Challenges in Resource Monitoring for Residential ... - NESL Projects
Challenges in Resource Monitoring for Residential Spaces Yan Wang ∗
Younghun Kim, Thomas Schmid, and Mani B. Srivastava
College of Biomedical Engineering & Instrument Science Zhejiang University, P.R. China
Networked and Embedded Systems Lab., Electrical Engineering Department, University of California, Los Angeles, U. S. A.
[email protected]
∗ This work was done while visiting UCLA
{kimyh,thomas.schmid, mbs}@ucla.edu Abstract
1
Buildings consume approximately 73% of the total electrical energy, and 12% of the potable water resources in the United States. Even a moderate reduction in this sector results in significant monetary and resource savings. Finegrained resource monitoring is regarded as one technology that could help consumers and building owners to understand, and thus reduce, their resource waste. In this paper, we discuss challenges emerging from these fine grained resource monitoring systems through an empirical study of long-term monitoring data of a residential space. We collected synchronous water and electricity usage over 3 months from a single family house. Using a matched filter mechanism we detect several water and electrical events happening in the house, showing that with simple mathematical tools, these data traces reveal already a lot of information about the consumption patterns. We further discuss challenges in fine grained load monitoring using the main power meter, advocating that synchronous water and power traces help to disambiguate several power consumers. In addition, our analysis revealed interesting privacy implications while monitoring a household’s resource consumption at high time resolution, a problem that could easily hamper the successful adaptation of these technologies.
The ever increasing concern about resource conservation prompts researchers to find techniques to make our consumption sustainable for future generations. Various perspectives on sustainable development exist, such as developing renewable energy sources, improving efficiency, and minimizing resource consumption. Recently researchers claimed that fine grained resource monitoring has a positive impact on conservation. Promising spatio-temporally fine grained monitoring techniques have emerged in various research settings. For example, Kim et al. [8] developed a proof-of-concept fine grained water monitoring system that tracks the pipe-level water flow in real time. Froehlich et al. [6] developed a single-point whole house water activity monitor by exploiting the change in water pressure during use. In [9] Kim et al. presented an indirect appliance level power monitoring system by leveraging wireless sensors including magnetic, acoustic and light sensors. And Jiang et al. [7] developed a radio enabled plug-in type meter that monitors the real-time power consumption of an attached device. We expect that such systems will become common place, since already similar technologies start to hit the commercial market. The major resources in a home are electricity, water, natural gas, and heating oil. Residential and commercial buildings together account for approximately 12% of the potable water and 73% of total electricity in the US [1, 14]. Modest conservation improvements in these sectors can have a significant impact. Interestingly, each resource production and supply chain has a complex interplay. For example, in the US, the water consumed to generate electricity accounts for 42% of the total water consumption in the building sector [13]. This means that consuming electricity implicitly consumes water too. Conversely, water purification and transportation systems consume electricity. The American public water supply and treatment facilities consume about 56 billion kWh per year, which is enough electricity to power over 5 million homes [14]. Moderate savings of potable water results in significant financial benefits. The US Environmental Protection Agency states that 3 trillion gallons of water could be saved each year if every household in the US decreased its water consumption by 30 percent [14]. This results in a dollar-volume saving of more than $18 billion a year. Water conservation is of even greater financial significance as increasing purifi-
Categories and Subject Descriptors H.4 [Information Systems Applications]: Miscellaneous
General Terms Experimentation, Human Factors, Measurement
Keywords Energy Efficient Homes, Resource Monitoring, Sustainability
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. BuildSys’09, November 3, 2009, Berkeley, CA, USA. Copyright 2009 ACM 978-1-60558-824-7 ...$5.00
Introduction
Toilet Bin Water Usage Pattern
Laundry Machine Water Usage Pattern
0.3
Sprinkling System Water Usage Pattern
0.18
1.4
Water Consumption: 4.4L
Water Consumption: 120L
Water Consumption: 4kL
0.16
1.2
0.25
0.2
0.15
0.1
Water Flow [L/s]
Water Flow [L/s]
Water Flow [L/s]
0.14 0.12 0.1 0.08 0.06
1
0.8
0.6
0.4
0.04 0.05
0.2
0.02 0
0
10
20
30
40
50
60
0
0
200
400
600
Time[s]
(a) Toilet Bin
800
1000
1200
1400
1600
Time[s]
1800
2000
0
0
1000
2000
3000
4000
5000
6000
7000
8000
Time[s]
(b) Laundry Machine
(c) Sprinkler System
Figure 1: Water consuming activities in a homes are sporadic events. However, many of consumption activities have a unique pattern, which makes it possible to detect particular events using a pattern matching algorithm.
cation cost is being compounded by rapid demand growth. In conclusion, to make buildings sustainable, it is necessary to monitor not just electricity, but several other resources as well. One start is the US Green Building Council’s (USGBC) Leadership in Energy and Environmental Design (LEED) program. The LEED guidelines put balanced emphasis on both electricity and water efficiency; and their interplay. In this paper, we discuss our experience on real-time electricity and water monitoring in a single family house. We collected 3 months worth of full house-hold synchronous power and water consumption data, which thus allows us to analyze both resources at the same time. We present several data analysis mechanisms, including main water meter aided usage detection, activity inference using diurnal statistics, exploitation of dual-consumption patterns to inference activity, and privacy implications. Using this analysis, we identified research opportunities and challenges in fine grained resource monitoring system design for buildings. Our contributions are three fold: (1) We present a fixture level water consumption profiler using the main water meter. (2) We identify challenges in fine grained load monitoring using the main power meter given the very high load dynamics in today’s home. (3) We show that it is possible to detect an appliance’s load by observing the water and power consumption at the same time, since many appliances consume several resources together. With this, we advocate that it is beneficial to monitor all the used resources, and not just electricity.
2
Non-Intrusive Resource Monitoring using a Single Sensor
We instrumented a single family house in Southern California with an ultrasonic water flow meter (Shenitech STUF200H) and a main power monitor (TED-1001). We collected the power and water consumption at a fine-grained rate of 1 Hz. While this installation is still costly, these systems will become more commodity as the Automatic Meter Reading Association [3] and the American Water Works Association [2] are deploying more and more smart meters with real-time wireless meter capabilities for billing and smart-grid application purposes. Therefore, we expect that the data analysis performed in this section can be leveraged without extra hardware cost in the near future. The analysis is performed using several pattern recogni-
tion techniques to automatically identify and track device level resource usage, while demonstrating the possibility of information fusion between water and power use to improve detection accuracy. In addition, we identify limitations of this approach by showing that the consumption trace is very complex and dynamic.
2.1
Fine Grained Water Monitoring using Pattern Matching
In a typical household, the number of water fixtures is limited. In addition, water consuming activities are sparse in time compared to power consuming activities. Many of the water fixtures have a distinct water usage pattern, because they are mechanically or electro-mechanically controlled, e.g. toilet, dish washer, laundry machine, and sprinkler systems. In addition, water related activities are sporadic in time, unique enough to be distinguishable in time-series data. These patterns can be identified and extracted at the fixture level using different types of pattern recognition algorithms (Figure 1). Similar to the Non-Intrusive Load Monitoring technique (NILM) [10], we use these unique patterns to identify the use of individual water fixtures. More specifically, we can perform an automatic data annotation using a matched filter with the real-time trace of the water consumption in the house. Using the matched filter, we were able to detect several different water activities including toilet flushes, lawn sprinkler system, and laundry machine activities. Figure 2 shows such a detection trace when the toilet was flushed. Its detection accuracy is about 90%. Several useful pieces of information can be extracted from this data. The easiest of which is how much water got consumed by flushing the toilet. More interestingly, and not as obvious, we can infer how many people stay in the house at a current day. For example, Figure 3 shows a transition region where two interesting events happened in the house. Abrupt changes in the number of flushes indicate that several people have visited, resulting in a much higher number of flush events (from 60 to 90 for a short term visit on day 14). In addition, the number of flushes changed from around 20 to 45, indicating a long-term visit by two persons. Incidentally the house-owner changed at the same time the toilet bins to more water efficient ones. It is interesting to observe that the number of flushes have increased after that installation. After interrogating the owner we found out that the residents flush more often than before, as they are less concerned about water usage now!
Toilet Flush Event for Five Days
Number of Flushes in a house 100 90 80
Day 1
Before New Toilet Bin Installation
70
Counts
Day 2
60 50 40
Day 3
30
After New Toilet Bin Installation
20
Day 4 10 0
Day 5 4
6
8
10 12 14 Time[h]
16
18
20
22
Several other useful pieces of information can be extracted with reasonable accuracy and resolution using just the main water meter. Figure 4 illustrates how much water the residents used for toilet flushes and laundry machine. The bottom plot in Figure 4 shows the total diurnal water usage over a month from June to July. The spikes in the plot indicate that the sprinkler system ran, accounting for about 2000 or 40001 liters of water per run. This simple matched filter approach works very efficiently for large water consuming events, or events that have a very distinct usage pattern. For example, the lawn sprinkler system is the biggest water consumer in a household, and detecting this event is very easy using this pattern matching mechanism. It results in a 100% detection rate. However, for smaller events, such as a faucet usage, and compound events, where several systems consume water at the same time, we may still want to use more sensors than just the main meter, to ensure accuracy. A rather straight-forward cost-accuracy trade-off can be considered in this application. For example, if we are only interested in routine water usage, we can simply use matched filtering on the main water monitor trace. If more visibility is necessary, one can spend more money and add extra sensors, thus gaining resolution with better accuracies and confidence.
Challenges in Fine Grained Power Monitoring using a Single Sensor
Unlike water activities, electrical energy use in a house has compound characteristics. It is partially due to the number of electrical appliances compared to the number of water fixtures (70 to 80 appliances vs. less than 20 fixtures). In addition, electrical appliances often consume a certain amount of static power, just by being plugged in. Compared to water fixtures, this would mean a constant drip of water. This makes the problem of power monitoring in a house much harder. Although recent advances in sensor networks allow to 1 The
10
15
20
25
24
sprinkler system has two different watering patterns, one consuming 2000, the other 4000 liters.
Figure 3: The number of flushes provides information regarding house occupancy. Water Consumption for Toilet Flushes and Laundry Machine Water Usage[L]
2
Figure 2: A simple matched filter can reliably detect the toilet flush event. The same technique is applicable to other water activities that have a unique usage pattern, including dish-washer, laundry machine, and sprinkler systems. By detecting these events, coarse-grained, yet real-time water usage can be estimated.
2.2
5
Time[Days]
1000 Toilet Flush Laundry Machine
800 600 400 200 0
0
5
10
15
20
25
Time[Days] Daily Water Consumption in a House(Mid June to July) Water Usage[L]
0
0
5000 Others Sprinkler
4000 3000 2000 1000 0
0
5
10
15
20
25
Time[Days]
Figure 4: Water Usage Breakdown is possible with reasonable accuracy.
have multiple sensor nodes at reasonable cost, the monitoring system may quickly become prohibitively expensive for this non-critical domestic application. For example, if we want to monitor the power consumption of every appliance, we need to have 70 to 80 sensors, one for each appliance, if we want to use a systems like described in [7, 9]. Even though each sensor could potentially cost as few as $10, the overall cost would quickly rise to $700. In addition, nontrivial installation, calibration, association and data aggregation is an outstanding issue in the UbiComp community [4]. Alternative approaches have limitation too. For example, it is difficult to use a system as described by Patel et al. [11] or Leeb et al. [10] because many electrical appliances have variable power consumption [9]. And even in a small one family house, we already observed many compound events (appliance on/off events), one of the challenges that makes NILM so difficult. It is possible to use NILM [10] techniques, but a huge burden is put on the feature extraction process and system training. While it is possible to break down the power consumption for large consumers, like the HVAC system, smaller appliances are very hard to extract. This is problematic given the trend that small appliances consume more and more of the total power consumption [1]. Figure 5 shows the load characteristics including abrupt power transition and their duration. Each appliance tends to have a unique pattern;
Synchronous Water/Power Usage(Laundry Active) Water Usage [L/s]
12
2.5
11 10
2
9
1.5 7 6 1 5 4 0.5
3 2
0
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
1
Power Transition[kW]
Figure 5: An instance of features in non-intrusive load monitoring: power transition and duration. Since features of many appliances are very close, the use of NILM has limited capabilities to disambiguate small appliances. Histogram of Power Transition
0.1 0.05
0.2 0.15 0.1
0 23
23.1
23.2
23.3
23.4
23.5
23.6
23.7
23.8
23.6
23.7
23.8
Time[Hour] 3
2.5
2
1.5 23
23.1
23.2
23.3
23.4
23.5
Time[Hour]
Figure 7: This synchronous water and power trace indicates that power and water usage has a unique pattern as the laundry machine runs a predefined routine. This pattern depicts its normal operation, Filling Water Tank, Washing, Draining, Filling Water Tank, Rinsing, and Final Spin. Note that this trace is from a controlled setting where the residents didn’t run other appliances so the power consumption pattern can be observed. It is usually not possible to see this clean power consumption trace due to other appliances turning on and off, making automatic annotation by looking only at power numbers difficult. Daily Power Consumption in a House(Mid June to July)
0.05 120
0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Power Transition[kW]
Figure 6: This histogram shows the number of power transitions over 4 days. Since too many small power transitions happen, a NALM approach is limited.
Power Usage[kWh]
Normalized Count
0.15
Power Consumption [kW]
Duration[h]
8
0.2
100 80 60 40 20 0
0
5
10
15
20
25
Time[Days]
unique average power consumption, time-of-use, inter power state transition. At the same time, the difference among appliances is very subtle, which makes features flock together in the feature space. The resolution of the legacy meter is limited, thus making reliable detection difficult (Figure 6).
2.3
Fine Grained Power Monitoring with Side Information
Monitoring several resources at home improves the accuracy in appliance level power monitoring. Some appliances, often major resource consumers, use several resources concurrently, e.g. gas oven, oil air heater, laundry machine, dish washer, etc. For example, a laundry machine uses water and electricity at the same time in a very deterministic way. Thus, analyzing both water and electricity at the same time helps us understand and detect when the laundry machine runs. Figure 7 shows the total water and power consumption of a house when the laundry machine is running. We can see that the resource consumption has a specific pattern as it runs through pre-programmed cycles. By observing these two traces, it is possible to detect when the laundry machine is active even if the power consumption trace is dynamically changing. In addition, monitoring all the resources in a house, even just at a main meter level, could reveal other interesting correlations amongst devices, thus allowing another dimension for possible sustainability optimizations.
3
Discussion and Future Direction
To make fine grained monitoring widely applicable, several challenges have to get solved. First, researchers may want to investigate trade-offs among various techniques and
Figure 8: Daily power consumption from mid June to mid July. Power usage went up due to the use of the air conditioning system, resulting in about 40kWh overhead per day.
develop a modular approach that incorporates existing technologies. Appropriate design in terms of cost, usability, accuracy, reliability, and device maintenance issues need to be considered. User adaptation is another issue. As Beckmann et al. [4] pointed out, usability of a system is of great concern. While the initial cost associated with hardware becomes negligible over long-term, non-trivial installation, maintenance, device association, and privacy concerns make the system prohibitively expensive for general users. Since the value of the application is relatively small compared to the burden lying on the users, further studies on end-user adaptation and user experience need to be conducted. When we look at the resource consumption in a building, we have to consider that the highest level of decision is made by people, since they have to occupy and use the space. Thus, excellent information support to the end-user is impeccable to maximize the utility of such a fine-grained monitoring system. The buildings can be seen as an ecosystem where residents make decisions on resource consumption with limited information. Providing the end-user with right choices and more valuable information will help them to make the right decisions on their consumption, without impacting their comfort or utility. Therefore, we need to consider effective and efficient data processing and visual-
Synchronous Water/Power Usage(Bathroom Use)
Efficiency, the Metric for Resource Consumption
Many resources are used to regulate a user’s comfort level given certain ambient conditions. A good example is the HVAC system that keeps temperature and humidity at a certain level so occupants feel comfortable. In addition, a sprinkler system is used to maintain the lawn in good condition. Purely monitoring the total resource consumption can be misleading. Depending on ambient conditions, the same amount of resource consumption means different efficiencies. For example, ambient temperature and relative humidity drastically change evaporative losses. Therefore, the time of day a lawn sprinkler is used can have vastly different efficiencies to actually water the plants and grasses. In addition, soil moisture level often determines the right time and amount of water the lawn needs. However, more often than not, a sprinkler system does not take these measurements into account, and a predefined watering pattern will put out the same amount of water everywhere. Another example can be made with heating and cooling of an indoor room. The difference between indoor and outdoor temperature determines heat loss or gain, and a proper setting of the thermostat depending on outdoor temperature can result in efficient HVAC usage. Although straight forward, by combining resource consumption information with ambient conditions, it is possible to let utility consumers understand the different aspects of their resource usage. More precisely, it can tell them how efficient their use is, or where and how they waste resources.
3.2
Information Fusion Mechanism
Buildings have very complex, diverse load and resource consumption characteristics. The key question is whether every monitoring technique scales. The sole use of only direct[7], indirect[8, 9], or infrastructure-mediated[6] sensing techniques cannot be sufficient to cover all the load and resource demands. At some point, a sophisticated inference engine that incorporates all these sensing techniques will be needed. We briefly touched upon this possibility for the laundry machine case, though further investigation is needed in how other resources could help to identify particular appliances, or how this fusion could be automated.
3.3
Privacy Implication
Home monitoring applications reveal interesting privacy implications. For example, Srinivasan et al. [12] showed that detecting packet exchanges in a home monitoring system can reveal privacy prone information such as occupancy. Indeed, Fogarty et al. [5] actively use the signals from the plumbing system to infer water activities. Similar privacy prone information can be extracted from the power and water traces. While analyzing the water consumption, we discovered that every now and then, someone flushes the toilet twice. We asked the home owner, and he said that this must be a long-term visitor. In addition to the double-flushing, we also discovered that in one instance, someone who flushed only once, did not use the bathroom sink, while the person flushing twice uses the sink for a long time (Figure 9). At the same time, the synchronous power data revealed that one does not
0.2
Flush Once
0.15 0.1
Flush Twice
Sink On
0.05
Power Consumption[kW]
3.1
Water Usage[L/s]
ization, which is tailored to specific user demand.
0
0
200
400
600
800
1000
1200
800
1000
1200
Time[s] 2.5
2
1.5
Lamp On 0
200
400
600
Time[s]
Figure 9: The flush pattern recognition mechanism finds that someone in the house tends to flush the toilet twice. The synchronous power data additionally revealed that one person did not turn on the light, while the other did.
use the bathroom light while the other turned on the light. This implies that by observing specific patterns in the bathroom, we can infer not only when and how an occupant uses the bathroom but also who uses it. Even a simple matched filter can reveal interesting observation on users resource usage patterns and abnormal conditions, which may be useful information for healthcare concerns. However, this information may be too private for some users. Once a user realizes just how much can be inferred from these consumption traces, they might stop using them in fear of privacy concerns. Therefore, we carefully need to consider a balance between privacy and application objective. The big question becomes on how to lower the resolution so people in a house feel safe to provide information while still ensuring the application benefits?
3.4
Data Management and Representation
One problem of fine-grained resource monitoring is the vast amount of data collected. In our 3 month run, we already collected more than 300MB worth of data. However, unlike multi-media data, this time-series data is hard for users to understand, unless proper representation and summaries are provided. Appropriate data visualization and representation need to be considered to make this information effective. Although many utility consumers are solely concerned with their monetary benefits, utility consumers’ motivations are very diverse as Woodruff et al. pointed out [15]. Although not exclusive, they found three main motivators that bright green initiatives have, such as counterculture biocentric activism, American frontier self-reliance and rugged independence, and trend-focused utopian optimism. These are all well perceived values in the US, as one can observe by the success of Al Gore’s “An Inconvenient Truth”. Different perspectives and interpretations about one’s resource consumption may have stronger motivation to people motivated by these different values. In that, a more personalized data representation and interpretation will show and motivate the end-user where to change what. Many different power meters today on the market incorporate the cost of energy into their calculations. Cent-ameter and TED are only a few examples that offer various
options on cost display, and can even cope with variable power rates often used by power utility companies. However, several other metrics can be used to quantify this resource consumption. Since most of the electricity comes from non-renewable resources, about 85% in the US, a rough projected Carbon emission estimate can be calculated for the consumption of electricity. For example, what does it imply if you consume 1kWh in Southern California in terms of CO2 emissions, or in terms of indirectly used water to produce that energy? For example, the US Western Interconnect consumes 16.7L per 1 kWh, compared to the Texas Interconnect’s 1.6L per 1 kWh [13]. Indeed, long term benefits and morals play an important role. Even if a sheer green decision sacrifices many monetary benefits, and sometimes even comfort level, many green thinking people prefer to make the right choice [15]. Some examples are using natural ventilation, biking to work, use costly but environmentally friendly grocery goods, etc. For someone who cares more about water conservation, projected water consumption for electricity is a more useful metric than just pure power numbers or CO2 emissions.
3.5
Developing Interactive Technology
As many sources of uncertainty exist, it will be too expensive to make a fully autonomous fine-grained resource monitoring system. Interactive technology can help to lower the complexity of the monitoring system. As we have seen in Section 2.1, several features can be effectively extracted from the users if they input the time of use of an appliance. Gradual adaptation of the system can be done in this way, where the user teaches the system of what pattern means which activity to the end user. Such a technique would allow users to actively participate in resource management by forcing them to look at the collected data. Several auxiliary information sources may add more values to this application. A good way for users to easily access and manipulate data becomes essential. For example, in Southern California, a major source of potable water consumption in homes is the sprinkler system. Watering the lawn has several implications. Residents want to maintain their lawn in good shape to maximize their property value. At the same time, they want to follow regional regulations. For example, watering the lawn on Mondays is prohibited in Los Angeles, and residents in that city may want to get some feedback on this regulation. Since this addresses a user’s concern, the user may want to actively input needs and intent.
4
Conclusion
Resource monitoring is a promising application that helps make buildings more sustainable. We showed that various pieces of information such as fixture level water consumption, occupancy, or appliance status can be extracted from the real time water and power traces. However, to maximize the utility and broad user adaptation, researchers should carefully address challenges in the monitoring system design. We showed through examples that having synchronous data improves accuracy of device level resource consumption. Yet, further research on information fusion from various sensors and monitoring systems is needed. In addition,
non-trivial installation, maintenance, data association and aggregation need to be considered to successful user adaptation of these new and exciting technologies.
5
Acknowledgements
This material is based upon work supported by the NSF under award # CNS-0627084, CCF-0820061, and CNS0905580, and by the UCLA Center for Embedded Networked Sensing. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding agencies.
6
References
[1] Annual energy outlook 2009, doe/eia-0383(2009). Tech report, U.S. Department of Energy, 2009. [2] American Water Works Association. C706-96(R05) : AWWA standard for direct-reading remote-registration systems for cold-water meters. STC, 2005. [3] Automatic Meter Reading Association. http://www.amra-intl.org/. [4] C. Beckmann, S. Consolve, and A. LaMarca. Some assembly required: Supporting end-user sensor installation in domestic ubiquitous computing environments. In Ubicomp, 2004. [5] J. Fogarty, C. Au, and S. E. Hudson. Sensing from the basement: A feasibility study of unobtrusive and low-cost home activity recognition. In ACM UIST, 2006. [6] J. Froehlich, E. Larson, T. Campbell, C. Haggerty, J. Fogarty, and S. Patel. HydroSense: Infrastructure-mediated single-point sensing of whole-home water activity. In UbiComp, 2009. [7] X. Jiang, S. Dawson-Haggerty, P. Dutta, and D. Culler. Design and implementation of a high-fidelity ac metering network. In ACM IPSN/SPOTS, 2009. [8] Y. Kim, T. Schmid, Z. M. Charbiwala, J. Friedman, and M. B. Srivastava. NAWMS: nonintrusive autonomous water monitoring system. In ACM SenSys, 2008. [9] Y. Kim, T. Schmid, Z. M. Charbiwala, and M. B. Srivastava. ViridiScope: Design and implementation of a fine grained power monitoring system for homes. In UbiComp, 2009. [10] S. Leeb, S. Shaw, and J. Kirtley, J.L. Transient event detection in spectral envelope estimates for nonintrusive load monitoring. IEEE Trans. on Power Delivery, Jul 1995. [11] S. N. Patel, T. Robertson, J. A. Kientz, M. S. Reynolds, and G. D. Abowd. At the flick of a switch: Detecting and classifying unique electrical events on the residential power line. In UbiComp, 2007. [12] V. Srinivasan, J. A. Stankovic, and K. Whitehouse. Protecting your daily in-home activity information from a wireless snooping attack. In UbiComp, 2008. [13] P. Torcellini, N. Long, and R. Judkoff. Consumpive water use for u. s. power production. National Renewable Energy Lab. Tech. Report, NREL/TP-550-33905, 2003. [14] U. S. EPA. http://epa.gov/watersense/index.htm. [15] A. Woodruff, J. Hasbrouck, and S. Augustin. A bright green perspective on sustainable choices. In CHI, 2008
Younghun Kim, Thomas Schmid, and Mani B. Srivastava
College of Biomedical Engineering & Instrument Science Zhejiang University, P.R. China
Networked and Embedded Systems Lab., Electrical Engineering Department, University of California, Los Angeles, U. S. A.
[email protected]
∗ This work was done while visiting UCLA
{kimyh,thomas.schmid, mbs}@ucla.edu Abstract
1
Buildings consume approximately 73% of the total electrical energy, and 12% of the potable water resources in the United States. Even a moderate reduction in this sector results in significant monetary and resource savings. Finegrained resource monitoring is regarded as one technology that could help consumers and building owners to understand, and thus reduce, their resource waste. In this paper, we discuss challenges emerging from these fine grained resource monitoring systems through an empirical study of long-term monitoring data of a residential space. We collected synchronous water and electricity usage over 3 months from a single family house. Using a matched filter mechanism we detect several water and electrical events happening in the house, showing that with simple mathematical tools, these data traces reveal already a lot of information about the consumption patterns. We further discuss challenges in fine grained load monitoring using the main power meter, advocating that synchronous water and power traces help to disambiguate several power consumers. In addition, our analysis revealed interesting privacy implications while monitoring a household’s resource consumption at high time resolution, a problem that could easily hamper the successful adaptation of these technologies.
The ever increasing concern about resource conservation prompts researchers to find techniques to make our consumption sustainable for future generations. Various perspectives on sustainable development exist, such as developing renewable energy sources, improving efficiency, and minimizing resource consumption. Recently researchers claimed that fine grained resource monitoring has a positive impact on conservation. Promising spatio-temporally fine grained monitoring techniques have emerged in various research settings. For example, Kim et al. [8] developed a proof-of-concept fine grained water monitoring system that tracks the pipe-level water flow in real time. Froehlich et al. [6] developed a single-point whole house water activity monitor by exploiting the change in water pressure during use. In [9] Kim et al. presented an indirect appliance level power monitoring system by leveraging wireless sensors including magnetic, acoustic and light sensors. And Jiang et al. [7] developed a radio enabled plug-in type meter that monitors the real-time power consumption of an attached device. We expect that such systems will become common place, since already similar technologies start to hit the commercial market. The major resources in a home are electricity, water, natural gas, and heating oil. Residential and commercial buildings together account for approximately 12% of the potable water and 73% of total electricity in the US [1, 14]. Modest conservation improvements in these sectors can have a significant impact. Interestingly, each resource production and supply chain has a complex interplay. For example, in the US, the water consumed to generate electricity accounts for 42% of the total water consumption in the building sector [13]. This means that consuming electricity implicitly consumes water too. Conversely, water purification and transportation systems consume electricity. The American public water supply and treatment facilities consume about 56 billion kWh per year, which is enough electricity to power over 5 million homes [14]. Moderate savings of potable water results in significant financial benefits. The US Environmental Protection Agency states that 3 trillion gallons of water could be saved each year if every household in the US decreased its water consumption by 30 percent [14]. This results in a dollar-volume saving of more than $18 billion a year. Water conservation is of even greater financial significance as increasing purifi-
Categories and Subject Descriptors H.4 [Information Systems Applications]: Miscellaneous
General Terms Experimentation, Human Factors, Measurement
Keywords Energy Efficient Homes, Resource Monitoring, Sustainability
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. BuildSys’09, November 3, 2009, Berkeley, CA, USA. Copyright 2009 ACM 978-1-60558-824-7 ...$5.00
Introduction
Toilet Bin Water Usage Pattern
Laundry Machine Water Usage Pattern
0.3
Sprinkling System Water Usage Pattern
0.18
1.4
Water Consumption: 4.4L
Water Consumption: 120L
Water Consumption: 4kL
0.16
1.2
0.25
0.2
0.15
0.1
Water Flow [L/s]
Water Flow [L/s]
Water Flow [L/s]
0.14 0.12 0.1 0.08 0.06
1
0.8
0.6
0.4
0.04 0.05
0.2
0.02 0
0
10
20
30
40
50
60
0
0
200
400
600
Time[s]
(a) Toilet Bin
800
1000
1200
1400
1600
Time[s]
1800
2000
0
0
1000
2000
3000
4000
5000
6000
7000
8000
Time[s]
(b) Laundry Machine
(c) Sprinkler System
Figure 1: Water consuming activities in a homes are sporadic events. However, many of consumption activities have a unique pattern, which makes it possible to detect particular events using a pattern matching algorithm.
cation cost is being compounded by rapid demand growth. In conclusion, to make buildings sustainable, it is necessary to monitor not just electricity, but several other resources as well. One start is the US Green Building Council’s (USGBC) Leadership in Energy and Environmental Design (LEED) program. The LEED guidelines put balanced emphasis on both electricity and water efficiency; and their interplay. In this paper, we discuss our experience on real-time electricity and water monitoring in a single family house. We collected 3 months worth of full house-hold synchronous power and water consumption data, which thus allows us to analyze both resources at the same time. We present several data analysis mechanisms, including main water meter aided usage detection, activity inference using diurnal statistics, exploitation of dual-consumption patterns to inference activity, and privacy implications. Using this analysis, we identified research opportunities and challenges in fine grained resource monitoring system design for buildings. Our contributions are three fold: (1) We present a fixture level water consumption profiler using the main water meter. (2) We identify challenges in fine grained load monitoring using the main power meter given the very high load dynamics in today’s home. (3) We show that it is possible to detect an appliance’s load by observing the water and power consumption at the same time, since many appliances consume several resources together. With this, we advocate that it is beneficial to monitor all the used resources, and not just electricity.
2
Non-Intrusive Resource Monitoring using a Single Sensor
We instrumented a single family house in Southern California with an ultrasonic water flow meter (Shenitech STUF200H) and a main power monitor (TED-1001). We collected the power and water consumption at a fine-grained rate of 1 Hz. While this installation is still costly, these systems will become more commodity as the Automatic Meter Reading Association [3] and the American Water Works Association [2] are deploying more and more smart meters with real-time wireless meter capabilities for billing and smart-grid application purposes. Therefore, we expect that the data analysis performed in this section can be leveraged without extra hardware cost in the near future. The analysis is performed using several pattern recogni-
tion techniques to automatically identify and track device level resource usage, while demonstrating the possibility of information fusion between water and power use to improve detection accuracy. In addition, we identify limitations of this approach by showing that the consumption trace is very complex and dynamic.
2.1
Fine Grained Water Monitoring using Pattern Matching
In a typical household, the number of water fixtures is limited. In addition, water consuming activities are sparse in time compared to power consuming activities. Many of the water fixtures have a distinct water usage pattern, because they are mechanically or electro-mechanically controlled, e.g. toilet, dish washer, laundry machine, and sprinkler systems. In addition, water related activities are sporadic in time, unique enough to be distinguishable in time-series data. These patterns can be identified and extracted at the fixture level using different types of pattern recognition algorithms (Figure 1). Similar to the Non-Intrusive Load Monitoring technique (NILM) [10], we use these unique patterns to identify the use of individual water fixtures. More specifically, we can perform an automatic data annotation using a matched filter with the real-time trace of the water consumption in the house. Using the matched filter, we were able to detect several different water activities including toilet flushes, lawn sprinkler system, and laundry machine activities. Figure 2 shows such a detection trace when the toilet was flushed. Its detection accuracy is about 90%. Several useful pieces of information can be extracted from this data. The easiest of which is how much water got consumed by flushing the toilet. More interestingly, and not as obvious, we can infer how many people stay in the house at a current day. For example, Figure 3 shows a transition region where two interesting events happened in the house. Abrupt changes in the number of flushes indicate that several people have visited, resulting in a much higher number of flush events (from 60 to 90 for a short term visit on day 14). In addition, the number of flushes changed from around 20 to 45, indicating a long-term visit by two persons. Incidentally the house-owner changed at the same time the toilet bins to more water efficient ones. It is interesting to observe that the number of flushes have increased after that installation. After interrogating the owner we found out that the residents flush more often than before, as they are less concerned about water usage now!
Toilet Flush Event for Five Days
Number of Flushes in a house 100 90 80
Day 1
Before New Toilet Bin Installation
70
Counts
Day 2
60 50 40
Day 3
30
After New Toilet Bin Installation
20
Day 4 10 0
Day 5 4
6
8
10 12 14 Time[h]
16
18
20
22
Several other useful pieces of information can be extracted with reasonable accuracy and resolution using just the main water meter. Figure 4 illustrates how much water the residents used for toilet flushes and laundry machine. The bottom plot in Figure 4 shows the total diurnal water usage over a month from June to July. The spikes in the plot indicate that the sprinkler system ran, accounting for about 2000 or 40001 liters of water per run. This simple matched filter approach works very efficiently for large water consuming events, or events that have a very distinct usage pattern. For example, the lawn sprinkler system is the biggest water consumer in a household, and detecting this event is very easy using this pattern matching mechanism. It results in a 100% detection rate. However, for smaller events, such as a faucet usage, and compound events, where several systems consume water at the same time, we may still want to use more sensors than just the main meter, to ensure accuracy. A rather straight-forward cost-accuracy trade-off can be considered in this application. For example, if we are only interested in routine water usage, we can simply use matched filtering on the main water monitor trace. If more visibility is necessary, one can spend more money and add extra sensors, thus gaining resolution with better accuracies and confidence.
Challenges in Fine Grained Power Monitoring using a Single Sensor
Unlike water activities, electrical energy use in a house has compound characteristics. It is partially due to the number of electrical appliances compared to the number of water fixtures (70 to 80 appliances vs. less than 20 fixtures). In addition, electrical appliances often consume a certain amount of static power, just by being plugged in. Compared to water fixtures, this would mean a constant drip of water. This makes the problem of power monitoring in a house much harder. Although recent advances in sensor networks allow to 1 The
10
15
20
25
24
sprinkler system has two different watering patterns, one consuming 2000, the other 4000 liters.
Figure 3: The number of flushes provides information regarding house occupancy. Water Consumption for Toilet Flushes and Laundry Machine Water Usage[L]
2
Figure 2: A simple matched filter can reliably detect the toilet flush event. The same technique is applicable to other water activities that have a unique usage pattern, including dish-washer, laundry machine, and sprinkler systems. By detecting these events, coarse-grained, yet real-time water usage can be estimated.
2.2
5
Time[Days]
1000 Toilet Flush Laundry Machine
800 600 400 200 0
0
5
10
15
20
25
Time[Days] Daily Water Consumption in a House(Mid June to July) Water Usage[L]
0
0
5000 Others Sprinkler
4000 3000 2000 1000 0
0
5
10
15
20
25
Time[Days]
Figure 4: Water Usage Breakdown is possible with reasonable accuracy.
have multiple sensor nodes at reasonable cost, the monitoring system may quickly become prohibitively expensive for this non-critical domestic application. For example, if we want to monitor the power consumption of every appliance, we need to have 70 to 80 sensors, one for each appliance, if we want to use a systems like described in [7, 9]. Even though each sensor could potentially cost as few as $10, the overall cost would quickly rise to $700. In addition, nontrivial installation, calibration, association and data aggregation is an outstanding issue in the UbiComp community [4]. Alternative approaches have limitation too. For example, it is difficult to use a system as described by Patel et al. [11] or Leeb et al. [10] because many electrical appliances have variable power consumption [9]. And even in a small one family house, we already observed many compound events (appliance on/off events), one of the challenges that makes NILM so difficult. It is possible to use NILM [10] techniques, but a huge burden is put on the feature extraction process and system training. While it is possible to break down the power consumption for large consumers, like the HVAC system, smaller appliances are very hard to extract. This is problematic given the trend that small appliances consume more and more of the total power consumption [1]. Figure 5 shows the load characteristics including abrupt power transition and their duration. Each appliance tends to have a unique pattern;
Synchronous Water/Power Usage(Laundry Active) Water Usage [L/s]
12
2.5
11 10
2
9
1.5 7 6 1 5 4 0.5
3 2
0
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
1
Power Transition[kW]
Figure 5: An instance of features in non-intrusive load monitoring: power transition and duration. Since features of many appliances are very close, the use of NILM has limited capabilities to disambiguate small appliances. Histogram of Power Transition
0.1 0.05
0.2 0.15 0.1
0 23
23.1
23.2
23.3
23.4
23.5
23.6
23.7
23.8
23.6
23.7
23.8
Time[Hour] 3
2.5
2
1.5 23
23.1
23.2
23.3
23.4
23.5
Time[Hour]
Figure 7: This synchronous water and power trace indicates that power and water usage has a unique pattern as the laundry machine runs a predefined routine. This pattern depicts its normal operation, Filling Water Tank, Washing, Draining, Filling Water Tank, Rinsing, and Final Spin. Note that this trace is from a controlled setting where the residents didn’t run other appliances so the power consumption pattern can be observed. It is usually not possible to see this clean power consumption trace due to other appliances turning on and off, making automatic annotation by looking only at power numbers difficult. Daily Power Consumption in a House(Mid June to July)
0.05 120
0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Power Transition[kW]
Figure 6: This histogram shows the number of power transitions over 4 days. Since too many small power transitions happen, a NALM approach is limited.
Power Usage[kWh]
Normalized Count
0.15
Power Consumption [kW]
Duration[h]
8
0.2
100 80 60 40 20 0
0
5
10
15
20
25
Time[Days]
unique average power consumption, time-of-use, inter power state transition. At the same time, the difference among appliances is very subtle, which makes features flock together in the feature space. The resolution of the legacy meter is limited, thus making reliable detection difficult (Figure 6).
2.3
Fine Grained Power Monitoring with Side Information
Monitoring several resources at home improves the accuracy in appliance level power monitoring. Some appliances, often major resource consumers, use several resources concurrently, e.g. gas oven, oil air heater, laundry machine, dish washer, etc. For example, a laundry machine uses water and electricity at the same time in a very deterministic way. Thus, analyzing both water and electricity at the same time helps us understand and detect when the laundry machine runs. Figure 7 shows the total water and power consumption of a house when the laundry machine is running. We can see that the resource consumption has a specific pattern as it runs through pre-programmed cycles. By observing these two traces, it is possible to detect when the laundry machine is active even if the power consumption trace is dynamically changing. In addition, monitoring all the resources in a house, even just at a main meter level, could reveal other interesting correlations amongst devices, thus allowing another dimension for possible sustainability optimizations.
3
Discussion and Future Direction
To make fine grained monitoring widely applicable, several challenges have to get solved. First, researchers may want to investigate trade-offs among various techniques and
Figure 8: Daily power consumption from mid June to mid July. Power usage went up due to the use of the air conditioning system, resulting in about 40kWh overhead per day.
develop a modular approach that incorporates existing technologies. Appropriate design in terms of cost, usability, accuracy, reliability, and device maintenance issues need to be considered. User adaptation is another issue. As Beckmann et al. [4] pointed out, usability of a system is of great concern. While the initial cost associated with hardware becomes negligible over long-term, non-trivial installation, maintenance, device association, and privacy concerns make the system prohibitively expensive for general users. Since the value of the application is relatively small compared to the burden lying on the users, further studies on end-user adaptation and user experience need to be conducted. When we look at the resource consumption in a building, we have to consider that the highest level of decision is made by people, since they have to occupy and use the space. Thus, excellent information support to the end-user is impeccable to maximize the utility of such a fine-grained monitoring system. The buildings can be seen as an ecosystem where residents make decisions on resource consumption with limited information. Providing the end-user with right choices and more valuable information will help them to make the right decisions on their consumption, without impacting their comfort or utility. Therefore, we need to consider effective and efficient data processing and visual-
Synchronous Water/Power Usage(Bathroom Use)
Efficiency, the Metric for Resource Consumption
Many resources are used to regulate a user’s comfort level given certain ambient conditions. A good example is the HVAC system that keeps temperature and humidity at a certain level so occupants feel comfortable. In addition, a sprinkler system is used to maintain the lawn in good condition. Purely monitoring the total resource consumption can be misleading. Depending on ambient conditions, the same amount of resource consumption means different efficiencies. For example, ambient temperature and relative humidity drastically change evaporative losses. Therefore, the time of day a lawn sprinkler is used can have vastly different efficiencies to actually water the plants and grasses. In addition, soil moisture level often determines the right time and amount of water the lawn needs. However, more often than not, a sprinkler system does not take these measurements into account, and a predefined watering pattern will put out the same amount of water everywhere. Another example can be made with heating and cooling of an indoor room. The difference between indoor and outdoor temperature determines heat loss or gain, and a proper setting of the thermostat depending on outdoor temperature can result in efficient HVAC usage. Although straight forward, by combining resource consumption information with ambient conditions, it is possible to let utility consumers understand the different aspects of their resource usage. More precisely, it can tell them how efficient their use is, or where and how they waste resources.
3.2
Information Fusion Mechanism
Buildings have very complex, diverse load and resource consumption characteristics. The key question is whether every monitoring technique scales. The sole use of only direct[7], indirect[8, 9], or infrastructure-mediated[6] sensing techniques cannot be sufficient to cover all the load and resource demands. At some point, a sophisticated inference engine that incorporates all these sensing techniques will be needed. We briefly touched upon this possibility for the laundry machine case, though further investigation is needed in how other resources could help to identify particular appliances, or how this fusion could be automated.
3.3
Privacy Implication
Home monitoring applications reveal interesting privacy implications. For example, Srinivasan et al. [12] showed that detecting packet exchanges in a home monitoring system can reveal privacy prone information such as occupancy. Indeed, Fogarty et al. [5] actively use the signals from the plumbing system to infer water activities. Similar privacy prone information can be extracted from the power and water traces. While analyzing the water consumption, we discovered that every now and then, someone flushes the toilet twice. We asked the home owner, and he said that this must be a long-term visitor. In addition to the double-flushing, we also discovered that in one instance, someone who flushed only once, did not use the bathroom sink, while the person flushing twice uses the sink for a long time (Figure 9). At the same time, the synchronous power data revealed that one does not
0.2
Flush Once
0.15 0.1
Flush Twice
Sink On
0.05
Power Consumption[kW]
3.1
Water Usage[L/s]
ization, which is tailored to specific user demand.
0
0
200
400
600
800
1000
1200
800
1000
1200
Time[s] 2.5
2
1.5
Lamp On 0
200
400
600
Time[s]
Figure 9: The flush pattern recognition mechanism finds that someone in the house tends to flush the toilet twice. The synchronous power data additionally revealed that one person did not turn on the light, while the other did.
use the bathroom light while the other turned on the light. This implies that by observing specific patterns in the bathroom, we can infer not only when and how an occupant uses the bathroom but also who uses it. Even a simple matched filter can reveal interesting observation on users resource usage patterns and abnormal conditions, which may be useful information for healthcare concerns. However, this information may be too private for some users. Once a user realizes just how much can be inferred from these consumption traces, they might stop using them in fear of privacy concerns. Therefore, we carefully need to consider a balance between privacy and application objective. The big question becomes on how to lower the resolution so people in a house feel safe to provide information while still ensuring the application benefits?
3.4
Data Management and Representation
One problem of fine-grained resource monitoring is the vast amount of data collected. In our 3 month run, we already collected more than 300MB worth of data. However, unlike multi-media data, this time-series data is hard for users to understand, unless proper representation and summaries are provided. Appropriate data visualization and representation need to be considered to make this information effective. Although many utility consumers are solely concerned with their monetary benefits, utility consumers’ motivations are very diverse as Woodruff et al. pointed out [15]. Although not exclusive, they found three main motivators that bright green initiatives have, such as counterculture biocentric activism, American frontier self-reliance and rugged independence, and trend-focused utopian optimism. These are all well perceived values in the US, as one can observe by the success of Al Gore’s “An Inconvenient Truth”. Different perspectives and interpretations about one’s resource consumption may have stronger motivation to people motivated by these different values. In that, a more personalized data representation and interpretation will show and motivate the end-user where to change what. Many different power meters today on the market incorporate the cost of energy into their calculations. Cent-ameter and TED are only a few examples that offer various
options on cost display, and can even cope with variable power rates often used by power utility companies. However, several other metrics can be used to quantify this resource consumption. Since most of the electricity comes from non-renewable resources, about 85% in the US, a rough projected Carbon emission estimate can be calculated for the consumption of electricity. For example, what does it imply if you consume 1kWh in Southern California in terms of CO2 emissions, or in terms of indirectly used water to produce that energy? For example, the US Western Interconnect consumes 16.7L per 1 kWh, compared to the Texas Interconnect’s 1.6L per 1 kWh [13]. Indeed, long term benefits and morals play an important role. Even if a sheer green decision sacrifices many monetary benefits, and sometimes even comfort level, many green thinking people prefer to make the right choice [15]. Some examples are using natural ventilation, biking to work, use costly but environmentally friendly grocery goods, etc. For someone who cares more about water conservation, projected water consumption for electricity is a more useful metric than just pure power numbers or CO2 emissions.
3.5
Developing Interactive Technology
As many sources of uncertainty exist, it will be too expensive to make a fully autonomous fine-grained resource monitoring system. Interactive technology can help to lower the complexity of the monitoring system. As we have seen in Section 2.1, several features can be effectively extracted from the users if they input the time of use of an appliance. Gradual adaptation of the system can be done in this way, where the user teaches the system of what pattern means which activity to the end user. Such a technique would allow users to actively participate in resource management by forcing them to look at the collected data. Several auxiliary information sources may add more values to this application. A good way for users to easily access and manipulate data becomes essential. For example, in Southern California, a major source of potable water consumption in homes is the sprinkler system. Watering the lawn has several implications. Residents want to maintain their lawn in good shape to maximize their property value. At the same time, they want to follow regional regulations. For example, watering the lawn on Mondays is prohibited in Los Angeles, and residents in that city may want to get some feedback on this regulation. Since this addresses a user’s concern, the user may want to actively input needs and intent.
4
Conclusion
Resource monitoring is a promising application that helps make buildings more sustainable. We showed that various pieces of information such as fixture level water consumption, occupancy, or appliance status can be extracted from the real time water and power traces. However, to maximize the utility and broad user adaptation, researchers should carefully address challenges in the monitoring system design. We showed through examples that having synchronous data improves accuracy of device level resource consumption. Yet, further research on information fusion from various sensors and monitoring systems is needed. In addition,
non-trivial installation, maintenance, data association and aggregation need to be considered to successful user adaptation of these new and exciting technologies.
5
Acknowledgements
This material is based upon work supported by the NSF under award # CNS-0627084, CCF-0820061, and CNS0905580, and by the UCLA Center for Embedded Networked Sensing. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding agencies.
6
References
[1] Annual energy outlook 2009, doe/eia-0383(2009). Tech report, U.S. Department of Energy, 2009. [2] American Water Works Association. C706-96(R05) : AWWA standard for direct-reading remote-registration systems for cold-water meters. STC, 2005. [3] Automatic Meter Reading Association. http://www.amra-intl.org/. [4] C. Beckmann, S. Consolve, and A. LaMarca. Some assembly required: Supporting end-user sensor installation in domestic ubiquitous computing environments. In Ubicomp, 2004. [5] J. Fogarty, C. Au, and S. E. Hudson. Sensing from the basement: A feasibility study of unobtrusive and low-cost home activity recognition. In ACM UIST, 2006. [6] J. Froehlich, E. Larson, T. Campbell, C. Haggerty, J. Fogarty, and S. Patel. HydroSense: Infrastructure-mediated single-point sensing of whole-home water activity. In UbiComp, 2009. [7] X. Jiang, S. Dawson-Haggerty, P. Dutta, and D. Culler. Design and implementation of a high-fidelity ac metering network. In ACM IPSN/SPOTS, 2009. [8] Y. Kim, T. Schmid, Z. M. Charbiwala, J. Friedman, and M. B. Srivastava. NAWMS: nonintrusive autonomous water monitoring system. In ACM SenSys, 2008. [9] Y. Kim, T. Schmid, Z. M. Charbiwala, and M. B. Srivastava. ViridiScope: Design and implementation of a fine grained power monitoring system for homes. In UbiComp, 2009. [10] S. Leeb, S. Shaw, and J. Kirtley, J.L. Transient event detection in spectral envelope estimates for nonintrusive load monitoring. IEEE Trans. on Power Delivery, Jul 1995. [11] S. N. Patel, T. Robertson, J. A. Kientz, M. S. Reynolds, and G. D. Abowd. At the flick of a switch: Detecting and classifying unique electrical events on the residential power line. In UbiComp, 2007. [12] V. Srinivasan, J. A. Stankovic, and K. Whitehouse. Protecting your daily in-home activity information from a wireless snooping attack. In UbiComp, 2008. [13] P. Torcellini, N. Long, and R. Judkoff. Consumpive water use for u. s. power production. National Renewable Energy Lab. Tech. Report, NREL/TP-550-33905, 2003. [14] U. S. EPA. http://epa.gov/watersense/index.htm. [15] A. Woodruff, J. Hasbrouck, and S. Augustin. A bright green perspective on sustainable choices. In CHI, 2008
