survey, they may be asked to choose from a scale of options. Generally, the standards suggest using a 7-point structure for thermal sensation (“hot” to “cold”), or thermal acceptability (“acceptable” to “unacceptable”). However, some researchers chose to create their new scales in the survey for different modeling purposes. The influence of different scale structures is not obvious as this kind of scale is divided in a reasonable way. A psychologist (G.A. Miller, 1956) has recommended limiting the options to 5-7 since human’s brain is hard to distinguish the difference between two continuous sensations. 14 Another important criterion needs to be considered in the survey is frequency. Not surveying enough will not get sufficient data, while surveying too often may burden occupants and occur dissatisfaction. The reviewed studies took different surveying intervals. Jiang and Yan (2016) surveyed participants every 10 minutes in their experiments. However, it is too short to achieve in real life since it can interrupt people’s daily tasks. Most studies limited the surveying interval to a few times a day (R. Rana et al., 2013), or several hours (A. Ghahramani, C. Tang, and B. Becerilk-Gerber, 2015). Determining a proper survey frequency can help to collect enough data required for reliable modeling but not bother occupants. 2.4.2 Data Selection Collected data should be disposed before modeling, because the data may be heterogeneous data sets that have different structures, granularity, and volume. Therefore, it is critical to deal with the raw data following a preset procedure. This process may include: (1) cleaning missing rows, outliers, and measurement errors since they may hinder the analysis; (2) scaling numerical data into a consistent range; (3) aggregating the data into statistically values (e.g., median, mean) or grouping into discrete categories (e.g., acceptable, unacceptable); (4) exploring new categories (e.g., age group, rate of temperature change) that may affect the models; (5) merging heterogeneous data according to same variables (e.g., time, temperature range); (6) splitting the data into different groups (e.g., control group, experimental group) for single-valuable exploring (Kim, Schiavon, and Brager, 2018). 2.4.3 Data Visualization IoT can help to collect large amounts of physical data in the real world, but people need to find a way to check them visually. For example, making various graphs is easily to understand data and find out key points. Other methods like a 2D map or a 3D model can also be used as the vector of data visualization. Sihombing and Coors (2018) built a web- based 2D map and 3D model visualize sensors mounted inside a building. They successfully located the indoor sensors and then read their measurement data, which could be used to remind occupants to raise energy consumption awareness. McCaffrey et al. (2015) developed a web-based graphical user interface (GUI) to connect a building model with building management system. McCaffrey and his teammates collected data from sensors based on IoT technology and displayed the data in the GUI, including CO2, electricity, temperature etc. Thus, it is possible for building managers, occupants, and stakeholders to get access to the building information with the GUI, which can help them make decision or improve energy-saving awareness. 2.4.4 Model Selection Before starting modeling, it is critical to choose a proper prototype of the model for the sake of better understanding personalized comfort model. There are a lot of traditional modeling methods, like Random Forest, Linear Regression, Bayesian Network, etc. and most of them have been applied to previous studies. However, personalized comfort model may explore to more non-traditional data for more accurate results. So, recently there is a new tendency to adopt machine learning into the modeling process. But researchers still need to choose a corresponding algorithm as well. The research only considers traditional statistical modeling approach due to limited ability. 2.4.5 Model Evaluation The purpose of model evaluation is to see how close the personalized comfort model to real-life spaces. It is the criterion that decides whether a model can accurately simulate and predict the actual thermal environment. If not, researchers need to consider possible errors during the modeling process or find a better modeling method to match again. Basically, a model evaluation may include: (1) Prediction Accuracy assesses how correct the predicted outcome to the real outcome; (2) Prediction Consistency assesses whether the model predictions can keep matching with the same sample during a long period of time; (3) Model Convergence assesses if the model can converge into a stable prediction. 2.4.6 Continuous Learning Both physical thermal condition and human’s sensation will change as sun moves. Season change also can affect the people’s perception of cooling or heating (J.F. Nicol et al., 1999). Therefore, personalized comfort model should adapt to different weather conditions, which means it can slightly change when new data comes in. previous researches suggested the following approaches to develop continuous learning of personalized comfort model: (1) delete statistically irrelevant values in the new data; (2) add more weight to recent data and decrease weight to historical data; (3) remove old data that gotten a long time ago (e.g., a month) when new data enters; (4) ready to relearn more new 15 data. Although many researchers proposed the way to continuous adopt new data, only Ghahramani et al (2015). tested it in their research. Hence, more efforts are needed in the future to update the personalized comfort model over time. Moreover, people also need to consider an effective way to save the growing volume of data sets since tons of data may be collected as more advanced sensors appear. 2.4.7 Case Study The KieranTimberlake (KT) office is located in Northern Liberties, a neighborhood of Philadelphia (Witold, 2017). It is two stories high, and the whole office is an open space filled with mobile workstations, adaptable meeting rooms, and flex space. This office uses a combination of active techniques and passive techniques together instead of air- conditioning, like heavy insulation, operable windows, and dehumidification. In order to test the effect of the new type of cooling system in Philadelphia, KT decided to build personalized thermal models to simulate the results. They installed hundreds of thermal sensors on the floors, walls, and roof to monitor the temperature. The firm developed its own mini-wireless sensors, which were called Pointelist, in the fabric of the building. Detecting temperature and humidity is only half of the data collection, they also ask people how they feel in such an office. About 120 KT employers were sent an online survey that asked people to choose their current thermal condition, including “much too cold,” “cold,” “comfortable” and so on. Staff also were asked their clothing and their physical location in the building. The questionnaire was sent by another product of the KT, called Roast. It is a cloud-based app for people to submit anonymous votes. This experiment met many problems. For example, the thermal transfer between the concrete and the air was not as effective as people anticipated; the indoor air cannot be adequate dehydrated during nighttime, which increased the “too hot” vote, etc. In conclusion, only relying on active and passive techniques but without air-conditioning is not enough for the thermal satisfaction in an office building. Another finding is that 80 percent of the staff still feel comfortable under 83 degrees, which is a little higher than the industry recommended standard from 68 to 78. This finding proves that the current office buildings have a huge potential to expand the HVAC setpoints but also keep comfort for most people. For those 20 percent of who felt “hot”, the firm set a special “Cool Room” for them to relax. “Cool Room” will keep the temperature from 72 to 74 degrees at any time. The firm also found that during the time when the indoor temperature exceeded the comfort standard, only 25 percent of the time is work hours. That means 75 percent of the time can use passive ways to meet comfort needs. So, these findings are very useful to deduct energy consumption. Finally, the principle Kieran said human’s behavior is a critical factor as well. He said people will add clothes or do small exercise to keep warm. But human’s act is hard to change, and behavioral adaptation is an uncertain factor to predict. So, the most challenging thing for energy saving may not be techniques but people themselves. 2.5 Summary Many previous studies have demonstrated that personalized thermal models can match and predict thermal condition for a large amount of spaces. However, there is no relative research focusing on university campus. That’s why building a campus-level thermal model is not only very critical to achieve thermal comfort on campus, but also an effective way to save more energy. Moreover, traditional thermal sensors do not support real-time reading, so people need to analysis and react after a long time. Nowadays, IoT technology can solve this problem, which enables real- time information exchange between sensors and the Internet. This is remarkable compared to only 10 years ago, but the progress of the Internet makes it possible to achieve more efficient collection and more accurate thermal model. 16 3. METHODOLOGY Based on the background and research study above, several research questions could be asked: (1) How does IoT sensor collect real-time temperature and humidity? (2) What percentage of occupied hours are rooms out of the USC recommended setpoint range as stated by temperature policies? (3) How can we evaluate the current temperature policies with physical thermal data and occupants’ feedback? (4) Can actual setpoints be determined from analysis of IoT data feeds? (5) How do IoT data help to improve thermal comfort? In order to answer these research questions, the building Watt Hall is instrumented as the “living lab” to provide continuous data logging during a period of 2 weeks. More specifically, the methodology could be divided into four phrases: (1) developing portable IoT sensors that can detect real-time temperature and humidity around it; (2) choosing several spaces in the Watt Hall to install assembled IoT sensors and turning the whole building into a “living lab”; (3) collecting occupants’ thermal feedback through the TrojanSense App; (4) analyzing both physical thermal conditions and people’s responses, then trying to improve the current temperature policies. The whole process involves many fields, like 3D modeling, circuit assembling, coding, etc. So, it requires a wide range of knowledge reserve. 3.1 Methodology Diagram The methodology is showed as a graph below. IoT sensors are controlled by code when connecting to computer. The status (e.g., on, sleep or off) could be changed, detecting intervals, measurement accuracy and so on. Then, a total of 18 IoT sensors would be installed in different classrooms or work stations in USC Watt Hall. The initiating IoT sensors can continuously transmit thermal data into the cloud through the campus-covered WiFi. At the same time, the TrojanSense App designed by the TrojanSense group is collecting faculties and students’ feedback. Based on the database, a thermal map could be built to show the thermal information on it. Moreover, it is convenient for students and faculty to search and look which classroom is more comfortable. When getting enough data, a series of data-driven comfort models that focus on each single space could be built. These models can guide the FMS to adjust the HVAC setpoints to meet occupants’ requirement. HVAC managers also could predict the routine temperature trend according to these models. If so, occupants could enjoy the desired thermal environment without any delay. Changing HVAC settings would result in new feedback data and supply the models in turn. As a result, the personalized comfort models would be improved and perfected over time. Also, some repeatable patterns could be generalized to other similar rooms and buildings. 17 Figure 10. Methodology 3.2 IoT Sensors IoT sensor is composed of a shell, a DHT22 sensor, an ESP32 microcontroller board, a battery box, and three AA batteries. It is small enough to carry and install in the corners while people will not mind. In a decade ago, IoT sensor cost much, but the cost has dropped significantly as the improvement of 3D printing techniques in the last five years. Nowadays, it is very easy to 3D-print the shell based on the 3D model, and any shape could be printed out as long as the model was built in modeling software. Thus, the research designed a shell that just can hold all required components. It is a square-shape box, which has approximately 3 inches length, 4 inches width, and 7 inches height. The DHT22 sensor is designed to expose on the outside, while other components will be covered in the box. The sensor is connected to the interior electric board by wires that going through a hide hole behind the sensor. The whole IoT sensor has only about 5 pounds, so it could be attached to a wall or the downside of a table and will not attract too much attention. The research designed and built a series of IoT sensors to collect physical thermal data. He initiated those sensors by coding on the Arduino. According to the code, the IoT sensors will collect real-time data and transmitted to the Adafruit IO at the first time. The transmission process yields to the MQTT protocol. Thus, people can log into paired account on the Adafruit website to check collected data, and he can generate graphs or download all data on the Adafruit. 18 Figure 11. IoT Sense Methodology 3.2.1 ESP32 Board ESP32 Feather Board is a microcontroller board produced by Adafruit, and it is made with the WROOM32 module. The board includes a built in USB-to-Serial converter, plenty of analog inputs, two analog outputs, and a dual-core ESP32 chip, etc. ESP32 Board supports both WiFi and Bluetooth, which ensures high-speed transmission by either way. The board was designed to advanced developers, so it will require some experience with microcontroller programming. 3.2.2 DHT22 Sensor DHT22 sensor is a low-cost temperature & humidity sensor. this kind of sensor has 0.5 degree’s error range, but it is enough for basic data logging. The sensor is made of two parts, a capacitive humidity sensor and a thermistor. There is also a very basic chip inside that does some analog to digital conversion and spits out a digital signal with the temperature and humidity. The digital signal is easy to be receive and read using any microcontroller. Basically, the nominal voltage of the sensor is between 3V to 5V, and 2.5mA max current is needed during conversion data. It can detect 0-100% humidity with 2-5% accuracy, and -40 to 80 degrees Celsius with 0.5 degrees Celsius accuracy. Therefore, the DHT22 sensor has a wide range of detection which is enough for the research’s need. 3.2.3 Arduino Arduino is an open-source electronics platform based on easy-to-use hardware and software. coding platform used to control all Adafruit microcontroller boards. The company advocates people, even those do not receive professional training, to assemble their own circuits. With hundreds of components and combination, people can make different electric gadgets and inventions. For example, a robot that can blow bubbles automatically; a shining skirt; or a paper airplane launcher etc. Arduino software has both web editor and downloading version, which can meet different users’ preferences. The software uses the Arduino programming language (based on programming language C) and it also support Wiring and JAVA. Therefore, the coding environment of the Arduino is very friendly and compatible. Also, the downloaded software has included many basic samples that may guide people to have a good start. Tons of courses and examples on the official community can provide instructive guidance as well. After writing code in the Arduino, clients can automatically check the error and any error would be marked the location in the code string. If all code is correct, users can easily upload it to their board and test it. 19 Figure 12. Arduino Code Basically, the code was composed of two parts. In the “setup” function, it will try to connect the WiFi by typing pre- set “WLAN_SSID” and “WLAN_PASS”. The “USC GUEST WiFi” was used as the uniform internet for all boards. In the “loop” function, it will attempt to connect MQTT, which is a back-stage management protocol for data processing. With the MQTT, we can transmit data to the Adafruit cloud paired with a preset account. There is a unique IO_key as an identification code for each Adafruit account, so do not worry collected data will be transmitted to others’ account. Then, comment “analogRead” will output the temperature and humidity detected by DHT22 sensor. After a series of calculations, the value will be transferred between different units, and transmitted to our computer finally. The detection interval is set to 900 seconds (15 minutes), since too frequent detection will run out the power quickly. IoT sensors were proved to run about 12-14 days with 15 minutes’ interval. Also, a deep-sleep mode was added into the code. That is the sensor will go into deep sleep during the 15 minutes’ break. If so, the circuit can run at a very low current, which can save a lot of power of batteries. There is a “restart” function that can restart the sensor when it “sleeps” over 30 minutes. The detection result is set to be correct to two decimal places. If the detected temperature is lower than 0 degrees Fahrenheit or the relative humidity is over 100%, it would be seen as measurement error and the sensor would redetect again after 3 seconds. A total of 5 times measurement errors will lead the system to “deep sleep” mode and waiting next detection in15 minutes. 3.2.4 MQTT MQTT (MQ Telemetry Transport) is a machine-to-machine IoT connectivity protocol invented by Dr Andy of IBM in 1999. It was designed for constrained devices and low-bandwidth, high-latency or unreliable networks. The design principles are to minimize network bandwidth and also ensure reliability. These principles make the protocol ideal of the emerging M2M or IoT world of connected devices. Nowadays, there are many wireless signals overlapped in a room. In order to receive and send data in a stable method, MQTT is needed to block other interfering signals and make sure the stability of one specific signal’s transmission. 3.2.5 Adafruit IO Adafruit IO is cloud service provided by the Adafruit company. It can store the data transmitted by IoT devices. A client has to create an account to receive data, and each account has a unique recognition code used for paring. When coding in the Arduino, people can type in the recognition code and upload into the IoT devices, then the collected data 20 would be received and showed in the Adafruit IO. Basically, a normal user has limitation of storage and transmission speed, but an updated VIP user has no limitation anymore. On the website, the Adafruit IO also provide “dashboard” for people to create analysis graphs online. Diverse tools could be applied to help users to better understand data. For example, Slider can see linear variations of a set of data; Map can show the location that data came from. Moreover, clients can download all data or several data according to some limits, which means the data selection could be done online. Figure 13. Adafruit IO Figure 14. Temperature Data in Adafruit IO 21 Figure 15. Humidity Data in Adafruit IO 3.3 Installation As discussed before, there are in total 18 spaces were installed the IoT sensors. Since there are more than 100 buildings and over 1000 rooms on USC campus, it is hard to detect the thermal data for every space. Making a small sample of 18 representative rooms in one building could reduce much workload. Also, chosen locations are usually popular and suitable for long-term studying, which makes them more valuable for data digging than other spaces. All locations are displayed in a list below. 22 Table 5. Chosen Spaces Watt Hall Spaces Mapping Number Location Number Location 1 MBS Corner 11 Studio 200 2 Upper Rosendin 12 Studio 208 3 Lower Rosendin 13 Studio 209 4 WAH 212 14 Arch Lib 5 WAH B1 15 3 rd Floor Balcony 6 HAR 101 16 Courtyard 7 HAR 102 17 2 nd Floor Office 8 WAH 322 18 WAH B1 Studio 9 WAH 324 19 10 ARCH Corner 20 23 Figure 16. 3D Model for Spaces Figure 17. Top View for Spaces 24 In total, there are 12 enclosing rooms, 4 open spaces, and 2 outdoor spaces. The spaces cover the whole Watt Hall and attaching Harris Hall from basement to the top floor. Most of them are work studios where many students and faculties aggregate. The principle of installation is trying to deploy the sensor in the center of the space, and the installation height is about 1~1.2 meters which is close to the human’s average height. At the same time, the sensor should be hided under the table or in the place where people will not mind in case someone takes it off. Furthermore, the relative administers of each room were alerted before installation, or they may report the sensor as an unknown hazardous article. After installation, it is also important to take a picture to record the exact location and to design a routine check. 3.4 Collection During the collection process, an Adafruit account used to connect MQTT can build several “Feeds” in the user personal page. First of all, matching the account’s recognition key with the MQTT, which needs manually typing the key in the code mentioned above. Then, naming “Temperature”, “Humidity” and “BetteryVoltage” feeds for each space respectively. Because the codes could be changed slightly to match each feeds’ name according different locations, the computer can receive all data from 18 places together and there is no mess. When all IoT sensors start work, collected data would be drawn as trend chart and every value could be traced underneath these charts. Moreover, a “Dashboard” could be built in Adafruit to add several feeds. If so, users could check and compare diverse thermal information between study spaces. 3.5 Feedback IoT sensors can get real-time thermal data, but occupants’ responses also matter. For the sake of improving thermal comfort, listening to students and faculties who work in the room should be a necessary procedure. Occupants’ answers can provide insight for which thermal conditions satisfy them, so they can be deployed across the whole campus map. Another advantage of capturing occupants’ opinions is helping to build the comfort models for different types of rooms. Survey questionnaire is the most normal way to collect people’s view, but traditional physical questionnaire takes a long time to collect, and people may feel tired on writing on paper. Therefore, the TrojanSense APP was used to collect feedback. In this app, users can choose their preferences and submit through their mobile phone. A total of 3 round of TrojanSense campaign were launched from November to January. In each campaign, after a negotiation with professors, they agreed to assign a small homework to their students. More specifically, students in the class were asked to download the TrojanSense App, and a list of spaces was sent to students in the class, then they were encouraged to vote during a specific period of time. 3.6 Data Analysis The expected data is a series of temperature and humidity data for each space. Basically, temperature and humidity could be showed in line plots. But they cannot be used for analysis directly, since a series of data in 24 hours is not representative in real world. For example, the spaces are not occupied during the nighttime, so the HVAC system does not work at night. Therefore, data processing is necessary before accurate analysis. The research assumed 9 am to 5 pm as the general occupied time for all spaces. Deleting the data that out of this period of time, then consecutive line plots would be transferred to periodical line plots. The research was trying to define a range of temperature setpoints which cover the most of occupied time. That is the actual setpoints for each space, and it would be compared with the USC temperature policies. As a result, the research can verify if the actual setpoints are involved in the preset setpoints, also this would help people to understand the possible gaps between reality and policies. At the same time, collecting outdoor temperature and humidity as the compare group. Temperature and humidity are consecutive data stream, but occupants’ feedback is discrete descriptive votes. In order to transfer descriptive votes to statistic data, a total of 13 numbers from “-6” to “6” are chosen to represent peoples’ feeling. The “-6” means people feel hot and they want to be much cooler, while “6” means people feel cold and they want to be much wormer. When the TrojenSense APP is collecting people’s feedback, it will record the time as well. So, it is possible to compare each vote with the thermal condition at that time. Basically, the experiment is expected to get five series of data as mentioned above, so pairing them by two and at least six types of analysis could be done: (1) the relationship between indoor temperature and votes; (2) the relationship between indoor humidity and votes; (3) the relationship between outdoor temperature and votes; (4) the relationship between outdoor humidity and votes; (5) the relationship between outdoor temperature and indoor temperature; (6) 25 the relationship between outdoor humidity and indoor humidity. These analyses can answer which factor can influence occupants’ satisfaction most, and how the outdoor environment affect indoor thermal condition. 3.7 Assessment After data analysis, a campus thermal map and some personalized thermal models could be concluded. The easiest way to assess whether these models work accurately is collecting more feedback from occupants in spaces. Collecting a group of votes without any change as the compare group and adjust the HVAC system setting according to concluded models, then collecting the same volume of votes as the compare group. If the adjusted data is obviously improved, which means the personalized thermal models work well. If not, it means the models are not accurate and need to examine the original data. The most possible reason for inaccurate modeling is lacking enough data. So, one advantage of this assessment is more collected data can help to improve the personalized thermal model itself. Even though the first-time modeling failed match the real station, it is possible to quickly iterate a new model again. As the iteration continues, the models would be more and more accurate. It must admit that 100% matching between personalized thermal models and actual situation is impossible. In real world, evaluation error always happens. Moreover, occupants’ votes may vary from one person to another. For example, the same thermal condition is acceptable for a strong male while it is a little cooler for a skinny female. It may happen that some extreme values in one assessment, which may affect the accuracy of results. Sanguinetti, Hybrid, and Vehicle (2016) got 87% accuracy rate after applied personalized thermal models. Therefore, in this experiment, the research assumed 80% accuracy as the threshold that verifies the thermal models. 3.8 Summary This experiment was designed to install 18 IoT sensors in different spaces around the Watt Hall on the USC campus. Controlled by code in Arduino, these sensors could detect the real-time temperature and humidity and send it to the cloud—Adafruit IO. So, people could log in paired account in Adafruit IO to collect the real-time thermal data for each space. Combined with outdoor temperature and sufficient occupants’ feedback collected by TrojanSense APP, it is possible to figure out the effect of outdoor thermal condition on indoor condition. Also, a series of data-driven comfort models that match and predict the prerequired thermal condition for each space could be concluded. As more as data be gotten, the thermal models would be more accurate. As a result, the conclusion could be sent to the FMS and help them to adjust the HVAC system’s setting according to different rooms, which may improve people’s thermal comfort and avoid energy waste. 26 4. RESULTS In this chapter, the research shows the results of the experiments with IoT sensors. A total of three rounds of experiments lasted from late October to early December. The TrojanSense Campaign started at Mar 2018, and it keeps running all the time. At the time of writing this thesis, more than 1000 feedback were received, 486 valid feedback among them, and 177 valid feedback came from 11 spaces that IoT sensors located, average 16 votes in each room and 1 vote in each day. However, such a small amount of data is not enough to support a convincing thermal comfort model. Also, there is no guarantee that everybody in the classroom submitted feedback, and models built with partial feedback may have user bias to the other occupants. Therefore, the research focused on the indoor thermal conditions and the contrast with the current temperature policies. 4.1 Research Questions • What is the conditioning mode for the Watt Hall in November? • Whether the current temperature policy has achieved the desired temperature point? • What fraction of occupied hours are indoor temperatures outside of the USC campus setpoint policy? • How does outdoor temperature affect indoor temperature in a room? • What is the balance point temperature for the Watt Hall? 4.2 Basic Analysis After 3 rounds of TrojanSense campaigns, a large amount of data was collected. These data could be shown in box and whisker plots by different places. These data were detected during different periods of time because of some objective factors, for example, some classrooms were used for classes, so the installation of sensors had to be postponed. However, all data were collected during November, and people can also easily compare the thermal conditions from temperature to relative humidity. Most spaces are between 70 to 75 Fahrenheit, which was included in the USC Temperature Policy. The classroom WAH 324 has the highest average temperature of 81 Fahrenheit, while the WAH Lower Rosendin has the lowest average temperature, which is about 69 Fahrenheit. Moreover, most spaces have a narrow range of temperature fluctuation, except the Courtyard, the WAH 322, and WAH 324. These three spaces have a wide range of 6~7 Fahrenheit fluctuation. Such a wide temperature fluctuation may cause discomfort, so people need to carefully adjust the HVAC system. Figure 18. Temperature Comparison between 18 Spaces 27 As for the humidity, the values vary a lot between different spaces. The HAR 102 has the highest average humidity of 48%, while the WAH B1 has the lowest average humidity of 18%. Furthermore, most spaces have a large fluctuation except the 2 nd Floor Office, the WAH Lower Rosendin, and B1 Studio. Since indoor humidity is affected by many factors including the number of occupants, the room’s location, and so on. It is hard to find a pattern of variation before knowing more spaces’ information. The box and whisker plots can only show the maximum value, minimum value, and the cover range of the interquartile range, but it did not show the variation trend. Therefore, after the overall comparison, the research also compared each space’s thermal condition. Every room was detected during the November and the data were shown in the line plots and the box and whisker plots. Also, all figures are only showing the occupied time, while the data during the unoccupied time were erased. Thus, the lines are intermittent. Figure 19. Relative-Humidity Comparison between 18 Spaces These spaces could be divided into three types according to their HVAC systems: 1) Central AC System, including the 2 nd Floor Office, the ARCH Corner, the Architecture Library, the HAR 101, the HAR 102, the MBS Corner, the Studio 200, the Studio 201, the Studio 209, the WAH 212, the WAH B1, the WAH Lower Rosendin, the WAH Upper Rosendin, and the B1 Studio; 2) Independent AC Unit, including the WAH 322 and the WAH 324; 3) No HVAC System, including the 3 rd Floor Balcony and the Courtyard. 4.2.1 Central AC System Spaces The 2 nd Floor Office is located at the east side of the Watt Hall and it contains about 10 people on weekdays. From November 15 th to 24 th , the temperature was kept stable around 75 Fahrenheit. The humidity also stayed around 40%. Moreover, the daily temperature variation kept constant, and the daily humidity variation varied within 5% except on November 19 th and 20 th , on which the humidity decreased 10%. 28 Figure 20. 2nd Floor Office Temp. Track Figure 21. 2nd Floor Office Humid. Track 29 Figure 22. 2nd Floor Office Temp. Range and Humid. Range The ARCH Corner is the north-east corner of the 3 rd floor in the Watt Hall. It is an open-space studio and about 30 students work here during the occupied time. The temperature increased from 73 to 75 Fahrenheit and then slightly decreased to 74 Fahrenheit. The humidity surged on November 17 th from 15% to 40%. Moreover, the daily temperature variation kept about 1 Fahrenheit, and the daily humidity variation fluctuated from 0 to 10%. An obvious variation (10%) happened on November 10 th , and 18 th ~ 20 th . Figure 23. ARCH Corner Temp. Track 30 Figure 24. ARCH Corner Humid. Track Figure 25. ARCH Corner Temp. Range and Humid. Range The Architecture Library is in the basement of the Watt Hall. It was surrounded by concrete and about 20 people work inside during occupied time. The temperature of the library was slightly rising from 71 to 73 Fahrenheit as time goes on, and the relative humidity surged on November 17 th as well. Furthermore, the daily temperature varied about 0.5 Fahrenheit on average, and the daily humidity varied from 0 to 10%. The biggest variation happened from November 18 th to 20 th . 31 Figure 26. Arch. Library Temp. Track Figure 27. Arch. Library Humid. Track 32 Figure 28. Arch. Library Temp. Range and Humid. Range The B1 Studio is located on the opposite side of the Architecture Library in the basement. It has the same finishing material as the Architecture Library, and it can contain about 50 people working at the same time inside. In general, the temperature declined from 73 to 71 Fahrenheit, while the humidity increased from 20% to 45% on November 16 th , and then it kept around 40% until November 26 th , when it decreased back to 30%. Moreover, the daily temperature variations are 1 Fahrenheit before November 21 st , then it almost kept constant. The daily humidity variation kept the same except on November 18 th to 20 th , on which it reached 10%. Figure 29. B1 Studio Temp. Track 33 Figure 30. B1 Studio Humid. Track Figure 31. B1 Studio Temp. Range and Humid. Range The HAR 101 is a classroom located on the 1 st floor of the Harris Hall, which is a 2 story-height building next to the Watt Hall. It is a big auditorium that can contain more than 100 people. The temperature significantly declined from 73 to 69 Fahrenheit. The humidity slightly increased from 45% to 55% at first, then it dived from 55% to 30% on November 8 th and lasted at 25% after that. In general, the daily temperature variations are about 1.5 Fahrenheit on November 2 nd to 4 th and November 9 th to 14 th . In addition to that, it kept almost constant. The daily humidity variations are 5% generally, and the biggest variation happened on November 10 th to 12 th . 34 Figure 32. HAR 101 Temp. Track Figure 33. HAR 101 Humid. Track 35 Figure 34. HAR 101 Temp. Range and Humid. Range The HAR 102 is located on the 1 st floor in the Harris Hall as well and it just next to the HAR 101. However, HAR 101 is much bigger than the HAR 102, which can contain about 30 people at the same time. The temperature declined 71 to 67 Fahrenheit, and the humidity also obviously decreased from 50% to 30%. Moreover, the daily temperature variation kept about 1 Fahrenheit before November 8 th , and then it increased to about 2 Fahrenheit after November 9 th . The daily humidity variation kept within 5% except on November 9 th , 11 th , and 12 th , on which the daily humidity varied around 15%. Figure 35. HAR 102 Temp. Track 36 Figure 36. HAR 102 Humid. Track Figure 37. HAR 102 Temp. Range and Humid. Range MBS Corner is located at the south-east corner on the 3 rd floor of the Watt Hall. It has the same size as the ARCH Corner. However, temperature and humidity are totally different. In general, the temperature kept about 74 Fahrenheit, while the humidity fluctuated in November. The trend of humidity slightly increased to 50% at first, then it went through a cliff dive to 15% on November 9 th , and it rose to 40% again after that. The daily temperature variations are between 2 to 4 Fahrenheit, and it reached the biggest value on November 12 th . The daily humidity variation kept within 5% in most times, but it reached about 15% on November 9 th to 10 th and 16 th to 19 th . 37 Figure 38. MBS Corner Temp. Track Figure 39. MBS Corner Humid. Track 38 Figure 40. MBS Corner Temp. Range and Humid. Range The Studio 200 is on the east side of the 2 nd floor of the Watt Hall. It has the same size as the B1 Studio. The temperature trend fluctuated around 72 Fahrenheit during the November, while the humidity went through a sudden surge on November 16 th from 20% to 40%. The daily temperature varied about 4 Fahrenheit on November 9 th and 19 th . In addition to that, it kept about 1 Fahrenheit during other times. The daily humidity variations are around 10%. Figure 41. Studio 200 Temp. Track 39 Figure 42. Studio 200 Humid. Track Figure 43. Studio 200 Temp. Range and Humid. Range The Studio 208 is located at the north-west corner on the 2 nd floor of the Watt Hall. It only has a half size of the Studio 200 and it can contain about 50 people. The temperature in the Studio 208 was about 76 Fahrenheit on November 9 th , then is dropped to 69 Fahrenheit and gradually bounced back to 70 Fahrenheit after that. The humidity trend started at 20% on November 9 th , and then it increased to 40% on November 16 th . Furthermore, the daily temperature variations are around 2 Fahrenheit, but it reached about 4 Fahrenheit on November 9 th and 20 th . The daily humidity variation kept about 10% during this month. 40 Figure 44. Studio 208 Temp. Track Figure 45. Studio 208 Humid. Track 41 Figure 46. Studio 208 Temp. Range and Humid. Range The Studio 209 is on the opposite side of the Studio 208 and they all have a similar size. Therefore, these two studios have almost same temperature and humidity trends. Figure 47. Studio 209 Temp. Track 42 Figure 48. Studio 209 Humid. Track Figure 49. Studio 209 Temp. Range and Humid. Range The WAH 212 is a small classroom located in the middle area on the 2 nd floor of the Watt Hall. It can only contain about 20 people at the same time. The temperature gradually declined from 74 to 72 Fahrenheit, and the humidity slightly increased from 25% to 45%. The daily temperature variation kept approximately 1.5 Fahrenheit on average, and the daily humidity variation is 10% before November 20 th , then it decreased to about 5%. 43 Figure 50. WAH 212 Temp. Track Figure 51. WAH 212 Humid. Track 44 Figure 52. WAH 212 Temp. Range and Humid. Range The WAH Lower Rosendin is a patio area on the 2 nd floor of the Watt Hall. 10 to 20 students may study here sometimes. The temperature started at 73 Fahrenheit on November 9 th , and it dropped to 69 Fahrenheit and then gradually bounced back to 71 Fahrenheit. Similar, the humidity declined from 30% to 15% on November 10 th , and then it slowly increased to 35%. The daily temperature variation kept about 1 Fahrenheit on average, and the biggest variation (4 Fahrenheit) happened on November 9 th . The daily humidity variations are around 10%. Figure 53. WAH Lower Rosendin Temp. Track 45 Figure 54. WAH Lower Rosendin Humid. Track Figure 55. WAH Lower Rosendin Temp. Range and Humid. Range The WAH Upper Rosendin is also a patio area that above the Lower Rosendin. Likewise, about 20 people may work here sometimes. The trend of temperature went through a slightly decline from 75 to 73 Fahrenheit on November 4 th and then kept around 74 Fahrenheit. The trend of humidity gradually rose from 35% to 50%, then it significantly dived to 20% on November 9 th and kept at the value. Moreover, the daily temperature variations are about 1.5 Fahrenheit. And the daily humidity varied about 10% on average, but there was almost no change between November 6 th to 8 th . 46 Figure 56. WAH Upper Rosendin Temp. Track Figure 57. WAH Upper Rosendin Humid. Track 47 Figure 58. WAH Upper Rosendin Temp. Range and Humid. Range The WAH B1 is located at the basement of the Watt Hall and it can contain about 50 people taking a class at the same time. The general temperature kept around 74 Fahrenheit. The humidity dropped from 50% to 15% on November 9 th , then it bounced back to 40% on November 16 th . The daily variations fluctuate for both temperature and humidity. The daily temperature almost kept the same between November 9 th to 12 th , but it varied about 2 Fahrenheit on other days. Similarly, the daily humidity changed about 15% on November 9 th , 10 th , 15 th , and 16 th . In addition to that, it changed under 5%. Figure 59. WAH B1 Temp. Track 48 Figure 60. WAH B1 Humid. Track Figure 61. WAH B1 Temp. Range and Humid. Range 4.2.2 Independent AC Unit Spaces The WAH 322 is located at the south side of the Watt Hall, and it is a small faculty office that contains only 1 person inside in most times. The temperature trend of the space is stable at around 76 Fahrenheit, while the temperature trend dived from 50% to 20% on November 9 th , then it gradually increased back to 40%. The daily humidity variation varied about 10 Fahrenheit on every single day. As for the humidity, the daily variation kept constant except the November 9 th and November 12 th . There were obviously humidity declines on these two days. 49 Figure 62. WAH 322 Temp. Track Figure 63. WAH 322 Humid. Track 50 Figure 64. WAH 322 Temp. Range and Humid. Range The WAH 324 is an activity room next to the room WAH 322. In most times, no one works inside but sometimes, a group of 6 or 7 people may discuss here. Like the WAH 322, the WAH 324 has a stable temperature trend around 76 Fahrenheit and a large daily temperature variation about 7 Fahrenheit. And the humidity trend dropped from 50% to 20% on November 9 th as well, and it kept at 20% after that. Figure 65. WAH 324 Temp. Track 51 Figure 66. WAH 324 Humid. Track Figure 67. WAH 324 Temp. Range and Humid. Range 4.2.3 Outdoor Spaces The 3 rd Floor Balcony is located at the east side of the Watt Hall and it only has 2 square feet. It is an aisle connecting the MBS Corner and the Faculties’ offices, so no people stay here most times. The general tendency slightly decreased from 62 to 60 Fahrenheit, and the humidity surged from 30% to 70% on November 16 th , then it slightly fluctuated around 65%. Since the 3 rd Floor Balcony is an outdoor space, so there was a significant fluctuation on both temperature (2.5 Fahrenheit) and humidity (15%) in every single day. 52 Figure 68. 3rd Floor Balcony Temp. Track Figure 69. 3rd Floor Balcony Humid. Track 53 Figure 70. 3rd Floor Balcony Temp. Range and Humid. Range The WAH Courtyard is surrounded by the Watt Hall and the Harris Hall. It is also an open space for several students studying or relaxing here during the occupied time. The temperature trend decreasingly descended from 62 to 58 Fahrenheit. The humidity trend looks like a “V” shape, and it dived from 65% to 40% on November 16 th , then it quickly bounced back to 60% on November 28 th . In addition, the daily temperature variation kept about 2 Fahrenheit, and the daily humidity variations are about 10%. Figure 71. WAH Courtyard Temp. Track 54 Figure 72. WAH Courtyard Humid. Track Figure 73. WAH Courtyard Temp. Range and Humid. Range In the Figure 62-67, the average indoor temperatures of the WAH 322 and the WAH 324 are much higher than other spaces, that is because these two rooms were controlled by independent AC units instead of central AC unit. Therefore, their room temperatures are decided by occupant’s preferences rather than school temperature policies. Since their average room temperatures are higher than 76 Fahrenheit, which is the preset value according to policies (especially the WAH 324 was usually unoccupied and the AC unit usually kept stopped). So, it could be concluded that the whole Watt Hall was in cooling mode in November. In summary, it is possible to figure out the temperature and humidity zone that cover most data for each space. Thus, people can easily find out what is the actual thermal condition compared to the school’s preset temperature policies. Also, placing IoT sensors in rooms with different AC systems can help to determine the cooling mode for the Watt Hall. 4.3 Comprehensive Analysis All analyses above are based on the one-dimensional objective data on field, but people cannot know the connection between multidimensional data and how people react. In order to understand what a specific thermal environment means to people; the research needs to consider occupants’ feedback and the relative data at that time. Also, the comprehensive analysis between outdoor temperature and indoor temperature can help to explain more about overcooling or overheating. 55 4.3.1 Occupants’ Feedback To put objective data in context with subjective response data, the researcher was provided with a .csv file of subjective response data from the TrojanSense database. The .csv file contained 1000 responses collected from Oct to Dec. However, about half of the feedback came from the spaces without IoT sensors. For the rest of the data in the detecting spaces, some data are out of the occupied time, and some are invalid data. Therefore, there are only 177 valid data that could be used for the present analysis. The researcher also divided the 13 grades of votes into 5 groups by every 3 degrees (See Table 6). There are 75% of all votes choosing warmer, and only 10% votes choosing cooler, with the remaining votes choosing neutral. Overall, these responses suggest that most participants perceived the spaces to be over-cooled. Also, the 1-3 Deg. Warmer accounts for about 50% total votes, that means most spaces need to be heated 1-3 degrees higher than the preset indoor temperature. Table 6. Occupants’ Feedback Distribution Generally, analyzing occupants’ feedback need enough valid data to support. The more data could be collected, the more accurate result may be gotten. After comparing, the 5 spaces with the most feedback are chosen to analyze in this case. These available spaces include the MBS Corner, the WAH Upper Rosendin, The Lower Rosendin, the WAH 212, and the WAH B1. The research will take the MBS Corner as an example to show how to conduct a comprehensive analysis. 4.3.2 MBS Corner The MBS Corner is the space that collected the most valid feedbacks, so it is representative to show the effect of the experiment. The research took as the thermal data from October 17 th to November 14 th since occupants’ feedbacks are mainly distributed on this period of time. Combining temperature track and feedback together into one chart can help to easily find out their connection. Before analysis, it is necessary to delete the data outside of the occupied time (9 am~5 pm) as well. 56 Figure 74. MBS Corner Temperature Figure 75. MBS Corner Temperature + Feedback 57 Figure 76. MBS Corner Work-Time Temperature + Feedback In Figure 82, the histogram represents occupants’ feedback. The number above each column means how many votes people voted, and the color of the column represents how people wish to change the temperature. The red is deeper the occupants want to be warmer; the blue is deeper the occupants want to be cooler. When the color shows green, it means the occupants felt neutral. The temperature trend of the MBS Corner kept stable within 74 to 78 Fahrenheit from October 17 th to November 14 th . Comparing with occupants’ feedback, people tended to feel cool when the indoor temperature below 74 Fahrenheit, so they wanted to be warmer. When the indoor temperature reached about 76 Fahrenheit, most voted people felt just right. Therefore, it could be concluded that 76 Fahrenheit is a comfortable temperature, and people can confirm that the comfort zone is between 75 Fahrenheit to 77 Fahrenheit for the MBS Corner according to this chart. Figure 77. MBS Corner Humidity 58 Figure 78. MBS Corner Humidity + Feedback Figure 79. MBS Corner Work-Time Humidity + Feedback Similarly, analyzing the relationship between relative humidity and occupants’ feedback with the same method. Figure 85 illustrates people may feel warm or cool even at the same humidity level. There is no clear evidence shows the inner connection between humidity and feedback in this case. 4.4 Outdoor Temperature In addition to the indoor thermal condition, the researcher also considered the effect of the outdoor environment. Downloading the official climate data from the nearest meteorological station and using the box and whisker plots to show the average outdoor temperature for each day. Even though the outdoor temperature was below than the school’s heating set-point temperature (68 Fahrenheit), the internal heat source, including electric lighting, mechanical equipment, and body heat may offset the need of heating. As mentioned, the Watt Hall was in cooling mode in 59 November, so the desired temperature is 76 Fahrenheit according to the school temperature policies. Then subtracting 76 from the room temperature and showing the daily degrees-of-overcooling for each space (See Figure 80-93). All data is in working hours (9 am – 5 pm). Since the independent-AC-unit spaces (the WAH 322 and the WAH 324) and outdoor spaces (the 3 rd Floor Balcony and the Courtyard), their room temperatures do not comply with the school policies so that they are not considered in this analysis. • 2 nd Floor Office: 60 Figure 80. 2nd Floor Office Temperature Comparison • ARCH Corner: 61 Figure 81. ARCH Corner Temperature Comparison • Arch Library 62 63 Figure 82. Arch Library Temperature Comparison • B1 Studio 64 Figure 83. B1 Studio Temperature Comparison • HAR 101: 65 66 Figure 84. 2nd HAR 101 Temperature Comparison • HAR 102: 67 Figure 85. HAR 102 Temperature Comparison • MBS Corner: 68 Figure 86. MBS Corner Temperature Comparison • Studio 200: 69 70 Figure 87. Studio 200 Temperature Comparison • Studio 208: 71 Figure 88. Studio 208 Temperature Comparison • Studio 209: 72 73 Figure 89. Studio 209 Temperature Comparison • WAH B1: 74 Figure 90. WAH B1 Temperature Comparison • WAH Lower Rosendin: 75 76 Figure 91. WAH Lower Rosendin Temperature Comparison • WAH Upper Rosendin: 77 Figure 92. WAH Upper Rosendin Temperature Comparison • WAH 212: 78 79 Figure 93. WAH 212 Temperature Comparison The room temperatures are visually affected by the outdoor temperatures, but it is not for sure. Validating this effect by conducting correlation analysis, which is a statistical method that measures the direction and strength of the linear relationship between two variables. P-value is used to show the result of the analysis and it is a quantized value between 0 and 1. When the P-value of two variables is smaller, it means the correlation between the two variables are more significant. In general, a P-value that is smaller than 0.05 is considered as a “significant relationship”. Calculating the P-values for all spaces and show as below (See Table 7). 80 Table 7. Correlation Analysis Results As shown in the Table 7, the indoor temperatures of the ARCH Corner, the Courtyard, the HAR 101, and the HAR 102 were significantly affected by the outdoor temperature in November. Among them, the Courtyard (outdoor space), the HAR 101, and the HAR 102 have extremely significant correlation since their P-values are below than 0.01. For these spaces, the lower outdoor temperature changed, the more severe the indoor overcooling would be. For the other spaces, the outdoor temperature may be a factor that affect the indoor temperature but there are more other factors, for example the insulation, construction material, volume of the room, and more. 4.3.1 Degree Hours Degree hours are defined here by summing the differences between actual indoor temperature and desired temperature. It can help to quantize the effect of overcooling. Since all data are filtrated by the working hours, which is from 9 am to 5 pm, so the daily degree hours could be calculated by average degrees-of-overcooling multiply 8 hours. In total, the degree hours for all spaces could be concluded as below (See Table 8): 81 Table 8. Overcooling Degree Hours for 14 Spaces As shown in the Table 8, all spaces in the Watt Hall are overcooled in November, but they are overcooled for different degree-hours. The 2 nd Floor office was slightly overcooled space for only 94 degree-hours. The HAR 102, the Studio 208, and the Studio 209 were the most severe overcooling spaces since they all were overcooled more than 500 degree- hours. The Table 8 illustrates that even though all spaces in the same building are overcooled, the overcooling conditions could be different. Therefore, it is better to monitor the individual thermal condition for room to room, and IoT sensors could be used to quantize each room’s overcooling conditions. 82 Table 9. Overcooling Hour Percentage The Table 9 shows the percentage of overcooling occupied hours that below than 76 Fahrenheit. Since the detecting interval of the IoT sensors is 15 minutes, so the overcooling hours could be calculated by counting the number of occupied data and then divide 4. In summary, this table illustrates that most rooms in the Watt Hall were overcooled during occupied hours in November. A total of 9 spaces were 100% overcooled and the other spaces had more than 95% occupied hours overcooled except the WAH 322 and the WAH 324, which are controlled by independent AC unit. 4.4 Summary In summary, the researcher collected the indoor temperature and humidity for 18 spaces in the Watt Hall with the IoT sensors and he downloaded the outdoor temperature through the nearest meteorological station. Also, he was provided 177 valid occupants’ feedback from the TrojanSense team. After the basic analysis and comprehensive analysis, several findings are provided as below: • What is the conditioning mode for the Watt Hall in November? The average indoor temperatures for spaces that controlled by the Central AC system are below than the WAH 324, which is a space that usually doesn’t have air conditioning, and most spaces in the Watt Hall are controlled by the Central AC system. Therefore, it could be deduced that the Watt Hall was in cooling mode in November. 83 • Whether the current temperature policy has achieved the desired effect? The current temperature policy is cooled spaces into 76 Fahrenheit in cooling season. However, the research found that occupants complained that most spaces in the Watt Hall were overcooled from feedback. It also be validated by calculating the degrees-of overcooling and overcooling hours with IoT data. Moreover, the overcooling conditions are different from space to space as well. • What fraction of occupied hours are indoor temperatures outside of the USC campus setpoint policy? After calculation, the room temperature that is below than 76 Fahrenheit accounts for more than 95% occupied time (9 am to 5 pm) for most spaces, and 9 spaces were overcooled all the occupied time. • What is the degree of overcooling? The degrees-of-overcooling for spaces in the Watt Hall fluctuates from -1.30 to -6.27 degrees, and the average degree is -3.68 degrees. • How does outdoor temperature affect indoor temperature in a room? After analyzing the correlation between outdoor and indoor temperatures, the research found that the effect of outdoor temperature to indoor temperature are different from space to space. This kind of correlation are not significant for most rooms since there are many other factors may affect the indoor temperature. The HAR 101 and the HAR 102 are only spaces that could be considered as “significantly affected” by the outdoor temperature according to correlation analysis 84 5. DISCUSSION This chapter will discuss three significant findings according to the results of the three rounds of experiments at first: (1) Why is it important to have room-level air/humidity sensing? (2) What benefits are there from IoT data that could not be obtained from a regular zone-level thermostat data stream? (3) How can IoT data help inform decision-making? Then, the research will evaluate the whole workflow and possible validation method. At the last, the researcher will talk about the current limitations of the workflow. Also, the research will present some possible improvements according to these limitations. 5.1 Significance of the Thesis • Why is it important to have room-level air/humidity sensing? As analyzed in the Chapter 4, the different rooms may have different overcooling conditions even though they are controlled by one central AC unit. The degrees-of-overcooling may vary from several to dozens of degrees according to the current school temperature policies. And more than 95% occupied hours in most Watt Hall spaces were dominated by overcooling. However, the situation is opposite in some spaces, like the WAH 322 and the WAH 324 which are controlled by independent AC units. The two rooms have only 40% and 20% occupied hours were overcooled separately. If people only look at the general thermal condition of the whole building, they may conclude a wrong result that the building is in comfort zone, which ignore the extreme overcooling in some rooms. What is worse, the misleading conclusion may aggravate the overcooling condition, reduce occupants’ comfort level, and waste unnecessary energy. IoT sensors have the advantages including small, portable, attachable, and real-time. So that they could be installed in any rooms and would attract too much attention. With the campus-level free WiFi, IoT sensors can detect and transfer the real-time data which could be used to monitor the room-level thermal conditions for each room. That makes it possible for the FMS to adjust the HVAC system individually according to different overcooling or overheating conditions. Furthermore, the room-level IoT data can also help to determine both individual and general balance point temperature. Since the balance point temperatures are different between rooms, it is hard to confirm the general value for the whole building. Compared with empirical method, the IoT data from spaces can narrow the range of balance point temperature and deduce the value more accurately. • What benefits are there from IoT data that could not be obtained from a regular zone-level thermostat data stream? The regular method to build a thermal comfort model is based on zone-level thermostat data. However, people always have personalized needs about the surrounding work environment. Even under the same thermal conditions, someone wants to be warmer, while others want to be drier. It is hard to satisfy diverse requirements without advanced technology and hardware. The emergence of the IoT technology makes it possible to provide room-level data. More data based on individual spaces can be used to build more accurate comfort models and guide people to improve the work environment at the room level. It is hard for now to adjust the surrounding thermal condition around individual person in public spaces although IoT sensors could be carried by people. But the room-level adjustment can already meet many people’s needs. One question is how many IoT sensors are appropriate in one room. In general, the more IoT sensors in one room, the more accurate results could be gotten, but the analyzing process would be more complex. The researcher suggests that the number of IoT sensors should be considered with the headcount in the room and the area/volume of the room. Confirming one number according to the number of people (e.g. 1 sensor for every 50 people) and calculating the other number according to the area/volume (e.g. 1 sensor for 1000 square feet), then comparing these two numbers and taking the larger one as the final result. Of cause, the decision maker needs to balance the cost and expected effect as well. Moreover, the IoT sensor should be placed in the middle of the preplanned space as much as possible, but it is better to put it on the less obvious location in case of losing. • How can IoT data help inform decision-making? The IoT data could be seen as the source of “water”, and the analyzing process is the canal leading water to farmland. So, guiding the FMS to improve adjustment decision is irrigation. The final goal is to improve the campus comfort level and improve the work environment of students and personnel on campus. As discussed in the Chapter 4, a large 85 volume of physical data could be analyzed and processed to more secondary data, like degrees-of-overcooling, overcooling hours, and so on. A lot of diagrams would be plotted during the process as well. Those secondary data and diagrams would deduce a conclusion, which is a summary of the running patterns of the objective world. Therefore, it could be used to guide the future decision-making and improve the current situation. The occupants in these spaces would have a better thermal comfort with a more rational decision, and the overall energy use efficiency would be improved as well. For example, the research collected the indoor temperature/humidity of 18 spaces in the Watt Hall and outdoor temperature through the nearest meteorological station. By comparing the indoor temperature and the current school temperature policies, the researcher concluded that most spaces in the Watt Hall were overcooled and the overcooling condition for each room are different. It informs that the FMS should adjust the HVAC system individually instead of a uniform preset temperature. Then, plotting the average indoor temperature and outdoor temperature into one diagram, it could be deduced that the current balance point temperature of the whole building is about 57 Fahrenheit and desired balance point temperature is about 58 Fahrenheit through linear regression analysis. It is a more accurate result compared with empirical method. Since the balance point temperature is the outdoor temperature when the indoor heat gains equal to heat losses, it can help the FMS to adjust the thermal condition of the whole building macroscopically, then they can calibrate the individual values according to each data set microscopically. 5.2 Evaluation of the Workflow The whole workflow could be divided into three phases: developing IoT sensors, collecting data, analyzing data and then informing decision-making. It is corresponding to the phases of the experimental research: experimental preparation, conducting an experiment, experimental analysis, and guiding practical work. If anyone of those phases has problems, the results would not be accurate to the actual condition and affect the analysis result. There is always possible to improve the workflow on the hardware. For example, using more accurate instruments, changing to plug- in power, decreasing detecting internals, and so on. On the other hand, it is inevitable to make errors during the analyzing process, like the calculation of overcooling hours, the deduction in linear regression analysis. Also, there are more factors that may affect the indoor temperature and occupants’ sensations, including the construction material, built year, insulation, occupants’ clothing insulation, occupants’ metabolic activities, etc. it is impossible to 100% match the analysis models with the objective world theoretically. But there is no need to pursue “100% match”, it is enough when the experimental result could be used to guide practical decision-making. 5.3 Validation One way to validate the analysis result is by contrast test. For example, adjusting the HVAC system of the Watt Hall according to a higher balance point temperature (e.g. 60 Fahrenheit) for the similar period of time, then monitoring the indoor thermal conditions with IoT sensors in the same spaces. After collecting enough data, comparing the overcooling conditions with previous results. If the overcooling conditions are significantly improved, it validates that the IoT data can be used to guide decision-making. Otherwise, it would need to consider more factors during the decision-making process. Occupants’ participation is an effective supplement to the validation. People’s preferences could be used to assess the overcooling condition of spaces. However, different persons’ sensations may be affected by many factors, like clothing insulation, psychological states, metabolic activities, and more. So, the effect of assessment would not as accurate as objective data. Another condition for validation is the adjustment of HVAC systems, which needs to collaborate with the FMS in USC. That has not been implemented because of the limitation of data volume and time. This research only collected and analyzed the data in the Watt Hall from October to November, so the result only represents the situation of the Watt Hall from late fall to winter. It is required more data around the whole campus for a whole year to conduct campus-level analysis. Therefore, there is no intact set of data that could be referenced by the FMS for now, and the validation has to be planned as the future work. 5.4 Limitations There are two core elements in the workflow, one is the instruments and the other one is data. So, limitations are generated from anyone of these two elements. For the instruments, there will always have more accurate sensors coming out as the development of the technology. Thus, the research was only talking about the problems when current sensors are in good condition. 86 5.4.1 Insufficient Participation Occupants’ feedback is one important part of the personalized comfort models. In theory, at least two data can determine a comfort zone, but it is not accurate enough for decision-making. The more data were collected, the more accurate comfort models could be gotten. However, lacking participants is a significant problem in this experiment. A total of 177 valid data limited the analysis as well as the development of comfort models. The reasons why people would not actively participate in voting include: (1) they do not need to since most spaces’ environments actually are sustainable. A human body can adjust itself according to the surrounding environment, so people would not aware of cold or hot as long as the indoor thermal condition does not beyond the threshold; (2) they cannot get the instant improvement as they submit feedback. The process of this project is collecting and analyzing data, then the data would be used for adjusting HVAC systems, which means there is a time-delayed adjustment after occupants’ voting. So, people will feel it is not working and they would not do it again when they were not satisfied for the first time; (3) submitting feedback is not a high-frequency action for most people. Compared to eating and going to the restroom, reporting sensation is a very low-frequency incident. Thus, people do not have the habit to open APP and submit their feedback. They just cannot remember to do this unless they are obviously uncomfortable. In terms of these possible reasons, there are some strategies to improve participation. For example, (1) sending a reminding email to remind students to vote. It is better to work with the Students Office and send a group email that every student can receive; (2) giving some small gifts for people who submitted feedback. A practical way is awarding people some points in the TrojanSense APP after each submission, and sufficient points could be used to exchange gifts that cater to students’ hobbies; (3) asking for professors’ help about assigning a voting mission to students. The required assignment can help to develop students’ voting habit, even though they do not feel uncomfortable. 5.4.2 Limited Battery Life Since the IoT sensors are run by battery power, they will die after 12-14 days. Thus, the researcher needs to replace the batteries about every two weeks. Although better batteries can sustain a longer time, it is still not a permanent solution to avoid a lot of labor work. What is worse, different sensors may die at a different time even though they started at the same time because of diverse physical environments, which increases the difficulty of battery replacement. Also, the researcher cannot 24-hours keep looking at the operating status of all sensors. Even though the MQTT has a function that sending a reminder to people when any sensor dies, it is hard to replace the batteries instantly. The result is that collected data would be intermittent and there are short breaks every two weeks. Therefore, the experiments ran three rounds from October to December and each round lasted for full battery life. A strategy is changing the battery to plug to supply power. It is a much more effective way to sustain the sensors constantly. But people need to consider whether there is an available socket in each space. Furthermore, a plug-in sensor would attract more attention, so the researcher also needs to figure out how to prevent losing sensors for a long time. 5.4.3 WiFi Accident The IoT sensors upload thermal data to the cloud through the campus-ranged free WiFi. So, a power outage or WiFi break may block the sensors sending data. Although this is a very small possibility incident, it did happen in the second round of detection in November. After a short-time power outage, almost all sensors stopped uploading data even after the power recovered. These sensors need to be restarted manually after a WiFi break. This is a latent threat to the stability of data collection, and one possible solution is adding a reconnection function in the code that controls sensors. 5.4.4 Contradictory Feedback There may be more than one person’s feedback at the same time, and the researcher needs to figure out one result that represents a group of occupants’ opinions. It is a critical problem that two or more occupants have the opposite opinion in the same space. This is not a new problem; many studies have discussed this for a long time since disagreement always happens in a public space. In essence, this is a problem that how to balance personal interests and public resource. The research only calculates the mean value of collecting data as the group’s preference for the sake of convenience. However, it may be affected by extreme values, for example, the result of one “+6” and three “-1” is “1”, which represents “1 degree warmer” but actually more people feel cool. Another strategy is listening to the majority rather than the minority. This seems reasonable since a group’s benefit should be prior to personal benefit. Nevertheless, it may introduce biases to other relatively few people, for example, the result of three “+1” and two “- 87 3” is “+1” according to the majority principle, which represents “1 degree warmer” but actually almost half people feel much more sensitive to hot. In this case, the researcher presented a formula to calculate the comfort value for a group of people: 𝐶 𝑜𝑚𝑓 𝑜𝑟 𝑡 𝑉𝑎 𝑙𝑢𝑒 = 𝛼 ( + 6 ) 𝑝 1 + 𝛽 ( + 5 ) 𝑝 2 + ⋯ + 𝛾 ( − 6 ) 𝑝 13 𝑃 In this formula, numbers (from +6 to -6) represent the values that people voted, and α, β, … γ represent weighting factors corresponding to each value. The p 1, p 2 … p 13 represent the number of people that voted under each value, and the P means the total number of people. The weighting factors are determined by the control experiment. The experiment could like this: (1) organizing a group of people (the more the better) into a big room and making sure that the indoor temperatures in every corner of this room are the same and they could be controlled simultaneously. Every person in this room has a voting controller with +6 to -6 buttons and there is a screen shows the real-time voting results; (2) changing the indoor temperature and asking people to vote after every adjustment. Then setting the room temperature to a relative comfort level that most people feel neutral (voted “0”) and kicking out the people that didn’t feel neutral; (3) turning down the room temperature 1 degree and letting the left people vote again. Then recording the percentage of people that voted “+1”, which is the weighting factor of “-1”; (4) repeating this process for all weighting factors and then finishing this formula. Although the research did not conduct this experiment, it could be inference that the weighting factors of “3” or “4” are bigger than “6” or “1”. For example, assuming this formula is: 𝐶 𝑜𝑚 𝑓𝑜 𝑟 𝑡 𝑉𝑎𝑙𝑢 𝑒 = 0 . 3 ( + 6 ) 𝑝 1 + 0 . 4 ( + 5 ) 𝑝 2 + 0 . 6 ( + 4 ) 𝑝 2 + 0 . 7 ( + 3 ) 𝑝 2 + 0 . 6 ( + 2 ) 𝑝 2 + 0 . 4 ( + 1 ) 𝑝 2 + 0 . 4 ( − 1 ) 𝑝 2 + 0 . 6 ( − 2 ) 𝑝 2 + 0 . 7 ( − 3 ) 𝑝 2 + 0 . 6 ( − 4 ) 𝑝 2 + 0 . 4 ( − 5 ) 𝑝 2 + 0 . 3 ( − 6 ) 𝑝 13 𝑃 There is a voting result collected at the same time: one “+6”, one “+4”, and three “-1”, then the final comfort value should be: 𝐶 𝑜𝑚𝑓 𝑜𝑟 𝑡 𝑉𝑎𝑙 𝑢 𝑒 = 0 . 3 ( + 6 ) × 1 + 0 . 6 ( + 4 ) × 1 + 0 . 4 ( − 1 ) × 3 1 + 1 + 3 = 0 . 6 The result means turning on 0.6 degrees would be comfortable to most people. However, the results of “average method” and “majority method” are 1.4 and -1, which may lead to more uncomfortable and consumed more energy. 5.5 Summary The research conducted three rounds of experiment with IoT sensors and analyzed the IoT data, several significant findings include: (1) running room-level temperature/humidity sensing is important since it can monitor extreme overcooling or overheating conditions and reduce building energy waste; (2) the IoT data can benefit the indoor thermal condition individually and maximize occupants’ work efficiency compared to regular zone-level thermostat data; (3) IoT data can help decision makers to figure out the better decision from the building level to the room level. The research follows the basic process of experimental research: experimental preparation, conducting an experiment, experimental analysis, and guiding practical work. However, the research did not validate the analyzing result since validation needs to be conducted by collaborating with the FMS in USC. It is better to run this experiment as long as enough to conclude more accurate results. Moreover, there are several limitations that restricted the accuracy of analyzing results, including insufficient participants, limited battery life, WiFi accident, and contradictory feedback. Many strategies are provided to avoid these limitations, but they have to be included in future work because of limited time, instruments, and work forth. 88 6. CONCLUSION This chapter will summarize improvements to the current workflow and discuss the possible future work that may improve the research into a higher field, including more factors of people’s sensations, the energy impact, man- machine interaction, and the application of machine learning. After that, it will conclude the whole thesis. 6.1 Improvement to the current workflow There is no doubt that more accurate and advanced instruments will come out as the development of the technology. Updating the instruments in the workflow can always get better results, but people also need to consider the total cost. Moreover, there are many limitations that restricted the accuracy of comfort models, including insufficient participants, limited battery life, WiFi accident, and contradictory feedback. Many strategies are possible to improve the current workflow, for example, encouraging more people to participate; changing the battery power to plug power; figuring out a weighting formula to evaluate contradictory feedback. 6.2 Future work In terms of the future work, there are four possible fields to discuss: (1) the factors of people’s feedback; (2) the energy impact of comfort models; (3) the efficiency of data collection; (4) the application of AI. 6.2.1 Consider More Factors Right now, the research only investigated the connection between thermal condition (including indoor temperature, indoor humidity, and outdoor temperature) and occupants’ feedback. However, there are many factors that may affect human’s sensation, like gender, age, race clothing insulation, metabolic rate, etc. if possible, collecting every person’s data in these aspects when they are submitting feedback. Then, a large amount of data could be used for more complex analysis, and people can better understand occupant’s preferences. 6.2.2 Energy Impact One of the ultimate goals for building comfort models is saving more energy. The energy performance and comfort level are important factors when people consider about the temperature setpoints. They should be kept in a balance rather than just focusing only one side. The research did not calculate the energy use for thermal comfort models because it involves many works, including building 3D models, introducing factors, validation and adjustment. 6.2.3 Man-Machine Interaction Basically, there three phrases of the Internet of Things: (1) transmitting information from instruments to the Internet; (2) transmitting information from the Internet to instruments; (3) transmitting information between instruments and the Internet. Right now, the research only achieves the first phrase for IoT, that is transmitting thermal data from the sensors to the Internet. But the IoT sensors are working by preset rules, they cannot react to the instant feedback. For example, these sensors can only detect the room thermal condition by every 15 minutes, but there may be small distinction between periodical room temperature and the temperature when receiving feedback. A possible strategy is updating the IoT Sensor and making it smarter so that human and the instrument can exchange information. As a result, the IoT sensor would return the instant thermal data when people submit feedback. 6.2.4 Machine Learning As the fast development of technology, Machine Learning and Big Data have become the trends of the future. These techniques can significantly improve the work efficiency and accuracy especially for experimental research. As discussed above, the more could be gotten, the thermal comfort models would be more accurate. There are more than 10,000 data only for three months in this research, so it is hard to image how big the data set would be for a longer time. Moreover, processing and analyzing so many data is a huge work that consumes a large amount of labor and time. Therefore, introducing Machine Learning into this process can process the same amount of data within several seconds, and it also can predict and generate a more accurate comfort model. 6.3 Summary Nowadays, people have higher and higher requirements about work environment. Instead of a uniform adjustment for all spaces in a building, a personalized monitoring and analyzing based on each space would maximize occupants’ thermal comfort and improve energy efficiency. As the improvement of technology, IoT makes it possible that running 89 a room-level temperature/humidity sensing, which is important since it can monitor extreme overcooling or overheating conditions and reduce building energy waste. Also, the IoT data can benefit the indoor thermal condition individually and maximize occupants’ work efficiency compared to regular zone-level thermostat data. Furthermore, IoT data can help decision makers to figure out the better decision from the building level to the room level. The USC campus consumes a large amount of energy, but many rooms didn’t reach the temperature policies’ standard according to the HOBO tests. Because of a lack of appropriate data and means to analyze and interpret, the FMS staff cannot dynamically adjust HVAC system. The research collected the indoor temperature and humidity for 18 spaces in the Watt Hall with the IoT sensors and collected about 177 valid occupants’ feedback with the TrojanSense APP. After analysis, it verified the former hypothesis that many spaces were overcooled in November, and most occupants in these spaces wanted to be 1-3 degrees warmer than the current HVAC set points. 90 REFERENCE [1] Barnaby, F. (1987). Royal Swedish Academy of Sciences Our Common Future : The " Brundtland Commission " Report Published by : Springer on behalf of Royal Swedish Academy of Sciences Our Common Future- The " Brundtland Commission " Report Synopsis. Springer, 16(4), 217–218. [2] “What is sustainability”. Retrieved from the Global Footprints website: www.globalfootprints.org. [3] Adam, Z. (2017, June 12). Students develop NASA flight instruments to study ozone depletion over Midwest. Science that soars. Retrieved from: www.seas.harvard.edu/news/2017/06/science-that-soars. [4] Judy, L. (2011, July 27). Researchers use wireless technology to speed transition to electric vehicles. UCLA Newsroom. Retrieved from: newsroom.ucla.edu/stories/rajit-gadh-electric-vehicle-research-211170. [5] University of Southern California. (2013). USC Graduate Admission. Retrieved from the Wayback Machine website: https://web.archive.org/web/20131029204247/http://www.usc.edu/admission/graduate/learn/ [6] University of Southern California. (2008). All Enrolled Students, by Class Level [PDF document]. Retrieved from the Wayback Machine website: https://web.archive.org/web/20080625134601/http://www.usc.edu/private/factbook/2008/all_byclass_08.pdf [7] The Unlimited Carbon Assistance Network (UCAN). (2018). University of Southern California Greenhouse Gas Emissions Inventory Summary 2014-2017 [PDF document]. Retrieved from https://green.usc.edu/files/2018/08/USC_GHG_emissions_survey2018_S-web-V_081518.pdf [8] Advancement of Sustainability in Higher Education (AASHE). (2018). 2018 Sustainable Campus Index [PDF document]. Retrieved from http://www.aashe.org/wp-content/uploads/2018/08/SCI-2018.pdf [9] “Mission Statement” Retrieved from the USC website: https://about.usc.edu/policies/mission-statement/ [10] University of Southern California. (2015). Sustainability 2020 [PDF document]. Retrieved from: https://green.usc.edu/2020-plan/ [11] “Thermal Energy Storage” Retrieved from the USC website: https://facilities.usc.edu/leftsidebar.asp?ItemID=462 [12] Gershenfeld, N., Krikorian, R., & Cohen, D. (2004). The internet of things. Scientific American (Vol. 291). https://doi.org/10.1038/scientificamerican1004-76 [13] Almirall, E., & Wareham, J. (2011). Living Labs: Arbiters of midand ground-level innovation. Technology Analysis and Strategic Management, 23(1), 87–102. https://doi.org/10.1080/09537325.2011.537110 [14] “Living Laboratory Projects” Retrieved from the UCLA website: https://www.sustain.ucla.edu/living-laboratory- projects/ [15] Hewltt, A. (2015, Jan 12). More important than money? Environmental health benefits inspire people to cut back on electricity. UCLA Newsroom. Retrieved from: http://newsroom.ucla.edu/releases/more-important-than- money-environmental-health-benefits-inspire-people-to-cut-back-on-electricity [16] Contreras, J. (2017, March 8). Is your room too hot or cold? Retrieved from the Energy Conservation Office of UC Davis website: https://eco.ucdavis.edu/blog/welcome-thermoostat [17] J.G. Allen, P. MacNaughton, J.G.C. Laurent, S.S. Flanigan, E.S. Eitland, J.D. Spengler, Green buildings and health, Curr. Environ. Health Rep 2 (2015)250–258, http://dx.doi.org/10.1007/s40572-015-0063-y. [18] C. Talon, N. Goldstein, Smart Offices: How Intelligent Building Solutions Are Changing the Occupant Experience, Navigant Consulting, Inc, 2015. [19] D. Li, C.C. Menassa, V.R. Kamat, Personalized human comfort in indoor building environments under diverse conditioning modes, Build. Environ. 126 (2017) [20] L. Jiang, R. Yao. Modelling personal thermal sensations using C-Support Vector Classification (C-SVC) algorithm, Build. Environ. 99 (2016) 98–106, http://dx.doi.org/10.1016/j.buildenv.2016.01.022. [21] Q. Zhao, Y. Zhao, F. Wang, J. Wang, Y. Jiang, F. Zhang, A data-driven method to describe the personalized dynamic thermal comfort in ordinary office environment: from model to application, Build. Environ. 72 (2014) 309–318, http://dx.doi.org/10.1016/j.buildenv.2013.11.008. [22] Olesen BW. International standards for the indoor environment. Indoor Air 2004; 14:18e26. [23] ISO Standard 7730. Moderate thermal environmentsddetermination of the pmv and ppd indices and specification of the conditions for thermal comfort. International Organisation for Standardisation; 1994. [24] Nicol JF, Humphreys MA. Thermal comfort as part of a self-regulating system. Building Research and Practice (Journal of CIB) 1973;6(3):191e7. [25] Humphreys MA, Nicol JF. The validity of iso-pmv for predicting comfort votes in every-day thermal environments. Energy and Buildings 2002;34(6): 667e84. 91 [26] M. Schweiker, X. Fuchs, S. Becker, M. Shukuya, M. Dovjak, M. Hawighorst, J. Kolarik, Challenging the assumptions for thermal sensation scales, Build. Res. Inf. 45 (2017) 572–589, http://dx.doi.org/10.1080/09613218.2016.1183185. [27] G.A. Miller, The magical number seven, plus or minus two: some limits on our capacity for processing information, Psychol. Rev. 63 (1956) 81. [28] R. Rana, B. Kusy, R. Jurdak, J. Wall, W. Hu, Feasibility analysis of using humidex as an indoor thermal comfort predictor, Energy Build. 64 (2013) 17–25, http://dx.doi. org/10.1016/j.enbuild.2013.04.019. [29] A. Ghahramani, C. Tang, B. Becerik-Gerber, An online learning approach for quantifying personalized thermal comfort via adaptive stochastic modeling, Build. Environ. 92 (2015) 86–96, http://dx.doi.org/10.1016/j.buildenv.2015.04.017. [30] J.F. Nicol, I.A. Raja, A. Allaudin, G.N. Jamy, Climatic variations in comfortable temperatures: the Pakistan projects, Energy Build. 30 (1999) 261–279, http://dx. doi.org/10.1016/S0378-7788(99)00011-0. [31] W. Rybczynski. (2017, Nov 29). Kieran Timberlake’s Cool Experiment. Architect. Retrived from https://www.architectmagazine.com/design/kierantimberlakes-cool-experiment_o [32] Sanguinetti, A., Hybrid, P., Vehicle, E., & Davis, U. C. (2016). TherMOOstat : Occupant Feedback to Improve Comfort and Efficiency on a University Campus. ACEEE Summer Study 2016, (Paciuk 1990), 1–12. [33] Sihombing, R., & Coors, V. (2018). LINKING 3D BUILDING MODELS, MAPS AND ENERGY-RELATED DATA IN A WEB-BASED VISUALIZATION SYSTEM, IV(October), 4–5. [34] Mccaffrey, R., Coakley, D., Corporation, M. E., Keane, M., & Melvin, H. (2015). Development of a Web-based BMS Data Visualisation Platform Using Building Information Models, (July). https://doi.org/10.13140/RG.2.1.4169.5840 [35] Utzinger, Michael; Wasley, James. "Vital Signs Curriculum Materials Project" (PDF). UC Berkeley. College of Environmental Design. Archived from the original (PDF) on 12 June 2012. Retrieved 25 November 2014. 92 APPENDIX Source Code of Arduino int br = 0; unsigned long startMillis; unsigned long currentMillis; const unsigned long period = 1000; //the value is a number of milliseconds, ie 1 second #define uS_TO_S_FACTOR 1000000 /* Conversion factor for micro seconds to seconds */ #define TIME_TO_SLEEP 900 /* Time ESP32 will go to sleep (in seconds) */ RTC_DATA_ATTR int bootCount = 0; #include <WiFi.h> #include "Adafruit_MQTT.h" #include "Adafruit_MQTT_Client.h" // DHT sensor stuff #include <dht.h> dht DHT; #define DHT22_PIN 21 float temp = 0; float humidity = 0; // Temperature Sensor #define DHTPIN 21 //check the temperature pin from the adafruit #define DHTTYPE DHT22 //type of sensor // Battery pin #define VBATPIN A13 // pin for reading battery voltage /************************* WiFi Access Point *********************************/ 93 #define WLAN_SSID "USC Guest Wireless" //wifi name #define WLAN_PASS "" //wifi password /************************* Adafruit.io Setup *********************************/ #define AIO_SERVER "io.adafruit.com" #define AIO_SERVERPORT 1883 // use 8883 for SSL, 1883 for unsecure #define AIO_USERNAME "buildingscience" //username for adafruit iot dashboard #define AIO_KEY "1281bb70b3f277a12c018fe75d6f464017fa46a2" //secret key for adafruit iot dashboard /************ Global State (you don't need to change this!) ******************/ // Create an ESP8266 WiFiClient class to connect to the MQTT server. WiFiClient client; // Setup the MQTT client class by passing in the WiFi client and MQTT server and login details. Adafruit_MQTT_Client mqtt(&client, AIO_SERVER, AIO_SERVERPORT, AIO_USERNAME, AIO_KEY); /****************************** Feeds ***************************************/ // match end with "KEY" from adafruit IO Adafruit_MQTT_Publish RoomTemperature = Adafruit_MQTT_Publish(&mqtt, AIO_USERNAME "/feeds/node-3.temp"); Adafruit_MQTT_Publish RoomHumidity = Adafruit_MQTT_Publish(&mqtt, AIO_USERNAME "/feeds/node-3.humid"); Adafruit_MQTT_Publish BatteryVoltage = Adafruit_MQTT_Publish(&mqtt, AIO_USERNAME "/feeds/node-3.v"); // Bug workaround for Arduino 1.6.6, it seems to need a function declaration // for some reason (only affects ESP8266, likely an arduino-builder bug). void MQTT_connect(); void setup() { Serial.begin(115200); 94 //Connect to WiFi access point. //Serial.println(); //Serial.println(); //Serial.print("Connecting to "); //Serial.println(WLAN_SSID); // first check time to see how long this will take startMillis = millis(); br = 0; WiFi.begin(WLAN_SSID, WLAN_PASS); // try to connect Serial.println("Trying to connect to WiFi..."); // ok this basically stalls until wifi connection is made... could take forever !!!!!!!!!!!!!!!!!!!!!!! while (WiFi.status() != WL_CONNECTED) { delay(500); Serial.print("."); currentMillis = millis(); if (currentMillis - startMillis > 60000) { br = 2; break; } } Serial.println(); Serial.println("WiFi connected"); Serial.println("IP address: "); Serial.println(WiFi.localIP()); delay(3000); // Ensure the connection to the MQTT server is alive (this will make the first // connection and automatically reconnect when disconnected). See the MQTT_connect // function definition further below. 95 MQTT_connect(); int detectcount = 0; int chk = DHT.read22(DHT22_PIN); temp = DHT.temperature; temp = temp * 1.8; temp = temp + 32; humidity = DHT.humidity; Serial.print("TempF: "); Serial.println(temp); Serial.print("Humidity: "); Serial.println(humidity, 1); while (temp < 0 || humidity > 100) { Serial.println("temp error... taking another measurement"); detectcount++; temp = DHT.temperature; temp = temp * 1.8; temp = temp + 32; humidity = DHT.humidity; Serial.print("detectcount = "); Serial.println(detectcount); delay(3000); if (detectcount > 5) { sleep(); } } Serial.print("TempF: "); Serial.println(temp); Serial.print("Humidity: "); Serial.println(humidity, 1); 96 float measuredvbat = analogRead(VBATPIN); measuredvbat = measuredvbat * 2; // we divided by 2, so multiply back //measuredvbat = measuredvbat * 3.3; // Multiply by 3.3V, our reference voltage measuredvbat = measuredvbat / 1024; // convert to voltage Serial.print("VBat: " ); Serial.println(measuredvbat); RoomTemperature.publish(temp); //publish temperature RoomHumidity.publish(humidity); //publish humidity BatteryVoltage.publish(measuredvbat); //publish voltage delay(3000); sleep(); unsigned long waitMillis; unsigned long stopMillis; waitMillis = millis(); delay(1800000); stopMillis = millis(); if (stopMillis - waitMillis > 1800000) { Serial.println("going to RESTART .............................."); //try restart //https://techtutorialsx.com/2017/12/03/esp32-arduino-software-reset/ ESP.restart(); } } // end br = 0 test void loop() { } void sleep() { Serial.println("going to sleep now ..."); 97 esp_sleep_enable_timer_wakeup(TIME_TO_SLEEP * uS_TO_S_FACTOR); esp_deep_sleep_start(); } // Function to connect and reconnect as necessary to the MQTT server. // Should be called in the loop function and it will take care if connecting. void MQTT_connect() { int8_t ret; // Stop if already connected. if (mqtt.connected()) { return; } //Serial.print("Connecting to MQTT... "); uint8_t retries = 3; while ((ret = mqtt.connect()) != 0) { // connect will return 0 for connected //Serial.println(mqtt.connectErrorString(ret)); //Serial.println("Retrying MQTT connection in 5 seconds..."); mqtt.disconnect(); delay(5000); // wait 5 seconds retries--; if (retries == 0) { // basically die and wait for WDT to reset me //while (1); break; } } //Serial.println("MQTT Connected!"); }
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Real-time simulation-based feedback on carbon impacts for user-engaged temperature management
PDF
Enhancing thermal comfort: air temperature control based on human facial skin temperature
PDF
Enhancing thermal comfort: data-driven approach to control air temperature based on facial skin temperature
PDF
Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance
PDF
A BIM-based visualization tool for facilities management: fault detection through integrating real-time sensor data into BIM
PDF
Exploration for the prediction of thermal comfort & sensation with application of building HVAC automation
PDF
Developing a data-driven model of overall thermal sensation based on the use of human physiological information in a built environment
PDF
Mitigating thermal bridging in ventilated rainscreen envelope construction: Methods to reduce thermal transfer in net-zero envelope optimization
PDF
Natural ventilation in tall buildings: development of design guidelines based on climate and building height
PDF
IEQ, sleep quality, and IoT: meta-analysis on improving IEQ and sleep quality using IoT
PDF
Smart buildings: employing modern technology to create an integrated, data-driven, intelligent, self-optimizing, human-centered, building automation system
PDF
Multi-occupancy environmental control for smart connected communities
PDF
Occupant-aware energy management: energy saving and comfort outcomes achievable through application of cooling setpoint adjustments
PDF
Nursing comfort study: are hospitals performing for nurses? IEQ analysis of working conditions inside hospitals
PDF
Impacts of indoor environmental quality on occupants environmental comfort: a post occupancy evaluation study
PDF
Impacts of building performance on occupants' work productivity: a post occupancy evaluation study
PDF
Leaf module panel system: novel vertical wall wind energy production
PDF
Energy performance of different building forms: HEED simulations of equivalent massing models in diverse building surface aspect ratios and locations in the US
PDF
Single & multi-family housing: the complex behind the simple: a kit of parts for minimally invasive interventions
PDF
Adaptive façade controls: A methodology based on occupant visual comfort preferences and cluster analysis
Asset Metadata
Creator
Dong, Zhengao
(author)
Core Title
Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus
School
School of Architecture
Degree
Master of Building Science
Degree Program
Building Science
Publication Date
07/24/2019
Defense Date
05/10/2019
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Internet of things (IoT),OAI-PMH Harvest,overcooling,temperature setpoint policy,thermal comfort
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Konis, Kyle (
committee chair
), Choi, Joon-Ho (
committee member
), Noble, Douglas E. (
committee member
)
Creator Email
varydza@gmail.com,zhengaod@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-189796
Unique identifier
UC11662779
Identifier
etd-DongZhenga-7602.pdf (filename),usctheses-c89-189796 (legacy record id)
Legacy Identifier
etd-DongZhenga-7602.pdf
Dmrecord
189796
Document Type
Thesis
Format
application/pdf (imt)
Rights
Dong, Zhengao
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
Internet of things (IoT)
overcooling
temperature setpoint policy
thermal comfort