Experimental and data collection methods for a large-scale smart grid deployment: Methods and first results

Energy 65 (2014) 462e471

Contents lists available at ScienceDirect

Energy journal homepage: www.elsevier.com/locate/energy

Experimental and data collection methods for a large-scale smart grid deployment: Methods and first results Joshua D. Rhodes a, *, Charles R. Upshaw b, Chioke B. Harris b, Colin M. Meehan c, David A. Walling d, Paul A. Navrátil d, Ariane L. Beck e, Kazunori Nagasawa b, Robert L. Fares b, Wesley J. Cole f, Harsha Kumar g, Roger D. Duncan h, Chris L. Holcomb e, Thomas F. Edgar f, h, Alexis Kwasinski g, Michael E. Webber b, h, ** a

Department of Civil, Architectural and Environmental Engineering, The University of Texas at Austin, 301 E. Dean Keeton St, Stop C1700, Austin, TX 78712-0273, USA b Department of Mechanical Engineering, The University of Texas at Austin, 204 E. Dean Keeton Street, Stop C2200, Austin, TX 78712-1591, USA c Environmental Defense Fund, 301 Congress Avenue, Suite 1300, Austin, TX 78701, USA d The Texas Advanced Computing Center, The University of Texas at Austin, 10100 Burnet Road (R8700), Austin, TX 78758-4497, USA e Pecan Street Inc., 3925 West Braker Lane, Austin, TX 78759, USA f Department of Chemical Engineering, The University of Texas at Austin, 200 E. Dean Keeton St., Stop C0400, Austin, TX 78712-1589, USA g Department of Electrical Engineering, The University of Texas at Austin, 2501 Speedway, Stop C0803, Austin, TX 78712-1684, USA h Energy Institute, The University of Texas at Austin, 2304 Whitis Ave, Stop C2400, Austin, TX 78712-1718, USA

a r t i c l e i n f o

a b s t r a c t

Article history: Received 26 July 2013 Received in revised form 25 October 2013 Accepted 2 November 2013 Available online 5 December 2013

This paper has two objectives: 1) to describe the experimental and data collection methods for a largescale smart grid deployment in Austin, Texas, and 2) to provide results based on those data. As of October 2012, the test bed was comprised of 1) 250 homes concentrated in a single neighborhood all built after 2007, and 2) 160 homes distributed throughout Austin with ages ranging from 10 to 92 years old. This experiment includes 200 electric monitoring systems (15-s resolution), 211 electric monitoring systems (1-min), 182 gas meters (2-cubic foot), and 51 water meters (1 gallon) and many of the monitored homes also have energy audits and homeowner surveys. The test bed also includes 185 rooftop PV (photovoltaic) installations and 50 electric vehicles in the same neighborhood. Data streams were automated and gathered at a supercomputing facility at UT-Austin yielding 250 GB (2.95
109 records) of data in the first year. This paper describes the baseline study and monitoring methods, characterizes the study participants, and provides some first results about residential energy use. These results include a negative correlation between energy use and knowledge about energy as well as a possible positive correlation between energy use and some rebates. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Energy Smart grid Residential energy use Natural gas Water

1. Introduction 1.1. Motivation and brief history The electric utility industry is rapidly changing, especially in the area generally called the “smart grid” [1]. While the definition of smart grid varies, it usually refers to the integration of information,

* Corresponding author. ** Corresponding author. Department of Mechanical Engineering, The University of Texas at Austin, 204 E. Dean Keeton Street, Stop C2200, Austin, TX 78712-1591, USA. E-mail addresses: [email protected] (J.D. Rhodes), [email protected] (M. E. Webber). 0360-5442/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.energy.2013.11.004

sensors, meters, automated controls, distributed generation, resource storage, and many other technologies relevant to the distribution grid of utilities. Though there is much speculation about the future of the smart grid, there are few controlled experiments providing rigorous field data on the deployment of these technologies. A multi-institutional smart grid demonstration project in Austin, Texas is filling this knowledge gap by gathering and analyzing novel datasets that have the potential to be valuable for grid planning and understanding how customers will interface with new devices, information, and price signals. The project includes partners from the utility, academic, business, and environmental sectors and seeks to perform an unprecedented level of monitoring and analysis of the technologies and behaviors that are relevant to

J.D. Rhodes et al. / Energy 65 (2014) 462e471

the future of the electric industry. The project is a public-private partnership between Pecan Street Inc. (a non-profit research and development organization), The University of Texas at Austin (including the Cockrell School of Engineering and the Texas Advanced Computing Center), Austin Energy (the local municipal electric utility), the Austin Chamber of Commerce, Austin Water Utility, and the Environmental Defense Fund (a non-profit environmental advocacy group). 1.2. Introduction to the research projects This manuscript covers two related research projects: the Smart Grid Demonstration Project at Mueller (Mueller project) funded by the U.S. Department of Energy and the National Science Foundation and the Home Energy Research Project (Duke project) funded by the Doris Duke Charitable Foundation. The Mueller project was originally funded through the ARRA (American Recovery and Reinvestment Act) as a smart grid demonstration project in Austin e 1 of 99 projects funded by the act. The homes that are part of the Mueller (pronounced “Miller”) project are located on the site of Austin’s former municipal airport, close to central Austin. The homes that are part of the Duke project are located throughout the greater Austin area. The homes selected to take part in these research projects received monitoring equipment that captures electricity use on less than or equal to 1 min intervals for the whole home and 6 to 22 subcircuits and major appliances. A subset of homes also received natural gas and water use monitoring equipment that measure whole-home consumption in 2 cubic foot and 1 gallon increments, respectfully. Fig. 1 shows a cartoon schematic of monitored systems in a home. Monitoring systems were selected to be passive and nearly invisible in the home to minimize homeowner awareness and interaction, so that the collected data would offer an undistorted baseline representation of energy use. Participants in the Duke and Mueller projects received a free home energy audit overseen by project researchers, and will receive, at the conclusion of the multi-year study, a confidential and detailed report on the energy usage of their home and its major circuits and appliances for each season of the year. The initial yearlong monitoring phase of the Mueller project was completed in February 2012 and preliminary research related to the project has

463

already been published [2e4]. To date, published research has centered on the home energy audits and residential surveys undertaken at the outset of the Duke project. This manuscript intends to serve as the seminal methods paper for the baseline phase and for all future work. 1.2.1. Mueller project The Mueller redevelopment area is located at the former site of the Robert Mueller Municipal Airport (þ30 170 47.6600 , 97 410 55.8300 ) (see Fig. 2). The site contains 711 acres located approximately 1.5 miles from The University of Texas at Austin and 3 miles from downtown Austin. As of October 2012, the development had 750 single-family homes that were all built after 2007, and 643 apartments completed in 2011. The development is planned to eventually include 5700 households, comprised of single and multi-family units. Planning for the Mueller Redevelopment Area (Mueller) began in 2000 [5], with construction beginning in 2007. Mueller was selected as the test bed for this research project because of its location, the relative uniformity of new homes, and the developer’s requirement to build energy efficient homes and buildings. According to the website for Mueller, “Every single Mueller building, both residential, retail and commercial will meet standards for green building established by the Austin Energy Green Building Program and the U.S. Green Building Council’s LEED certification” [6]. As of October 2012, 303 energy monitoring systems had been deployed in homes in Mueller. These homes are relatively new, most utilize gas for heating, and were built by a small number of builders. Also, 185 of the monitored homes have PV (photovoltaic) systems, totaling over 1.2 MW, which to the authors’ knowledge is the highest residential PV density in the U.S. for retrofitted systems. A majority of these PV systems have both south and west facing arrays, the purpose of which will be discussed later in this manuscript. Over 70 homes within Mueller are expected to use EVs or PHEVs by 2013 (with Level 2 charging capability), again the highest residential concentration known to the authors in the U.S. As of October 2012, 50 homes had EVs or PHEVs. 1.2.2. Duke project The Duke Project studies residential resource use in Austin’s older housing stock, monitoring usage in homes more than 10 years

Fig. 1. A cartoon schematic of a house that is interacting with the electric, gas, and water grids in a dynamic way shows some of the flows and components. The information overlay allows for the HEMS (Home Energy Management Systems) to interact with all the appliances in the home, including large loads such as the air-conditioner and EVSE (electric vehicle supply equipment). Not every water, gas, or electric line is depicted.

464

J.D. Rhodes et al. / Energy 65 (2014) 462e471

Fig. 2. The Mueller development area is located approximately 3 miles from downtown Austin, Texas (map created by the authors).

old that have undergone Austin’s ECAD (Energy Conservation Audit and Disclosure) Ordinance audits [7] (see Section 4 for a more complete description of the audits). The Duke Project consists of detailed monitoring of home resource usage for approximately 100 homes in the greater Austin area. Unlike those in Mueller, the homes in the Duke project are far from homogeneous in age and building standards. The oldest home in the Duke study was built in 1920, whereas the oldest home in the Mueller project was built in 2007. Also, the heating and air-conditioning systems vary much more in type and efficiency. The Duke project also contains 5 allelectric homes whereas all homes in Mueller have natural gas. The homes are being studied on the same parameters and resolution as the Mueller homes to provide a comparison between new homes designed for higher efficiency and older, typically less efficient homes. 1.2.3. Summary In total the 410 houses in the study represent a diverse set that will allow researchers to capture elements of energy use beyond just technological differences, including elements such as human behavior impacts on energy usage. Further, the inclusion of new (Mueller) homes to older (Duke) existing homes, both consisting of typically sized units, will allow for comparisons within and between the groups. However, the demographics of the study participants (discussed in the Supplementary Materials) are not a representative cross-section of the entire population and their selfselection bias will have to be considered in any forthcoming analysis. The current and future deployment of hardware is summarized in Sections 3 and 4. 2. Previous smart grid demonstration and home monitoring projects To date there have been numerous other studies of smart grid demonstration projects, pricing and feedback surveys, and residential energy and water consumption [8e16]. However, in the U.S., few of these investigations have focused on the customer side of

the meter. The following project summaries highlight previous studies that are relevant to the demonstration project discussed in this manuscript. These projects are not the full extent of the prior work, but are merely meant to offer historical and scientific context for the work presented here. The Center for Environmental Studies at Princeton University performed a landmark energy use study in Twin Rivers, New Jersey from 1972 to 1976 [9]. The study included monthly billing information for 200 homes, 31 of which had identical floor plans and were instrumented to collect electricity and gas data, and 3 of which were instrumented with a host of temperature and other sensors. The natural gas usage was measured on a 15 min basis; electricity meters captured data for the whole home, water heater, air-conditioner, refrigerator, and electric range all on a 20 min basis. In addition to usage records, the study collected information on people’s attitudes toward the economy, energy, and comfort as a means of predicting energy use. Finally, this study was also one of the first to examine the effects of providing energy usage feedback, and showed that daily feedback produced a 10e15% reduction in energy use. They also found that the new homeowners had virtually unrelated energy use patterns as compared to the previous owners. The Florida Solar Energy Center performed a study of 10 newly built Habitat for Humanity homes from 1994 to 1995 [10]. The homes were all identical in size, were all-electric, and outfitted with identical appliances. Electricity consumption data were taken at 15 min intervals on the whole home level, and for the major appliances. Temperature and humidity measurements were also collected from each house. The study demonstrated the impact of indoor temperature settings on energy use. The study did not provide any feedback to the participants, nor did it test the impact of any energy management technologies or advanced pricing structures. Since the study only contained 10 lower-income households, it did not provide a statistically robust dataset for drawing broad conclusions, but did provide much needed data on a typically underrepresented group in energy studies. The Laredo Customer Choice and Control Program was a demand-side management program administered by Central

J.D. Rhodes et al. / Energy 65 (2014) 462e471

Power and Light, and ran from 1995 to 1997 in the city of Laredo, Texas [11]. The study consisted of 650 homes with energy management systems, and 325 homes that were part of a control group that did not receive systems. This project was a utility-driven experiment to determine the costs and impacts of demand-side management systems. The utility installed a completely new advanced two-way network, and communicated with in-home energy management systems to provide time-of-use pricing and other information. The home energy management systems offered customer controlled load management features that allowed them to adjust consumption based on pre-programmed settings selected by the occupant. The utility also used the system for direct load control, which was an optional program offered to homeowners for an additional discount. The results of the project demonstrated that time-of-use rate structures, coupled with providing information and control capability to the customer, could be effective in shifting electricity use off peak, but not necessarily at reducing overall annual energy consumption. The project described in the manuscript is different in several ways: 1) the scale (number of homes), 2) the ability to leverage the internet for data collection, 3) diversity of homes and occupants, 4) novel appliances (PV, EVs, etc.), and 5) climate/location. In contrast with previous research, this research project will leverage the internet for data collection and consumer feedback, which was not available during previous projects. Furthermore, this project will look at the impact of high concentrations of advanced technologies including green-built homes, solar PV, electric vehicles, home energy management systems, energy storage, and fuel cells on residential developments and on the grid at large scale. This project will also be able to leverage the large amount of information (introduced in Sections 3 and 4) to analyze the impact of residential construction, human behavior, knowledge, and choice, and the impact of various incentives on overall resource consumption, both total and temporal. 3. Methods: direct measurements Direct measurements encompass quantitative data that have been collected on-site with an installed system of monitors and measured data pulled from other sources, such as solar insolation data. The main direct measurements provide electricity, gas, and water usage data and are captured with varying degrees of granularity and precision by different metering technologies, as summarized in Table 1. The primary focus of the first phase of the project was to collect electricity, gas, and water consumption data on homes, both inside and outside of Mueller, to develop an expected baseline profile of consumption. As of October 2012, monitoring systems had been deployed at approximately 400 homes (some homes have multiple systems), with a growing list of homeowners who are official participants awaiting system deployment. The majority of these homes are located in Mueller, including 185 in Mueller with PV (photovoltaic) systems (5 in Duke), many with both south and west facing arrays. The other 100 houses with systems are outside of Mueller and dispersed throughout the city. The baseline is a subset of the total project comprised of 100 homes inside of the Mueller community (the Mueller baseline group) and 100 homes outside of Mueller (the Duke baseline group). The baseline will provide a reference point for the effects of deployment of home energy management systems, including some with varying levels of user feedback. Other direct measurements, such as weather, solar insolation, wholesale electricity prices, and power plant emissions, are collected from various outside sources. Weather and solar insolation data are collected for home energy use and solar generation analyses, while price and emissions data

465

Table 1 The distribution of meters and monitoring systems deployed (and planned as of October 2012) shows a diverse set of systems are deployed at a large number of homes. Devices deployed

Resolution

Mueller

Landis þ Gyr Focus AX electric meter Landis þ Gyr Focus AXR electric meter Badger E-Series water meter Itron Centron SR gas meters eGauge system Check-It system Incenergy system HEMS

5 min

25

15 min 1 gal 2 c.f. 1 min. Variable 15 s Variable

Duke

Future Mueller

Future Duke

2

500

150

250

160

500

300

26 100 155 9 100 132

25 82 43 5 100 19

26 400 200 9 0 132

32 100 100 5 0 19

are collected to assess the houses’ impact on the energy market and the environment. The following subsections describe the characteristics of the systems used and/or data collected for electricity, natural gas, water, solar insolation, weather, wholesale electric pricing, and emissions data. 3.1. Electricity The bulk of the data collected thus far have been electricity data. Electricity measurements fall into three primary categories, based on the different systems that collect the data: 1. Utility meter readings (Austin Energy, Landis þ Gyr AXR meters) 2. Whole home and subcircuit readings (Incenergy, eGauge, CheckIt) 3. Whole-home and PV monitoring system readings (eGauge, Check-It) The systems each offer a different level of accuracy and sampling resolution. The Austin Energy meter captures average power draw on 15 min intervals. The Incenergy electricity measurement systems allow subcircuit data to be captured on a 15 s resolution. The PV monitoring systems (eGauge, Check-It) capture whole home, subcircuit, and PV generation data, nominally on a 1 min interval. 3.1.1. Data resolution Since electricity measurements have been the primary data collected so far, they cover the entire baseline population. Austin Energy monitors all residential homes with wireless meter reading using Landis þ Gyr Focus AXR electric meters and has a dedicated backhaul network to collect the data remotely. Austin Energy meter data include 15-min, daily, and monthly cumulative energy consumed (kWh) readings. These values were then translated into average real power demand (kW) for comparison with the other electricity data. The data are collected and managed by Austin Energy, then sent to the TACC (Texas Advanced Computing Center) in batches. Homes with PV have an additional meter on the circuit containing the PV system, so these houses have two Austin Energy meters and three datum fields: “To Grid”, “From Grid”, and “PV Generated.” The Incenergy subcircuit data are collected at every home in the baseline, as it was the first system deployed as part of both the Mueller and Duke studies. This system captured current measurements from both legs (live wires) of the main panel and 6 prioritized subcircuits on 15 s intervals using current transformers installed inside the circuit breaker panel. Circuit prioritization focused on large power drawing circuits and isolated appliances, starting with the air-conditioning compressor circuit as top

466

J.D. Rhodes et al. / Energy 65 (2014) 462e471

Fig. 3. Sub-figures a, b, c, and d show example electricity data for the three different in-home monitoring systems installed. Fig. 3(a) is an example of whole home and subcircuit data collected with the Incenergy systems. Fig. 3(b) contains a sample of data collected from a Check-It energy monitoring system. Fig. 3(c) shows example data collected for one home with the eGauge system. Lastly, Fig. 3(d) provides a sample of the eGauge data for 14 homes averaged together to demonstrate the smoothing of aggregated generation and demand for a group of homes.

priority, and often including the refrigerator, microwave, and large home entertainment systems. These higher granularity data provide a more complete and detailed view of energy use in the home than the whole home 15 min measurements, which is essential for load disaggregation and appliance profiling. Load disaggregation is discussed further in the Supplementary Materials. The CT (current transformers) capture the absolute average current over the interval via an encoder inside the homes’ circuit breaker panels. The encoder then wirelessly communicates the readings back to a gateway connected to the home’s wireless router. The data are then sent via the internet to Incenergy for processing. In order to convert the current to approximate real power consumption, the current measurements are multiplied by the RMS voltage measured at the time of installation and are also multiplied by a manufacturers’ recommended adjustment factor for differently sized current transformers (see Supplementary Materials). All homes in the baseline project received Incenergy systems. In addition to the Austin Energy metering, homes receiving rebates for PV systems as part of this project also have additional whole-home electricity monitoring systems. However, homes that had solar PV before the project began might not have received these additional systems. Most homes have eGauge monitoring systems, while a small number of homes have monitoring systems from Check-It. The eGauge systems collect readings from both whole home legs (120 V, 60 Hz feed) and the PV circuit on a 1 min interval. The eGauge system captures both the current and voltage for each leg, and other measurements such as power factor. There are 47 homes in the baseline that have eGauge systems. There were also 16 homes that had PV before the baseline started that do not have any additional monitoring system beyond the Incenergy system. The Check-It system takes measurements every second, but stores the data in averaged variable intervals depending on several conditions. If the real power measured on a circuit changes by more

than 100 W or 10% of the present value if the present value is between 10 and 100 W, it begins a new interval. Thus, any significant change is recorded. The Check-It system also allows for more subcircuit monitoring than the Incenergy and eGauge systems, with a maximum of 16 subcircuit measurements. Samples of the electric power data streams that are being collected are shown in Fig. 3. 3.2. Natural gas Smart gas meters were used beginning in October 2011 to monitor natural gas usage. As of October 2012, 100 homes in Mueller and 82 homes in the Duke project were monitored. Monitoring natural gas usage is integral to understanding energy consumption for homes in the study, as most homes in Mueller were built with natural gas heating, cooking, and water heating. The natural gas data are collected via an Itron ERT transmitter from an Itron gas meter and sent to the existing Incenergy gateway in the home. The meter replaces the conventional meter (provided by Texas Gas Service, the local retail gas utility), and is a utilityrated gas meter. The gas meter registers a change of cumulative consumption at the time when the last marginal 2 cubic foot (or greater quantity) of natural gas passes through the meter. Standard cubic feet of gas is used because that is the conventional standard for billing. The meter reading frequency is every 15 s, which is a report of the current cumulative total as read by the meter. In order to estimate consumption over a time-span, the difference in cumulative totals is taken, which renders a 2 cubic foot pulse at the time step when the meter registered the change, as shown in Fig. 4. 3.3. Water Water consumption data are the third family of direct measurements collected as part of the study. There are 26 homes in

J.D. Rhodes et al. / Energy 65 (2014) 462e471

Mueller and 25 homes in the Duke project that have had smart water meters installed. The water meter installations began May 2011, with the full roll-out completed June 2012. The remote water metering system uses a Badger E-Series water meter, coupled with an Itron ERT transmitter to communicate usage data back to the Incenergy gateway in the home (see Fig. 5). The Badger meter replaces the existing mechanical water meter for the home and now functions as the meter for Austin Water Utility’s billing purposes. The meter registers every gallon of water consumed and sends the cumulative consumption data every 15 s (see Fig. 6). The data processing is very similar to that of the natural gas meter, since the rate of consumption is estimated by dividing the increase in the cumulative total by the span of time between the two readings. The water data are at the whole home usage level only, so the usage from specific internal water loads is not directly known. There is interest in trying to determine, either by disaggregation or by direct measurement, the water consumption for landscape irrigation, but that is still an area of future work.

467

Fig. 4. An example figure of natural gas usage for a typical home shows that usage is clustered around certain events (e.g. cooking or taking a hot shower) with long periods of very low or no use registered.

3.4. Other direct datasets Other direct datasets, including solar insolation data, local and regional weather data, ERCOT (electricity pricing data), and power plant emissions data have also been collected in order to compliment the datasets measured by this project. They are summarized in more detail in the supplementary materials document associated with this manuscript. 4. Methods: indirect measurements In addition to these direct measurements, indirect static measurements of the physical states of homes and the attributes and attitudes of those living in them have also been collected. Indirect measurements include home energy audits and survey data. Datasets include Austin’s Energy Conservation Audit Disclosure (ECAD) audits, which includes over 12,000 energy audits from homes and commercial facilities in Austin, Texas and advanced energy audits performed on homes in the Mueller and Duke studies that probed deeper than the ECAD audits. Lastly, surveys of residents in the study were conducted to assess aspects that direct measurements and energy audits could not address.

furnace age, furnace capacity, furnace efficiency, steady-state temperature drop across the cooling coil, recommendations to improve HVAC performance, number of doors, envelope penetrations sealed state, major appliances with age, and types of plumbing fixtures. A sample blank ECAD audit is included in the supplementary materials. Using this audit as a template, a more thorough audit (Section 4.2) was created that was performed on a smaller number (N ¼ 200) of homes that are also being monitored for their resource use. 4.2. Enhanced energy audits The enhanced energy audits were performed on the 200 homes of the baseline study (100 in Mueller and 100 in Duke). The enhanced audit’s 304 fields cover the same material as the ECAD audits (Section 4.1), and also include more detailed information, including measurements of envelope leakage via blower door tests at 15, 30, and 50 Pa, window sizes, and orientation. These audits provided fields necessary to perform specific analyses such as building energy modeling and optimization via software platforms such as EnergyPlus [17]. A sample blank enhanced energy audit is included in the Supplementary Materials.

4.1. ECAD 4.3. Surveys Austin is unique in that it is one of the few cities in the world that requires energy audits to be performed on buildings before they can be sold. Austin’s Energy Conservation Audit Disclosure ordinance [7] mandates the performance of an energy audit on all residential and commercial buildings that are greater than 10 years old at the point of sale. Since the ordinance went into effect in 2007, there have been approximately 12,000 audits performed on buildings in Austin (see Fig. 7). Audits are a static representation of the state of efficiency of the home. These audits are a useful dataset from which to benchmark the homes monitored in the smart grid projects. Previous work by some of the authors on the utility of these audits has shown potential to significantly reduce energy use and peak demand by homes in Austin Refs. [2,3]. The audits have 275 fields that characterize the physical state of the home including home size, number of stories, type of foundation, utilities serviced, home type, type of cooling system, type of heating system, thermostat type, type of windows, type of window shading (if any), recommendations for window shading, attic insulation level and type, type of HVAC ducts, duct blaster results (% air leakage) at low pressure (25 Pa), location of air handler, airconditioner (A/C) age, A/C nominal capacity, A/C efficiency,

Internet-based surveys were distributed to homeowners in the study and to homeowners interested in joining for the next phase

Fig. 5. This picture shows the first smart water meter being installed at a home in the Duke project. The meter is the device with the LCD display and is connected to the antenna.

468

J.D. Rhodes et al. / Energy 65 (2014) 462e471

The questions in this survey were designed to allow for comparisons to previous survey studies [9,18]. This survey will be required of homeowners who wish to participate, and an updated version will be sent out annually to capture any major changes in the house or inhabitants. A sample of the survey is included in the supplementary materials. 5. Residential energy use regression analysis 5.1. Regression analysis introduction

Fig. 6. This graph shows an example of water use for a single home over a 24 h period. Note the punctuated periods of use in the morning, then again at night, separated by a long period of semi-regular 1 gallon increments at roughly even intervals. The regular 1 gallon water draw could be a singular event (e.g. a toilet flush), or many smaller timedistributed water draws that have to add up to trip the meter (e.g. an ice maker).

of the study. SurveyGizmo, an online survey tool, was used to administer the survey. Over 250 people completed the entire survey (including some non-baseline participants) which resulted in an 82% and 72% response rate for Mueller and Duke, respectively. The surveys were designed to acquire information from the homeowners such as how long they have lived in the home, age of all resident(s), income, education level of all persons in the home, heating and cooling set points, hot water set point, average number of showers taken, and information corresponding to types, and number of appliances used in the home. In addition to surveying some aspects of the home’s physical state, the survey asked questions to gauge participants’ general knowledge of electricity, water, and gas usage. General electricity, water, and gas usage questions pertained to the U.S. energy consumption fuel mix, knowledge of peak demand, and order of magnitude estimates of how much water, energy, and power a typical Austin home consumes. These data have been compiled and included in the same database as the direct home measurements and audits.

The integration of uniquely detailed quantitative and qualitative datasets described in this paper allow for novel analysis. Two different regression analyses are used with the data to assess both the factors driving total energy use and the effectiveness of residential energy retrofits. In the first part of this analysis, data from this project are used to assess the relationship between survey and audit results and residential energy use. While there are studies that have analyzed macro-level data from the EIA (Energy Information Administration) [19] to understand residential energy use [20e22], there is little analysis that has had this level of detail, particularly on homes in a hot and humid climate such as Austin, TX. This preliminary work seeks to fill this knowledge gap. The second analysis seeks to quantify the benefit of energy retrofits on homes in the study. There have been studies that have looked at the effect of residential energy retrofits on total energy use, but they have typically been limited to the billing (month) level or utilized building energy simulation software [23e26]. Most past studies have lacked the data to quantify the effect of retrofits on measured energy use at a finer granularity. 5.2. Project data used in these analyses The data used in the total energy use analysis consist of survey and audit data (Section 4) and energy use data (Section 3). Total energy use data used were gathered from a subset of 41 homes for

Fig. 7. A distribution of residential ECAD audits as of March 2011 shows a wide dispersal throughout the city limits of Austin (map created by the authors).

J.D. Rhodes et al. / Energy 65 (2014) 462e471

the year 2011. The survey and audit data fields used as explanatory variables (regressors) included the home-specific data: number of levels, number of bedrooms, year built, home size, number of children and adults, thermostat set points, energy and water quiz scores, education level and income. The data used in the retrofit analysis consist of daily electricity use (kWh) data for 28 homes for the period from January 02, 2010 to November 07, 2012 (27,532 total observations). Half of these homes received various energy retrofits during those three years. Care was taken to make sure that there was enough data (at least a spring-summer or summer-fall time period) before and after the retrofit(s) to support analysis consisting of adequate seasonal variation. Not all homes received the same mix of retrofits, nor did homes receive them at the same time. The retrofits performed were part of a municipal electric utility (Austin Energy) residential retrofit rebate program and some of the homes involved happened to be in the Duke project group [27]. The retrofit analysis data are panel data, which consist of multiple individuals (in this case homes) where each individual also contains its own time series data [28]. Dummy variables (0 ¼ no retrofit, 1 ¼ received retrofit) were introduced to indicate when a home received a retrofit and weather effects were normalized by the inclusion of cooling degree days (CDD) and HDD (heating degree days) [29,30]. 5.3. Regression methodology Multiple linear regression is “a method that summarizes how the average values of a numerical outcome variable vary over subpopulations defined by linear functions of predictors” [31]. This analysis seeks to determine how total yearly energy use is related to static values found in the surveys and audits. This model used the lm linear model package of the statistical analysis tool R [32]. The basic multiple linear regression model is given in Equation (1):

Yi ¼ bXi þ a þ εi ;

(1)

where Yi is the amount of yearly energy used (kWh) for a given home, b is the vector of fit regression coefficients, Xi is the set of explanatory variables, a is a constant or intercept, and εi is the error associated with the estimation of the energy use for that home. Panel regression allows one to estimate the effect of explanatory variables on multiple individuals. This analysis used the plm panel data estimators package in R. Typically, panel regression analysis must check if the estimation method used is consistent with the data. Results of Hausman tests indicated that a fixed-effects estimator was the best option [33]. The fixed effects model is similar to a multiple linear regression for each home d the set of regression coefficients b will be the same, but each will have its own constant, or intercept. The basic panel regression model is given in Equation (2):

Yit ¼ bXit þ ai þ εit ;

(2)

where Yit is the amount of energy used (kWh) for a given home i on a given day t, b is the vector of fit regression coefficients, Xit is the set of explanatory variables, ai is a constant associated with home i, and εit is the error associated with the estimation of the energy use for that home and day. 5.4. Results The output from the multiple linear regression model described above is given in Table 2.

469

In Table 2 Intercept is the constant term of the regression, however in this model its interpretation is of no value (the “energy use” associated with zeros for all other coefficients, i.e. a home built in year 0, of 0 square foot, etc.), the Year_Built coefficient estimate is the effect of increasing the construction date of the home by one year, all else held constant, the Condition_Sqft coefficient estimate is the effect of increasing the size of the home by 1 square foot, Number_Kids is the effect of increasing the number of children in the home by 1, Number_Adults is the effect of increasing the number of adults in the home by 1, Income is the effect of increasing an income bracket (Section 4.3), Water_Knowledge_Score is the effect of a 1 point increase in the score on the water knowledge quiz included in the survey, and Energy_Knowledge_Score is the effect of a 1 point increase in the score on the energy knowledge quiz included in the survey (see the Supplementary Materials for a copy of the survey). The model was built using best practices from regression texts [31]. All the regression coefficients in the final model had the expected signs even if they were not significant (Number_Kids and Income). The coefficient Year_Built indicates that newer homes (those built in more recent years), on average consume less energy than older homes. Home size (Condition_Sqft) also has a positive relationship with energy use, as do the number of children and adults (Number_Kids and Number_Adults). Interestingly, higher scores (Water_Knowledge_Score and Energy_Knowledge_Score) on the water and energy quiz in the survey (indicating more knowledgeable participants) were correlated with reduced energy use. This finding suggests that education might be effective in reducing in residential energy usage. The output from the panel regression model is given in Table 3. In Table 3, the coefficient estimate of HDD is the estimated effect of an additional heating degree day on daily energy use, CDD is the estimated effect of an additional cooling degree day on daily energy use, Solar.Shading is the estimated effect of upgrading solar shading of windows, Air.Seal is the estimated effect of reducing the outdoor air infiltration rate of the home (weatherstripping and sealing cracks in the façade), Attic.Insul is the estimated effect of increasing the amount of attic insulation, HVAC is the estimated effect of upgrading the home’s HVAC equipment, and Other.System is the estimated effect of upgrading some non-HVAC appliances to Energy Star versions or acquiring new Energy Star appliances. All the coefficient estimates have the expected sign with the exception of Other.System. The data did not indicate if the appliance in question was an upgrade or a new addition to the home. The data simply stated when an Energy Star appliance was purchased. Since the coefficient is positive and significant, it would seem to imply that the purchases, on average, were for new additions to the homes’ set of appliances. A possible scenario being that as an older, less efficient refrigerator was replaced by a new Energy Star unit,

Table 2 Multiple regression output for the relationship between survey and audit data and total yearly energy use. Explanatory variable Coefficient Std. Error t-value Two-tailed Significance estimate P-test level Intercept Year_Built Condition_Sqft Number_Kids Number_Adults Income Water_Knowledge_ Score Energy_Knowledge_ Score

273,641 136 7.6 1096 2083 1115 1858 1540

77,209 3.54 38.9 3.51 0.86 8.80 748 1.47 920 2.26 776 1.44 783 2.38 673

2.29

0.001 0.001 4.69E-10 0.153 0.031 0.160 0.024

** ** ***

0.029

*

* *

0 ‘***’, 0.001 ‘**’, 0.01 ‘*,’ 0.05 ‘.’, 0.1 ‘ ’ 1. Adjusted R2: 0.7874, n ¼ 41.

470

J.D. Rhodes et al. / Energy 65 (2014) 462e471

Table 3 Panel regression output for fixed-effects model showing the estimated impact of energy use retrofits on daily energy use using the whole time period. Explanatory variable

Coefficient estimate

Std. error

t-value

Two-tailed P-test

Significance level

HDD CDD Solar.Shading Air.Seal Attic.Insul HVAC Other.System

1.57 2.68 0.53 1.56 1.70 4.09 3.78

0.03 0.02 1.02 0.63 0.66 1.02 1.19

51.66 156.61 0.52 2.46 2.57 4.02 3.19

<2.2E-16 <2.2E-16 0.602 0.014 0.010 5.91E-05 0.001

*** *** * * *** **

dominated climate. Austin Energy has a tiered rebate system with increasing rebates for higher efficiency HVAC units. Data from Austin Energy’s rebate program indicate that the average rebate associated with upgrading HVAC equipment is $450 for an estimated rebate-cost to the utility of $0.020/kWh. The average capital cost of the retrofit was $6517 leading to a final homeowner capital cost of $6067 and a homeowner-cost of $0.271/kWh (Table 4). However, this result might not be fully accurate if the HVAC unit was in need of replacement and code required a unit of higher efficiency than the unit being replaced.

0 ‘***’, 0.001 ‘**’, 0.01 ‘*’, 0.05 ‘.’, 0.1 ‘ ’ 1. Adjusted R2: 0.5099, n ¼ 28.

5.5. Discussion

the former was moved to the garage where it continued to be used. The energy efficiency rebate might have offset a larger increase in energy use associated with the purchase of a less efficient appliance, but the data do not allow us to test this hypothesis. The only variable that was not significant was Solar.Shading. The sign of the coefficient is as expected and according to regression texts, it is convention to leave such variables in the model [31]. There were three retrofits that showed significant energy reductions: Air.Seal, Attic.Insul, and HVAC. The retrofits showed, on average, a 1.56, a 1.70, and a 4.09 daily kWh reduction, respectively, holding all else constant. Also, these retrofits are expected to last 30, 40, and 15 years, respectively [34]. Data from Austin Energy’s rebate program [27] indicate that the average rebate associated with air sealing is $241 for an estimated rebate-cost to the utility of $0.014/kWh (over the expected lifetime of the retrofit). This rebate-cost was calculated using Equation (3):

Results from both regression analyses reveal some interesting results. While most of the results from the first regression are intuitive, the significant correlation of reduced energy use with increased energy and water knowledge is interesting. Survey questions about residential and national resource use were deployed to assess the homeowner’s knowledge with the hypothesis that it would significantly effect choices and behavior related to energy use. This result might lend some support for increased energy and water education campaigns. The retrofit analysis provided results that utilities can use to assess the value of residential retrofit rebates as compared to the cost of acquiring energy on the wholesale market. Average yearly wholesale electricity costs (Supplementary materials) for 2011e 2012 were $0.037/kWh. The model indicates that the current level of rebates is cost effective for the utility for all three retrofits, and could possibly be increased. Austin Energy has a five-tiered residential electricity rate structure based on consumption, partitioned by summer and non-summer seasons [35]. The realized residential rate (per kWh costs) increases as one uses more energy in a month from $0.065/kWh for winter use less than 500 kWh to $0.161/kWh for all use in excess of 2500 kWh in the summer. Considering an average of $0.113/kWh for residential electric service, both the airsealing (Air.Seal) and added attic insulation (Attic.Insul) seem to make economic sense for the homeowner. It is difficult to infer much about the value of the homeowner cost associated with HVAC replacements as it is not known if the HVAC system was at its end of life at the time of the upgrade. Overall, these datasets can be used to examine detailed research questions. More analyses, including south vs. west solar, emissions impacts, clustering of residential energy profiles, electric load disaggregation, electric vehicle impact, natural gas vehicle impact, dynamic electricity pricing plans, energy storage, and residential water use characterization are found in the Supplementary Materials.

Crebate ¼

Ravg

bretrofit
365
Lretrofit

;

(3)

where Crebate is the “rebate-cost” to the utility ($/kWh), Ravg is the average rebate of the retrofit ($), bretrofit is the coefficient of the retrofit given in the panel regression (kWh/day), and Lretrofit is the lifetime of the retrofit (years). Data from Austin Energy also revealed that the average capital cost of the retrofit Air.Seal was approximately $744 for a final (to-homeowner) capital costs of $503. This leads to a homeowner-cost of $0.029/kWh. This homeowner-cost was calculated using Equation (4):

Chomeowner ¼

FCavg

bretrofit
365
Lretrofit

;

(4)

where Chomeowner is the “homeowner-cost” ($/kWh), FCavg is the average final capital cost to the homeowner ($), bretrofit is the coefficient of the retrofit given in the panel regression (kWh/day), and Lretrofit is the lifetime of the retrofit (years). Austin Energy also offers rebates for increasing attic insulation up to R-38. Data from Austin Energy indicate that the average rebate associated with increasing attic insulation is $163 for an estimated rebate-cost to the utility of $0.007/kWh. The average capital cost of the retrofit was $1037 leading to a final homeowner capital cost of $874 and a homeowner-cost of $0.035/kWh. The highest reduction in daily energy use came from upgrading the HVAC system, which is not surprising given the local cooling

6. Conclusions The scope of this smart grid project extends across electric, gas, and water resource usage for over 400 homes in Austin with a range of age, size, appliance composition, and socio-economics of occupants. Data collected includes five different HEMS (Home Energy Management Systems) systems, AMI electric meters, AMR gas and water meters, audit data, utility billing data, and annual survey data, as well as external datasets for weather, solar radiation,

Table 4 Summary of costs and rebates associated with residential energy retrofits and Austin Energy’s retrofit rebate program. Retrofit

Average capital cost

Average rebate

Average final cost

Energy saved (kWh)

Rebate cost ($/kWh)

Homeowner cost ($/kWh)

Air.Seal Attic.Insul HVAC

$744 $1037 $6517

$241 $163 $450

$503 $874 $6067

17,082 24,820 22,393

$0.014 $0.007 $0.020

$0.029 $0.035 $0.271

J.D. Rhodes et al. / Energy 65 (2014) 462e471

electricity pricing, and emissions data. All homes with PV installations have monitoring systems capturing usage and generation at 1-min granularity. Most homes with EVs have a Level 2 charger with whole-home and extensive subcircuit monitoring at 1-min granularity in order to capture detailed usage and charging patterns. For all homes in Mueller, the transformer number is tracked, allowing for exploration of the grid impacts of highly concentrated solar PV and EV to be analyzed in the context of the distribution system. This project will provide data from many analyses (see the Supplementary Materials for some examples of future work) on varied smart grid related topics. Our initial results indicate that there is an inverse correlation between energy and water knowledge and residential energy use and that residential energy retrofits such as air-sealing and increasing attic insulation are cost effective not only for homeowners but the local utility providing rebates as well. The culmination of these efforts to date are a sizable, curated, one-of-a-kind dataset covering many use cases that might be widely adopted in the near future, including dense deployments of solar PV systems and EVs in both newly-constructed neighborhoods, as well as existing neighborhoods with older infrastructure. It is the author’s hope that this dataset will prove valuable for utilities, cities, private industry, and the scientific community by providing insights into how the home resource profiles (total and temporal) might change in response to the adoption of new and existing smart grid technologies and appliances, such as HEMS, PV, EVs, and storage, at scale. Acknowledgments The authors would like to thank the United States Department of Energy, the National Science Foundation IGERT program, the Texas State Energy Conservation Office, the Doris Duke Charitable Foundation, Austin Energy, Austin Water Utility, Texas Gas Service, and Pecan Street Inc. for their direct and in-kind sponsorship of this project. The authors would also like to thank the study participants and the Mueller neighborhood groups for their support. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.energy.2013.11.004. References [1] Blumsack S, Fernandez A. Ready or not, here comes the smart grid! Energy 2012;37(1):61e8. http://dx.doi.org/10.1016/j.energy.2011.07.054. URL: http://www.sciencedirect.com/science/article/pii/S0360544211005287. [2] Rhodes JD, Stephens B, Webber ME. Using energy audits to investigate the impacts of common air-conditioning design and installation issues on peak power demand and energy consumption in Austin, Texas. Energy Build 2011;43(11):3271e8. http://dx.doi.org/10.1016/j.enbuild.2011.08.032. URL: http://www.sciencedirect.com/science/article/pii/S0378778811003823. [3] Rhodes JD, Stephens B, Webber ME. Energy audit analysis of residential airconditioning systems in Austin, Texas. ASHRAE Trans 2012;118(1):143e50. [4] Rhodes JD, Nagasawa K, Upshaw C, Webber ME. The role of small distributed natural gas fuel cell technologies in the smart energy grid. In: Proceedings of the ASME 2012 6th International Conference on energy Sustainability & 10th fuel cell Science, Engineering and Technology Conference. American Society of Mechanical Engineers; 2012. [5] History d Mueller Austin. URL: http://www.muelleraustin.com/history; 2013. [6] Frequently asked questions d Mueller Austin. URL: http://www. muelleraustin.com/faqs; 2013. [7] About the energy conservation audit and disclosure (ECAD) ordinance. URL: http://www.austinenergy.com/aboutus/environmentalinitiatives/ordinance/ index.htm; 2013. [8] SmartGrid.gov: Home. URL: http://www.smartgrid.gov/; 2013. [9] Socolow RH. The Twin Rivers program on energy conservation in housing: highlights and conclusions. Energy Build 1978;1(3):207e42. URL: http:// www.sciencedirect.com/science/article/pii/0378778878900038.

471

[10] Monitored energy use patterns in low-income housing in a hot and humid climate. ESL-HH-96-05-38. In: Tenth Symposium on improving building systems in hot and humid climates. 1679 Clearlake Road, Cocoa, Florida 32922: Florida Solar Energy Center; 1996. [11] Tech. Rep. The Laredo Pilot program. Central Power & Light Company; 1998 [12] Faruqui A, Sergici S. Household response to dynamic pricing of electricity-a survey of the empirical evidence. Tech. Rep. 1134132. The Brattle Group; 2010. [13] Lutz J. Water and energy wasted during residential shower events: findings from a pilot field study of hot water distribution systems. Tech. Rep. LBNL5115E. Berkeley, CA 94720: Lawrence Berkeley National Laboratory; 2011. [14] Kofler MJ, Reinisch C, Kastner W. A semantic representation of energy-related information in future smart homes. Energy Build 2012;47:169e79. [15] Saldanha N, Beausoleil-Morrison I. Measured end-use electric load profiles for 12 Canadian houses at high temporal resolution. Energy Build 2012;49(0): 519e30. http://dx.doi.org/10.1016/j.enbuild.2012.02.050. URL: http://www. sciencedirect.com/science/article/pii/S0378778812001429. [16] Sadineni SB, Boehm RF. Measurements and simulations for peak electrical load reduction in cooling dominated climate. Energy 2012;37(1):689e97. http://dx.doi.org/10.1016/j.energy.2011.10.026. URL: http://linkinghub. elsevier.com/retrieve/pii/S0360544211006852. [17] Online BEopt e NREL. URL:, http://beopt.nrel.gov/; 2012. [18] Allen D, Janda K. The effects of household characteristics and energy use consciousness on the effectiveness of real-time energy use feedback: a pilot study. Tech. Rep. American Council for an Energy-Efficient Economy; 2006 [19] Residential energy consumption survey (RECS) e energy information administration. URL: http://www.eia.gov/consumption/residential/; 2013. [20] Garbacz C. A model of residential demand for electricity using a national household sample. Energy Econ 1983;5(2):124e8. http://dx.doi.org/10.1016/ 0140-9883(83)90019-1. URL: http://www.sciencedirect.com/science/article/ pii/0140988383900191. [21] Hirst E, Goeltz R, Carney J. Residential energy use. Energy Econ 1982;4(2):74e 82. http://dx.doi.org/10.1016/0140-9883(82)90024-X. URL: http://www. sciencedirect.com/science/article/pii/014098838290024X. [22] Kaza N. Understanding the spectrum of residential energy consumption: a quantile regression approach. Energy Policy 2010;38(11):6574e85. http:// dx.doi.org/10.1016/j.enpol.2010.06.028. URL: http://www.sciencedirect.com/ science/article/pii/S030142151000491X. [23] Guiterman T, Krarti M. Analysis of measurement and verification methods for energy retrofits applied to residential buildings. ASHRAE Trans 2011;117(2): 382e94. [24] Xu P, Xu T, Shen P. Energy and behavioral impacts of integrative retrofits for residential buildings: what is at stake for building energy policy reforms in northern China? Energy Policy 2013;52(0):667e76. http://dx.doi.org/10.1016/ j.enpol.2012.10.029. URL: http://www.sciencedirect.com/science/article/pii/ S0301421512008877. [25] Ma Z, Cooper P, Daly D, Ledo L. Existing building retrofits: methodology and state-of-the-art. Energy Build 2012;55(0):889e902. http://dx.doi.org/ 10.1016/j.enbuild.2012.08.018. URL: http://www.sciencedirect.com/science/ article/pii/S0378778812004227. [26] Guler B, Fung AS, Aydinalp M, Ugursal VI. Impact of energy efficiency upgrade retrofits on the residential energy consumption in Canada. Int J Energy Res 2001;25(9):785e92. http://dx.doi.org/10.1002/er.721. URL: http://doi.wiley. com/10.1002/er.721. [27] Home performance with energy STARRebate levels. URL: http://www. austinenergy.com/energyefficiency/Programs/Rebates/Residential/ HomePerformancewithEnergyStar/index.htm; 2013. [28] Hansen BE. EconometricsIn Self. 12 ed.. URL: http://www.ssc.wisc.edu/ wbhansen/econometrics/; 2013. [29] Valor E, Meneu V, Caselles V. Daily air temperature and electricity load in Spain. J Appl Meteorol 2001;40(8):1413e21. http://dx.doi.org/10.1175/15200450(2001)040<1413:DATAEL>2.0.CO;2. URL: http://journals.ametsoc.org/ doi/abs/10.1175/1520-0450\%282001\%29040\%3C1413\%3ADATAEL\%3E2.0. CO\%3B2. [30] Quayle RG, Diaz HF. Heating degree day data applied to residential heating energy consumption. J Appl Meteorol 1980;19(3):241e6. http://dx.doi.org/ 10.1175/1520-0450(1980)019<0241:HDDDAT>2.0.CO;2. URL: http:// journals.ametsoc.org/doi/abs/10.1175/1520-0450\%281980\%29019\% 3C0241\%3AHDDDAT\%3E2.0.CO\%3B2. [31] Gelman A, Hill J. Data analysis using regression and multilevel/hierarchical modelsIn vol. 625 of analytical methods for social research. 11 ed. Cambridge: Cambridge University Press; 2006, ISBN 9780511790942; 2006. http:// dx.doi.org/10.1017/CBO9780511790942. URL: http://books.google.com/ books?id¼c9xLKzZWoZ4C. [32] The comprehensive R archive network. URL: http://cran.us.r-project.org/; 2013. [33] Hausman JA. Specification tests in econometrics. Econometrica 1978;46(6): 1251e71. URL: http://www.jstor.org/stable/1913827. [34] Standard for the calculation and labeling of the energy performance of lowrise residential buildings using the HERS index; 2013. [35] Austin energy rates. URL: http://www.austinenergy.com/aboutus/rates/index. htm; 2013.