Data Driven Energy Efficiency in Buildings

Nipun Batra1, Amarjeet Singh1, Pushpendra Singh1, Haimonti Dutta2

Venkatesh Sarangan3, Mani Srivastava4

1Indraprastha Institute of Information Technology Delhi, India {nipunb, amarjeet, pushpendra}@iiitd.ac.in2 CCLS Columbia {haimonti@ccls.columbia.edu}

3 Innovation Labs, Tata Consultancy Services {venkatesh.sarangan}@tcs.com4 UCLA {mbs@ucla.edu}

ABSTRACTBuildings across the world contribute significantly to the over-all energy consumption and are thus stakeholders in grid op-erations. Towards the development of a smart grid, utili-ties and governments across the world are encouraging smartmeter deployments. High resolution (often at every 15 min-utes) data from these smart meters can be used to understandand optimize energy consumptions in buildings. In additionto smart meters, buildings are also increasingly managedwith Building Management Systems (BMS) which controldifferent sub-systems such as lighting and heating, ventila-tion, and air conditioning (HVAC). With the advent of thesesmart meters, increased usage of BMS and easy availabil-ity and widespread installation of ambient sensors, there isa deluge of building energy data. This data has been lever-aged for a variety of applications such as demand response,appliance fault detection and optimizing HVAC schedules.Beyond the traditional use of such data sets, they can be putto effective use towards making buildings smarter and hencedriving every possible bit of energy efficiency. Effective useof this data entails several critical areas from sensing to de-cision making and participatory involvement of occupants.Picking from wide literature in building energy efficiency,we identify five crust areas (also referred to as 5 Is) for real-izing data driven energy efficiency in buildings : i) instru-ment optimally; ii) interconnect sub-systems; iii) inferreddecision making; iv) involve occupants and v) intelligent op-erations. We classify prior work as per these 5 Is and dis-cuss challenges, opportunities and applications across them.Building upon these 5 Is we discuss a well studied problemin building energy efficiency - non-intrusive load monitoring(NILM) and how research in this area spans across the 5 Is.

1. INTRODUCTIONBuildings across the world contribute significantly tototal energy usage. Figure 1 shows the contribution ofbuildings to energy consumption in India, USA, China,Korea and Australia [34, 41, 93, 35, 94]. Rapid rateconstruction of buildings presses the need to look intoimproving energy efficiency in buildings with a goal ofdecreasing overall energy footprint. Thus, towards thevision of sustainability in the age of dwindling natu-

India USA China Korea Australia0

10

20

30

40

50

%en

ergy

cont

ribu

tion

Figure 1: Contribution of buildings to overall energy con-sumption

ral resources, buildings need to be made more energyefficient.

“Measure twice and cut once”- so goes the old adage.With this adage in mind and the goal of understand-ing building energy efficiency, collecting building energydata assumes prime importance. Traditionally, buildingenergy data included monthly electricity bills collectedmanually by the utility companies and other sporadicdata such as energy audits. Owing to the manual andsporadic nature of this collected data, the available datais very sparse and provides limited insights into buildingenergy efficiency.

Buoyed by the success of data sets such as MNIST [67]in instigating machine vision research, previous study [62]suggests that research in building energy domain canbe spurred by availability of data sets. While previ-ously deemed improbable, collection of such data setshas become increasingly common due to low cost sens-ing devices enabling high resolution and automateddata collection. Furthermore, governments and utili-ties across the world have started rolling out smart me-ters1 with an aim of developing a smart grid. Beyond

1UK: https://www.gov.uk/government/policies/helping-households-to-cut-their-energy-bills/supporting-pages/smart-meters;Australia: http://uemg.com.au/customers/your-electricity/smart-meter-rollout.aspx

1

arX

iv:1

404.

7227

v3 [

cs.O

H]

2 S

ep 2

014

the envisioned applications by the government and theutilities, this data can also be used for developing anunderstanding into building energy. Smart meters typ-ically report electricity consumption to the utility andto the end consumer at rates ranging from a reading ev-ery second to a reading every hour. Commercial build-ings are also increasingly managed with Building Man-agement Systems (BMS) which control different sub-systems such as lighting and HVAC. BMS are computerbased control systems for controlling and monitoringvarious building systems such as HVAC and lightingand are typically used to manage commercial buildings.BMS sense several spatially distributed points acrossthe building to monitor parameters required for controlaction. Buildings are also commonly equipped with am-bient sensors for monitoring parameters such as light,temperature and humidity. These ambient sensors areoften coupled with security systems to raise intruderalarm or to maintain healthy ambient conditions viathermostat or BMS control. With the advent of smartmeters, increased usage of BMS and ease of availabil-ity and installation of ambient sensors, there is a now adeluge of building energy data.

While collecting building energy data is easier thanbefore, deployments can get increasingly hard to man-age, especially when large number of sensors are de-ployed. However, previous research [57, 17, 90] hashighlighted that building energy data can be obtainedat different spatio-temporal granularities. This high-lights the need for optimal instrumentation that canhave different connotations for spatial and temporal do-main. For spatial domain, optimal instrumentation in-volves monitoring at a subset of locations while stillbeing able to accurately predict at all the desired lo-cations. For the temporal dimension, optimal instru-mentation involves sampling at a lower resolution whilestill being able to predict at a higher temporal resolu-tion. Desired application may choose the optimal set ofsensors considering cost-accuracy tradeoffs across thesedifferent granularities.

Traditionally, there exist multiple building subsys-tems such as security, networking, HVAC and light-ing, each performing their own operations in isolation.However, buildings are a unified ecosystem and optimaloperations would require interconnecting these sub-systems. The combined information from the differentsub-systems is greater than sum of individual informa-tion from each of the systems.

Once the systems are interconnected, data comingfrom diverse systems can be used to improve upon thedecision making for optimal building operations. In-ferred decision making can help in identifying ineffi-ciencies, raising alerts and suggest optimizations. Tra-ditionally data from within the system has been usedfor simple decision making e.g. motion sensor based

Figure 2: Data driven building energy efficiency sits atthe intersection of behavioral science, data science, con-trol systems and sensor networks

lighting control. Furthermore, utility companies previ-ously relied on customers’ phone calls to detect poweroutages. However, utility companies can now leveragesmart meter data from different homes to quickly detectpower outages and plan accordingly. Temporal patternsin the electricity consumption data [4, 17, 43] can beused to detect faulty operations, predict load profilesand optimize operations accordingly. Spatial patternsin the electricity consumption can identify different elec-trical sub-systems and study their effect on aggregateconsumption. The vast set of analysis possible furthernecessitates the importance of data sets collected fromthe real world which can be to used to simulate the effectof inferred decisions before performing optimizations.

Buildings constitute ecosystems involving interactionsbetween the occupants, the physical and the cyber world [48,89]. The presence of occupants, each with their individ-ual preferences makes the ecosystem even more com-plex. Traditionally, the control decisions for buildingoperation are taken at a central facility level assumingcertain desired operating conditions without involvingthe occupants in decisions regarding desired conditions.Such policies are bound to be energy inefficient andmay cause occupant discomfort at times [15]. More-over, previous literature [75] suggests that energy un-aware occupant behavior can add upto one-third to abuilding’s energy performance. However, when empow-ered with actionable feedback, occupants may save upto15% energy [27]. Building occupants can also provideuseful data such as comfortable temperature and lightintensity levels, which can be used to optimally schedulethe HVAC systems. Thus, various optimizations in thebuilding energy ecosystem can be enabled by involvingoccupants.

All of optimal sensing from interconnected subsys-tems, resulting in decisions inferred from the rich dataset

2

while involving the occupants will overall result in in-telligent operations. Such intelligent operations al-ready exist but miss out one or more aspects discussedpreviously i.e. either they do not involve the occupantsor are taken at a subsystem level without accountingfor data from other subsystems in operation.

Thus, based on literature in building energy domain,we have identified these five crust areas also called five Iswhich are as follows: i) instrument optimally; ii) inter-connect sub-systems; iii) inferred decision making; iv)involve occupants and v) intelligent operations. Fig-ure 2 summarizes the relationship amongst the 5 Is ofbuilding energy which span across broad fields such assensor networks, behavioral sciences, data science andcontrol science. The interconnected nature of these 5Is makes the overall optimization of building energy ef-ficiency a complex problem. For solving such a complexproblem of intelligent building operations, rich data fromacross the diverse systems, together with smart algo-rithms for data inference and participatory engagementof occupants is critical. Such rich datasets are now be-coming a reality [62, 7, 17, 12]. Further, rich researchexists in from recent times that address algorithms fordata inference and understanding behavioral aspects ofoccupants feedback [50, 27, 24]. This work bring to-gether these different aspects into a survey while point-ing out the big opportunities that exist along each of the5 Is together with those existing at the system level.

2. EXAMPLE SCENARIOHaving briefly discussed the various facets of data drivenenergy efficient buildings, we now discuss an examplescenario covering all of these aspects. We base ourtheme around the classic paper on Pervasive comput-ing [89].

John goes to bed at 11 PM in the night. He sets thealarm on his smartphone at 5 AM when he intends to gofor a jog. However, his health has not been at the bestsince the last few days affecting his sleep. He decidesto inform his alarm clock that it should wake him upat 5 AM if he is able to sleep by 11:30, else wake himup after he has completed 6 hours of his sleep. John’ssmart surround sensors capture his sleep patterns andcommunicate to the alarm to ring at 6 AM. His sleepdisturbances, coughing are also captured, archived andemailed to John for sending to his doctor. John goesto the refrigerator to grab some cold water. The refrig-erator understands that the person who has come topick up water is John. John’s heath system remindedhis refrigerator that John has been advised to refrainfrom cold water. John gets ready and leaves home foroffice at 9 AM. The security system, door and motionsensor register this movement. This information is com-municated to John’s thermostat which starts rampingdown. His lights also turn off automatically sensing his

absence. John has a meeting at 10 AM and thus hewould only go to his cabinet to keep his bag. Basedon the number of expected attendees, the thermostat inthe meeting room starts ramping up, in order to achievethe desired temperature in time for the meeting.

John’s washing machine at his home is scheduled torun for 1 hour before John returns in the evening. Hiswashing machine interacts with the grid and predictsthe best time to run when the load on the grid is low isnoon time when the electricity prices are low. Around3 PM, winds carry away the clouds and there is brightsunshine. John’s solar system starts producing elec-tricity. Part of this DC produce is directly fed to DCappliances in John’s home and the surplus is stored ina battery. John figured that he can save more moneyby locally consuming his generated solar energy as op-posed to selling it to the grid. In the meanwhile, someof the artificial lights in John’s office turn off in lieuof the sunlight available. John gets an email from theNILM system installed at his home about his disaggre-gated monthly consumption. The system recommendsthat the tungsten based lighting in his home is very in-efficient and eating up 30% of his bill. If John wouldreplace the same with more efficient LED based light-ing, his overall bill would go down by 15%. Consideringthe cost of replacement, his ROI period would be lessthan 6 months, after which he would be able to save 10USD a month.

We now analyse the above scenario and see how eachsmall piece in this giant picture is a reality today. Sys-tems such as iSleep [42] have explored using smart-phones and motion sensors for sleep quantity and qual-ity detection. In order to wake up John after he hascompleted 6 hours of sleep, his smartphone detects hissleep pattern. His smartphone contains an app whichdetects his cough patterns [66] and uploads the mostcritical data to his doctor. When John wakes up andgoes to the refrigerator, systems for energy apportion-ment [44] ascertain that it is John and not someone elsewho is trying to draw cold water. The refrigerator isconnected to John’s smartphone over IP and informshim that cold water could be injurious to his health.Several years of research in occupancy detection usingvariety of sensing modalities [2, 68, 60, 57] are able todetect that John’s home is unoccupied when John leavesfor his office. Smarter occupancy driven thermostatshave been proposed in recent literature [68, 87] whichcontrol and save energy based on occupancy predictionand external temperature, when John leaves for his of-fice. By interconnecting soft sensors [98] such as of-fice chat client and meeting software, the thermostat inJohn’s personal cabinet does not ramp up since he has ameeting in the board room. Conference room manage-ment sensor system [78] in the boardroom is alert to thenumber of occupants and drives the HVAC accordingly.

3

FacebookDaily

Smart metersHomes: 5mn

Monthly

LHCDaily

SDSYearly

0

100

200

300

400

500Si

zein

TB

Figure 4: Comparing the data deluge in building en-ergy with social networking (Facebook), particle physics(Large Hadron Collider(LHC)) and astronomy (SloanDigital Sky (SDS))

Meanwhile in John’s home his washing machine turnson at 1 PM. Since this is an off-peak period, electricity isavailable at much cheaper rates at this time in compar-ison to the rest of the day. John scheduled his washingmachine to run for one hour when the electricity wouldbe cheap, which was enabled by demand response strat-egy [13]. Due to availability of sunlight after 3 PM, DCappliances in John’s home switch from the utility andconsume raw DC power from the solar panel [38]. Thelighting control system [30] in John’s office also sensesthe bright sunshine and saves power while maintainingcomfortable lighting levels inside the office. The smartmeter installed at John’s home is regularly collectinghis electricity data and periodically sends John the dis-aggregated breakdown of power by appliances [43].

While many of these individual systems have beenexplored in the past in isolation and often in researchsettings, their application to the real world and theirinterconnection to complete the picture remains largelyuntested. The stepping stone to making such complexscenario a reality is the high resolution multimodal datacollected from each of these subsystems, interconnectedtogether, with algorithms that can do efficient opera-tions while accounting for John’s preferences.

3. BUILDING ENERGY DATA“Data is the new oil”. Data science has brought about aparadigm shift in the way problems are solved in myr-iad applications. From social networks to astronomyand subatomic physics, data has brought a new revolu-tion enabling answering important questions. We nowbriefly discuss how data can play a pivotal role towardsthe development of energy efficient building.

Traditionally, building energy data was collected onlyfor billing purposes. Figure 3 shows the electricity billsof residential apartments in London, UK and Delhi,India. Usually, data for such bills is collected once

a month manually by the utility companies. The billshown in Figure 3b only provides the units of energyconsumed during the billing period. Commercial enti-ties owing to their heavy consumption are often pricedon a time of day based pricing. Thus, in addition to thetotal units consumed, commercial entities are also pro-vided with units consumed in peak and non-peak hours.In both of these cases, the consumption information iscoarse and provides limited actionable insights.

With the advent of the smart meter and some of theother sensors discussed in Section 1, more building en-ergy data is now available than ever before. While theconventional process followed by utilities would gener-ate a single reading per home per month, smart meterscollecting data once every 15 minutes would be collect-ing 3000 times more data [26]. Smart meters are oftencapable of collecting data at higher rates of once everyminute which can amount to 240 TB of data collectedfrom 5 million homes a month [9]. Figure 4 contraststhe volume of this smart meter data collected with otherapplications which have greatly benefited by the datadeluge. The availability of such building energy data en-ables one to answer several important questions, someof which are presented below:

• Consumers can be informed in real time abouttheir consumption. This may be seen analogousto the cell phone alerts we get after every datatransaction or call made. This may greatly helpthe end users to not only understand their usagebut also act on it to conserve energy.

• Utilities earlier used to rely on receiving phonecalls from aggrieved customers for detecting out-ages. However, with smart meter data readilyavailable, utilities can leverage it for speedier out-age detection and allocate resources efficiently toreduce downtime.

• Figure 5 shows the time of day energy consump-tion of a building in IIIT Delhi campus. Whensuch detailed information is made available to theend users, they can take initiatives towards shift-ing their loads to normal or off-peak hours, whenelectricity is cheaper. Such measures also benefitthe utility as the peak demand is reduced.

• Figure 6 shows the trend in energy consumptionfrom a home in New Delhi [17]. This decreasingenergy usage can be attributed to decreasing tem-peratures and hence decreasing use of air condi-tioners. Developing an understanding into the cor-relations between weather and energy usage canallow better HVAC control. Such information ispresent in the UK bill (Figure 3a). However, with-out detailed analysis the same does not translateinto actionable savings.

4

(a) London, UK (b) Delhi, India

Figure 3: Electricity bills from residential apartments in London, UK and Delhi, India. [Best viewed in color]

1Jan

2Jan

3Jan

4Jan

5Jan

6Jan

7Jan

0

50

100

150

200

250

300

Ene

rgy

(KW

h)

normaloff peakpeak

Figure 5: Time of day energy usage for a building in IIITDelhi campus

Oct2013

Nov2013

Dec2013

Jan2014

Feb2014

0

200

400

600

800

1000

1200

1400

Bill

Am

ount

inIN

R

Figure 6: Historical bill collected from the home used iniAWE data set [17]

Towards the realization of many of the above men-tioned questions, building energy data requirements needto be categorized as per the application scenario. Wefirstly discuss the different temporal and spatial resolu-tions at which such data can be collected.Temporal resolution: In accordance with the intendedapplication, building energy data is collected at differenttemporal resolutions. Energy audits are performed onceevery few years [53] and provide key insights into severalaspects of buildings including energy efficiency and com-fort levels. Such audits are costly and require significantinstrumentation. At a lower resolution of once a monthutility companies collect total energy consumption. Asdiscussed previously this provides limited insights tothe end user. Commercial entities often monitor theirpower factor daily as they are liable to be penalized ifthe power factor drops below a certain threshold. Mostof the above considered data collection is largely man-ual and sporadic. Monitoring at higher resolutions of 1hour or lesser requires automated mechanisms such asinstallation of smart meters. With the national smartmeter roll outs across many countries as discussed inSection 1 and initiatives such as Greenbutton2, collect-ing and accessing electricity data has become much eas-ier for the end consumers. Certain applications such asnon-intrusive load monitoring (NILM) [43] may requirehigh frequency data with sampling rates of more than athousand samples every second [40, 7]. We summarizethe temporal variations in building energy data collec-tion in Table 1.Spatial resolution: Buildings consist of multiple loca-tion contexts such as floors, rooms, wings. The utility(water, electricity, gas) networking in buildings also fol-low different topologies introducing the spatial contextacross different nodes in the flow. Thus, different lo-cation contexts and utility topologies provide differentspatial resolution to observe building energy and related

2http://www.greenbuttondata.org/

5

Application RateEnergy auditing once every few years

Electricity billing once a monthPower factor checking once a day

Automated meter reading once 15 minutes or lessHigh frequency NILM several KHz

Table 1: Temporal variations in building energy datacollection

phenomena. The data collected at different spatial reso-lutions can provide a varying degree of information [47].For instance, an electric meter connected at the supplyfrom the grid provides a single point of easy to gatherdata. This data is also used by the utilities for billinginformation. For finer information, electric meters canbe installed at sub-system level (e.g. HVAC, Genset incommercial buildings and circuits at residential build-ings) or at appliance level. As we traverse down theelectrical load tree (Figure 7) in a building, i.e. supplyfrom grid to subsystems such as HVAC to individualappliances such as printers, information content is en-hanced. However, this additional information comes atthe cost of increased monetary and time cost of sensingand maintaining additional instrumentation.

Having discussed the spatio-temporal categorizationof building energy data and its sources, we now discussdifferent sensing modalities which exist within build-ings and can provide additional insights alongside en-ergy data.

1. Building management systems (BMS): Us-ing open standards such as BACNet, the data col-lected by the BMS can be used for building energyanalysis. BMS systems can provide rich data forbuilding energy applications and can also be usedfor control applications based on proposed opti-mizations. Typically, BMS control HVAC, light-ing and security systems and may use a variety ofsensors such as:

• cameras, motion, RFID for security and ac-cess control

• temperature for HVAC monitoring

• light for lighting control

• motion and RFID for occupancy monitoring

2. Water meters: Several building sub-systems com-prise of a rich interaction between electricity andwater. For instance, the central HVAC systems in-volve heating or chilling water to heat or cool theair respectively. Further, water treatment plantshave a significant energy footprint. In several build-ings, electrical pumps are often used to store thewater in tanks for times when water is not avail-able. Swimming pools and water sprinklers also sitat the intersection of energy and water. Thus, wa-ter meter information measuring parameters such

as flow and pressure may provide significant hintsin understanding energy usage of such buildingsub-systems.

3. Ambient sensors: Ambient conditions signifi-cantly impact the working environment [75]. Hence,many of ambient parameters such as humidity, lightlevel, temperature are monitored for maintaininggood working conditions. While some of the mon-itoring equipment are controlled centrally by theBMS, others such as smart thermostats can existas independent units.

4. Access and security systems: These systemslog entries into rooms and can be used to estimatethe occupancy levels in near real-time.

Having discussed the data deluge and its categoriza-tion, we move towards instrumentation systems for ob-taining such data.

4. INSTRUMENT OPTIMALLYHaving discussed the importance of data in realizingbuilding energy efficiency, we discuss how to efficientlyobtain such data. Firstly, we discuss general sensor de-ployments followed by discussion in the specific contextof building energy as per the spatio-temporal catego-rizations which we discussed previously in Section 3.

Sensor deployments have been well studied in the lit-erature and are understood as means of providing datafor improved operations across a variety of applicationdomains [5]. Based on previous work in sensor networkdomain [14, 29, 5], we summarize the design goals ofsuch instrumentation as follows:

• Low power consumption: It is desired that thedeployed sensors and/or controllers do not con-sume significant power impacting their life (whenpowered from a battery) or utility (when poweredfrom mains).

• Wide network coverage: For collecting spa-tially distributed information, more than one endcontroller or node may be deployed. Often, theseend nodes relay the collected data to a central nodeor gateway for storage or further processing. Thedeployment must ensure that all nodes are withinthe communication range of at least one node andtogether cover the whole area that is to be moni-tored, under different real world scenarios. Often,multi-hop strategies are used to increase networkcoverage.

• Robust: The real world presents many unfore-seen challenges which the sensor deployment mustaccount for. It is often desired that the system iscapable of easy healing and recovery.

6

• Ease of deployment and maintenance: Easeof deployment and maintenance outside the con-trolled settings has always been an important chal-lenge and design goal of sensor network deploy-ments.

Sensor deployments for building energy monitoringinvolve some key peculiar features when compared tosensor deployments in general. Many of these featureshave made collecting building energy data a non-trivialtask. Previous literature [45, 17, 15] identifies these keydifferences, some of which we highlight as follows:

• Ugly retrofitting: While monitoring energy con-sumption is important, the occupants are also con-cerned about the building aesthetics.

• Hostility of building environments: In gen-eral, building environments are less hostile whencompared to external world deployments. How-ever, accidental wire snapping and sensor mishan-dling are common.

• Buildings may become inaccessible after in-stallation: Since people work inside office build-ings and stay in residential buildings, buildings of-ten become inaccessible as the occupants resumenormal life once the installation phase is complete.This makes debugging and iterative research diffi-cult in the context of buildings.

• Occupant interaction drops with time: Afterthe initial interest, occupants usually get busy intheir routine lives and forget to provide regularfeedback pertinent to the deployment.

• Wireless spectrum may be clogged: Manyuser appliances and gadgets operate in the samefrequency as some of the deployed sensors. Thisis due to the limited availability of license-free fre-quency bands3 (which also vary across differentcountries). Limited frequency bands results in mostsensors being designed to communicate in 2.4 GHzband, which is also used in household WiFi routers.This can cause interference and reduced user expe-rience for the occupants and sensor data loss [17].

Sensor deployment architecture for building energymonitoring typically involve end sensor nodes commu-nicating with a central controller. This controller isprimarily responsible for data aggregation and sensornode control. We now discuss the communication andautomation protocol standards used for monitoring andcontrol in building energy domain.

3Information about license-free bands in India: http://www.wpc.dot.gov.in/faq.asp

Meter

CircuitBreaker

Appliances

Cost andAccuracy

Ease ofDeployment

Figure 7: Cost-accuracy optimization in obtaining build-ing appliance level data

4.1 Communication and automation protocolsOver the past few years, many specialized interfaces forcommunication and automation have been standardizedfor building energy monitoring and control. Firstly, wediscuss about building automation protocols. BuildingAutomation and Control Network (BACnet)4 is a non-proprietary protocol used for automating and control-ling HVAC, lighting, security and several other buildingsystems. BACnet fundamentally consists of collectionof objects called properties (an object is an informationunits such as temperature reading) and services for de-vice discovery and accessing and setting these proper-ties. It provides various data and physical links suchas Ethernet and RS-232 [82]. The Modbus protocol de-veloped by Modicon Inc. in the 1970s is widely usedfor building automation. Like BACnet, Modbus is alsoan open standard. Modbus provides client-server com-munication via its messaging structure. Recently, Mod-bus/TCP has been developed allowing Modbus com-munication over TCP/IP networks, in addition to al-lowing for serial communication [82]. Similarly, Lon-Work and LonTalk are proprietary protocols developedby Echelon corporation. Extensive comparison of thesecommunication standards has been well studied in theliterature [54, 31, 82].

The term home automation is often used in the con-text of residential buildings to discuss building automa-tion. Residential buildings usually have simpler electricwiring and smaller physical area in comparison to thecommercial buildings. Due to these two reasons, severalhome automation protocols exploit existing power lineinfrastructure which work well over small distance elec-trical circuit (such as in homes), but may work poorlyin commercial buildings. Powerline protocols such asInsteon and X10 have been widely used for home au-tomation despite the low bandwidth and packet colli-sion. However, these protocols were not designed for

4http://bacnet.org

7

Protocol RF band(MHz)

Range(m)

Bitrate(kb/s)

Routing

ZigBee 868/915/2400 10-100 20/40/250 mesh,tree,source

6LoWPAN 868/915/2400 10-100 20/40/250 RPL

ZWave 868/908/2400 30-100 9.6/40/200 source

Insteon 904 45 38.4 simulcast

X10 310/433 100 20 simulcast

Table 2: Comparison of home automation protocols.Mostly referred from [39]

high frequency monitoring and need to be extended.Irwin et al. [49] work around these shortcomings by cre-ating smart polling strategies to maximize energy mea-sured with least packet drop in their energy monitoringsystem based on Insteon.

In the other end of the spectrum, several protocols op-erating on wireless nodes are commonly used. Zigbee5

and its close variant 802.15.4 present networking stackssimilar to the layered internet protocol suite. Zigbeeis intended to be low power and some of its key fea-tures include AES-128 security, multiple star topologiesand personal area networks (PAN) and device discoveryamong others. Similarly, ZWave supports mesh topol-ogy and is commonly used in home automation.

The internet of things (IoT) revolution has led to adirected effort on IP based sensor data monitoring. Thisallows physical sensors to be exposed as HTTP servicesallowing an integration with the ubiquitous internet andsimultaneously allowing modern web standards basedrich applications. The IoT revolution has also led to thepopularity of constrained application protocol (CoAP)and IPv6 over Low power Wireless Personal Area Net-works(6LoWPAN). CoAP implements the HTTP RESTscheme without much of HTTP’s complexities and sup-ports only datagrams in the transport layer. 6LoWPANdefines header and encapsulation format for commu-nication with 802.15.4 networks. Table 2 summarizescomparison of different home automation protocols.

4.2 Criteria for optimalityHaving discussed the important differences in buildingenergy domain from the generalized sensor network do-main, we discuss the issue of optimal instrumentationand how these differences are to be incorporated. Wepick up electricity monitoring in a residential home todiscuss the criteria for optimal instrumentation. As dis-cussed previously in Section 3, electrical wiring in build-ings create a natural hierarchy across different spatialresolutions. Figure 7 shows these different spatial lev-els at which building energy data for a residential homemay be obtained. As we go from the meter to the cir-

5https://www.zigbee.org/Specifications/ZigBee/Overview.aspx

cuits and to the appliances, cost of instrumentation andmaintenance increases. However, going this route fromthe meter to the appliances also adds significant infor-mation at each level. The optimal spatial resolution forinstrumentation must be decided in accordance to theintended application.

For the same application of electricity monitoring,sensors with different sampling frequencies exist. As thesampling frequency increases, so does the cost of the in-strumentation and consequently the cost of maintainingmore data. For some applications such as energy dash-boards [4], low sample data might suffice, whereas forapplications such as high frequency NILM analysis [40],one may need to sense more than a thousand samplesevery second.

4.3 Challenges and OpportunitiesBesides being peculiar in certain ways, building en-

ergy sensing also brings in a peculiar set of challengesand opportunities, which we discuss as follows:

• Indirect sensing: Owing to the challenges inmaintaining and scaling sensor deployments, re-cent research [98] proposed using existing sensorinfrastructure such as Wi-Fi access point coupledwith soft sensors such as office messenger for en-ergy management of green buildings. Such re-search highlights the need to look into indirectlymonitoring the desired parameter via the use ofside channel information or information from otherexisting context sources, which can help avoid newinstrumentation. On the same lines, Kim et al.proposed ViridiScope [58], a system which uses in-direct sensing modalities such as magnetic, lightand acoustic for power measurement of differentappliances. Similarly, Schoofs et al. propose An-not [91], a system to automatically annotate elec-tricity data using wireless sensor networks. Simi-larly, Barker et al. [12] show the relationship be-tween a refrigerator’s power consumption and in-ternal temperature. They key opportunity lies inunderstanding the causal relationship in data mea-sured by different sensors to optimally sense a phys-ical phenomena. However, this remains a challeng-ing task owing to the per indirect sensor calibra-tion required [58].

• Remote configuration: Buildings usually be-come physically inaccessible to researchers afterthe initial deployment phase of their study. Asa result debugging becomes hard. Previous re-search [14] has suggested that reconfiguration ofsensor nodes should be preferred over reprogram-ming. Previous work [71] has also mitigated thisproblem via remote debugging over IP. This chal-lenge can be approached from various angles. On

8

one front, human computer interaction (HCI) re-search can help identify mechanisms of allowinghome occupants to easily configure their deploy-ment (which is considered a highly technical taskcurrently). From another angle, this problem hasalso been mitigated to a certain extent by using anIP backbone for all data communication. Anotherapproach could involve distributed debugging andconfiguration often used in swarm based robotics.

• Unreliable network: Previous studies [17, 45]have highlighted unreliable network as one of theissues in building sensor deployments. Local buffer-ing on the controller or the sensor (such as on de-vice EEPROM) have been used as quick fix solu-tions to these issues. The opportunity here is tore-think about new architectures that can mitigatedata loss due to unreliable network.

• Optimal sensor placement: Finding the opti-mal number of sensor and their placement is a fun-damental Wireless Sensor Network (WSN) task.While we took an application driven discussionon optimality criterion earlier, this problem canalso be studied theoretically. This problem maybe viewed as a spatial optimization problem overthe number of sensors, communication costs andsensor information. Krause et al. [63] presented aGaussian Processes based non-parametric proba-bilistic model for spatially interesting phenomenaand connection links. The key opportunity lies inreducing the disjoint between application drivenempirical and theoretical research that is currentlycommonplace in the building energy space.

5. INTERCONNECT SUBSYSTEMSBuildings consist of multiple sub-systems such as net-working, utility (electricity, water and gas) and secu-rity. Traditionally, many of these sub-systems are opti-mized for operation in isolation. However, since build-ings are a unified ecosystem, optimal operations acrossthe building would require understanding the relation-ships amongst these sub-systems and interconnectingthem. Previous work has shown that the sum of in-formation obtained from each of these sub-system isgreater than each one of them individually considered.We now discuss subsystem interconnection for occu-pancy based lighting and HVAC control.

As we shall discuss in Section 7, occupancy levels havea significant impact on building energy consumption.Previous studied have looked into occupancy inferenceusing a variety of system interconnections. For instance,Thanayankizil et al. [98] fuse information from network-ing equipment such as Wi-Fi access points with Officechat and meeting software for detecting occupancy foroptimizing HVAC operations. On the same lines of us-

Laptop activity

Employee badge

Network Chat client

Online calendar

Motion sensors

Door contact sensors

Electricitymeter

Firewall

Detect occupancy to control HVAC and lighting

Lighting

HVAC

Figure 8: Different systems (top: using existing infras-tructure and soft sensors [98], middle: using motion anddoor sensors [2] and bottom: using electricity metersand network firewall [57]) interconnected to deduce oc-cupancy for optimized lighting and HVAC control.

ing existing infrastructure, Kim et al. [57] interconnectelectricity data with firewall traffic to detect occupancy.Lu et al. [68] and Scott et al. [92] obtain occupancy pat-terns from Passive infrared sensors (PIR) and door sen-sors to control the home thermostat to conserve electric-ity in the absence of occupants. Similarly, Agarwal etal. [2] and Batra et al. [15] use motion and door sensorsfor occupancy detection in office settings. On similarlines, Delaney et al. [64] discuss interconnecting motionsensor, light sensors and lighting control circuit for effi-cient lighting controls accounting for the availability ofsunlight and occupant presence. Figure 8 shows variousinterconnected systems to detect occupancy for lightingand HVAC control.

Apart from subsystem interconnection for occupancybased control, there exists active interest in using thenow ubiquitous smartphones for various energy appli-cations, which we do not cover here in interest of space.Another active opportunity exists in exploiting the in-ter relationship between water and energy sub-systemsin a building (discussed in Section 6).

Interconnection of various sub-systems has been madepossible due to the fact that many systems can exposetheir data over IP, as discussed in Section 4. Further-more, various middleware systems [8, 3, 1] have beenproposed in the recent past which greatly aid in col-lecting data from diverse sensors by writing wrappersand exposing the data in open formats. Many suchmiddleware systems aim to expose buildings as a uni-fied system of individual sub-systems, whereby, end userapplications can be developed exploiting multiple sub-systems.

5.1 Challenges and opportunitiesWe now discuss some challenges and opportunities

pertinent to interconnecting building sub-systems.

• Complexity of writing applications: Build-ings are not currently programmable in a mean-ingful sense because each building is constructed

9

with vertically integrated, closed subsystems andwithout uniform abstractions to write applicationsagainst [28, 100]. Further, applications developedfor one building are often not portable to otherbuildings due the vast differences in buildings andthe underlying sensors and control systems. Tomitigate this, Krioukov et al. propose BAS [65], abuilding application stack designed specifically toease writing portable building applications whichare agnostic of underlying hardware and the com-munication interfaces (such as Modbus, BACnet).They introduce software engineering principles tothe building domain and allow applications to bebuilt on top of a rich query interface. Haggertyet al. discuss Building operating system services(BOSS) [28] which can leverage BAS query lan-guage and other features to develop portable fault-tolerant applications. While the challenge of ap-plication portability has been addressed to someextent via BAS and BOSS, it remains an open op-portunity to unify existing building energy appli-cations to adhere to such stacked designs. Addi-tionally, it must be highlighted that similar to gen-eral purpose software applications, this software-oriented view of buildings is prone to low qualityand ad-hoc methodologies calling for developmentof building application quality metrics.

• Proprietary communication protocols: Thetask of writing hardware agnostic applications, asdiscussed above, is made complex due to severalof these devices communicating using proprietaryprotocols. For each such end point, a custom wrap-per has to be written on an end device which canaccumulate the data and make it available in anopen data format (or allow for control interface us-ing common interface formats). Different buildingsub-systems may employ different vendor lockedsolutions making interconnection even more diffi-cult. This calls for the development of open com-munication standards enabling decoupled network-ing stacks. Often this job is simplified by introduc-ing a gateway device that converts the proprietaryinterface (such as Zwave) into IP interface. How-ever, this adds up to the overall cost and mainte-nance. As previously discussed in Section 4, theshift towards IP based communication can miti-gate this problem.

• Inter department communication gap: Whiledifferent sub-systems may employ different ven-dor locked solutions for their custom application,a greater challenge in interconnecting sub-systemsis that of the communication gap existing betweendifferent departments. The involvement of humansin this task makes it inherently difficult. While

non technical in nature, this concern can limit op-timal interconnections since individuals often havein-depth understanding about their own domainwhile interconnections would require understand-ing across different domains. As an example, HVACsystems typically are taught to mechanical engi-neering experts while the expertise of software stacksfor data communication and storage often remainswith computer science experts.

• Unstructured data: Many of the sub-systemsdiscussed above are managed by field experts. Sig-nificantly important data collected from these ex-perts such as CAD layouts, notes from the respec-tive departments need to be transformed into use-ful metadata. In the past, Krioukov et al. [65]have used image processing techniques to extractchillers and other building sub-systems from a CADdrawing. In general, this exercise is both non-trivial and time consuming as the data being un-structured does not adhere to any fixed schema.To mitigate this requires establishment of schemasfor buildings across these sub-systems, some ofwhich have been studied in the past.6

• Data and control access: By interconnectingsub-systems, an application built on top can lever-age data access and control across these sub-systems.However, each sub-system may need to comply toits specifications for data access and control. Thischallenge can be mitigated via establishment ofrules for data and control access [8, 28] for differentapplications and sub-systems. While theoreticallyfeasible, it remains a complicated challenge to ad-dress, in context of the above stated challenges.

6. INFERRED DECISION MAKINGTraditionally, data from within the system has beenused for simple decision making and rule of the thumbapproaches. However, once the systems are intercon-nected, there is a deluge of data coming from diversesystems. This data can be used to improve upon thedecision making for optimal building operations. In-ferred decision making can help in identifying inefficien-cies, raising alerts and suggesting optimizations. Forexample, earlier lighting control was performed in anautomated way based on a fixed motion sensor time-out. However, this does not incorporate ambient lightlevels and the motion sensor timeout is static, as de-cided at the time of audit or installation. Based oninterconnecting light sensor data with finer grainer oc-cupancy data, previous systems [30] have discussed pos-sible energy savings by inferring optimal lighting lev-els. In Table 3 we summarize several similar optimiza-6One such example is gbXML http://www.gbxml.org/aboutgbxml.php

10

tions made possible by inferred decision making acrosscommon building energy applications. We now discussin detail an upcoming opportunity in building energywhich is facilitated by the interconnection between en-ergy an water.

6.1 Water-Energy nexusWater and energy are two critical indicators of sustain-ability in an organization [52]. While we have largelydiscussed building energy, the inter-relationship betweenenergy and water prompts a detailed discussion. Thewater-energy nexus7 is defined as the relationship be-tween energy and water encompassing the following twodimensions: i) energy usage in water treatment ii) waterusage for energy generation. In the context of buildings,the water-energy nexus is an important topic encom-passing inter-disciplinary research and has been attract-ing attention in the research community as evidenced byworkshops such as the ACM Buildsys. We now discussthe water-energy nexus in two different settings: 1)largeindustrial campus; 2)residential apartment.

Kadengal et al. [52] study the inter-relationships be-tween energy and water inside a campus and answer thefollowing questions based on the collected data: i) Whatare the trade-offs between the external water footprintof a campus and its internal energy footprint of wa-ter? ii) Are improvements in either footprint realizablein practice? iii) Does reducing the consumption of onewater grade have more impact on the energy consump-tion than other water grades? and iv) Does rainwaterharvesting help reduce a facility’s energy footprint.

In residential settings, previous research [59] has re-marked on the fact that simultaneously observing boththe electricity trace and the water trace can reveal finerinformation than when each of the system is analyzed inisolation. Figure 9 shows the water pump rate from ahome in Delhi, India [17]. Due to low water pressureand non-availability 24X7, water motors are used topump water to an overhead tank. This figure capturesthe motor’s energy-water relationship. While turningon the water motor increases the pumping rate from 1l/min to 20 l/min, it incurs an additional expense of 1horsepower. In a different application, Prodhan et al.discuss HotWaterDJ [83] which is a system for smartwater heating reducing the energy consumption of wa-ter heaters.

6.2 Inference approach categorizationProgressive research in the field of data science andbuilding energy has led to newer inference approachesbeing developed to cater building energy applications.Comparison of these approaches is an application spe-cific task. However, certain categorizations need tobe studied for a fair comparison of these inference ap-

7http://en.wikipedia.org/wiki/Water-energy_nexus

Applica-tion

Without inferreddecisions

With inferred decisions

Lightingcontrol

Turn off lights aftera threshold intervalin which no motionhas been detected

Fine control of light based onamount of ambient light anddetailed occupancy informa-tion for light level control tomatch optimal conditions

HVACscheduling

Fixed 9 AM to 5 PM Scheduling based on occu-pancy

Utilitypoweroutagedetection

Rely on phone callsfrom distressed con-sumers several min-utes after an outage

Observe realtime data fromsmart meters and deploy cor-rective actions

HVACchillertempera-ture setpoint

Fix a setpoint as perone time audit

Correlate HVAC consump-tion with ambient tempera-ture and working conditions,season and occupancy levels

Utilityconsumercom-plaintsfor highpricing

Customers usedto call the util-ity company andcomplain regardingextorbitant bills

Customers can view theirconsumption in realtime andset targets exceeding whichthey can be easily alerted

Table 3: Common building energy applications whichcan be optimized by inferring decisions

06:30:00

06:31:00

06:32:00

06:33:00

06:34:00

06:35:00

06:36:00

06:37:00

06:38:00

06:39:00

0

100

200

300

400

500

600

700

800

Mot

orPo

wer

(W)

Wat

er1

leve

nts

Figure 9: Observing the energy water nexus in a residen-tial apartment. The water motor increases the pumpingrate at the cost of 1 horsepower

11

proaches. While this categorization is true across datascience in general, we present a summary to understandthe design space of inference algorithms in the contextof building energy.

• Centralized vs Distributed: In centralized ap-proaches the data resides and is processed on asingle machine. The data deluge calls for a needto dwell into distributed systems where efficient in-ference and learning can be done via the use of dis-tributed algorithms performed on data stored ondifferent machines/nodes. We illustrate the differ-ence between these two approaches via the prob-lem of appliance fault detection. Traditionally, forsolving this problem, sensor nodes measure variouspower features (e.g. active and reactive compo-nents) at the outlet level and relay them to a cen-tral controller. The central controller maintains amapping between a node and the appliance it isconnected to. The fault detection algorithm run-ning at the central controller detects faults basedon the data relayed by different nodes. However,since fault in an appliance is almost always inde-pendent of faults in other appliances, this problemhas an inherent distributed characteristic. Ganu etal. [36] take a distributed approach for this prob-lem, whereby, each sensor node decides whetherthe connected appliance is faulty or not. This ap-proach does not require data transmission to a cen-tral controller and can be also exploited for localcontrol whereby the sensor node can individuallyindicate faulty operation.

• Supervised vs Unsupervised: Supervised learn-ing involves collection of ground truth/labeled datafor training. As discussed in Section 4, this canbe expensive and non-trivial. Unsupervised ap-proaches where learning does not involve labeleddata and labels are discovered as opposed to ex-plicitly being specified are more attractive. To il-lustrate the difference between the two categories,we consider the task of finding fixtures (both elec-tricity and water) in a home. Typical supervisedapproaches involve instrumenting each fixture witha sensor for collecting labeled events (for exam-ple [95, 58]). The training data is then used forlearning classifiers. On the other hand, Srinivasanet al. [96] use smart meter data along with datafrom already existing security systems for unsu-pervised event clustering and labeling for fixturediscovery.

• Online vs Offline: Another categorization of in-ference mechanisms which is widely prevalent inrecommender systems and search engines is on-line versus offline models. Unlike offline models,online models can adapt to new data on the fly

without the need to relearn the entire model fromscratch. As the size of data increases, online mod-els become more relevant. However, often onlinealgorithms can not have a long term vision builtusing offline collected data and hence may need totake locally optimal decisions that may be globallynon-optimal.

We believe that the optimal category of inference andlearning approaches should be online, distributed andunsupervised algorithms. This is due to the challengeswhich exist in inferred decision making which we high-light now.

6.3 Challenges and opportunities

• Collection of ground truth: For quantifyingimprovements or effect of various proposed strate-gies, one needs to collect ground truth data. Fur-thermore, every new building on which evaluationis to be performed requires instrumentation. Dueto this, scalability of the application is greatly hin-dered. The opportunity lies in developing easy toconfigurable and extensible building simulators (orinterfaces to existing ones. Recently, Bernal et al.presented MLE+ [74]) which leverages the high-fidelity building simulation capabilities of Ener-gyPlus and the scientific computation and designcapabilities of Matlab for controller design. Thepresence of such simulators can also mitigate thechallenge previously faced in the research commu-nity when two different strategies were performedunder different settings and could not thus be di-rectly compared.

• Need for online and tractable algorithms:Computationally intractable algorithms for per-forming inference and learning scale poorly withthe size of the data. As discussed in Section 3,data size in building energy domain is increasingrapidly. Thus, the traditional offline intractablealgorithms need to be complemented with theirtractable and online equivalents for timely action-able insights.

• Testing in real environments: Even thougheffective simulators may be developed, testing thevalidity of the inference produced in the real worldis hard to realize. This is due to the complicationsarising out of closing the loop requiring interfacewith the existing control system.

7. INVOLVE OCCUPANTSModern buildings are designed to provide occupant com-fort, security and energy efficiency, among others. Ow-ing to these design goals, buildings form a complex

12

Human Building

Computer

Sensing

ControlFeedback

Visualiza

tion

Load

Comfort

Figure 10: Human-Building-Computer interaction(HBCI)-architecture to bridge the gap between the dig-ital and physical worlds [Figure adapted from [48]]

ecosystem consisting of physical, cyber and human com-ponents and their interactions [89]. Occupants form anintegral part of these energy efficient buildings ecosys-tem and their presence makes this ecosystem even morecomplex as occupants have individual preferences. Hsuet al. [48] coin the term Human-Building-Computer-Interaction (HBCI)] to describe the inter-relationshipsbetween human beings, buildings and computers. Fig-ure 10 shows the HBCI as proposed by the authors.

Traditional systems missed out on one or more com-ponent in the HBCI framework while trying to aim foroptimal environment. The occupants and their prefer-ences were mostly not taken into account and the de-cisions for energy efficiency were taken centrally. How-ever, the HBCI framework helps in formalizing the opti-mization problem involving human comfort and energysavings. Occupants’ behavior plays a significant factorin this optimization problem. Previous literature [75]suggests that energy unaware behavior can add uptoone-third to a building’s energy performance. On theother hand, Darby et al. [27] suggest that providing theoccupants feedback on energy usage can provide up to15% savings. Thus, both occupant comfort and energyefficiency can be considered as joint goals of a smartbuilding and an optimal sweet point balancing theirtrade offs is desired.

7.1 HBCIWe now discuss in detail HBCI and how the different

interactions relate with energy efficiency and the otherIs. As per Figure 10, there are four primary occupantinteractions in the HBCI system which are as follows:

1. Occupants provide feedback for better com-putation: Building occupants can provide valu-able information for improving efficiency (feedbackin Figure 10) in a variety of ways. Previous stud-ies have instrumented human bodies for a variety

of applications such as healthcare [51, 77]. How-ever, human wearable sensors and the collecteddata thereof, are highly intrusive and are not in-cluded in the scope of this paper. Building oc-cupants can also indirectly provide their locationin the building. Several studies in the past haveused cell phone data to localize an occupant inthe building [10, 97, 76]. This can greatly help inactivity recognition, culminating in accurate en-ergy apportionment [44] and in several other en-ergy saving measures such as modifying HVAC pa-rameters to account for occupancy and turning offlights where no one is present [32]. This interac-tion between occupants and the computing worldrs can be seen as an extension of optimal instru-mentation, where the occupants and their pos-sessions (such as cell phones) can themselves actas sensors reducing the necessity for dense sensingdeployments.

2. Computation to provide feedback and sug-gestions to occupants: Personalized recommen-dation systems have been well studied in the en-tertainment sector. A similar opportunity lies inthe building energy domain as well. Energy dash-board [4] systems have been proposed in the pastto make the users aware of their energy footprint.Providing personalized energy consumption infor-mation has also been made possible via energyapportionment strategies [44, 57]. It is believedthat providing occupants with detailed feedbackon their consumption can help reduce their energyfootprint by induced behavioral changes [24, 27].Recently, Batra et al. [17] and Ganu et al. [36] dis-cuss systems for appliance fault detection. Whenappliance faults are detected, it is upto the occu-pants to take proper action when informed aboutpossible optimizations.

3. Occupants load buildings: Occupants occupythe physical space in the building and utilize var-ious resources. Several occupant actions usuallytranslate into energy consumption inside buildings.For instance, in a residential building, occupantsmay use different electrical appliances such as tele-vision or oven. In commercial buildings, occupantsmay interact with servers and other resources. Theoccupancy on the building often directly relateswith the energy consumption of the building. De-cisions may be inferred to decide the optimal oper-ating conditions in a building based on occupancyload. Such schemes ensure that occupant comfortis not compromised and at the same time energyefficient measures are considered.

4. Expecting comfort from building: As discussedearlier in this section, control actions were previ-

13

ously taken mostly at a central facility level andwere totally agnostic of individual preferences. How-ever, by providing feedback to the building man-agement system about their preference, individu-als can better control their environment translat-ing to both higher comfort and increased energyefficiency. Previous research (personalized light-ing [64] and participatory HVAC control [33]) hasproposed systems for participatory and individualHVAC and lighting control.

7.2 Challenges and opportunitiesWe now discuss some challenges and opportunities in

involving occupants in the context of building energy.

• Privacy concerns: Empowering the occupantswith more information about their activities of dailylife and consequently impact on energy consump-tion has significant privacy concerns. This maygreatly reduce the involvement of occupants whomay feel threatened about the amount of data be-ing collected and the potential information thatcan be inferred. This reaction may be similar tothe reaction of users to internet companies collect-ing a lot of user data for targeted advertisement.Several activities can be easily inferred even froma whole home smart meter. Further, when seen inconjunction with other sensors commonly presentin buildings, much finer grained information canalso be revealed. Molina et al. [72] address thischallenge via a privacy enhancing scheme for pre-serving metering goals of a smart meter while pro-tecting occupant information. Thus, there is aneed to develop privacy preserving architecturesto allow meeting the goal of building energy moni-toring while still maintaining occupant privacy [8].

• Indifferent occupant attitude: Previous research[57] suggests that if the occupants do not directlypay for their electricity bills, such as in commer-cial offices and academic institutes, they may notbe enthusiastic towards electricity savings. Undersuch circumstances, indifferent occupant attituderemains a big challenge and needs to be accountedfor in future work. It is also possible that occu-pants may initially show interest and may provideregular feedback for improving the system. How-ever, with time the users may fall back to theirusual schedules and thus, sustained occupant en-gagement remains an open challenge. Novel hu-man computer interaction (HCI) [81] techniquescan be used for better occupant engagement.

8. INTELLIGENT OPERATIONSAll of optimal sensing from interconnected subsystems,resulting in decisions inferred from the rich dataset while

involving the occupants will result in intelligent opera-tions. Such intelligent operations already exist but missout one or more aspects discussed previously i.e. ei-ther they do not involve the occupants or are takenat a sub-system level without accounting for data fromother sub-systems in operation. Realizing energy effi-ciency requires striking the perfect match between thefive Is. For instance, a system having advanced controlcapabilities but poor inference capabilities is unlikelyto result in energy efficiency. Likewise, a system havingadvanced inference capabilities, but poor control capa-bilities will only be able to propose efficiency and notrealize it.

We base our discussion in this section by discussing asimplified view of a smart grid (which excludes the roleof government and regulators) and the role of build-ings in it. Our simplified smart grid setup has threekey activities: 1) generation 2) transmission and 3)consumption. Of these three activities, buildings weretraditionally involved only in electricity consumption.However, buildings are increasingly equipped with gen-eration capabilities through renewable resources (e.g.solar). Since the renewables involve generation that istypically weather dependent and hence is time vary-ing, such sources bring forth an interesting optimizationperspective - if the production is more than what canbe consumed, how best to utilize the remaining excessgenerated power. These topics are usually studied un-der the umbrella field of automated demand responsewhich we briefly discuss now.

Demand response is defined as - “Changes in elec-tric usage by end-use customers from their normal con-sumption patterns in response to changes in the priceof electricity over time, or to incentive payments de-signed to induce lower electricity use at times of highwholesale market prices or when system reliability isjeopardized” [11]. While generally energy efficiency isconsidered to be closely related with energy conserva-tion, studying demand response allows us to presentan alternative understanding. We had discussed ear-lier in Section 3 that commercial buildings are billed ontime-of-day based pricing. This is due to the fact thatover and above the average load, the utilities have toexpend a lot of resources and money to cater to peakdemand. Thus, in order to encourage consistent loadprofiles (with minimum time variation), the utilities in-troduce variable pricing- non peak hours have cheaperelectricity rates as compared to peak hours. Buildingenergy research community and the industry has lookedinto this problem and has come up with several waysof peak demand flattening. We present two such ap-proaches as follows.

Firstly, we discuss SmartCap [13], a scheme basedon the well known scheduling algorithms used in realtime operating systems. The authors discuss dividing

14

electrical loads into two categories- background loadswhich can run without human intervention (such as re-frigerator, air conditioner) and interactive loads (suchas television, stove, microwave, lighting). Since schedul-ing interactive loads may interfere with daily life of theoccupants, only background loads are rescheduled toalleviate peak demand. Each load is modeled with asimple slack, or time required for it to complete its op-erations (e.g. for a refrigerator the operation goal maybe to reach the set temperature within the next hour).

In contrast, companies such as Stem have proposedbattery storage schemes for reducing peak consumption.The battery is charged from the grid during the off-peakperiod and is discharged to lower the demand from thegrid during peak period, thereby, reducing the overallpeak demand and flattening the power draw. The com-pany claims to reduce up to 20% of the overall bills bythis storage based mechanism to lower peak demand.Integration of local generation via renewables (e.g. so-lar) with such batteries remains an interesting open pos-sibility, which becomes more difficult due to the timevarying nature of electricity produce.

While theoretically such operations seem feasible, theyrequire an elaborate control system setup. In the caseof SmartCap like approaches, a control system needsto interact with the inference engine which decides theslack for each appliance. Correspondingly, the oper-ating condition for each appliance has to be decided.Subsequently, the control also has to abide by certainlimits such as limited the rate of appliance state tog-gling to prevent damage. In the second case (Stem),where additional batteries are provisioned, the controlsystem must have the capability of efficiently changingthe power supply from the grid to the battery and viceversa. Due to the existence of such complications andrisks, research systems in the past have usually onlypresented proof of concept intelligent operations as op-posed to realizing these in the real world.

8.1 Challenges and opportunitiesWe now discuss some of the challenges and opportu-

nities in performing intelligent operations.

• Significant up front costs: To realize the fullpotential of such intelligent operations, there maybe significant up front cost. In the case of bat-tery based solutions, one needs to invest heavilyin managing additional equipment. In case of ap-proaches such as SmartCap one needs to add con-trol capability to existing monitoring equipment.Additionally, each appliance must be individuallyinstrumented.

• Security breach: To facilitate schemes such asSmartCap, appliance owners can take a step to-wards making their appliances more IoT ready.

Appliances can allow their configuration over stan-dard HTTP based interfaces. However, taking astep towards controlling appliances for peak de-mand and especially more so over standard inter-faces like HTTP, exposes them to cyber attacks.Since most of the end devices usually have lowcomputing and memory capabilities, they are lim-ited in terms of computational capabilities neededfor advanced encryption systems. Thus, allowingintelligent operations while maintaining securityaspects in resource constrained end device remainsa challenge to be addressed in the future.

• Complex control environment: Controlling op-erations of a building encompasses a variety of sys-tems needing highly sophisticated controls. Thus,introducing optimizations into such a complex sys-tem is highly risky. While simulations may be ef-fectively used, the real world may bring up unfore-seen challenges. Moreover, the controls which maywork well in one building, may miserably fail inanother. An analogous example may be the crash-ing of Ariane V spacecraft 8. The spacecraft usedthe software from its previous version, but had up-graded hardware. The faster velocity attained bythe new hardware caused numerical overflow andthe control system of the spacecraft failed caus-ing auto destruction. While the empirical researchmay promise exciting results, it could miss one ormore cases which could cause a disaster similar tothat of Ariane V. This calls for the development oftheoretical proofs of valid and risk averse controland operation and their integration into existingsimulators.

9. APPLICATIONSHaving discussed each of the 5 Is in detail, we nowpresent a summary study of recent applications in thebuilding energy domain in this section. Most of theseapplications span across the 5 Is as discussed above andare summarized in Table 4.

We specifically discuss in detail about the well studiedproblem of non-intrusive load monitoring (NILM) anddiscuss how it spans across the 5 Is.

9.1 Non-intrusive load monitoringNon-intrusive load monitoring (NILM) or energy disag-gregation is the process of breaking down the energymeasurement observed at a single point of sensing intoconstituent loads [43]. Figure 11 shows disaggregatedconsumption across a day for different appliances fromthe iAWE [17] data set. In 2006, Darby et al. [27] sug-gested that providing detailed electricity informationfeedback to end users can lead to 5-15% electricity usage

8http://www.around.com/ariane.html

15

00:00:00 03:00:00 06:00:00 09:00:00 12:00:00 15:00:00 18:00:00 21:00:00Time

0

1000

2000

3000

4000

5000

Pow

er(W

)

air conditionerair conditionerentertainment unitfridgelaptop computerwater filterunaccounted

Figure 11: Disaggregated consumption [Best viewed in color]

reduction by behavioral change. Recently, Chakravartyet al. [24] performed a study across more than 300 usersin California and observed mean reduction of approx.15% when disaggregated information and real time elec-tricity information is provided to end users.

With the advent of smart metering infrastructure, asdiscussed in Section 1, NILM has had a lot of renewedinterest. A number of startups9 which provide itemizedelectrical usage to households over cloud based serviceshave emerged recently. The recent interest has also ledto the release of data sets meant to aid NILM research.Of these REDD [62], BLUED [7], AMPds [70], UK-DALE [56], ECO [20], Sust-data [80], GREEND [73],COMBED [19], BERDS [69] and Pecan [46] have beenspecifically released for NILM like applications. Apartfrom these data sets, other data sets such as Trace-base [85], HES [102], SMART* [12] and iAWE [17] canbe used for several other building applications, includ-ing NILM. We now discuss the NILM problem acrossthe five Is.Instrument optimally: NILM involves breaking downthe aggregate meter data measured at a single point intoconstituent appliances or sub-meters lower in the elec-trical tree. Thus, instrumentation for NILM involvesmeasuring electrical parameters at both the aggregateand the sub-metered level (desired for ground truth).The aggregate power readings are typically measuredusing smart meters. In special cases, when high fre-

9Such as Bidgely, Neurio, Plotwatt

quency data (more than 1 KHz) is required, sophisti-cated data acquisition systems (DAQs) are used. Sub-metered power data is typically measured either at thecircuit level using current transformer (CT) based sen-sors or at appliance level using appliance sensors.Interconnect sub-systems: While in the classic NILMproblem, interconnection across different data streamsis not studied, some studies use the extra informationfrom other modalities to improve disaggregation. Pre-vious work [86, 55] indicate the correlation between en-ergy usage and external temperature. Berges et al. [22]correlate occupancy sensors with electrical data to iden-tify potential savings in unoccupied rooms. Similarly,other ambient sensors can also be interconnected withelectrical sensors to obtain additional insights into thedisaggregation problem [96].Inferred decision making: The vanilla use case ofthe NILM implementation provides an itemized break-down of the electrical load. This inference problem canbe viewed as an inverse classification problem. Previ-ous work has related this problem to source separationwhich is a well studied problem in sound processing.Hart et al. [43], in their seminal work on NILM, pro-posed a simplistic combinatorial based and a simpleedge detection based NILM algorithm. Both these algo-rithms form the foundation behind many of the state-of-the-art algorithms. Markovian analogues of combina-torial optimization formulation led to factorial hiddenMarkov [37] among other hidden Markov models. Sev-

16

eral NILM approaches have been proposed in the recentpast [79, 50, 84, 16, 61, 6] and a rich overview has beencaptured in several recent work as well [103, 23, 101,18]. It must be pointed that most of the prior researchincludes supervised methods in a centralized offlinesetup.Involve occupants: As discussed earlier in this sec-tion, providing itemized feedback to end users has shownto reduce their end consumption. Thus, the occupantscan be involved by not only providing them with item-ized billing, but also providing actionable suggestionson top. Another interesting occupant involvement mayarise from devising novel techniques for ground truthcollection, which eliminates the need for appliance me-tering.Intelligent operations: Real time control actions tosave electricity based on disaggregated information re-quire complicated interactions with the control systems.Current literature is thin on the aspect of automatedcontrol beyond the usual feedback that NILM systemsprovide.

9.1.1 Challenges and opportunitiesIn this section we briefly mention some of the chal-

lenges and opportunities in NILM research.

• Need for extensive deployment: For certifyingan algorithm on a previously unseen home, appli-ance level data must be collected from that home.

• Computationally expensive approaches: Manyapproaches are computationally expensive and thusmodel only the high energy consuming appliances.Owing to this, detailed information about low en-ergy consuming appliances is often not made avail-able. The computationally intractable approachesalso limit the application in the real world setting.

• Supervised methods require sub-metered data:Supervised NILM approaches train on the sub-metered data and create a model for each appli-ance. This requires ground truth instrumentation,motivating the need for development of novel un-supervised learning mechanisms.

10. CONCLUSIONS AND FUTURE WORKOptimizing building energy usage remains an area ofconcern in light of dwindling natural resources. Dueto this concern, efforts are being concentrated to de-velop an understanding into buildings. These effortshave led to a deluge of data coming from a variety ofsensors. This availability of data is changing the waywe consume our electricity information. In this paperwe highlighted some of the applications enabled by thisdata availability, such as early power outage detection,peak load reduction, electricity consumption reduction.

Based on our literature survey of research in build-ing energy domain, we identified five crust areas en-abling data centric energy efficient buildings: i) instru-ment optimally ii) interconnect sub-systems iii) inferreddecisions iv) involve occupants and v)intelligent opera-tions. Across each of these five areas we present thestate-of-the-art, the core challenges and the opportuni-ties in the field. Finally, we categorize different build-ing energy applications as per these five Is and discussnon-intrusive load monitoring, a well studied problemin building energy domain, in greater detail.

11. REFERENCES[1] K. Aberer, M. Hauswirth, and A. Salehi. The global sensor

networks middleware for efficient and flexible deployment andinterconnection of sensor networks. Ecole PolytechniqueFdrale de Lausanne (EPFL), Tech. Rep.LSIR-REPORT-2006-006, 2006.

[2] Y. Agarwal, B. Balaji, R. Gupta, J. Lyles, M. Wei, andT. Weng. Occupancy-driven energy management for smartbuilding automation. In Proceedings of the 2nd ACMWorkshop on Embedded Sensing Systems forEnergy-Efficiency in Building, pages 1–6. ACM, 2010.

[3] Y. Agarwal, R. Gupta, D. Komaki, and T. Weng.Buildingdepot: an extensible and distributed architecture forbuilding data storage, access and sharing. In Proceedings ofthe Fourth ACM Workshop on Embedded Sensing Systemsfor Energy-Efficiency in Buildings, pages 64–71. ACM, 2012.

[4] Y. Agarwal, T. Weng, and R. K. Gupta. The energydashboard: improving the visibility of energy consumption ata campus-wide scale. In Proceedings of the First ACMWorkshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 55–60. ACM, 2009.

[5] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, andE. Cayirci. A survey on sensor networks. Communicationsmagazine, IEEE, 40(8):102–114, 2002.

[6] K. Anderson, M. Berges, A. Ocneanu, D. Benitez, andJ. Moura. Event detection for non intrusive load monitoring.In Proceedings of 38th Annual Conference on IEEEIndustrial Electronics Society, pages 3312–3317, 2012.

[7] K. Anderson, A. Ocneanu, D. Benitez, D. Carlson, A. Rowe,and M. Berges. Blued: A fully labeled public dataset forevent-based non-intrusive load monitoring research. InProceedings of 2nd KDD Workshop on Data MiningApplications in Sustainability, pages 12–16, Beijing, China,2012.

[8] P. Arjunan, N. Batra, H. Choi, A. Singh, P. Singh, and M. B.Srivastava. Sensoract: a privacy and security aware federatedmiddleware for building management. In Proceedings of theFourth ACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 80–87. ACM, 2012.

[9] Autogrid. The energy data deluge, 2013.[10] P. Bahl and V. N. Padmanabhan. Radar: An in-building

rf-based user location and tracking system. In INFOCOM2000. Nineteenth Annual Joint Conference of the IEEEComputer and Communications Societies. Proceedings.IEEE, volume 2, pages 775–784. Ieee, 2000.

[11] V. M. Balijepalli, V. Pradhan, S. Khaparde, and R. Shereef.Review of demand response under smart grid paradigm. InInnovative Smart Grid Technologies-India (ISGT India),2011 IEEE PES, pages 236–243. IEEE, 2011.

[12] S. Barker, A. Mishra, D. Irwin, E. Cecchet, P. Shenoy, andJ. Albrecht. Smart*: An open data set and tools for enablingresearch in sustainable homes. In Proceedings of 2nd KDDWorkshop on Data Mining Applications in Sustainability,Beijing, China, 2012.

[13] S. Barker, A. Mishra, D. Irwin, P. Shenoy, and J. Albrecht.Smartcap: Flattening peak electricity demand in smart homes.In Pervasive Computing and Communications (PerCom),2012 IEEE International Conference on, pages 67–75. IEEE,2012.

[14] G. Barrenetxea, F. Ingelrest, G. Schaefer, and M. Vetterli.The hitchhiker’s guide to successful wireless sensor networkdeployments. In Proceedings of the 6th ACM conference onEmbedded network sensor systems, pages 43–56. ACM, 2008.

[15] N. Batra, P. Arjunan, A. Singh, and P. Singh. Experienceswith occupancy based building management systems. In

17

Intelligent Sensors, Sensor Networks and InformationProcessing, 2013 IEEE Eighth International Conference on,pages 153–158. IEEE, 2013.

[16] N. Batra, H. Dutta, and A. Singh. INDiC: ImprovedNon-Intrusive load monitoring using load Division andCalibration. In International Conference of MachineLearning and Applications, Miami, FL, USA, 2013.

[17] N. Batra, M. Gulati, A. Singh, and M. B. Srivastava. It’sDifferent: Insights into home energy consumption in India. InProceedings of the Fifth ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, 2013.

[18] N. Batra, J. Kelly, O. Parson, H. Dutta, W. Knottenbelt,A. Rogers, A. Singh, and M. Srivastava. NILMTK: An OpenSource Toolkit for Non-intrusive Load Monitoring. In ACMe-Energy, Cambridge, UK, 2014 (in press).

[19] N. Batra, O. Parson, M. Berges, A. Singh, and A. Rogers. Acomparison of non-intrusive load monitoring methods forcommercial and residential buildings. arXiv:1404.3878, 2014.

[20] C. Beckel, W. Kleiminger, R. Cicchetti, T. Staake, andS. Santini. The eco data set and the performance ofnon-intrusive load monitoring algorithms. In Proceedings ofthe First ACM International Conference on EmbeddedSystems For Energy-Efficient Buildings. ACM, 2014.

[21] A. Beltran, V. L. Erickson, and A. E. Cerpa. Thermosense:Occupancy thermal based sensing for hvac control. InProceedings of the 5th ACM Workshop on Embedded SystemsFor Energy-Efficient Buildings, pages 1–8. ACM, 2013.

[22] M. Berg’es and A. Rowe. Appliance classification and energymanagement using multi-modal sensing. In Proceedings of theThird ACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, BuildSys ’11, pages 51–52,New York, NY, USA, 2011. ACM.

[23] K. Carrie Armel, A. Gupta, G. Shrimali, and A. Albert. Isdisaggregation the holy grail of energy efficiency? the case ofelectricity. Energy Policy, 52:213–234, 2013.

[24] P. Chakravarty and A. Gupta. Impact of energydisaggregation on consumer behavior. 2013.

[25] D. Chen, S. Barker, A. Subbaswamy, D. Irwin, and P. Shenoy.Non-intrusive occupancy monitoring using smart meters. InProceedings of the 5th ACM Workshop on Embedded SystemsFor Energy-Efficient Buildings, pages 1–8. ACM, 2013.

[26] B. Ciara. The smart grid data deluge, June 2011.[27] S. Darby. The effectiveness of feedback on energy

consumption. A Review for DEFRA of the Literature onMetering, Billing and direct Displays, 2006.

[28] S. Dawson-Haggerty, A. Krioukov, J. Taneja, S. Karandikar,G. Fierro, N. Kitaev, and D. Culler. Boss: building operatingsystem services.

[29] S. Dawson-Haggerty, S. Lanzisera, J. Taneja, R. Brown, andD. Culler. @ scale: Insights from a large, long-lived applianceenergy wsn. In Proceedings of the 11th internationalconference on Information Processing in Sensor Networks,pages 37–48. ACM, 2012.

[30] D. T. Delaney, G. M. O’Hare, and A. G. Ruzzelli. Evaluationof energy-efficiency in lighting systems using sensor networks.In Proceedings of the First ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, pages61–66. ACM, 2009.

[31] D. Dietrich, D. Bruckner, G. Zucker, and P. Palensky.Communication and computation in buildings: A shortintroduction and overview. Industrial Electronics, IEEETransactions on, 57(11):3577–3584, 2010.

[32] F. Englert, I. Diaconita, A. Reinhardt, A. Alhamoud,R. Meister, L. Backert, and R. Steinmetz. Reduce the numberof sensors: Sensing acoustic emissions to estimate applianceenergy usage. In Proceedings of the 5th ACM Workshop onEmbedded Systems For Energy-Efficient Buildings, pages1–8. ACM, 2013.

[33] V. L. Erickson and A. E. Cerpa. Thermovote: Participatorysensing for efficient building hvac conditioning. 2012.

[34] M. Evans, B. Shui, and S. Somasundaram. Country report onbuilding energy codes in india. Pacific Northwest NationalLaboratory, 2009.

[35] M. Evans, H. C. UMD, B. Shui, and S. Lee. Country report onbuilding energy codes in republic of korea. Pacific NorthwestNational Laboratory, 2009.

[36] T. Ganu, D. Rahayu, D. P. Seetharam, R. Kunnath, A. P.Kumar, V. Arya, S. A. Husain, and S. Kalyanaraman.Socketwatch: An autonomous appliance monitoring system.

[37] Z. Ghahramani and M. I. Jordan. Factorial hidden markovmodels. Machine learning, 29(2-3):245–273, 1997.

[38] S. K. Ghai, Z. Charbiwala, S. Mylavarapu, D. P.Seetharamakrishnan, and R. Kunnath. Dc picogrids: a case for

local energy storage for uninterrupted power to dc appliances.In Proceedings of the fourth international conference onFuture energy systems, pages 27–38. ACM, 2013.

[39] C. Gomez and J. Paradells. Wireless home automationnetworks: A survey of architectures and technologies. IEEECommunications Magazine, 48(6):92–101, 2010.

[40] S. Gupta, M. S. Reynolds, and S. N. Patel. Electrisense:single-point sensing using emi for electrical event detectionand classification in the home. In Proceedings of the 12thACM international conference on Ubiquitous computing,pages 139–148. ACM, 2010.

[41] M. A. Halverson, B. Shui, and M. Evans. Country report onbuilding energy codes in the united states. Pacific NorthwestNational Laboratory, 2009.

[42] T. Hao, G. Xing, and G. Zhou. isleep: unobtrusive sleepquality monitoring using smartphones. In Proceedings of the11th ACM Conference on Embedded Networked SensorSystems, page 4. ACM, 2013.

[43] G. W. Hart. Nonintrusive appliance load monitoring.Proceedings of the IEEE, 80(12):1870–1891, 1992.

[44] S. Hay and A. Rice. The case for apportionment. InProceedings of the First ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, pages13–18. ACM, 2009.

[45] T. W. Hnat, V. Srinivasan, J. Lu, T. I. Sookoor, R. Dawson,J. Stankovic, and K. Whitehouse. The hitchhiker’s guide tosuccessful residential sensing deployments. In Proceedings ofthe 9th ACM Conference on Embedded Networked SensorSystems, pages 232–245. ACM, 2011.

[46] C. Holcomb. Pecan Street Inc.: A Test-bed for NILM. InInternational Workshop on Non-Intrusive Load Monitoring,Pittsburgh, PA, USA, 2012.

[47] D. Hong, J. Ortiz, K. Whitehouse, and D. Culler. Towardsautomatic spatial verification of sensor placement in buildings.2013.

[48] J. Hsu, P. Mohan, X. Jiang, J. Ortiz, S. Shankar,S. Dawson-Haggerty, and D. Culler. Hbci:human-building-computer interaction. In Proceedings of the2nd ACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Building, pages 55–60. ACM, 2010.

[49] D. Irwin, S. Barker, A. Mishra, P. Shenoy, A. Wu, andJ. Albrecht. Exploiting home automation protocols for loadmonitoring in smart buildings. In Proceedings of the ThirdACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, BuildSys ’11, pages 7–12,New York, NY, USA, 2011. ACM.

[50] M. J. Johnson and A. S. Willsky. Bayesian NonparametricHidden Semi-Markov Models. Journal of Machine LearningResearch, 14:673–701, 2013.

[51] E. Jovanov, A. Milenkovic, C. Otto, and P. C. De Groen. Awireless body area network of intelligent motion sensors forcomputer assisted physical rehabilitation. Journal ofNeuroEngineering and rehabilitation, 2(1):6, 2005.

[52] J. Kadengal, S. Thirunavukkarasu, A. Vasan, V. Sarangan,and A. Sivasubramaniam. The energy-water nexus incampuses. In Proceedings of the 5th ACM Workshop onEmbedded Systems For Energy-Efficient Buildings,BuildSys’13, pages 15:1–15:8, New York, NY, USA, 2013.ACM.

[53] A. Kamthe, V. Erickson, M. A. Carreira-Perpinan, andA. Cerpa. Enabling building energy auditing using adaptedoccupancy models. In Proceedings of the Third ACMWorkshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 31–36. ACM, 2011.

[54] W. Kastner, G. Neugschwandtner, S. Soucek, andH. Newmann. Communication systems for building automationand control. Proceedings of the IEEE, 93(6):1178–1203, 2005.

[55] A. Kavousian, R. Rajagopal, and M. Fischer. Determinants ofresidential electricity consumption: Using smart meter data toexamine the effect of climate, building characteristics,appliance stock, and occupants’ behavior. Energy, 55(0):184 –194, 2013.

[56] J. Kelly and W. Knottenbelt. Uk-dale: A dataset recording ukdomestic appliance-level electricity demand and whole-housedemand. arXiv preprint arXiv:1404.0284, 2014.

[57] Y. Kim, R. Balani, H. Zhao, and M. B. Srivastava. Grangercausality analysis on ip traffic and circuit-level energymonitoring. In Proceedings of the 2nd ACM Workshop onEmbedded Sensing Systems for Energy-Efficiency inBuilding, pages 43–48. ACM, 2010.

[58] Y. Kim, T. Schmid, Z. M. Charbiwala, and M. B. Srivastava.Viridiscope: design and implementation of a fine grainedpower monitoring system for homes. In Proceedings of the

18

11th international conference on Ubiquitous computing,pages 245–254. ACM, 2009.

[59] Y. Kim, T. Schmid, M. B. Srivastava, and Y. Wang.Challenges in resource monitoring for residential spaces. InProceedings of the First ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, pages1–6. ACM, 2009.

[60] W. Kleiminger, C. Beckel, T. Staake, and S. Santini.Occupancy detection from electricity consumption data. InProceedings of the 5th ACM Workshop on Embedded SystemsFor Energy-Efficient Buildings, pages 1–8. ACM, 2013.

[61] J. Z. Kolter and T. Jaakkola. Approximate Inference inAdditive Factorial HMMs with Application to EnergyDisaggregation. In Proceedings of the InternationalConference on Artificial Intelligence and Statistics, pages1472–1482, La Palma, Canary Islands, 2012.

[62] J. Z. Kolter and M. J. Johnson. REDD: A public data set forenergy disaggregation research. In Proceedings of 1st KDDWorkshop on Data Mining Applications in Sustainability,San Diego, CA, USA, 2011.

[63] A. Krause, A. Singh, and C. Guestrin. Near-optimal sensorplacements in gaussian processes: Theory, efficient algorithmsand empirical studies. The Journal of Machine LearningResearch, 9:235–284, 2008.

[64] A. Krioukov, S. Dawson-Haggerty, L. Lee, O. Rehmane, andD. Culler. A living laboratory study in personalizedautomated lighting controls. In Proceedings of the ThirdACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 1–6. ACM, 2011.

[65] A. Krioukov, G. Fierro, N. Kitaev, and D. Culler. Buildingapplication stack (bas). In Proceedings of the Fourth ACMWorkshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 72–79. ACM, 2012.

[66] E. C. Larson, T. Lee, S. Liu, M. Rosenfeld, and S. N. Patel.Accurate and privacy preserving cough sensing using alow-cost microphone. In Proceedings of the 13th internationalconference on Ubiquitous computing, pages 375–384. ACM,2011.

[67] Y. LeCun and C. Cortes. Mnist handwritten digit database.AT&T Labs [Online]. Available: http://yann. lecun.com/exdb/mnist, 2010.

[68] J. Lu, T. Sookoor, V. Srinivasan, G. Gao, B. Holben,J. Stankovic, E. Field, and K. Whitehouse. The smartthermostat: using occupancy sensors to save energy in homes.In Proceedings of the 8th ACM Conference on EmbeddedNetworked Sensor Systems, pages 211–224. ACM, 2010.

[69] M. Maasoumy, B. M. Sanandaji, K. Poolla, and A. S.Vincentelli. Berds-berkeley energy disaggregation data set.

[70] S. Makonin, F. Popowich, L. Bartram, B. Gill, and I. V. Bajic.AMPds: A Public Dataset for Load Disaggregation andEco-Feedback Research. In IEEE Electrical Power andEnergy Conference, Halifax, NS, Canada, 2013.

[71] D. McIntire, K. Ho, B. Yip, A. Singh, W. Wu, and W. J.Kaiser. The low power energy aware processing (leap)embedded networked sensor system. In Proceedings of the 5thinternational conference on Information processing insensor networks, pages 449–457. ACM, 2006.

[72] A. Molina-Markham, P. Shenoy, K. Fu, E. Cecchet, andD. Irwin. Private memoirs of a smart meter. In Proceedings ofthe 2nd ACM workshop on embedded sensing systems forenergy-efficiency in building, pages 61–66. ACM, 2010.

[73] A. Monacchi, D. Egarter, W. Elmenreich, S. D’Alessandro,and A. M. Tonello. Greend: An energy consumption dataset ofhouseholds in italy and austria. arXiv preprintarXiv:1405.3100, 2014.

[74] T. X. Nghiem. MLE+: a Matlab-EnergyPlus co-simulationinterface. http://www.seas.upenn.edu/~nghiem/mleplus.html.

[75] T. A. Nguyen and M. Aiello. Energy intelligent buildingsbased on user activity: A survey. Energy and buildings,56:244–257, 2013.

[76] V. Otsason, A. Varshavsky, A. LaMarca, and E. De Lara.Accurate gsm indoor localization. In UbiComp 2005:Ubiquitous Computing, pages 141–158. Springer, 2005.

[77] C. Otto, A. Milenkovic, C. Sanders, and E. Jovanov. Systemarchitecture of a wireless body area sensor network forubiquitous health monitoring. Journal of Mobile Multimedia,1(4):307–326, 2006.

[78] K. Padmanabh, A. Malikarjuna, V, S. Sen, S. P. Katru,A. Kumar, S. P. C, S. K. Vuppala, and S. Paul. isense: Awireless sensor network based conference room managementsystem. In Proceedings of the First ACM Workshop onEmbedded Sensing Systems for Energy-Efficiency in

Buildings, BuildSys ’09, pages 37–42, New York, NY, USA,2009. ACM.

[79] O. Parson, S. Ghosh, M. Weal, and A. Rogers. Non-intrusiveload monitoring using prior models of general appliance types.In Proceedings of the 26th AAAI Conference on ArtificialIntelligence, pages 356–362, Toronto, ON, Canada, 2012.

[80] L. Pereira, F. Quintal, R. Goncalves, and N. J. Nunes.Sustdata: A public dataset for ict4s electric energy research.In ICT for Sustainability 2014 (ICT4S-14). Atlantis Press,2014.

[81] J. Pierce and E. Paulos. Beyond energy monitors: Interaction,energy, and emerging energy systems. In Proceedings of theSIGCHI Conference on Human Factors in ComputingSystems, CHI ’12, pages 665–674, New York, NY, USA, 2012.ACM.

[82] J. Piper. Bacnet, lonmark and modbus: How and why theywork. http://facilitiesnet.com/buildingautomation/article/BACnet-LonMark-and-Modbus-How-and-Why-They-Work--7712.Accessed: 2014-03-25.

[83] M. A. Prodhan and K. Whitehouse. Hot water dj: Savingenergy by pre-mixing hot water. In Proceedings of the FourthACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 91–98. ACM, 2012.

[84] D. Rahayu, B. Narayanaswamy, S. Krishnaswamy, C. Labbe,and D. P. Seetharam. Learning to be energy-wise:Discriminative methods for load disaggregation. In 3rdInternational Conference on Future Energy Systems, pages1–4, 2012.

[85] A. Reinhardt, P. Baumann, D. Burgstahler, M. Hollick,H. Chonov, M. Werner, and R. Steinmetz. On the Accuracy ofAppliance Identification Based on Distributed Load MeteringData. In Proceedings of the 2nd IFIP Conference onSustainable Internet and ICT for Sustainability (SustainIT),pages 1–9, 2012.

[86] Richard de Dear and Melissa Hart. Appliance ElectricityEnd-Use: Weather and Climate Sensitivity. Technical report,Sustainable Energy Group, Australian Greenhouse Office,2002.

[87] A. Rogers, R. Wilcock, S. Ghosh, and N. R. Jennings. Ascalable low-cost solution to provide personalized homeheating advice to households. In Proceedings of the FourthACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 211–212. ACM, 2013.

[88] M. Saha, S. Thakur, A. Singh, and Y. Agarwal. EnergyLens:Combining Smartphones with Electricity Meter for AccurateActivity Detection and User Annotation. In FifthInternational Conference on Future Energy Systems(e-Energy), pages 289–300. ACM, 2014.

[89] M. Satyanarayanan. Pervasive computing: Vision andchallenges. Personal Communications, IEEE, 8(4):10–17,2001.

[90] A. Schoofs, D. T. Delaney, G. M. O’Hare, and A. G. Ruzzelli.Copolan: non-invasive occupancy profiling for preliminaryassessment of hvac fixed timing strategies. In Proceedings ofthe Third ACM Workshop on Embedded Sensing Systems forEnergy-Efficiency in Buildings, pages 25–30. ACM, 2011.

[91] A. Schoofs, A. Guerrieri, D. T. Delaney, G. O’Hare, and A. G.Ruzzelli. Annot: Automated electricity data annotation usingwireless sensor networks. In Sensor Mesh and Ad HocCommunications and Networks (SECON), 2010 7th AnnualIEEE Communications Society Conference on, pages 1–9.IEEE, 2010.

[92] J. Scott, A. Bernheim Brush, J. Krumm, B. Meyers,M. Hazas, S. Hodges, and N. Villar. Preheat: controlling homeheating using occupancy prediction. In Proceedings of the13th international conference on Ubiquitous computing,pages 281–290. ACM, 2011.

[93] B. Shui, M. Evans, H. Lin, W. Jiang, B. Liu, B. Song, andS. Somasundaram. Country Report on Building EnergyCodes in China. Pacific Northwest National Laboratory, 2009.

[94] B. Shui, M. Evans, and S. Somasundaram. Country report onbuilding energy codes in Australia. Pacific NorthwestNational Laboratory, 2009.

[95] V. Srinivasan, J. Stankovic, and K. Whitehouse. Watersense:Water flow disaggregation using motion sensors. InProceedings of the Third ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, pages19–24. ACM, 2011.

[96] V. Srinivasan, J. Stankovic, and K. Whitehouse. Fixturefinder:Discovering the existence of electrical and water fixtures. InProceedings of the 12th international conference onInformation processing in sensor networks, pages 115–128.ACM, 2013.

19

[97] M. Sugano, T. Kawazoe, Y. Ohta, and M. Murata. Indoorlocalization system using rssi measurement of wireless sensornetwork based on zigbee standard. Target, 538:050, 2006.

[98] L. V. Thanayankizil, S. K. Ghai, D. Chakraborty, and D. P.Seetharam. Softgreen: Towards energy management of greenoffice buildings with soft sensors. In Communication Systemsand Networks (COMSNETS), 2012 Fourth InternationalConference on, pages 1–6. IEEE, 2012.

[99] T. Weng, B. Balaji, S. Dutta, R. Gupta, and Y. Agarwal.Managing plug-loads for demand response within buildings. InProceedings of the Third ACM Workshop on EmbeddedSensing Systems for Energy-Efficiency in Buildings, pages13–18. ACM, 2011.

[100] T. Weng, A. Nwokafor, and Y. Agarwal. Buildingdepot 2.0:An integrated management system for building analysis andcontrol. In Proceedings of the 5th ACM Workshop onEmbedded Systems For Energy-Efficient Buildings, pages1–8. ACM, 2013.

[101] M. Zeifman and K. Roth. Nonintrusive appliance loadmonitoring: Review and outlook. IEEE Transactions onConsumer Electronics, 57(1):76–84, 2011.

[102] J.-P. Zimmermann, M. Evans, J. Griggs, N. King, L. Harding,P. Roberts, and C. Evans. Household Electricity Survey. Astudy of domestic electrical product usage. Technical ReportR66141, DEFRA, May 2012.

[103] A. Zoha, A. Gluhak, M. A. Imran, and S. Rajasegarar.Non-intrusive load monitoring approaches for disaggregatedenergy sensing: A survey. Sensors, 12(12):16838–16866, 2012.

20

Authors Year Keyapplication

Instrumentation Interconnect Infer Involveoccupants

Intelligentoperations

Delaney et al.[30]

2009 Lighting control Light sensor, mo-tion sensor

Light and motionsensor

Infer savingswhen unoccupiedor light abovethreshold

- Suggests inte-grating energysaving algorithminto controlsystem

Hay et al. [44] 2009 Energy appor-tionment

Electric meters,diary entries,door security,appliance clampmeters

Combine electricmeters, door se-curity and diaryentries

Apportion in-dividual energyusage by un-derstandingcontext andapportionmentpolicy

Provide detailedapportioned en-ergy

-

Lu et al. [68] 2010 Smart thermo-stat

Thermostat, mo-tion and door

Determine occu-pancy from com-bining door andmotion data

Optimized tem-perature setting

Occupants estab-lish ground truthby taking notes

Pro-actively con-trol thermostatset point

Kim et al. [57] 2010 Energy appor-tionment

Circuit panelelectricity, fire-wall

electricity andfirewall

Find the causal-ity between IPand circuit levelenergy consump-tion

Provide appor-tioned energyusage and userfeedback

-

Weng et al. [99] 2011 Demand re-sponse

Smart plugs,door, motionsensors

door and motion Load sheddingpotential andoccupancy

Enlist occupantpriority for loadactuation

Automaticallyturn off loadsin response todemand

Srinivasan et al.[95]

2011 Water disaggre-gation

Household flowmeter, motionsensors

Combine motion,room metadataand flow

Bayes net to la-bel flow events

Provide roomand fixture infor-mation; may actupon data fromfeedback

Provide feedbackto occupants

Krioukov et al.[64]

2011 Personalizedautomatedlighting

access existingBACNet

Lighting controland monitoringand electricityover BACNet

Estimated en-ergy saving

Occupants haveaccess to web ap-plication to con-trol and monitorlight usage

Control lightbased on humaninput and someuser thresholds

Schoofs et al.[90]

2011 Non-invasive oc-cupancy profil-ing

Electricity me-ters, VLAN

VLAN, electric-ity

Correlate VLANand electricityconsumption

Collect occupantentry and exittimes

Dynamicallyoptimize HVACbased on occu-pancy inferredfrom VLAN

Kamthe et al.[53]

2011 Energy auditing camera sensors - Tune previousmodel fromsimilar layoutbuilding accord-ing to occupancyinformation mea-sured for smallduration

- Optimize base-line buildingstrategies

Prodhan etal. [83]

2012 Smart waterheating

Temperature andwater flow sen-sors

Water flow andtemperature tofind

Learn tempera-ture and lag timefor fixtures

Occupants con-figure comfortknobs for differ-ent fixtures

Choose fixturetemperature anddelay

Thanayankizilet al. [98]

2012 Opportunisticcontext sources

Soft sensors-WiFi, accesscards

Fuse differentcontext sources

Employee loca-tion and roomoccupancy

Access cardswipe, laptopmonitoring

HVAC optimiza-tion

Barker et al.[13]

2012 Peak load re-duction

Smart meter, In-steon

- Slack time forappliances

- Schedule back-ground loadswhile maintain-ing operationrequirements

Rogers et al.[87]

2013 Personalizedheating advice

Temperature log-ger

Indoor tempera-ture and outsidetemperaturefrom weatherstation

Thermostatmodel andsavings fromchanging opera-tions

Provide action-able feedback tooccupants

Chen et al. [25]and Kleimingeret al. [60]

2013 Occupancy pre-diction fromsmart meterdata

Smart meters - Infer occupancyfrom variouspower featuresobtained fromsmart meters

- Suggested: in-tegrate withenergy efficiencysolutions relyingon occupancy

Beltran etal. [21]

2013 Occupancysensing

thermal and mo-tion sensors

Motion sensortriggers thermalsensors whenmotion detected

Use classifiersand filteringfor predictingoccupancy

- Optimize HVACcontrol based onoccupancy

Saha et al. [88] 2014 Energy Appor-tionment

Smart Metersand smartphone(WiFi, audio)

Combines smartmeter data withsmartphone sen-sor data

What,whom,where, when ofappliance usage

Provide detailedapportioned en-ergy information

Suggests usingcontext infor-mation fromsmartphoneswith smartmeter data

Table 4: Illustration of multiple building energy applications and how they span across the 5 Is

21