Experiences with Sensors for Energy Efficiency in ... - CiteSeerX
Experiences with Sensors for Energy Efficiency in Commercial Buildings Branislav Kusy1 , Rajib Rana1 , Phil Valencia1 , Raja Jurdak1 , and Josh Wall2 1
CSIRO, Autonomous Systems, Brisbane, QLD, Australia, [email protected] 2 CSIRO, Energy Technology, Newcastle, NSW, Australia
Abstract. Buildings are amongst the largest consumers of electrical energy in developed countries. Building efficiency can be improved by adapting building systems to a change in the environment or user context. Appropriate action, however, can only be taken if the building control system has access to reliable real-time data. Sensors providing this data need to be ubiquitous, accurate, have low maintenance cost, and should not violate privacy of building occupants. We conducted a three-year study in a mid-size office space with 15 offices and 25 people. Specifically, we focused on sensing modalities that can help improve energy efficiency of buildings. We have deployed 25 indoor climate sensor nodes and 41 wireless power meters, submetered 12 electric loads in circuit breaker boxes, logged data from our building control system and tracked activity on 40 desktop computers. We summarize our experiences with the cost, data yields, and user privacy concerns of the different sensing modalities and evaluate their accuracy using ground-truth experiments.
1
Introduction
Climate change and greenhouse gas emissions are forcing society to reevaluate energy efficiency practices. Buildings consume a large portion of the total energy in industrial countries and most of this energy is provided by conventional fossil-based power plants. As a result, numerous techniques for improving energy efficiency of commercial buildings have been proposed in recent years. Examples include lighting systems and window shaders that react to changing daylight and occupancy conditions [15], software products that manage and control power consumed by PC workstations [7], and Heating Ventilation and Air Conditioning (HVAC) systems that adapt to occupancy, electricity price, and weather forecast [3,6,4]. Many of these systems rely on real-time sensing of environmental data and user context. Future building control systems will no doubt deploy and integrate numerous sensors in their core functionality. The advantage of the sensor and actuator convergence in building control systems is that individual sensors can be reused for different tasks. For example, power-meters can be used to detect
occupancy of offices based on power consumption of desktop PCs and LCD monitors [9]. Thermal comfort of people can be estimated using office temperature and user feedback [5]. The overhead of installation and maintenance of building sensors is thus expected to be amortized by the costs of operating the whole building. Despite the exploration of multiple sensing modalities for promoting energy efficiency in buildings, there is still limited work on comprehensive comparison of their effectiveness. The goal of this work is to quantify advantages and disadvantages of different sensing modalities in commercial office buildings. We collected data over a three year period from environmental sensor nodes with temperature, light, humidity, and presence sensors deployed in every office in our building and from power-meters measuring energy usage of office appliances. We also contracted electricians to install submeters on load circuits and obtained access to data from our building control system. Finally, we developed a PC application that tracked activity of the users on desktop computers and allowed us to remotely survey users about their thermal comfort. We evaluate installation and operational costs of the different building sensors. We found that data yields of low cost sensors tend to decrease over time due to hardware failures. Battery operated devices introduce frequent black-out periods due to lack or negligence of maintenance. Software-based sensing solutions are also surprisingly difficult to maintain, despite existing IT policies of pre-installing the software on new computers. Permanently powered commercial products, on the other hand, perform reliably on a long-term basis. We also evaluate accuracy of the different sensors by conducting ground truth experiments. We mostly focus on detecting office occupancy and estimation of thermal comfort of individual users. Our findings show that power-meters are 10-20% less accurate in detecting office occupancy than passive infrared sensors and suffer from high false positive rates. PC activity loggers perform better than power-meters, but are generally rejected by users due to privacy concerns. Interestingly, users accept power-meters connected to their desktop PCs despite us disclosing their use for occupancy detection. Finally, we conducted several human-in-the-loop experiments aiming at energy usage reduction of HVAC and office appliances. Our empirically-based simulations indicate that 42% reduction in HVAC energy usage can be achieved over a centrally managed HVAC with static space temperature setpoints. We also found that presenting energy efficiency data to users can lead to behavior change and energy reduction, achieving 10% energy savings on office appliances. Our findings help justify the need for fine-grained continuous sensing of office occupancy, office climate, and energy usage for building-scale systems.
2
Experimental Setting
We built a comprehensive system for collecting data relevant to energy efficiency and energy usage in commercial buildings. Specifically, we focused on collecting information about (1) presence of people in their offices, (2) energy usage of office
Sensing'
Personal' Personal Climate Climate'Meter' Monitor
Wireless Power Personal' Meter Power'Meter'
ComfortSENSE™' PC'Applica1on'
Aggregate' Power'Meter'
Building'Control' System'
Processing' Thermal' Comfort'
Occupancy'
Climate' Data'
Energy' Usage'
Control' HVAC%% Control%
Informa1on' PC/LCD% Control%
User% Behavior%
Climate,% Energy% Sta8s8cs%
Energy% Usage/ Efficiency%
Fig. 1. Left: Data from low level sensors can be used to drive adaptive environment control and improve our understanding of the building energy usage. Right: Office building used for our deployment. Icons depict locations of sensors in offices.
appliances and the HVAC, and (3) subjective perception of thermal comfort of users. Figure 1 (left) provides an overview of our architecture that includes multiple sensor inputs within a typical office building. Real-time collection of this data is important for closing the loop between building control, user behavior and comfort, and changes in the environment. We deployed our system on a single floor of an office building that included 3 single-occupancy offices, 10 double-occupancy offices, 1 conference room, and 2 open-plan student offices (see Figure 1 right). We describe each of our 5 sensing techniques in more detail in the following text. Personal Climate Monitors (Climate) and Power Meters (Power). We built two sensor platforms, climate meters and power meters, based on our inhouse sensor network board. The board uses Intel 8051 processor and sub 1GHz Nordic radio and is optimized for low power and small size. The nodes self-organize to provide a multi-hop wireless mesh and run on a single battery charge for approximately 2 months. We sampled climate and power meters every 60 seconds and transmitted the data over the wireless network to a central database. The climate monitors also transmitted their current battery voltage level for maintenance purposes. We deployed one climate monitor per user, most frequently next to the user’s LCD monitor. We measured office climate and occupancy using temperature, humidity, light, and motion sensors. The power meters measured current, voltage, power, and cumulative energy consumption of electric appliances at a power socket level (similar to [9]). We augmented off-the-shelf power meters with our wireless sensor board to enable real-time wireless data collection of the energy data. We were limited to 2 power meters per user, thus we generally measured an aggregate consumption of a
desktop PC and LCD monitor with one power meter and used the other power meter for external hard drives, laptops, and fans. Commercial Power Meters (BCM). We installed Powerlogic PM9C power meters in breaker circuits (thus we call this modality BCM) to measure the aggregate energy usage of 12 electric circuits in our building, including user office sockets, lighting, hot-water heaters, and printers. We connected the meters to our internal Ethernet network and periodically recorded voltage, current, and power on each of the electrical circuits every 60 seconds in the database. Building Control and Management System (BMS). Our building is controlled by Tridium JACE controllers connected together by an internal modbus network. The access to the controller is restricted to our property managers who use it on a daily basis, primarily to inspect and fine-tune performance of the heating, ventilation, and air-conditioning (HVAC) system. We have obtained access to the system and used standard Open Building Information Exchange (oBIX) RESTful web interface to log data in the database and to control temperature in the offices during user trials. Our office space is logically divided into 4 air-conditioning zones and we used a python script to collect 11 sensor inputs per air-conditioning zone, including information about chillers, heaters, supply fans, indoor temperature, and space temperature setpoint that the HVAC is currently set to. Unlike the fine-grained office-level temperature data obtained by the personal climate monitors, BMS has one temperature sensor per zone. PC Application (CSENSE, SURVEY). Finally, we used ComfortSENSETM PC application [17] to track user activity on desktop computers and to administer user surveys. The application runs as a background process on desktop computers and can be controlled from a central server to display user surveys on all registered computers. Surveys focus mainly on user thermal comfort and their satisfaction with HVAC performance. Additionally, the application tracks key-strokes and mouse events on the host operating system and records this information every 5 minutes in the database. Users can opt-out of activity monitoring in case of privacy concerns.
3
Analysis of Empirical Data
We summarize data that we collected over the course of our study in Figure 2. The rows in the figure denote different sensing modalities (see Section 2). The columns show the total amount of data received from all nodes for a given modality (top figure) and the total number of nodes that sent at least one data point (bottom figure) for a given month. For clarity, we normalize each row by the maximum value found for that modality over the whole experiment. The normalization factors, shown as bar-graphs in Figure 2, vary for each modality, as the number of sensor nodes and sampling rates are different. For example, the dark blue color corresponds to 20 million and 290 sensor readings for the BCM and SURVEY* modalities, respectively. We started our experiment by deploying climate, power, and BCM meters in September 2010. Overall, 23 climate domes, 41 power meters, and 12 BCM me-
!"#$%&'()'*%+,(&'-%./0+1,'
!"#$%%&
!"#$%'&
!"#$%(&
!"#$%)&
!"#$%*&
!"#$%&'()'!(*%+'
!"
#!"
$!"
%!"
&!"
'!"
Fig. 2. Summary of data that we downloaded over three years for all sensor modalities. Figures on the left show number of data points for each month, normalized to the maximum value over all months for each row. Figures on the right show the normalization factor (i.e., the maximum value over all months). Note the sensor readings are in a logarithmic scale, while the number of nodes are in linear scale.
ters were deployed. We started experimenting with HVAC and ComfortSENSETM in November 2011 and started logging BMS data in May 2012. Installation and Maintenance Cost. By far the most expensive was the initial deployment period. BCM meters cost us approximately $12k, including the cost of professional installation in our building. Manufacture and deployment of the wireless sensor platforms cost approximately $12k, or about $200 per node, including batteries and enclosures. Clearly, software-based solutions or solutions that utilize existing infrastructure are advantageous over new sensors. We obtained the BMS, and CSENSE modality for free, as they are amortized by the costs of running the whole building. In terms of the on-going maintenance, the only significant cost in terms of dollars was replacing the rechargeable batteries in CLIMATE nodes after one and half years. While we spent minimal time maintaining BCM and BMS modalities, maintaining CLIMATE, POWER, and CSENSE modalities was more difficult. CLIMATE and POWER nodes break down frequently as sensors or connectors fail and people tend to uninstall ComfortSENSETM application after a few months. Data Yields. It is clear that the best performing sensors in terms of node failures are BCM and BMS. This is commercially tested hardware that is periodically maintained by our property managers. The lower yields in the sensor reading
23*++"
CSIRO, Autonomous Systems, Brisbane, QLD, Australia, [email protected] 2 CSIRO, Energy Technology, Newcastle, NSW, Australia
Abstract. Buildings are amongst the largest consumers of electrical energy in developed countries. Building efficiency can be improved by adapting building systems to a change in the environment or user context. Appropriate action, however, can only be taken if the building control system has access to reliable real-time data. Sensors providing this data need to be ubiquitous, accurate, have low maintenance cost, and should not violate privacy of building occupants. We conducted a three-year study in a mid-size office space with 15 offices and 25 people. Specifically, we focused on sensing modalities that can help improve energy efficiency of buildings. We have deployed 25 indoor climate sensor nodes and 41 wireless power meters, submetered 12 electric loads in circuit breaker boxes, logged data from our building control system and tracked activity on 40 desktop computers. We summarize our experiences with the cost, data yields, and user privacy concerns of the different sensing modalities and evaluate their accuracy using ground-truth experiments.
1
Introduction
Climate change and greenhouse gas emissions are forcing society to reevaluate energy efficiency practices. Buildings consume a large portion of the total energy in industrial countries and most of this energy is provided by conventional fossil-based power plants. As a result, numerous techniques for improving energy efficiency of commercial buildings have been proposed in recent years. Examples include lighting systems and window shaders that react to changing daylight and occupancy conditions [15], software products that manage and control power consumed by PC workstations [7], and Heating Ventilation and Air Conditioning (HVAC) systems that adapt to occupancy, electricity price, and weather forecast [3,6,4]. Many of these systems rely on real-time sensing of environmental data and user context. Future building control systems will no doubt deploy and integrate numerous sensors in their core functionality. The advantage of the sensor and actuator convergence in building control systems is that individual sensors can be reused for different tasks. For example, power-meters can be used to detect
occupancy of offices based on power consumption of desktop PCs and LCD monitors [9]. Thermal comfort of people can be estimated using office temperature and user feedback [5]. The overhead of installation and maintenance of building sensors is thus expected to be amortized by the costs of operating the whole building. Despite the exploration of multiple sensing modalities for promoting energy efficiency in buildings, there is still limited work on comprehensive comparison of their effectiveness. The goal of this work is to quantify advantages and disadvantages of different sensing modalities in commercial office buildings. We collected data over a three year period from environmental sensor nodes with temperature, light, humidity, and presence sensors deployed in every office in our building and from power-meters measuring energy usage of office appliances. We also contracted electricians to install submeters on load circuits and obtained access to data from our building control system. Finally, we developed a PC application that tracked activity of the users on desktop computers and allowed us to remotely survey users about their thermal comfort. We evaluate installation and operational costs of the different building sensors. We found that data yields of low cost sensors tend to decrease over time due to hardware failures. Battery operated devices introduce frequent black-out periods due to lack or negligence of maintenance. Software-based sensing solutions are also surprisingly difficult to maintain, despite existing IT policies of pre-installing the software on new computers. Permanently powered commercial products, on the other hand, perform reliably on a long-term basis. We also evaluate accuracy of the different sensors by conducting ground truth experiments. We mostly focus on detecting office occupancy and estimation of thermal comfort of individual users. Our findings show that power-meters are 10-20% less accurate in detecting office occupancy than passive infrared sensors and suffer from high false positive rates. PC activity loggers perform better than power-meters, but are generally rejected by users due to privacy concerns. Interestingly, users accept power-meters connected to their desktop PCs despite us disclosing their use for occupancy detection. Finally, we conducted several human-in-the-loop experiments aiming at energy usage reduction of HVAC and office appliances. Our empirically-based simulations indicate that 42% reduction in HVAC energy usage can be achieved over a centrally managed HVAC with static space temperature setpoints. We also found that presenting energy efficiency data to users can lead to behavior change and energy reduction, achieving 10% energy savings on office appliances. Our findings help justify the need for fine-grained continuous sensing of office occupancy, office climate, and energy usage for building-scale systems.
2
Experimental Setting
We built a comprehensive system for collecting data relevant to energy efficiency and energy usage in commercial buildings. Specifically, we focused on collecting information about (1) presence of people in their offices, (2) energy usage of office
Sensing'
Personal' Personal Climate Climate'Meter' Monitor
Wireless Power Personal' Meter Power'Meter'
ComfortSENSE™' PC'Applica1on'
Aggregate' Power'Meter'
Building'Control' System'
Processing' Thermal' Comfort'
Occupancy'
Climate' Data'
Energy' Usage'
Control' HVAC%% Control%
Informa1on' PC/LCD% Control%
User% Behavior%
Climate,% Energy% Sta8s8cs%
Energy% Usage/ Efficiency%
Fig. 1. Left: Data from low level sensors can be used to drive adaptive environment control and improve our understanding of the building energy usage. Right: Office building used for our deployment. Icons depict locations of sensors in offices.
appliances and the HVAC, and (3) subjective perception of thermal comfort of users. Figure 1 (left) provides an overview of our architecture that includes multiple sensor inputs within a typical office building. Real-time collection of this data is important for closing the loop between building control, user behavior and comfort, and changes in the environment. We deployed our system on a single floor of an office building that included 3 single-occupancy offices, 10 double-occupancy offices, 1 conference room, and 2 open-plan student offices (see Figure 1 right). We describe each of our 5 sensing techniques in more detail in the following text. Personal Climate Monitors (Climate) and Power Meters (Power). We built two sensor platforms, climate meters and power meters, based on our inhouse sensor network board. The board uses Intel 8051 processor and sub 1GHz Nordic radio and is optimized for low power and small size. The nodes self-organize to provide a multi-hop wireless mesh and run on a single battery charge for approximately 2 months. We sampled climate and power meters every 60 seconds and transmitted the data over the wireless network to a central database. The climate monitors also transmitted their current battery voltage level for maintenance purposes. We deployed one climate monitor per user, most frequently next to the user’s LCD monitor. We measured office climate and occupancy using temperature, humidity, light, and motion sensors. The power meters measured current, voltage, power, and cumulative energy consumption of electric appliances at a power socket level (similar to [9]). We augmented off-the-shelf power meters with our wireless sensor board to enable real-time wireless data collection of the energy data. We were limited to 2 power meters per user, thus we generally measured an aggregate consumption of a
desktop PC and LCD monitor with one power meter and used the other power meter for external hard drives, laptops, and fans. Commercial Power Meters (BCM). We installed Powerlogic PM9C power meters in breaker circuits (thus we call this modality BCM) to measure the aggregate energy usage of 12 electric circuits in our building, including user office sockets, lighting, hot-water heaters, and printers. We connected the meters to our internal Ethernet network and periodically recorded voltage, current, and power on each of the electrical circuits every 60 seconds in the database. Building Control and Management System (BMS). Our building is controlled by Tridium JACE controllers connected together by an internal modbus network. The access to the controller is restricted to our property managers who use it on a daily basis, primarily to inspect and fine-tune performance of the heating, ventilation, and air-conditioning (HVAC) system. We have obtained access to the system and used standard Open Building Information Exchange (oBIX) RESTful web interface to log data in the database and to control temperature in the offices during user trials. Our office space is logically divided into 4 air-conditioning zones and we used a python script to collect 11 sensor inputs per air-conditioning zone, including information about chillers, heaters, supply fans, indoor temperature, and space temperature setpoint that the HVAC is currently set to. Unlike the fine-grained office-level temperature data obtained by the personal climate monitors, BMS has one temperature sensor per zone. PC Application (CSENSE, SURVEY). Finally, we used ComfortSENSETM PC application [17] to track user activity on desktop computers and to administer user surveys. The application runs as a background process on desktop computers and can be controlled from a central server to display user surveys on all registered computers. Surveys focus mainly on user thermal comfort and their satisfaction with HVAC performance. Additionally, the application tracks key-strokes and mouse events on the host operating system and records this information every 5 minutes in the database. Users can opt-out of activity monitoring in case of privacy concerns.
3
Analysis of Empirical Data
We summarize data that we collected over the course of our study in Figure 2. The rows in the figure denote different sensing modalities (see Section 2). The columns show the total amount of data received from all nodes for a given modality (top figure) and the total number of nodes that sent at least one data point (bottom figure) for a given month. For clarity, we normalize each row by the maximum value found for that modality over the whole experiment. The normalization factors, shown as bar-graphs in Figure 2, vary for each modality, as the number of sensor nodes and sampling rates are different. For example, the dark blue color corresponds to 20 million and 290 sensor readings for the BCM and SURVEY* modalities, respectively. We started our experiment by deploying climate, power, and BCM meters in September 2010. Overall, 23 climate domes, 41 power meters, and 12 BCM me-
!"#$%&'()'*%+,(&'-%./0+1,'
!"#$%%&
!"#$%'&
!"#$%(&
!"#$%)&
!"#$%*&
!"#$%&'()'!(*%+'
!"
#!"
$!"
%!"
&!"
'!"
Fig. 2. Summary of data that we downloaded over three years for all sensor modalities. Figures on the left show number of data points for each month, normalized to the maximum value over all months for each row. Figures on the right show the normalization factor (i.e., the maximum value over all months). Note the sensor readings are in a logarithmic scale, while the number of nodes are in linear scale.
ters were deployed. We started experimenting with HVAC and ComfortSENSETM in November 2011 and started logging BMS data in May 2012. Installation and Maintenance Cost. By far the most expensive was the initial deployment period. BCM meters cost us approximately $12k, including the cost of professional installation in our building. Manufacture and deployment of the wireless sensor platforms cost approximately $12k, or about $200 per node, including batteries and enclosures. Clearly, software-based solutions or solutions that utilize existing infrastructure are advantageous over new sensors. We obtained the BMS, and CSENSE modality for free, as they are amortized by the costs of running the whole building. In terms of the on-going maintenance, the only significant cost in terms of dollars was replacing the rechargeable batteries in CLIMATE nodes after one and half years. While we spent minimal time maintaining BCM and BMS modalities, maintaining CLIMATE, POWER, and CSENSE modalities was more difficult. CLIMATE and POWER nodes break down frequently as sensors or connectors fail and people tend to uninstall ComfortSENSETM application after a few months. Data Yields. It is clear that the best performing sensors in terms of node failures are BCM and BMS. This is commercially tested hardware that is periodically maintained by our property managers. The lower yields in the sensor reading
23*++"
