1 IMPROVING RESIDENTIAL MISCELLANEOUS ELECTRICAL LOAD MODELING By JOSEPH M. BURGETT A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTO R OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013
2 2013 Joseph M. Burgett
3 To my loving wife Jill
4 ACKNOWLEDGMENTS I would like to thank Dr. Chini, Dr. Sullivan, Dr. Oppenheim, Dr. Sriniva san and Dr. Ling for taking the time to serve on my committee and assist with this research. I would like to give a special thanks to Dr. Chi ni for being my guide and mentor throughout my academic career but soon my colleague. I would also like to give Dr. Sullivan special thanks for continually broadening my perspective and reminding me of what is important Thanks to Dr. Oppenheim, Dr. Srinivasan, and Ling for bringing their expertise to this research and guiding me in its development. In addition to my committee, I would also like to thank the M.E. Rinker, Sr. School of Building Construction staff for their support S pecial appreciation goes to all of the households who participated in the study. Wit hout your help the study would not have nearly as much depth or value. Special thanks to Jamie Bullivant at ThinkTank Energy Products Inc. for providing the Watts Up? Pro ES data loggers used in this study at a substantial discount. I would most import antly like to thank my wife, Jill, for all of her love and support. She has shared in the efforts to pursue my degree equally if not more. She is the glue that holds our family together and gives so much of herself without asking for anything in return. My wife Jill is by far the best decision I have ever made I am immeasurabl y blessed to have her in my life. I will never be able to pay her back for all that she has done for me but I look forward to spending a lifetime trying.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 11 LIST OF ABBREVIATIONS ................................ ................................ ........................... 13 ABSTRACT ................................ ................................ ................................ ................... 14 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 16 Energy Use in the Residential Market ................................ ................................ ..... 16 Problem Statement ................................ ................................ ................................ 17 Purpose of the study ................................ ................................ ............................... 17 Structure of Dissertation ................................ ................................ ......................... 18 Chapter Summary ................................ ................................ ................................ ... 19 2 LITERATURE REVIEW ................................ ................................ .......................... 21 Overview and Justification of Need ................................ ................................ ......... 21 Definitions and Nomenclature of Miscellaneous Electrical Loads ........................... 22 Past Studies Modeling MELs ................................ ................................ .................. 24 Moorefield Study ................................ ................................ .............................. 24 Moorefield study: preparation phase ................................ ......................... 25 Moorefield study: data collection phase ................................ .................... 26 Moorefield study: data analysis phase ................................ ...................... 27 Srinivasan Study ................................ ................................ .............................. 27 Energ y Simulation ................................ ................................ ............................ 29 Porter Study ................................ ................................ ................................ ..... 32 Porter study methodology ................................ ................................ .......... 33 Porter study results ................................ ................................ .................... 34 Current Methods of Calculating MELs ................................ ................................ .... 35 Home Energy Saver ................................ ................................ ......................... 36 Building America ................................ ................................ .............................. 37 Building America performance analysis procedures ................................ .. 38 Building America performance analysis procedures for existing homes .... 39 Building America Analysis Spreadsheets ................................ ................... 42 ELCAP study ................................ ................................ .............................. 43 Building America research benchmark definition ................................ ....... 44 Building America house simulation protocol ................................ ............... 45 Residential Energy Se rvices Network ................................ .............................. 47
6 Home energy rating system (HERS) index ................................ ................ 48 Parker study ................................ ................................ ............................... 49 Parker study refrigerators ................................ ................................ ........ 52 Parker study televisions ................................ ................................ .......... 53 Parker study ceiling fans ................................ ................................ ......... 53 Roth study ................................ ................................ ................................ .. 54 Roth study methodology ................................ ................................ ............ 55 California residential appliance saturation study ................................ ........ 58 Residential Energy Consumption Survey ................................ ................................ 59 Energy Data Sourcebook for the US Residential Sector ................................ .. 60 Energy data sourcebook for the US residential sector refrigerators ......... 60 Energy data sourcebook for the US residential sector televisions ........... 61 Energy data sourcebook for the US residential sector miscellaneous electrical loads ................................ ................................ ........................ 61 Developing and testing low power mode measurement methods .............. 62 Existing MEL Energy Efficiency Measures ................................ .............................. 62 Unplug the Appliance ................................ ................................ ....................... 63 Sma rt Power Strips ................................ ................................ .......................... 64 Smart Power Strips with Occupancy Sensors ................................ .................. 65 Timers ................................ ................................ ................................ .............. 65 Occupancy Sensors for Task Lighting ................................ .............................. 65 Energy Dashboards ................................ ................................ .......................... 65 Whole House Switch ................................ ................................ ........................ 66 Changing Occupant Behavior ................................ ................................ ................. 68 Chapter Summary ................................ ................................ ................................ ... 71 3 METHODOLOGY ................................ ................................ ................................ ... 83 Overview ................................ ................................ ................................ ................. 83 Objective ................................ ................................ ................................ ................. 83 Residential Energy Consumption Survey ................................ ................................ 84 Use of RECS to Calculate MELs ................................ ................................ ............ 84 Calculation for Appliances Addressed in the RECS ................................ ......... 84 Addressing Days Away from the Home in MEL Calculation ............................. 85 Calculation for Appliances Not Addressed in the RECS ................................ ... 86 Creation of Model through Stepwise Regress ion ................................ .................... 86 Statistical Software Package SPSS ................................ ................................ 87 Non linear Independent Variables ................................ ................................ .... 88 Validation of Model Overview ................................ ................................ ................. 88 Validation of the Model Survey ................................ ................................ ...... 89 Validation of the Model Data Loggers ................................ ........................... 91 Validation of the Model Monitoring of the Test Homes ................................ .. 92 Validation of the Model Compiling of Information Collected from Test Homes ................................ ................................ ................................ ........... 93 Validation of the Model Comparing Actual MEL with New Model Prediction ................................ ................................ ................................ ...... 93 Validation of the Model Defining Success ................................ ..................... 94
7 Chapter Summary ................................ ................................ ................................ ... 95 4 CALCULATION OF THE MELS FROM THE 2009 RECS ................................ ...... 97 Calcula tion ................................ ................................ ................................ .............. 97 Residual MEL Defined ................................ ................................ ............................ 97 Developing of the MEL Calculation ................................ ................................ ......... 98 Kitchen Appliances Microwaves ................................ ................................ .... 99 Kitchen Appliances Coffee Maker ................................ ............................... 100 Kitchen Appliances Toaster ................................ ................................ ......... 100 Home Entertainment Television Peripheral Devices ................................ ... 101 Home Computing ................................ ................................ ........................... 102 Rechargeable Po rtable Appliances and Electronic Devices ........................... 104 Well Water Pump ................................ ................................ ........................... 105 ................................ ................. 1 06 ..................... 106 Chapter Summary ................................ ................................ ................................ 107 5 RESULTS ................................ ................................ ................................ ............. 117 Overview of Findings for All Households ................................ .............................. 117 Analysis of Independent Variables ................................ ................................ ........ 120 Trending of Independent Variables ................................ ................................ 120 Descriptive Statistics ................................ ................................ ...................... 121 Testing for Significance ................................ ................................ .................. 123 Correlation of Independent Variables ................................ ................................ ... 126 Housing type ................................ ................................ ................................ .. 127 AIA climate zon e and Building America climate regions ................................ 128 Home size and number of household members ................................ ............. 129 Home Business ................................ ................................ .............................. 130 Income and Education ................................ ................................ .................... 131 Other independent variables ................................ ................................ .......... 131 Stepwise Regression ................................ ................................ ............................ 132 Errors in Variable ................................ ................................ ............................ 133 Stepwise Models ................................ ................................ ............................ 133 Multicollinearity of Model ................................ ................................ ................ 134 Components of Explanatory Equation ................................ ............................ 135 Example of Explanatory Equation ................................ ................................ .. 137 Hot Tubs ................................ ................................ ................................ ......... 138 Improved Standard Deviation Using the New Model ................................ ............ 139 Validation of Model ................................ ................................ ............................... 140 ................................ .......................... 140 Survey ................................ ................................ ................................ ............ 141 ................................ ................................ ....................... 141 Survey Response ................................ ................................ ........................... 142 Survey Response Income ................................ ................................ ........... 143 Skewness of the Test Houses ................................ ................................ ........ 144
8 Data Loggers ................................ ................................ ................................ .. 144 Comparing Recorded Energy Consumption with Published UEC data .......... 145 Calculating MEL for Test Houses ................................ ................................ ... 148 Similarity in Test Homes Where the HERS Model was the Better Predictor ... 149 Calibration of the New Model ................................ ................................ ................ 150 Statistical Review of Calibrated New Model ................................ ................... 151 Test Homes with Highest Deviation from Calibrated New Model ................... 152 Chapter Summary ................................ ................................ ................................ 153 6 WHOLE HOUSE SWITCH ................................ ................................ .................... 187 Overview ................................ ................................ ................................ ............... 187 Whole House Switch Defined ................................ ................................ ............... 189 Residential Energy Consumption Survey ................................ .............................. 190 A ssumptions Used to Calculate the WHS Energy Savings ................................ ... 191 Calculation for Appliances Addressed in the RECS ................................ ....... 192 Calculation for Appli ances Not Addressed in the RECS ................................ 193 Effectiveness of WHS ................................ ................................ ........................... 194 Testing of WHS Calculations ................................ ................................ ................ 195 Chapter Summary ................................ ................................ ................................ 197 7 CONCLUSIONS ................................ ................................ ................................ ... 203 Overview ................................ ................................ ................................ ............... 203 Key Conclusion #1: Occupant Characteristics are Better Predictors of MELs ..... 203 Key Conclusion #2: The New Model ................................ ................................ .... 204 Mod est Improvement in Accuracy for Modeling Average MEL ....................... 205 Significant Improvement in Individual Home Modeling ................................ ... 205 Hot Tubs ................................ ................................ ................................ ......... 206 Key Conclusion #3: Published UEC Values Compared to Recorded Consumption ................................ ................................ ................................ ..... 207 Future Studies ................................ ................................ ................................ ...... 208 Chapter Summary ................................ ................................ ................................ 209 APPENDIX A SURVEY ................................ ................................ ................................ ............... 212 B STATISTICAL ANALYSIS AND TERMINOLOGY ................................ ................. 225 C PEARSON CORRELATION TABLE ................................ ................................ ..... 231 LIST OF REFERENCES ................................ ................................ ............................. 232 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 242
9 LIST OF TABLES Table page 2 1 Building America goals for existing homes by climate type ................................ 72 2 2 .................. 73 2 3 Parker study television energy use table ................................ ............................ 74 4 1 Informa tion available from RECS ................................ ................................ ..... 109 4 2 UEC used in MEL calculation ................................ ................................ ........... 110 4 3 Toaster oven UEC values ................................ ................................ ................. 111 4 4 Weighted average of energy use by mode for set top boxes. ........................... 111 4 5 Weighted average rechargeable portable appliance energy use ...................... 112 4 6 Weighted average rechargeable electronic devices energy use ....................... 112 4 7 Appliance energy consumption in HERS Index as compared with RECS based model. ................................ ................................ ................................ .... 113 4 8 ................................ ..... 114 5 1 Appliances not included in RECS model but added to new model ................... 155 5 2 Occupant characteristics reviewed for significance ................................ .......... 156 5 3 Descriptive statistics of independent variables ................................ ................. 157 5 4 ANOVA table for all independent variables ................................ ...................... 158 5 5 Comparison of housing type with size and income ................................ ........... 160 5 6 Independent variables in stepwise regression model in order of highest to lowest explanatory power ................................ ................................ ................. 161 5 7 Model summary ................................ ................................ ................................ 162 5 8 ANOVA table for models A E ................................ ................................ ......... 163 5 9 Multicollinearity diagnostic ................................ ................................ ................ 164 5 10 Coefficients of models A E ................................ ................................ ............ 165
10 5 11 Estimation of the MEL of the RECS respondents using the HERS and new model ................................ ................................ ................................ ................ 165 5 12 Survey results: house characteristics ................................ .............................. 16 6 5 13 Survey results: occupant characteristics ................................ ......................... 167 5 14 Comparison between averaged recorded energy consumption and published UE C. ................................ ................................ ................................ ................. 168 5 15 Comparison between energy use of individual appliances and published UEC. ................................ ................................ ................................ ................. 168 5 16 Calculated MEL using new MEL model ................................ ............................ 169 5 17 New model and HERS model comparison with actual MEL from test houses .. 170 5 18 New Model in order of MEL magnitude ................................ ............................ 171 5 19 Calibrated new model and HERS model comparison with actual MEL from test houses ................................ ................................ ................................ ....... 172 5 20 Confidence interval for calibrated new model and test houses ......................... 173 5 21 Six test house with highest deviation from model ................................ ............. 174 5 22 Estimation of the MEL of the RECS respond ents using the HERS and calibrated new model ................................ ................................ ........................ 174 6 1 Hours assumed that the WHS would be activated ................................ ............ 198 6 2 Power usage information for appliances included in RECS .............................. 198 6 3 Power usage information for appliances not included in RECS ........................ 199 6 4 Average MEL and WHS s avings potential from RECS data ............................. 199 6 5 Effectiveness of WHS by occupant group ................................ ........................ 200 6 6 Validating WHS with test houses ................................ ................................ ...... 201 7 1 Correlation between occupant characteristics and MEL ................................ ... 211
11 LIST OF FIGURES Figure page 2 1 Sample installation of Watts Up? Meter ................................ .............................. 75 2 2 ................................ ................................ .............. 75 2 3 Interior residential equipment profile ................................ ................................ ... 76 2 4 MELs normalized energy use profile ................................ ................................ .. 76 2 5 Equation 3 and table 303.4.1.7.1(1) from 2006 Mortgage Industry National Home Energy Rati ng System Standards ................................ ............................ 77 2 6 Residual miscellaneous electrical loads from Parker study ................................ 78 2 7 Ceiling fan diversi ty profile ................................ ................................ .................. 79 2 8 Energy consumption by the miscellaneous electrical load from R oth study ....... 80 2 9 Red Jacket Water Products 12G series pump sizing cha rt ................................ 81 2 10 Distribution of RECS calculated MEL ................................ ................................ 82 3 1 Watts Up? Pro data logging device ................................ ................................ .... 96 5 1 Breakdown of calculated MEL for all RECS respondents by percentage ......... 175 5 2 Comparison between the age of the home and MEL ................................ ........ 176 5 3 Distribution of MEL standardized residuals ................................ ...................... 176 5 4 Homogeneity of variances for all predicted MEL values ................................ ... 177 5 5 Comparison between the type of home and MEL ................................ ............. 178 5 6 Comparison between AIA climate zone and MEL ................................ ............. 178 5 7 Comparison betw een Building America climate region and MEL ..................... 179 5 8 Comparison between home size and MEL ................................ ....................... 179 5 9 Comparison between home size and individual MEL appliances. .................... 180 5 10 Comparison between the total number of rooms in the home and MEL ........... 181 5 11 Comparison between n umber of bedrooms in the home and MEL ................... 181
12 5 12 Comparison between number of bedrooms in household and household members ................................ ................................ ................................ .......... 182 5 13 C omparison between number of people in the household and MEL ................ 183 5 14 Comparison between households with home business and MEL ..................... 183 5 15 .................. 184 5 16 Comparison between marital status and MEL ................................ .................. 184 5 17 Comp arison between the age of the householder and MEL ............................. 185 5 18 Comparison between the number of children in the household and MEL ......... 185 5 19 Comparison between households with members home during weekdays and MEL ................................ ................................ ................................ .................. 186 5 20 Comparison between the householders retirement status and MEL ................ 186 6 1 Z Wave enabled disconnectors ................................ ................................ ........ 202 6 2 Z Wave enabled control switches ................................ ................................ ..... 202
13 LIST OF ABBREVIATIONS BA Building America program DOE US D epartment of Energy EIA US Energy Information Agency HERS Home Energy Rating System MEL Miscellaneous Electrical Load RECS Residential Energy Consumption Survey RESNET Residential Energy Services Network WHS Whole House Switch
14 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy IMPROVING RESIDENTIAL MISCELLANEOUS ELECTRICAL LOAD MODELING By Joseph M. Burgett May 2013 Ch air: Abdol R. Chini Major: Design Construction and Planning Over the past 30 years, the intensity of all major energy use categories has decreased in the residential market with the exception of miscellaneous electrical loads (MELs). MELs include primar ily 120V plug loads and some hard wired loads. MELs stand alone as the only category in which energy intensity has steadily increased over time. While MELs constitute approximately 15% 25 use it is project ed to increa se to 36% by 2020. Despite the significant percentage of are the least researched energy end use category and most poorly modeled. The Home Energy Rating System ( HERS ) index is the most widely used residential energy rating sy stem and uses a simple square foot multiplier to model MELs. This study improves upon the HERS model by including occupant characteristics a s and explanatory equation from the Energy Infor Consumption Survey (RECS). The RECS has a very large sample size of 12,083 respondents who answered over 90 pages of questions related to home structure, appliances they own and demographical information. The informatio n provided by the respondents was used to calculate a MEL f or all the RECS households. A stepwise
15 regression process was used to create a model that included size of t he home, household income number of household members and presence of a home business t o predict the MEL. The new model was then tested on 24 actual homes to compare its predictive power with the HERS model The new model more closely predicted the actual MEL for 17 of the 24 test houses (71%). Additionally, the standard deviation or the on the RECS respondents. What this study found was that using occupant characteristics to supplement a square foot multiplier significantly increased the precision of MEL m odeling
16 CHAPTER 1 INTRODUCTION Energy Use in the Residential Market systems than in all other previous generations combined. Of these impacts, perhaps the most significant is g lobal climate change led primarily from the release of carbon dioxide ( CO 2 ) and other greenhouse gasses. The United States (US) is only behind China in their CO 2 emissions. The US Environmental Protection Agency reports that the generation of electricity is the leading source of CO 2 emissions and contributes 40% of all US greenhouse gas e s ( US EPA, 2013) The residential market has a high demand for electrical energy consuming nearly 38% of all US production (US EPA, 2013) While a significant consumer o f electricity, the residential market also has significant opportunity to increase its energy efficiency and reduce overall consumption. The residential market is the area of interest for this dissertation study. Energy use in a home is generally categori zed into six major end uses; heating, cooling, lighting, water heating, large appliances and miscellaneous electrical loads (MELs). The energy needed to heat and cool a home depends largely on climate type but typically uses approximately one third of the home s total energy use. Lighting and water heating also are significant energy uses each using between 10 and 20 percent of clothes washers/dryer s typically consume 15% of mentioned. Plug loads, hard wired loads, and all uncommon energy uses are captured in this category (Nordman and Sanchez, 2006)
17 Problem Statement Over the past 30 years, the intensity of all major energy use categories has decreased in the residential market with the exception of MELs. MELs stand alone as the single category in which energy intensity has steadily increased over time. Wh ile MELs constitute approximately 15% government sponsored study projects that this will increase to 36% by 2020 (Roth et al., MELs are the least researched energy end use category and most poorly modeled. Current practices use limited physical characteristic of the home such as square foot area and number of bedrooms to estimate MELs (RESNET, 2012; Hendron and Engebrecht 2010 ) This study will seek to improve current modeling practices by including occupant characteristics as part of the calculation. The primary objective of this research is to answer the question: nt practices of Purpose of the study The overarching purpose of this research is to support the efforts to reduce our dependence on fossil fuels and the reduction in global greenhouse gas emissions. Eliminating th e need for nonrenewable energy ranks as one of the most challenging and complicated problems facing our society. While this research will of course not solve this problem or even address it as a whole it will target a specific area of residential energy u se that is in great need for additional research. The first step in the solving of any problem is to understand it and the interactions of its parts. Furthering the state of knowledge about what influences the intensity of MELs and the interaction with t he
18 householder is the principle contribution of this study. This new knowledge will be u sed to better predict MELs. From a practical standpoint, better prediction s of MELs ha ve advantages at the city and individual home level. Better energy model s can l ead to improved projections of future energy demand from new developments. This can lead to better sizing of new power generation facilities. Another large scale advantage is that it can help improve the timelines of climate change models by more accurat ely predicting energy consumption. Perhaps more significantly, this research will more accurately predict MELs at the individual home level. When MELs are quantified, energy reduction programs such as the HERS rating system, can incentivize builders to d evelop measures to reduc e them. Also, if the goal of large scale development of zero net energy homes is to be realized, accurate forecasts of energy use must be created to correctly size solar panels or other on site power generation. For hyper efficien t homes, MELs can represent 50% or more of the total energy use so accurate models are critical ly important While modeling MELs is a critical first step in the process the reduction of energy use is the next logical step in the process. To further these efforts, a secondary objective of this dissertation is to evaluate a poorly understood energy efficiency measure (EEM) and determine its energy saving potential. Specifically, this study will evaluate the effectiveness of the Whole House Switch (WHS) EEM and determine its energy savings and cost effectiveness. Structure of Dissertation This dissertation is structured to provide the reader first with an overview of the current state of MEL modeling, then a detailed description of the research conducted in t his study and finally a summary of the conclusions that can be made from th e findings
19 Chapter 2 contains a summary of the relevant literature. Specifically, past studies on MEL modeling, appliance energy consumption, usage patterns and the current pract ice for modeling MELs is discussed. Chapter 3 provides a description of the methodology used to conduct this research. Topics such as the source of data, calculation of MELs, creation of the model and validation of the model are included. This study too k advantage of a large federal government survey to make statistical inferences about MELs. The actual MEL was not provided so a MEL value was calculated based on the information in the survey. The specifics of how the MELs were calculated are provided i n Chapter 4. Chapter 5 provides the reader a summary of the sta tistical inferences and trending of the data observed. This c hapter also shows the results of how well the new model predicted MELs on actual test homes with comparisons to current modeling p ractices. The whole house switch EEM was a secondary objective of the study and is summarized in Chapter 6. Chapter 7 is the conclusion chapter and summarized the key findings of the study. This c hapter also highlights gaps in the available data and pro vides recommendations for future studies. Chapter Summary The dependence on fossil fuel energy is one of the most significant challenges facing our society. Recognizing that current practices cannot be sustained indefinitely, scholars and practitioners are working together to develop ways of reducing our need for nonrenewable energy. This study advanced the effort by better understanding MELs in the residential market. This study showed the correlations between the occupant and their MEL intensity ulti mately leading to an improved way of modeling the load The chapters that follow provide the specifics of this dissertation study. The dissertation
20 takes a unique approach to modeling and provides valuable insights into an important area of research.
21 C HAPTER 2 LITERATURE REVIEW Overview and Justification of Need Over the past 30 years the intensity of all major energy use categories has decreased in the residential market with the exception of miscellaneous electrical loads (MELs). MELs stands alone as the single category in which energy intensity has steadily increased over time (Groves, 2009; KEMA, 2010; Moorefield et al., 2011; Nordman and Sanchez, 2006; Parker et al., 2011; US EIA, 2011d). The rapid expansion in the markets of home entertainment, p ersonal electronics, and convenience s low ing (Fanara et al., 2006; Groves, 2009 ; Brown et al., 2007 ; Roth et al., 2008a). Occupant behavior is the most critical component to MELs but remains one of the least researched areas of residential energy use (Hendron and Eastment, 2006; Lee et al., 2011; Masosa and Grobler, 2010; Nordman et al., 2006; Samarakoon and Soebarto, 2011; Steemers and Yun, 2009; Wang et al., 2011; Yamaguchi et al., 2011). MELs constitute approximately 15% 25% of a code compliant home energy use (Barley et al., 2008; Ecos Consulting, 2004; Porter et al., 2006; Roth et al., 2006; Sanchez et al., 1998; US DOE, 2012; Wentzel et al 1997). However, the MEL pe rcentage can be in excess of 50% for high efficiency homes (Barley et al., 2008; Brown et al 2007; Hendron and Eastment, 2006; Nordman and Sanchez, 2006; Steemers and Yun, 2009). Additionally, according to a report commission by the US Department of Ene rgy (DOE), MELs will grow to 36% of code compliant homes by 2020 (Roth et al., 2008b). The modeling and reduction of MELs is a key area of research for national energy reduction and goals for zero net energy homes
22 Definitions and Nomenclature of Miscel laneous Electrical Loads Before delving into past studies it is important to define the nomenclature of MELs so the specifics of th is dissertation are effectively communicated. When the science of building energy was in its early years the research effort s naturally gravitated to the areas of highest intensity. These areas are most commonly referred to as appliances. Studies were actual monitoring was performed typicall y measured the whole building consumption and the traditional end uses leaving the difference to be and Sanchez, 2006). As the research into miscellaneous loads lagged behind the traditional end uses so did standardized terminology. In the 1990s and 2000s MELs began to get more attention as their percentage of total energy use began to rise. Studies were conducted and the results published but each study needed to define the terminology used to avoid Electronics Come of Age: A Taxonomy for Miscellaneous and Low Power Products p rovide[d] definitions for key terms and concepts with the intent that future work can be more correctly and consistently reported and interpreted (Nordman and Sanchez, 2006). Nordman and Sanchez paper has been widely accepted as the defining work for M EL definitions and nomenclature (Dawson Harggerty, 2012; Groves, 2009; Meier et al., 2007; Moorefield et al., 2011; Southern California Edison, 2008). This research will use the terminology as defined by the Nordman and Sanchez (2006) paper with a minor r dissertation are defined below:
23 Product ly from main power (Nordman and Sanchez, 2006). A product is one specific appliance or piece of equipment. The product will have a specific manufacturers and model number. A product is not a line of appliances or equipment that serves similar function. Toasters as a group would not be defined as a product but a specific toaster manufactured by General Electric with a model number of 123 45 67890 would be. Product Type amount of common functionality, modes, and behavior (Nordman and Sanchez, 2006). In the example above, toaster s as a group would be defined as a product type. Traditional End Uses (Nordman and Sanchez, 2006). Plug Loads Any load that is powered by a normal AC receptacle. This is inclusive of major applianc es and electronics. Miscellaneous Electrical Loads (MELs) The Nordman and Sanchez paper This research will slightly dissertation MELs will be defined as all plug loads and hardwire loads not included with the traditional end uses, secondary refrigerators/freezers and televisions. Examples of MELs include microwaves, peripheral television products, security systems, spas, computers, and rechargeable consumer products.
24 Past Studies Modeling MELs Although MELs are one of the least researched areas of residential energy uses th is studies have been performed that have laid the ground work for this research. The me thodology of data collection from the Moorefield study is very similar to this research although the data collected was specific to commercial offices. The Srinivasan study used the approach of estimating MEL as a function of diversity factors and product type nominal power usage. The Porter study provides actual residential product energy use of 50 single family homes in California. The following paragraphs provide a summary of each of these studies. Moorefield S tudy One of the more significant past stu dies related to this research was a report 2011). The Moorefield study is one of the few studies to actually measure plug loads on a significant scale. The study was specific to California office buildings so the information dissertation; however the methodology used is of particular interest. The overarching obje ctive of the study was to document the type of plug loads being used in California office buildings, record the power use information of the plug loads and graph the usage patterns. The various steps to complete their
25 Moorefield study: p reparation p hase The first step in the Moorefield study was to identify and to prioritize the plug loads being included with the study. The taxonomy developed b y Nordman and Sanchez (2006) was use d to define the product list. Using existing literature the team develop ed a prioritization list of which devices were to be measured. Each plug load on the list was given a prioritization rating of High, Medium, Low or Do Not Meter. Plug loads that were expected to have an overall high energy use were given a higher priority to be monitored over plug loads expected to have an overall low energy use. Devices whose duty cycle or energy consumption was not well underst ood and devices with potential savings through automation were also given a high priority. The study specifically excluded large appliances (like refrigerators, dryers, ect.), air conditioning equipment and all hard wired loads. These devices were given on the prioritization schedule. The team allocated 60% of the available meters to The next step in the Moorefie ld study was to select participants for their study. To make the results as applicable to the widest range of users a diverse group of offices was recruited to participate in the study. Geographic factors like urban versus rural were considered along wit h the type of professional services being provided. Office size in terms of employees and square footage was also considered. The recruiting process included sending an initial letter of interest followed up with phone calls. The phone calls confirmed t he eligibility of the office and then a monetary incentive was offered to participate. Ultimately 47 participants were recruited all evenly distributed across the factors previously mentioned.
26 The third step in the Preparation Phase was the selection of t he metering device use d to collect the plug load data. The Moorefield study specifically listed the following characteristics required for the study: Ability to meter data from a single plug load and not an entire circuit Ability to record data at one min ute intervals Ability to store information for multiple weeks User friendly interface Wide range of power recording (<1 Watt up to 1,800 Watts) High accuracy Timely consumer availability ? ee Figure 2 1. The Watts Up Pro ES met all of the criterion mention ed above and was within their set price point. The study employed the use of 120 data loggers to collect 480 million data point from 470 products. Moorefield study : d ata c ollection p hase The first step in the Data Collection Phase was to perform walkthroughs of was to inventory all of the plug type s could be identified. Two researchers would physically observe each plug load This procedure was completed in 22 office buildings. The next step was to perform a similar buildings and create a list of all of the plug loads being used. Based on this list a subset of plug loads to be monitored was identified based on the prioritization schedule created in the Preparation Phase. The selected plug loads were monitored for two weeks with a
27 recording interval of one minute. Watts, volts, amps, volt amps, power factor and maximum wattage were all recorded. Moorefield study: d ata a nalysis p hase There were five main goals that the resear chers wanted to address with the data collected. The first goal was to determine the type of plug load devices being used in California offices. They were able to use the information collected in all 47 offices (22 ces and 25 from the metered offices) to create their list. The second goal was to determine the average power demand by mode (on, stand by, or off) of the product types. The third goal was to determine the percentage of time during the two week monitorin g period that each device spent in the various operational modes. The fourth goal was to determine the total energy used in each mode over the two week period and extrapolate annual energy use. The fifth goal was to track the real time energy consumption of the metered devices. Srinivasan Study This Srinivasan study was a comparison of existing methods for calculating plug load densities in K 12 schools with eighteen benchmark schools under two new categories; with computers and without computers (Srini vasan et al. 2011a). One of the existing approaches being reviewed is the ASHRAE standard 90.1 which is the reference guide for the US ASHRAE guide a flat 5.38W/m 2 is assumed for all plug loads in K 12 schools. This plug load intensity assumption has not changed since its original publication in 1989 (ASHRAE 90.1 1989). The value has been criticized as the saturation of computers, projectors, multimedia teaching tools and many other electronics have si gnificantly increased since 1989. This flat value also ignores specialty schools which place higher
28 emphasis on science and technology and as a result have a higher plug load density. The Srinivasan study calculates plug load density as a function of nom inal power of equipment and the diversity factor. It then compares it with existing approaches to estimate plug loads such as the ASHRAE standard previously mentioned. Other standards compared with the 18 benchmark schools are COMNET Guidelines, NREL, an 24. The Srinivasan study was conducted is six distinct steps. The first step was the selection of the schools to be included with the study. The study intentionally reviewed buildings that were either constructed or expected to be co nstructed after 2011 to include the modern designs that incorporate current teaching practices and technologies. Of the 18 schools included, nine were elementary, 2 were middle and seven were high school. The study included the plug load density of the c lassroom area only. The second step was the review of construction documents of the 18 schools to quantify the expected equipment being used. The equipment included projectors, laptops, desktop computers, smart boards, LCD monitors, VCR/DVD players and s urround sound systems. Approximately 30% of the classrooms surveyed contained computers. As expected the study found that the presence of computers dramatically influenced the plug load density and prompted the study to segregate the classrooms into two separate categories; with and without computers. In the third step plug load density from the 18 benchmark schools were calculated. Average energy consumption of the product type was collected from literature review. As mentioned earlier the plug load d ensity is based on nominal power consumption (determined through step 2 and the literature review of product data) and diversity factors. This
29 study used a flat diversity factor of .77 which is based on the recommendations of the COMNET Guidelines. The C significant assumption made by the researchers (Srinivasan, et al., 2011a). Detailed diversity factors are not available for K 12 schools. If diversity factors specific to K 12 schools were available a mor e accurate plug load model would have be en possible (Srinivasan, et al., 2011b). Step four evaluated the plug load density from the 18 benchmark schools with that of four existing approaches (COMNET Guidelines, ASHRAE 90.1, Title 24 and NREL). In step fi ve the existing approaches were compared with the benchmark schools with two types of classrooms; with and without computers. Step six involved creating an energy model using computer software. Separate models using all four existing approaches were crea ted and compared with the new benchmark values. The study found that the total energy use for schools was within 5% of the benchmark using the ASHRAE 90.1 plug load density values. The conclusion made was that while the saturation of new technology and e quipment has increased, efficiency of the equipment has increased at the same pace. Energy S imulation One of the ways that the Srinivasan study is different than the Moorefield study and the Porter study (which will be discussed shortly) is the use of ener gy simulation software. The Srinivasan study used the software package eQUEST which models the MEL as a function of kWh per square foot. In practice, MELs are rarely modeled alone and are only one piece of the entire energy model. Because this research can be used to enhance whole building energy models it is important to understand how energy modeling evolved to its current state.
30 Energy simulation is commonly thought of as a product of the computer age but its origins date back to the early 1920s. Andre Nessi and Leon Nisolle of France were the first to develop a calculation of transient heat flow known as the Response Factor Method (RFM) (Nessi and Nisolle, 1925). This method remained a widely accepted means of manually calculating heat flow into the 1960s (Brisken and Reque, 1956; Hill, 1957; Holden, 1963; Muncey, 1963; Pipes, 1957; Stewart, 1948). In the late 1960s the manual RFM calculation started to be replaced by computer programs which automated the process (Haberl and Cho, 2004; Kusuda, 19 69; Kusuda, 1970; Kusuda, 1971; Kusuda, 1974; Tupper et al., 2012). The RFM method is still used in energy modeling software today including the DOE 2.2 energy simulation engine. The DOE 2 platform evolved from several previous energy modeling programs from the 1970s. These programs included the Heating and Cooling Peak Load Calculation (HCC) program developed by the Automated Procedures for Engineering Program and the US Nati Cost Analysis Program (NECAP). Other packages such as the US Army Construction Thermodynamics (BLAST) program and Univers ity of Wisconsin Systems Simulation program (TRNSYS) were similar but developed independently. The California Energy Commission in partnership with the US Energy Research and Development Administration and the Lawrence Berkeley National Laboratory (LBNL) modified NECAP for civilian use and repackaged it as DOE 1 in 1978. The following year several upgrades were made and it was reissued as DOE 2 in 1979. The 1980s
31 was a time of improvement and refining of the existing software packages. BLAST was upgraded to BLAST 1.2 and then to 2.0. TRNSYS had updates nearly every year. Similarly DOE 2 was rereleased as DOE 2.1a, b, c, and d from 1981 to 1989. The (Tu pper 2012). In partnership with the US DOE and LBNL, James J. Hirsch and 2.1 which was renamed DOE 2.2 (International Building Performance Simulation Association, 2012). There was a disagre ement between JJH and the US DOE on how the program should be licensed which ultimately led to the partnership being dissolved. JJH continued to develop the DOE 2.2 and created the DOE 2.2 interface program eQUEST in 1999. JJH is still supporting and upd ating DOE 2.2 and eQUEST which are both available free to the public. The US DOE reallocated its funding to the development of a new energy simulation program called EnergyPlus. EnergyPlus built on the BLAST program whose funding was cut by the US Depar tment of Defense in 1995. Although the US Federal government financially supported the development of DOE 2.2 and EnergyPlus several been developed and are still widely used tod ay (Haberl and Cho, 2004; International Building Performance Simulation Association, 2012; Tupper, 2012). The next phase in energy modeling will be the integration with Building Information Modeling (BIM) software ( US GSA, 2012; Haberl and Cho, 2004; Int ernational Building Performance Simulation Association, 2012; Malin, 2008; Tupper, 2012). The use of BIM software enables the user to build a structure in the virtual world before it is built in the real world. BIM combines all of the engineering and arc hitectural
32 disciplines into a single computer file. BIM creates a 3D model of the proposed building where virtual walkthroughs can provide insights in the future functionality of the building. The 3D model can be extraordinarily detailed and allows for a much more collaborative and multidiscipline review of the design. Although the technology to fully integrate BIM with energy modeling software has not been completely developed, the industry is moving in this direction. Autodesk purchased the BIM softwar e developer Revit Technologies in 2002 and the energy modeling software developer Green Building Studio in 2008. Autodesk now provides Revit Conceptual Energy Analysis which can use either DOE g a similar path the BIM software developer Bentley Systems Inc. purchased Hevacom Ltd. in 2008 to provide an energy modeling component to its software (Malin, 2008). BIM resembles the building energy modeling programs of the 1980s in that its computing p ower and acceptance in the industry are rapidly expanding ( US GSA, 2012; Haberl and Cho, 2004; International Building Performance Simulation Association, 2012; Malin, 2008; Tupper, 2012). Porter Study A very significant past study related to this research was conducted by Ecos Consulting for the California Energy Commission The Porter study dential consumer] products was simply unavailable (Porter et al., 2006). The Porter study was conducted loads (Porter et al., 2006). Many of the
33 procedures, data type collected and analysis of this dissertation are similar to the Porter study. Porter study m ethodology The first phase of the Porter study was to develop their list of sampled homes. The researchers had access to a list of approximately 900 homes that had previously participated in an energy research conducted by Eco Consulting (authoring company). The 900 homes were reasonably well distributed throughout the state of California This pool of households was spec ifically targeted because it was felt that the households would be more likely to participate in additional energy research studies than randomly selected ones These 900 homes were contacted by phone and asked to participate in a survey. Survey responde nts were incentivized by being offered $10 for their participation. If the survey respondents met the stud criteria then they were asked if they would be willing to participate in the onsite MEL monitoring phase Participants were offered $100 for eve ry site visit made. Each participant would receive a minimum of $200 to compensate them for the initial set up of the monitoring devices and then the retrieval of the equipment. Ultimately 50 homes were selected to be continually monitored. Being able to loads that had a high overall energy use either through high individual intensity or through market saturation. Liter ature review was used to determine which plug loads had high expected energy use and that their usage patterns were not well understood. HVAC equipment and major appliances ( white goods ) from the study. Four categor ies were created to prioritize the MELs being monitored.
34 The prioritization schedule ranked the various MELs as High, Medium, Low or Zero Priority. Based on literature review entertainment items such as televisions, set top boxes and video game consoles were expected to have a high energy draw. Additionally, information technology devices such as computers, printers and wireless routers were also expected to have a high energy demand. Both entertainment items and information technology devices were give n Devices with a rechargeable battery such as cell phones, Bluetooth headsets and electric toothbrushes such as coffee makers, radios, clocks, and wi ne coolers These Priority With the sample homes identified and the prioritization schedule created the Porter study team was prepared to start the plug load monitoring. Teams of three would visit the house and install the data logger s. The quantity of MELs ranged between the various homes in the study. As a result not all home required the same number of data loggers to record the available information. On average, 17 data loggers were used per house. However, some homes used as f ew as five and others as many as 35 depending on the quantity of plug loads on the prioritization schedule found in the home. The plug loads were monitored for a seven day periods and data were logged in intervals of 1 minute. Data loggers from Brand El ectronics were custom designed to meet the specification of the study ( Figure 2 2 ) Electricity billing data were also collected from the home owner. Porter study r esults The Porter study found that on average homes in the study used between 1,069 and 1,2 07 kWh per year for their MELs costing the home owner approximately $150
35 Energy Consumption Survey (RECS) this represents approximately 9% of a typical US home but is mo re that 15% of an average California home. The Porter study provides detailed charts and graphical profiles of power averages by mode, duty cycle percentages and annual energy use for over 70 product types. A key contribution that the study made is devel oping weekday and weekend energy use profiles for entertainment equipment, information technology equipment and all plug loads combined. Current Methods of Calculating MELs Miscellaneous electrical loads in homes are typically estimated in one of two way s. The first way is for the user to provide detailed information about what types of MELs are in the home and how often they will be used. A program is typically used to provide the homeowner with an extensive list of appliances. The homeowner will then indicate which of the appliances are used and approximately how often. The program will have default energy consumption values for the appliances selected so that it can calculate the total load. The program will have typical usage schedules, appliance saturation data, and energy use characteristics to use as defaults if the user does not input the home specific data. The US (http://hes.lbl.gov/consumer/) is a web based energy program which uses this approach and will b e described in more detail later (LBNL, 2012; Mills, 2008) This type of program can produce very accurate results because there is a high level of user specific data. Use of programs with detailed input information is more energy accounting than energy modeling and not the goal of this research.
36 The second way in which MELs are estimated is more in line with the research goal of this dissertation. If user specific MEL data is not known they are typically modeled with generic formulas based on national s urveys and consumption data. The (HERS) index (Fairey et al., 2006). The Energy Star program, LEED for Homes, the Florida Green Building Coalition and the EPAct 2005 federal t ax credit for high efficiency new homes all use the HERS index as the basis for their energy performance requirements. The HERS index is being highlighted because it uses a generic formula method for calculating MELs. A second major home efficiency progr am that uses a generic formula for calculating MELs is the Building America program. Both the HERS this c hapter. Home Energy Saver The Home Energy Saver website (http:// hes.lbl.gov/consumer/) is an interactive site designed by the Lawrence Berkley National Laboratories (LBNL) to help home owners make informed decisions about energy use (Mills, 2008) The Home Energy Saver website assumes the user is a layperson and start s with a minimum number of user inputs to generate a projected energy use profile and recommended energy efficiency measures (EEMs). However, the website allows for more detailed inputs to increase the accuracy of the results. If requested inputs are not available default The way that MELs are address ed by the Home Energy Saver website is by taking a list of the most common MELs and calculating their total load. This cal culation takes into consideration mode, duty cycle and maximum energy draw. The heat load is
37 also considered using hourly profiles and monthly load factors. The study uses approximately 75 MELs with detailed usage profiles to create the Home Energy Saver default MEL value. The usage profiles and power information came from several published studies (McMahon and Nordman, 2004; Ross and Meier, 2000; Sancehez et The total lo ad is then multiplied by the market saturation to calculate a total MEL value. The user can override any of the individual appliances in the list of 75 MELs as well as adjust the usage pattern to increase the accuracy of the energy prediction (LBNL, 2012) Building America The Building America (BA) program is a department within the US DOE. The program is an industry driven research group with the charter of improving American comfort ( US DOE, 2011a). develop market ready energy solutions that improve efficiency of new and existing US DOE, 2011b). The program takes advantage of national laboratories and r esearch teams to advance the adoption of improved energy efficiency technologies in all climate types within the US Efficiency improvements are evaluated at the individual and systems level and range from individual homes to entire communities. The specif ic by 2013 and 50% by 2016 in the Hot/Humid climate type ( US DOE, 2011b Table 2 1 ) To be able to reach these aggressive targets the BA program needed to first establish w America House Simulation Protocol (HSP) (Hendron and Engebrecht, 2010). The HSP
38 evolved from three key BA publications; Building America Performance Analysis Procedures (2004) Building America Performance Analysis Procedures for Existing Homes (2006) and Building America Research Benchmark Definition (2009) (Engebrecht and Hendron, 2010). These documents will be summarized in the paragraphs below. Building America performance analysis procedures The Building America Performance Analysis Procedures (BAPAP) was the first of three publications that lead to the current HSP and dealt exclusively with new construction. BAPAP established a baseline to measure the energy efficiency o f year, average whole building energy reduc The benchmark established with the BAPAP was very similar to the 1999 Home Energy Rating System (HERS) Reference Home (Fairey et al., 2006). Many standard building features such as duct leakage, water heater efficiency, and indoor lighting levels were provided. Fix hourly energy usage profiles were also established which were not permitted to be to describe mo re energy efficient designs to be compared with the benchmark (Hendron et al., 2004). The BAPAP provided instructions for how to calculate MELs. Major appliances such as refrigerators, clothes washer and dryers, dishwashers and ranges were disaggregated out of the MEL category. The energy use of each of these major appliances was calculated independently. The non major appliance MELs were lumped
39 and MELs is not just their dir ect energy use but the heat that they generate. The BAPAP established a sensible load fraction of .9 and a latent load fraction of .1. (These fractions are updated in future benchmark revisions.) The BAPAP established a rudimentary means of calculating and function solely o f the square footage of the home. (The current HSP uses square footage, number of bedrooms and a state multiplier to calculate plug loads.) Based on ree noteworth y that this is almost twice the current HERS index MEL model which uses .91kWh/sqft. A single hourly sched Lawrence Berkley National Laboratory study published in 2002 (Huang and Gu). Figure 2 3 demonstrates the In terior Residential Equipment profile used in the BAPAP (Hendron et al., 2004). Building America performance analysis procedures for existing homes The Building America Performance Analysis Procedures for Existing Homes (BAPAPEH) is the second of three docu ments in the evolution of the HSP. The BAPAPEH is in response to the energy savings potential of 101 million existing residential households and the BA goal of achieving 20 % reduction in energy use by 2015. The BAPAPEH is the compl e ment to the BAPAP that was focused on new construction The purpose of using BAPAPEH estimating the energy savings achieved by a package of retrofits or an extensive
40 rehabilitation of an existing home (Hendron, 2006). Although actual occ upant behavior will impact home energy uses significantly BAPAPEH is limited by providing standard typical behavior profiles to be used in the benchmark and prototype energy models (Hendron, 2006). Similar to the BAPAP, the BAPAPEH provides baseline buildi ng inputs to be used in the benchmark model. Building characteristics such as R values of insulation, air conditioning unit efficiencies and window fenestration are examples of the building inputs that are provided in the report. A key difference between the BAPAP and the BAPAPEH is that the latter provides varying inputs depending on the age of the structure. For example the benchmark air conditioner unit will have an efficiency of EER 7.5 if buil t between 1981 and 1991 or an EER 6.5 if built before 1981 De rating factors are also available to adjust the benchmark to reflect the frequency that maintenance was performed or the natural deterioration of building materials (Hendron, 2006). A key difference between the two reports is how they calculate MELs. The first and category into two new categories; Plug In Lighting and Miscellaneous Electric Loads. an d .66 and .02 respectively (formerly .9 and .1 in BAPAP). Perhaps the most significant change between the two reports with respect to MELs is the development of new hourly profiles. The BAPAP used a single hourly schedule for all major appliances and plug loads whereas the BAPAPEH provides individual profiles for refrigerators, clothes
41 washers and dryers, dishwashers, cooking ranges and MELs. The MEL profile is an aggregate for 100+ small plug loads commonly found in homes (Hendron, 2006). The large appliance profiles were based on the End Use Load and Consumer Assessment Program (ELCAP) study which will be discussed later (Pratt et al., 1989). However, the MEL ho urly profiles were derived by subtracting out the large appliance profiles from a theoretical all electric, 1,800sqft, 3 bedroom house in Memphis, Tennessee. The report on a residual, it is susceptible to greater systematic errors and may be less realistic than the profiles for major appliances (Hendron, 2006). This is a significant short coming of the BAPAPEH study but is addressed in the current HSP supported by the Mills study (2008). The Mills study will be addressed later in this c hapter (Hendron, 2006). The BAPAPEH provided for one additional improvement to how MELs are calculated. The new report revised the formula for how to calculate the gross MEL energy usag e. The previous version used a set value (1.67) that was multiplied by the finished floor area of the home. BA recognized that some loads were near constant despite the size of the house or number of occupants. Refrigerators are an example of this const ant load. BA also recognized that number of people in a home also impacted homes (Hend ron and Eastment, 2006). Additionally, BA was able to quantify that MEL intensity was also a function of geographic location. The new MEL calculation formula published in Hendron report accounted for all three of these factors. The formula published is s hown below. The present HSP uses the same three factors but the actual
42 multiplier values are different (HSP: MEL all electric = 1703 + 266 x N br + .454 x FFA) (Hendron, 2006). BAPAPEH : MEL = (2803 + 0.316 x FFA + 194 x N br ) x F s Where: MEL = miscellaneous electric loads for the pre retrofit case (kWh/yr) FFA = finished floor area (ft 2 ) N br = number of bedrooms F s = state multiplier (New York = .82, California = .77, Florida = .94, Texas = 1.11, all others = 1.00) Building America A nalysis S preadsheets As an alternative to using the MEL formula described above the user could elect to calculate MELs based on the BA Analysis Spreadsheets ( US DOE, 2011a; US DOE, 2011c). Using the Analysis Spreadsheets is an option not available under the BAPAP report and is a tool that gives the user much more flexibility in calculating expected MELs. The Analysis Spreadsheets contain product information and occupancy behavior assumptions. It uses this information to calculate the MELs for benchmark and high performance ener gy models. The MEL calculations are based on a list of over 120 of the most common appliances ( US DOE, 2011c). The appliance data and usage rates are based on nation wide studies and the averages of large sample sizes. However, of these 120 items less t han 10% of them are calculated entirely using studies provide an approximation of the exp ected MELs of the average home, updated and statistically defendable consumption data is clearly needed to increase the accuracy of
43 ELCAP study As mentioned earlier the ELCAP study was a critical addition to the BAPAPEH st udy. The Bonneville Power Administration is a utility provider in the Pacific Northwest. end use load data [from] residential and commercial buildings (Pratt et al., 1989 ). This particular study focused solely on existing single family, detached, site built, owner occupied houses with permanent electric space heating equipment. The base study included 288 homes that were selected from 4,000 randomly surveyed customers o ver a period of four years (1984 1988). At the time of the study most end use studies HVAC and hot water load was deducted from the total consumption and the residual was lump 11 different end use categories. The categories specifically included heating, hot water, cooling, ranges (cooking), refrigerators, clothes dryers, lights, special major appl iances (ie hot tubs, kilns, and workshops), freezers, clothes washers and dishwashers. The monitoring of the devices was done at the circuit level. Circuit level monitoring has the advantage of being more cost effective and convenient then individual app liance monitoring but it exchanges this convenience with less reliable data. Several of the appliances listed are on dedicated circuits so the integrity of the data is maintained. These appliances included the HVAC unit, water heaters, ranges, clothes dr yers and most special major appliances. The HVAC unit is actually an aggregate of cooling and heating but if summed together is isolated from the other end uses. The ELCAP study However, the remainin g devices likely have other devices on the same circuit and contaminated the data collected. This
44 same circuit unquestionably influenced the usage profiles. The BA PAPEH uses this study as the bases for the hourly profiles for the refrigerators/freezers, clothes washers, clothes dryers, dishwashers, and ranges. With the exception of the clothes dryer category, all of the data used to create the appliance profiles ar e in the category (Pratt et al., 1989). Building America research benchmark definition The Building America Research Benchmark Definition (BARBD) was the third of three publications that built the foundation for the current HSP. The BARBD furth ered the program in two distinct ways. The first was that it increased the accuracy and expanded the scope of building characteristics and occupancy schedules of the benchmark and prototype models. The second was where the BAPAP and BAPAPEH focused solel y on single family houses, the BARBD also included detached single family housing as well as multi family housing (Hendron and Engebrecht, 2009). In terms of MELs the BARBD had some significant additions from the previous two studies. One addition include d the hourly profile for a common laundry room to facilitate multi family buildings. However, all large appliances for attached and detached single family homes remained the same as what was indicated in the BAPAPEH report. A second significant differenc e in the way the BARBD deals with MELs is that it created three formulas to calculate MELs instead of only one indicated in the BAPAPEH (MEL = [2803 + 0.316 x FFA + 194 x N br ] x F s ). The three formulas divide MELs into Variable MELs, Fixed Misc Loads for an all electric house and Fixed Misc Loads for a mixed gas/electric house. The formulas are shown below (Hendron and Engebrecht, 2009).
45 Variable MELs = (1281 + 196 x N br + .345 x FFA) x F s Fixed Miscellaneous Loads (All Electric) = (319 + 53 x N br + .083 x FFA) x F s Fixed Miscellaneous Loads (Gas/Electric) = (150 + 25 x N br + .039 x FFA) x F s Where: MEL = miscellaneous electric loads for the pre retrofit case (kWh/yr) FFA = finished floor area (ft 2 ) N br = number of bedrooms F s = state multiplier (New York = .82, California = .77, Florida = .94, Texas = 1.11, all others = 1.00) Building America house simulation protocol The BAPAP, BAPAPEH and BARBD are all steps in the evolution to the Building America HSP. The HSP is intended to be the definitive protoco l for baseline and high rd multi year, whole partners perform design tradeoffs and calculate energy savings from homes that are built/remodeled as part of the program (Engebrecht and Hendron, 201 0). The protocol is divided into three sections. The first deals with physical components of new construction consistent with standard building practices in 2010. The second section is related to the physical components of existing homes. The final se ction provides operating conditions for both new construction and existing homes (Engebrecht and Hendron, 2010). Although many of the concepts and parameters from the BAPAP, BAPAPEH and BARBD were carried into the HSP several significant changes were made. Some of the nomenclature changed specifically with baseline and new high efficiency designs. High
46 reflect the energy code it incorporates. The benchmark before the HSP was based on a mid 1990s home however the B10 benchmark is based on the 2009 IECC energy code. Large appliances, lighting and MELs were also revised from a typical mid 19 90s home to 2010 standards. The HSP provides a fairly comprehensive list of building characteristics and parameters to be used in energy models (Engebrecht and Hendron, 2010). With relation to MELs the HSP advanced the previous three NREL studies in four key ways. First, the MEL formula was adjusted again. In the BARBD study MEL s were calculated by using two of three available formulas (Variable MELs and Fixed Misc Loads for an all electric house or Fixed Misc Loads for a mixed gas/electric house). The HSP eliminated the Variable MEL formula and incorporated it into the two remaining formulas which are shown below. Miscellaneous Loads (All Electric) = (319 + 53 x N br + .083 x FFA) x F s Miscellaneous Loads (Gas/Electric) = (1595 + 248 x N br + .426 x FFA) x F s Where: MEL = miscellaneous electric loads for the pre retrofit case (kWh/yr) FFA = finished floor area (ft 2 ) N br = number of bedrooms F s = state multiplier (New York = .82, California = .77, Florida = .94, Texas = 1.11, all others = 1.00) The second important change made in the HSP with relation to MELs is the adjustment of the sensible and latent load fraction s The HSP used a .734 and .2 sensible and latent load fraction s, respectively for both all electric and mix electric/gas homes. A third impo rtant contribution was the introduction of seasonal multipliers for major appliances and MELs. Before the HSP all loads were applied uniformly
47 throughout the year. The newly added seasonal multipliers adjust the loads to more accurately reflect actual oc cupant behavior. These multipliers were derived from the Mills study which will be discussed later in this c hapter (Mills, 2008). A fourth significant improvement made in the HSP with respect to MELs is the further disaggregation of MELs into smaller subc ategories with additional individual hourly profiles. These subcategories include freezer, home entertainment, waterbed The hourly profiles for all of these appliances w ere created from the Mills study (2008) with the exception of the ceiling fan profile which was derived from a Florida Solar Energy Center study (Parker et al., 2011). Both of these studies will be reviewed later. The MEL hourly profile found in the HSP is shown in Figure 2 4. The numerical data can be found in the BA Analysis Spreadsheets ( US DOE, 2011c). Two categories of MEL that are notably not disaggregated in the HSP are small kitchen appliances and home computing devices and represent a significa nt percentage of residential MELs (Engebrecht and Hendron, 2010; Roth et al. 2008b). Although not addressed in this hourly profiles for MELs (Engebrecht and Hendron, 2010). Res idential Energy Services Network In the early 1980s, the mortgage industry began to understand that there was a pay their mortgage. The industry created the National Sh elter Industry Energy Advisory Council to quantify the financial savings from an energy efficient home and to include this with the home owners lending capacity. Although well intentioned, the council could not get energy mortgage programs to be widely ac
48 uniform method of efficiency evaluation (RESNET, 2012). With partners from the National Association of State Energy Officials and the mortgage industry, the council was re created into the current Residential Energy Services Network (RESNET). The home energy ratings and to create a market for home energy rating systems and energy mortgages (RESNET, 2012). he Home Energy Rating System (HERS) score which evolved into the current HERS index. Home energy rating system (HERS) index The HERS index is a point based rating system that measures the energy which has a score of 100 (Fairey et al., 2006). Each point represents one percentage point of how energy efficient the home is. A home that is 15% more energy efficient than a standard code compliant home would receive a score of 85. A net zero energy home would receive the lowest possible score of zero. The home rating is provided by as a benchmark to compare the rated home. The HERS Reference Home is defined by the protocol outlined in the 2006 Mortgage Industry National Home Energy Rating System Standards (HERS standards) and subsequent revisions (RESENT, 2006). In previous RESNET rating standards only heating, cooling and hot water use w ere consider ed. Traditionally energy codes also limited what they address to these three end uses (Fairey et al., 2006). However, the 2006 HERS standards included appliances and lighting. Specifically, indoor lighting, outdoor lighting, refrigerators, clothes dryer s, clothes washers, televisions, dishwashers, ceiling fans, ranges and ovens, and residual MELs were all individually accounted for. Equation 3 and Table
49 303.41.1.7.1(1) found in C hapter 3 of the HERS standard provide the specific method for calculating t Reference Homes (all electric). Figure 2 5 exhibits an excerpt from the HERS standard. The HERS Equation 3 is found below : kWh per year = a + b*CFA + c*Nbr Where: values provided in Table 303.4.1.7.1(1) CFA = conditioned floor area Nbr = number of bedrooms The above e quation has three coefficients and two user inputs. The user inputs ultiplied by either of the user inputs. The HERS standard assumes that all homes have a fairly consistent minimum fixed MEL. This minimum fixed MEL is then adjusted up based on is best defined by everything that is not otherwise listed in the HERS standard, is calculated entirely based on the size of the home. The supposition of this research is that this is a shortcoming of the HERS standard as at a minimum there is an obvious link between number of occupants and how often smaller appliances and plug loads are used. Parker study An important step of this research is to underst and the specifics behind how Table 303.4.1.7.1(1) was created with particular interest in the 0.91 value for Residual The values of the coefficients have evolved since the introduction of the HERS standards but the current revision i s based on a study conducted jointly by
50 FSEC and NREL referred to here as the Parker study (Parker et al., 2011). The Parker study used two primary data sources to make statistical correlations between MELs and the three coefficients described above. The two primary sources were the 2005 RECS and a report commissioned by the US DOE and conducted by TIAX, LLC referred to here as the Roth study. The Roth study will be reviewed later in this c hapter. Both the RECS (2009) and the Roth study were also primar y contributors to this dissertation All nine categories of MELs in Table 303.4.1.7.1(1) of the HERS standard are directly addressed in the Parker study and the recommendation s provided were accepted without any alterations. All of the categories, with th and number of occupants with total energy use. RECS data was used to estimate the market saturation and proliferation of particular appliance charac teristics such as size, type, efficiency level and duration of use for the eight MEL categories. The RECS did not provide any direct information about the energy consumption of any specific appliance. The Roth study, as well as several other studies, wer e used to fill in this missing information for this dissertation Using the RECS, statistical inferences were drawn between size of the home and number of bedrooms. The 2005 RECS had a large sample size of over 4,000 survey respondents and was a good sam ple to represent the US housing stock. It is worth mentioning that the 2009 RECS expanded the sample size to over 12,000 survey respondents. Although eight of the nine MEL categories were calculated using very similar egory was calculated very differently. Using data from the Roth study, the Parker study listed the 26 highest consuming MELs with an
51 Figure 2 6 is a copy of Table 2 from the Parker study and shows all 27 items. The unit electrical consumption (UEC) provided in Figure 2 6 is the total annual energy consumption including the various modes of that product type. The saturation is how many of the product types are in each household. The Roth study was the pr imary source of UEC and saturation data. It is important to note that the saturation came from the 2001 RECS which is currently two generations out of date. All of the items in Figure 2 6 are specifically addressed in the 2009 RECS with the exception of clothes iron, vacuum cleaner, hair dryer, water bed, security system, clock radio and other miscellaneous. The UEC is then multiplied by the saturation to determine the average energy consumption for that product type. s also included which is approximately 20% of the whole consumption (329kWh/1,714kWh). Table 2 2 provides the breakdown for how 2 is not provided in the Parker study but was provided directly by con tacting the authors. All 27 items are summed together with an additional 10% added for expected advancements in peripheral home electronics and entertainment devices. The total Residual MEL is 1,714 for an s a 1,900sqft home with 2.8 bedrooms based on 2005 RECS and Census data. The average MEL load of There does appear to be a discrepancy between the data provided in the R oth study and the referenced source data. Figure 2 6 indicates that the UEC for well pumps is 862kWh and references the Roth study (TIAX Report) as its source. It is believed that the UEC published in the Parker study actually came from the California Re sidential
52 Appliance Saturation S tudy (RASS Volume 2) which estimates the UEC for single family houses to be 862kWh (KEMA, 2010). The Roth study does not address well pumps directly but republishes the findings from a report from Arthur D. Little Inc ( ADL) (Zogg and Alberino, 1998). ADL and subsequently the Roth study published the UEC for well pumps as 83.4kWh. Later in this c hapter when the ADL findings are discussed it will be noted that the 83.4kWh seems quantitatively low H owever it is far less than what is shown in the Parker study. The saturation of well pumps is relatively low (17%) performed for the well pump and described in Chapter 5 Parker study refrigerators The Parker study did a detailed review of the 2005 RECS to understand what impact home refrigerators have on the total energy used in a home. The study found that nearly every home in the US has a refrigerator and that there is a great deal of variation in energy use in this subcategory. The data showed that on average a US home uses 1,360kWh while modern refrigerators use 800kWh/year and units manufactured before 1980 used more than 2,000kWh. Two high level conclusions were drawn from thei r review of the RECS. First, 30% of single family detached homes use a secondary refrigerator. Second, many of the secondary refrigerators are older and less efficient than the primary refrigerator. The study indicated that over 50% of secondary refrige rators were at least 10 years old (Parker et al., 2011). The study provided a regression analysis and derived correlations between number of bedrooms and refrigerator quantity and size. An important distinction between the BA program and RESNET is that B A includes secondary refrigerator in MEL whereas RESNET excludes it.
53 Parker study televisions The Parker study also provided a calculation for estimating the electrical load for household energy use is consumed with televisions. After a review of the 2005 RECS the study statistically confirms two fairly intuitive conclusions that are important in the study of MELs. First, the primary television is typically larger and second is used more often than the other televisions in the home. However, number of televisions does not influence energy consumption in a meaningful way. Television use profile and what type of unit it is are far more influential to total energy consumed than th eir numbers As actual viewing hours and energy use of the television are not often known, the study uses a regression analysis to derive a correlation between number of bedrooms in a home and television energy use. The correlation is based on the 2005 R ECS which surveyed the number of televisions and asked survey participants to approximate their televisions viewing habits. Actual data logging of the individual appliance would likely have produced more accurate results but was not included with the RECS protocol. Table 2 3 demonstrates the total estimated power draw from television use as estimated by the Parker study. Similar to refrigerators, the BA program and the RESNET differ in their inclusion of televisions as a MEL. The BA program includes it as a MEL but RESNET accounts for it separately in their MEL formula and does not consider it a residual MEL. Parker study ceiling fans One of the goals of the Parker study was to better approximate the energy use of ceiling fans. Based on the 2005 RECS, 69% of homes have at least one fan. This is an increase of 27% from the 1997 RECS data. Based on the national survey there were
54 approximately 2.9 ceiling fans per home. This number is greatly influenced by geographic region (South 3.2 fans, Midwest 2.8 fans, Northeast 2.6 fans, and West 2.4 fans). Fans typically have variable speeds which influences energy use. The study, based on previous FSEC papers, estimated that fans on low speed used 10 20W, medium speeds 25 45W and on high speeds 75 95W. Energy use is of course influence d by the efficiency of the model. Energy Star rated fans are required to be a minimum of 20% more efficient than a comparable unit. (James et al., 19 96). The FSEC fan study monitored the use of fans in 400 Florida homes to determine if there was an impact on thermostat set points. The study found that ceiling fans were used 13.5 hours during week days and 14.2 hours on weekends. On average one third of the homes in the study reported leaving their fans on 24 hours a day. Based on the FSEC study the Parker study generated the ceiling fan diversity factor indicated in Figure 2 7. Ceiling fan is another example where the BA program includes the applia nce whereas RESNET calculates it separately from their residual MELs. Roth s tudy from the Parker study. The Parker study retrieved UEC and market saturation almost exclusively fr om a study commissioned by the US DOE referred to here as the Roth study (Roth et al., 2008b). The US Department of Energy has a long standing interest in MELs. This is particularly true for their Building Technology Program that has set a research goal of developing cost effective net zero energy homes by 2020. In pursuit of this goal the US DOE commissioned TIAX, LLC to draft a report to provide the current
55 Miscellaneous Electric Loads: Energy Consumption Characterization and Savings Potential in 2006 and Scenario Based Projections for 2020 (Roth et al., 2008b). The study breaks down MELs into 21 key product types and 9 secondary product types (Figure 2 8 ). The Roth stud y is a significant source of data for this research as it provides detail product type information which specifically includes average household electricity consumption for MELs, energy consumption per mode, market saturation, number of devices per home, a nd nominal power usage. The Roth study evaluates product type energy consumption in two ways. The first way that MELs are calculated is by taking the total MEL value and dividing it by the 115 million households in 2006. The study calls this the average household electricity consumption (average HEC). The second method for calculating MELs was to calculate secondary MELs like pool pumps or waterbed heaters. This method is cal led the typical household electricity consumption (typical HEC). For example the average HEC method would include 2.4 televisions and .03 water beds per home. The typical HEC method would calculate the MEL value as two televisions and zero water beds. T he two methods produced very similar results with a total difference of only 4%. Roth study methodology Th is dissertation used the same basic calculation for quantifying the total electrical load for each of the product types as the Roth study. First the annual usage pattern is multiplied by the power draw by mode to determine the devices annual Unit Electricity Consumption (UEC). The UEC is then multiplied by the residential stock to calculate the annual energy consumption. To determine the annual energy
56 consumption at the house level or household energy consumption (HEC) the number of product types per home is substituted for the residential stock number. Residential stock is the number of devices in all US residential buildings and was approximated usi (Roth et al., 2008b). Estimated power draw will vary by mode. Most of the product types reviewed in the Roth study ha d at least two modes, on and off, but many ha d a third low power / stand by mode. The Roth study estimated the power draw for each mode as part of its calculation. The power draw values came from a variety of measurement reports and a limited number of past TIAX, LLC studies. The usage patterns refer to the ho urs per week that a device operates in a particular mode. measurements of resid ential MEL usage patterns exist (Roth et al., 2008b). Most of the usage patterns used in th e Roth study came from past consumer research which device usage patterns [w ill] have the greatest uncertainty of any components to the [annual energy consumption] calculations for most MELs (Roth et al., 2008b). In addition to the 21 key product types and 9 secondary MELs the Roth study other appliances (Roth et al., 2008b). These initial estimates use other studies as the bases for their UEC calculation and then t UEC data available (Roth et al. 2008b). Both the HERS index and the BA analysis spreadsheet use the UEC data for these 71
57 appliances extensively. One appliance which was expected to have a high UEC but stood out as quantitatively low was the well pump The energy consumption data for US DOE in 1989 (Zogg and Alberino). The ADL study assumed that a residence would re quire 10 gpm flow rate at a pressure between 30 and 50 psi. No information is provided on the assumed well depth or pipe size. The pump size was also not provided but the study indicated that they assumed it to have an average energy consumption of 725W and be used 115hr/year (19minutes/day). The wattage and time and Alberino, 1998). The other estimates that the ADL report references are from a study by L BNL (Sanchez et al., US EIA, 1995) which had 380% and 173% higher UEC values respectively. The ADL study does not indicate the well depth assumed H owever The Water Systems Co uncil indicates that most wells are between 100 and 500 feet in depth (2003). The well pipe size is also not provided H owever for a 10gpm system (indicated in the ADL study) a 1 Service, 1979). Assuming a well depth of 300 40psi (service pressure = ), pipe ( ) 1/2 horse power (Red Jacket, 2012 Figure 2 9 ) The horse power for the well pump in the ADL study is not provided but the 725W best describes a horse power pump
58 (Xylem, 2012). Energy c the studies available vary widely in their UEC estimates. The review of the literature found that the UEC estimates for well pumps are as low as 83.4 (Roth, 2007; Zogg and Alberino, 1998) and as high as 862kWh (KEMA, 2010; Parker et al., 2011) with several studies in between (Meier and Greenberg, 1992; Sanchez et al., 1998; US EIA, 1995; Willson and Morrill, 1990). Despite the range, based on the calculation shown above the UEC findings of the ADL study seem to underestimate the energy consumption of the average home owner. Because of this, the well pump UEC recommended by the ADL study and repeated in the Roth study was not used in this dissertation in its calculation of the MEL s The revise d UEC and how it was used in the MEL calculation will be elaborated on in C hapter 4. California residential appliance saturation study The Roth study use d an extensive list of sources to collect their data H owever three major sources were the California Residential Appliance Saturation Study (RASS), spreadsheets. The RASS was sponsored by the Califo rnia Energy Commission and provides energy consumption data on 27 MELs. The data was collected by surveying over 24,000 California residents. The survey provided information on a large range of topics but specifically collected data on housing type, age of occupant, income of occupant, number of various product types, and usage patterns of product types. Although the data provided is specific to California, many extrapolations for a nation wide study can be made. The information of particular relevance for this research is the market saturation of product types, the hours used by mode and energy draw by mode
59 of product types. The RASS also provides statistical comparison between the occupant characteristics and product types energy usage (KEMA, 2010). Residential E nergy C onsumption S urvey Residential Energy Consumption Survey (RECS). Since 1978, the EIA has been collecting and publishing residential energy use data. The REC S use occupant the residential market ( US EIA, 2011a). This research will be using the mo st current 2009 RECS data. The RECS are published every 4 years so the next revised set of data will be from the survey conducted in 2013. Although the RECS information is used to represent the nation as a whole the surveys have traditionally only been co nducted in the four most populous states; Florida, Texas, California and New York. The number of states was increased in 2009 to include 16 states. In addition to the four most populated states, Pennsylvania, Illinois, Michigan, Georgia, New Jersey, Virg inia, Massachusetts, Tennessee, Arizona, Missouri, Wisconsin and Colorado were included with the survey. Based on the information collected the additional 12 states increased the accuracy of the survey to approximate the energy use of 63% of all homes and 64% of all US citizens ( US EIA, 2011b). The information is collected by interviewing randomly selected housing units. The housing unit sample set is first created by randomly selected counties. The selected counties are then divided into census blocks called segments. The segments are then randomly selected in which randomly selected housing units are then picked to
60 be interviewed. Interviewers trained by the EIA are then sent to the selected homes to personally interview the occupant. Each interview er is equipped with a laptop computer in which they record the answers to a standardized questionnaire. In addition to the occupant questionnaire the local utility company is also surveyed to collect specific data on the homes whose occupant was interview ed. The utility company provides the EIA with total quantities of energy used from electricity, natural gas, fuel oil, and propane as well as pricing information ( US EIA, 2011c). Energy D ata S ourcebook for the US R esidential S ector Although the informatio n may be dated, there is value in exploring the studies that the BA Analysis Spreadsheet used to generate its MEL estimate. The current version of the spreadsheet referenced 22 data sources. One of the key studies the BA Analysis Spreadsheets use is the Energy Data Sourcebook for the US Residential Sector (Wenzel et al., 1997). The purpose of this US DOE funded study was to create a single resource to provide residential energy modelers a comprehensive data source for all end use Unit Energy Consumption (UEC). The report specifically acknowledges its end use energy demand forecasting in the residential sector (Wenzel et al., 1997). The study specifically provides end use data for heating and cooling, water heating, refrigerators, freezers, dishwashers, clothes washers, clothes dryers, lighting, cooking, televisions and miscellaneous loads. For the purpose of this literature review the analysis of the work will focus o n the plug in loads (Wenzel et al., 1997). Energy data sourcebook for the US residential sector refrigerators The Wenzel et al. study found that there was a great deal of variation in the amount of electrical energy consumed with refrigerators. The vari ation comes from the
61 many differences in make, model, size and number of times the door is opened. In addition, many home owners own a second refrigerator which also skews the data significantly. The data for the study was collected in 1990 and at that t ime the average power consumed for a common refrigerator ranged from 1,144kWh/year to 1,338kWh/year. This data is not relevant for this dissertation due to the significant increase in performance required by current appliance standards. The study indicat es by 5 to 20 percent (Wenzel et al., 1997). Energy data sourcebook for the US residential sector televisions This study was published in 1997 and at this time the use of televisions accounted for 5% of all residential energy use (Wenzel et al., 1997). The Wenzel study found that the average color television consumed 513kWh/year. The amount of energy used for televisions will have great variation from one home to the n ext. Factors like size, energy efficiency, number of sets and hours on per day provide a significant range in power consumption. This study used the same assumptions that the Department of Energy used for television usage which was that the set was in ac tive use for 6hrs/day and in stand by mode for 18hr/day. At the time of the study there was an average of two set per household and the life expectancy of the unit was 11 years. Energy data sourcebook for the US residential sector miscellaneous electric al loads The Wenzel et al., study indicates that in 1995, 25% 33% of all residential energy use came from unspecified end uses or Miscellaneous Electrical Loads (MELs). A significant finding of the study was the trending of the MELs. The total resident ial
62 energy consumed for MELs in 1980 was 96TWh but in 1995 this figure nearly doubled to 180TWh. Developing and t esting l ow p ower m ode m easurement m ethods In 2002 the California Energy Commission sponsored a study by the Lawrence Berkeley National Labora tories to determine the quantity of residential energy being used by appliances in the stand by mode (McMahon and Nordman, 2004). The study measure d the power draw in stand by mode for 280 devices located in 8 single family homes. The sample size of home s was relatively small because the purpose of the study was to be a pilot study to organize and plan out a larger and more encompassing future study. Despite the limited scope of the study the finding indicate d that stand by power accounted for an average of 1,000kWh/year. The Roth study used its data in its estimates of various appliances UEC and was also used by this dissertation to generate a comprehensive list of residential plug loads. Existing MEL Energy Efficiency Measures Current practices for red ucing MELs fall into the two general categories of technical or behavioral (Mohanty, 2001). Perhaps the most obvious of the technical improvements is increased energy efficiency from advances in technology. Old style CRT computer monitors for example use twice the energy to operate than modern LCD monitors and four times as much as Energy Star rated units (Roth et al., 2008b). Equipping units with sleep or low power modes is another improvement that many appliances have adopted. However, having the opti on of a low power mode is only effective if the owner chooses to use it. Many manufacturers have encouraged reduced energy use by having low power setting enabled as the factory default. Providing energy guide labels that inform consumers of expected ene rgy cost has been very
63 successful with major appliances. Having smaller appliances list their energy consumption would similarly use market pressure to decrease energy use ( US IEA, 2001). Changing the behavior of the occupant is another way of reducing M ELs. Opower is a company that partners with utility providers to analyze households and provide them with comparisons between their energy use and their neighbors. Participation in the Opower program has yielded an average savings of 2.8% of the homes to tal energy consumption with some municipalities experiencing over a 6% decrease (Parker et al., 2011). Home automation through the use of timers and occupancy sensors is another way of reducing MELs. Energy dashboards and smart meters are devices or netw orks of devices that provide the householder real time feedback on their energy use. Several nationwide studies have shown that when householders are provided this instant feedback total energy consumption is reduced from 5 15% (Parker et al., 2011). S tand by power is the energy used by an appliance when it is not functioning or is in the off mode. It is sometimes called a phantom load, vampire draw or leaking electricity. Existing literature indicates that stand by power accounts for 2 23% of a hou Meier and Huber, 1997). This dissertation will test the effectiveness of a measure that saves energy by reducing stand by power. However other stand by power EEMs are available. Unplug the Applia nce Perhaps the most basic way of reducing standby power draw is to simply unplug the device when not in use. Although absolutely effective, this has not been widely embraced because of the inconvenience of continually plugging and unplugging appliances. This is especially true for home entertainment and home computing
64 equipment where the electrical receptacle is not always easily accessible. Unplugging appliances can be made more convenient by using a surge protector with an integrated switch where seve ral appliances can be disconnected all at once. However, here again easy access can still be an issue. Smart Power Strips Many home owners have found another way of reducing MELs by using smart power strips. Smart power strips look very similar to typica l home surge protectors. Smart power strips are used in locations where several related appliances are operated at the same time such as home entertainment and home computing areas. Each strip The other receptacles are for the peripheral equipment. Most models also include receptacles that are uncontrolled. The smart power strips can detect when the master appliance is in standby mode mostly from the reduced Wattage. When the reduced Wattage is d etected it shuts power off to the peripheral equipment. According to the Building America analysis spreadsheets the primary television is in standby mode for 16.9hrs/day. Using the power values provided by the analysis spreadsheets, if a VCR (4.5W/hr), c ompact stereo (7W/hr), power speakers (2W/hr) and the cable box (15W/hr) where plugged into a smart power strip the standby power reduction would be 176kWh/year. Assuming the Florida average electric rate of $.1165/kWh there would be an annual savings of just over 20 dollars a year. The Bits Limited 10 Outlet Energy Saving Surge Protector cost important clarification to make is that in this calculation the cable box was incl uded and had the highest standby draw. Many cable boxes have lengthy start up times which may make this EEM too inconvenient to justify the savings.
65 Smart Power Strips with Occupancy Sensors In the same vein as the smart power strips are the smart power s trips with occupancy sensors. These power strips use occupancy sensors instead of master appliances to determine when the peripheral outlets should be disconnected. The WattStopper Plug Load Power strip cost $90 on the WattStopper website. Timers The us e of timers is another way of reducing stand by power draw. Timers have the advantage of being relatively low cost (5 15 dollars) and can be specific to an individual appliance. To provide an order of magnitude of effectiveness assume a microwave use 3 W/hr in standby mode and you set the time to disconnect power from 11:00pm to 6:00am. That would provide an annual savings of 7.7kW/year and a simple payback period of 6 years. Occupancy Sensors for Task Lighting Depending on how you define MELs task ligh ting may or may not be included. However, occupancy sensors that screw into the Edison socket can help reduce wasted task lighting energy with very minimal installation effort. The inactivity timer is fully adjustable by the user. Energy Dashboards Energ y dashboards are electronic devices that track real time energy consumption. Most residential monitoring systems are wireless with one or more central monitoring stations. These systems monitor overall consumption at the main electrical panel with 2 curr ent transformer (CT) rings that clamp around the two leads into the panel. For individual plug loads, wireless current transmitters plug directly into a receptacle and monitors usage. All of the data collected is displayed on monitoring
66 stations with the intention of allowing the householder to make better use of their energy with real time consumption feedback. These stations usually provide current electrical consumption in terms of kWh and cost as well as the potential to provide historical consumptio n information (Wilson, 2008). There are no shortages of studies on the potential savings form Energy Dashboards. Nationwide studies indicate that providing this real time feedback will save the householder between 5% and 15% of their total energy use ( Bo nino et al., 2012; Darby, 2006; Parker et al., 2011; Roth and Brodrick, 2008; Stein, 2004; Wiggins et al., 2009). The area that is most undefined in energy dashboard studies is how occupant behavior impacts building energy performance (Lee et al., 2011). Changing occupant behavior will be discussed later in this c hapter. Whole House Switch Another EEM that householders have available to them is the Whole House Switch (WHS). The WHS is sometimes called the whole house control switch or master kill switch and as its name suggest control multiple devices from a single control point. The energy savings can come from two ways; reduction in stand by power and reducing the time when appliances are on but not being actively used. There are two types of WHS that will be elaborate d on. The first type of WHS will be referred to here as the load WHS As the name suggest the plug load WHS is limited to plug in type appliances. The basic setup has two components; disconnectors and a controller. The disconnec tor is plugged into an electrical receptacle and the appliance is then plugged into it. The disconnector acts as a switch to sever power to the appliance and is commanded by the controller. Multiple disconnectors can be programmed to the same controller. As the
67 installation is very simple with no modifications to the home required the plug load WHS is ideal for retrofits and renters. The circuit level WHS provides energy savings in a similar way as the plug load WHS but saves additional energy in 3 distinct ways. First, oftentimes non plug loads are left on mistakenly. A classic example is ceiling fan use. In the Parker study (one of the studies used in the Building America House S imulation Protocol), 400 Florida homes were monitored and found that over one in three homes left fans on 24hrs a day. The circuit level WHS will also reduce stand by power in hardwire loads such as GFCI receptacles. Each GFCI uses approximately .7W/hr t o operate (Engebrecht and Hendron, 2010). There will also be savings from other hardwire loads such as doorbells, intercoms and other low voltage systems. The third and perhaps largest advantage is that it will be able to reduce standby power loss on mor e appliances with fewer disconnectors. For example one circuit level disconnect in the kitchen could sever the power for the microwave, coffee maker, toaster, and portable radio instead of requiring a plug level disconnector for each appliance. There are two main disadvantages to the circuit level WHS. First, circuit level disconnectors are more involved in their installation and are beyond the abilities of most householders to install. This makes them more suited for new construction than retrofits. T e nergized but are on the circuit. For example, if the microwave clock is a service that the home owner is not willing to give up than no app liance on the circuit would enjoy any energy reduction.
68 Both the plug and circuit level WHS can be implemented using off the shelf technology. One possible line of products that could accomplish this is Z W ave compatible products although there are others such as Zigbee. Z W ave (and Zigbee) is a proprietary protocol that allows for multiple manufactures to be c ompatible with one another. Z W ave smart relays are available to shunt residential circuit breakers thus de energizing entire circuits. The Z W av e switches look like a standard wall switches and fits inside a single gang electrical junction box. Z W ave receptacle disconnectors look just like a normal receptacle but are controlled by the master Z W ave controller. Similarly there are plug s in disco nnectors that look like common surge protectors. There are also Z W ave compatible thermostats that can be linked to the system. Each Z W ave device is also a repeater. This means that once it receives a signal it repeats it for other Z W ave devices to re ad. As you add devices the larger your net work becomes making distance less and less of an issue. By comparisons with the other traditional end uses, there are relatively few EEM for miscellaneous electrical loads. The most effective way of reducing thes e loads is by improving the performance of the appliance through programs like Energy Star and EPEAT, reducing stand by power loss and changing occupant behavior. Changing Occupant Behavior The Parker et al., 2011 study points out that there are three wa ys in which MEL s can be reduced; individual component efficiency improvements, energy management systems (automation) and occupant behavior changes. Individual component efficiency is perha ps the most obvious and requires the various MEL s to operate more efficiently while provid ing the same level of services. The Energy Star program has been very successful at encouraging manufacturers to pro vide consumers more energy efficien t
69 appliance options. Energy management systems or automations use occupant sens ors, timers and programmable switches to mimic the behavior of an energy conscious householder. In the absence of an energy conscious householder savings can be seen. However, building automation systems will never be as efficient as a householder activ ely and effectively managing their own energy consumption (Bonino et al., 2012) Occupant sensors may turn the lights off after a few minutes of inactively but an energy conscious householder w ill turn the light off immediately. Programmable thermostats may save energy over a constant set point but will never operate as efficiently as a person who can adaptively adjust conditioning demand in real time Forcing automation on an uninterested householder will not have significant energy savings. To illustr ate, the programmable function of programmable thermostats, whose energy saving potential is well documented, are only used 58% of the time (Meyers et al., 20 10 wasted by poor ener gy management (Meyers et al., 20 10 ). The key in reducing energy consumption is by changing occupant behavior (Lee et al., 2011) Perhaps the EEM with the greatest im pact on householder behavior is the energy dash board. As stated earlier in this c hapter, energy dashboards have a history of reducing whole home energy consumption by 5 15% ( Bonino et al., 2012; Darby, 2006; Parker et al., 2011; Roth and Brodrick, 2008; Stein, 2004; Wiggins et al., 2009) Having the information is important in making energy efficient decisions, however; energy. P.C. Sterns stated that the energy use information must be credible, useful in ve at getting the attention of the
70 the informational content given, but the way in which the information motivates the consumer into action that is important (Wood and Newborough, 2007) How to motivate people to willingly behave in a desired manor is a subject discussed in more than a few management books, seminar s and college courses. The effectiveness of any method will be as diverse as the person on which it is imp lemented on. However, several motivational techniques are particularly applicable to energy savings. Monetary reward in the form of direct payment or differed energy cost is perhaps the most obvious motivational technique. Another motivational factor is emotional reward where a householder would feel good about making an environmentally responsible decision (Wood and Newborough, 2007) Katzev et al., 1983 found that emotional rewards do not significantly influence energy use patterns however People ca n be motivated to reduce energy consumption through competition (Bonino et al., 2012). Utility providers often send out notices to their customers comparing their consumption to their neighbors. Several studies have found that goal setting is also an eff ective way of motivating the householder to energy reduction targets (Bonino et al., 2012; Wood and Newborough, 2007). desire to conform to social norms can also be a motivating factor. At the Florida Museum of Natural History in Gainesville, Florida there is a kiosk with a digital map where the average power consumption of all local homes are displayed. Energy intensive homes are shown in red, while moderate homes are displayed in yellow with the lowest energy consuming homes highlighted in g reen. Public displays such as th is can be used to praise more efficient homes and discourage less efficient ones.
71 Chapter Summary Historically, energy intensity as a function of Watts/sqft has steadily decreased in all of the traditional end uses with the exception of miscellaneous electrical loads. MELs stands alone as the only end use that has continued to increase. The expansion of home entertainment, personal electronics, and convenience items are key contributors to this increase in energy use. Alt hough MELs represents between 15% poorly modeled of the end uses. This c hapter has provided a summary of the latest work being done in this area. In addition it provide d the current state of how MELs are modeled and a review of past studies that will be used to improve these models. Reducing MEL is a secondary purpose of this dissertation. This c hapter has provided a summary of how householders can currently reduce the ir MEL and lays the foundation for the testing of the whole house switch that will be elaborated on in more detail later in this dissertation
72 Table 2 1. Building America g oals for e xisting h omes by climate t ype Energy Savings Mixed/Hot Dry and Marine Mixed Humid and Hot Humid Cold (Includes Cold, Very Cold, and Subarctic) 30% 2012 2013 2014 50% 2015 2016 2017 ( US Department of Energy, 2011b)
73 Table 2 m c ategory from P arker s tudy MEL calculation End Use Name UEC (kWh/yr) Sat uration (Fraction) Energy (kWh/yr) Dehumidifier 400.0 0.15 60.0 Electric Blankets 120.0 0.28 33.5 Portable Air Cleaner 500.0 0.05 24.2 Answering Machine 28.0 0.84 23.5 Ground Fault Interrupt 7.0 3.00 21.0 Humidifier 100.0 0.17 17.0 Electric Clock 17.5 0.94 16.4 Doorbell 17.5 0.65 11.5 Slow Cooker 16.0 0.56 9.0 Digital Picture frame 88.0 0.02 8.8 Waffle Iron/Sandwich Grill 25.0 0.31 7.8 Garage Door Opener 29.8 0.26 7.8 Hot Plate 30.0 0.23 6.9 Blender 7.3 0.77 5.6 Exhaust Fan 15.0 0.35 5.2 A uto Engine Heaters 250.0 0.02 4.7 Men's Shaver 12.7 0.36 4.6 Timer 18.4 0.23 4.3 Garbage Disposer 10.0 0.39 3.9 Grow Lights 800.0 0.00 3.7 Sump/Sewage Pump 40.0 0.09 3.7 Deep Fryer 20.0 0.15 2.9 Window Fan 20.0 0.14 2.9 Floor Fan 8.1 0.35 2.8 Bott led Water Dispenser 300.0 0.01 2.8 Heat Tape 100.0 0.03 2.8 Air Cleaner Electric ( not mounted ) 54.8 0.05 2.7 Hair Setter 10.4 0.26 2.7 Desk Fan 8.1 0.31 2.5 Power Strip 2.6 0.94 2.5 Heating Pads 3.4 0.66 2.2 Stand Fan 8.1 0.27 2.2 Can Opener 3.3 0. 64 2.1 Home Medical Equipment 400.0 0.00 1.9 Hot Oil Corn Popper 2.5 0.10 1.8 Automatic Griddles 5.5 0.25 1.4
74 Table 2 2. Continue d End Use Name UEC (kWh/yr) Saturation (Fraction) Energy (kWh/yr) Hand Mixers 1.5 0.85 1.3 Electric grill 180.0 0 .01 1.2 Women's Shaver 12.4 0.10 1.2 Air Corn Popper 6.1 0.19 1.2 Electric Kettle 75.0 0.01 0.8 Hand Held Electric Vacuum 3.9 0.19 0.8 Instant Hot Water 160.0 0.00 0.7 Compactor 50.0 0.01 0.6 Curling Iron 1.0 0.52 0.5 Food Slicer 0.9 0.41 0.4 Stan d Mixers 1.3 0.21 0.3 Electric Knife 0.7 0.36 0.3 Juicer 0.4 0.04 0.0 Total "Other Miscellaneous" 329 Table 2 3. Parker s tudy television e nergy u se t able Bedrooms TV kWh/yr Bedrooms TV kWh/yr 1 463 7 858 2 561 8 898 3 636 9 933 4 705 10 966 5 762 11 994 6 814 12 1020 Source: Parker et al., 201 1
75 Figure 2 1. Sample installation of Watts Up ? Meter ( Photo courtesy of Moorefield et al., 2011) Figure 2 2 Brand e o ne m eter ( Photo courtesy of Porter et al., 2006)
76 Figure 2 3 Interior r esidential equipment p rofile (Hendron et al., 2004) Figure 2 4. MELs norm alized energy use profile (Hendron and Engebr e cht 20 10 )
77 Figure 2 5. Equation 3 and t able 303.4.1.7.1(1) from 2006 Mortgage Industry National Home Energy Rating System Standards (RESNET 2006)
78 Figure 2 6. Residual m iscellaneous e lectrical l oads from P arker s tudy (Parker et al., 2011 ) (C lothes iron, vacuum cleaner, hair dryer, water bed, security system, clock radio and other miscellaneous not addres sed in 2009 RECS.)
79 Figure 2 7 Ceiling f an d iversity p rofile (Parker et al., 201 1 )
80 Figure 2 8. Energy c onsumption by the m iscellaneous e lectrical l oad from Roth study (Roth et al., 2008b)
81 Figure 2 9. Red Jacket Water Products 12G s eries p ump s izing c hart (Red Jacket 2012)
82 Figure 2 1 0 Distribution of RECS calculated MEL
83 CHAPTER 3 METHODOLOGY Overview categorized into heating, cooling, ventilation, lighting domestic hot water, large appliance loads, and m iscellaneous e lectrical l oads (MELs). MELs include hard wired components that are provided by the builder such as a garage door opener, alarm system, and doorbell, as well as components that are selected b y the occupants such as a television, computer, and coffee maker. MELs are traditionally estimated as a function of square foot (sqft) area of a home with multipliers to adjust to specific criteria like location and number of occupants. The hypothesis of this research is that calculating MELs as a function of physical as well as the characteristics of the occupants is a more accurate way of modeling the energy use. This c hapter will describe in detail the methodology for testing this hypothesis. Objecti ve The primary objective of this research was to create a model for forecasting MEL energy use that is more accurate than current practices. The most widely accepted residential energy rating system in the US (HERS) i ndex. The HERS index uses a simple square foot multiplier to model MELs. This square foot multiplier is the most common practice used for modeling residential MELs. This study improved the accuracy and precision of the HERS index model by calculating en ergy use as a function of the characteristics of the occupants. A secondary objective of this study was to develop an energy efficiency measure (EEM)
84 Residential Energy Consumption Survey The principal data source used to creat e the new MEL model is the Residential Energy Consumption S urvey (RECS) ( US EIA, 2011d). Since 1978, the US Energy Information Administration has been publishing the RECS, which summarize thousands of in person interviews in combination with data collecte d from utility providers. The survey asks the householder over 90 pages of questions and provides detailed information about home construction characteristics, types of appliances in the home, occupant usage patterns, and demographical data. The individu al survey data is aggregated together for high level summaries, but the raw data of 12,083 individual survey results are available for custom statistical analysis. The RECS are conducted every four years and this study used the most current survey, which collected its data in 2009. Use of RECS to Calculate MELs As mentioned previously the RECS were the primary means to calculate the individual MEL for all 12,083 survey respondents. However, the RECS did not contain all of the information required to make the calculation. Specifically, the actual energy consumption for any individual appliance was not provided so the energy consumption of the appliances that the survey asked about needed to be estimated. In addition, the RECS did not survey all MEL appli ances. A summary of how the RECS data was used to calculate a MEL for all of the RECS respondents will be outlined below. For more detailed information on the MEL calculation please see C hapter 4 Calculation for Appliances Addressed in the RECS The REC S asks the survey respondent about appliances that make up, on average, about two
85 information that the RECS provides is whether the appliance was present in the home. For appliances reported to be present, the unit energy consumption (UEC) from the literature review was used to calculate the total load (Hendron and Engebrecht, 2010; Roth et al., 2008a; Roth et al., 2008b). The UEC is the estimated total annual energy used by the average person w ith a typical model appliance. It averages usage patterns over a range of people and weighs the energy consumption of the different appliance models based on their market penetration. Other appliances, such as television peripherals, computers, monitors, microwaves, rechargeable electronics, and rechargeable tools, have more detailed information collected by the RECS about the appliance type and usage. When present, this information was used to calculate the specific UEC for each individual survey respon dent. Addressing D ays A way from the H ome in MEL C alculation When calculating the MEL, time away from the house traveling is an important consideration. For appliances where published UEC data was used then average vacation time was included in the UEC s o no additional consideration was required. However, for the appliances where the UEC was computed for each survey respondent time spent away from the home was considered. The RECS does not provide information on number of days the householders were awa y so an assumption was used. The US Department of Labor estimates that the average paid time off for full time workers ranges from 7 18 days (1996). In addition, the average full time worker has between 7 and 9 paid holidays. Although not intended for recreational use, paid sick days, which range from 7 11 days, can be converted to paid time days off in many companies. Paid days off do not necessarily mean that the members of a household
86 are traveling of course. For this study, it is assumed that a ll householders are away from the house for 14 days per year. Calculation for Appliances Not Addressed in the RECS The RECS did not ask the respondents about all the appliances in their home. The appliances not asked about make up approximately one thir d of the average and Engebrech t, 2010). A list of the most common appliances not included in the RECS was created. The UEC of each of these appliances as found in a literature review (KEMA, 2010; Roth et al., 2008b; Sanchez et al., 1998) was then multiplied by its market saturation t o estimate the typical energy load of the appliance as a national average. Market saturation is defined as the average number of appliances per home (total appliance / all homes). Creation of Model through Stepwise Regression Once the individual MEL was calculated for each of the survey respondents a independent variables (occupant characte ristics) to determine which ones best explain the dependent variable (MEL in kWh). The regression study created a series of models each with increasingly large numbers of independent variables. The first model would only use one independent variable and have the lowest explanatory power (R 2 ). The last model would use all (or most) of the independent variables to explain the dependent variable and have the highest explanatory power. However, the explanatory power did
87 not increase linearly with the additi on of more independent variables. Each additional independent variable would have a diminishing explanatory value than the previous selected that was a balance between high explanatory power and simplicity to use in application. Statistical Software Package SPSS The stepwise regression analysis for this study was done using the statistical packaged as Statistics Package for the Social Sciences in 1968 by Norman H. Nie, Dale H. Bent and C. Hadlai Hull. SPSS is one of the widest used software packages available and is currently used with survey companies, government agencies, social research base applications include descriptive statistics, bivariate statistics, linear regression and factor analysis although additional add ins are available to increases its capabilities. SPSS use s a two dimensional table format where ro w s contain cases (in this study there are 12,083 individual cases for each household) and the columns indicate the measurement (in this study the measurements are occupant characteristics like age, income and MEL). In 2009 the company and all rights to SPSS were purchased by IBM. SPSS is the software package that is currently used at the University of Florida statistics department and is taught to both the undergrad and master s level students. The Unive rsity has an agreement with SPSS making the yearly license available to students at a substantially reduced rate (Argyrous, 2011).
88 Non linear Independent Variables It should be noted that the stepwise regression is a linear regression process. This means that the dependent variable (MEL in kW h ) is explained with linear independent variables (occupant characteristics). An example of a linear independent variable is the income of a householder. Averaged out over a large sample size, the higher the income t he householders have the more energy they use. However, not all of the independent variables are linear. For example, age, number of household members, and number of children are non linear. W ith age, energy use continues to increase until householders are in their retirement years in which energy consumption starts to decline. To account for non was added to the list of independent variables that could be used in the stepwise regression. The original n on linear variable was not removed but could be tempered by the squared variable. Validation of Model Overview Once a MEL model was created, it was (partially) validated with 24 real world test homes. It is important to note that 24 test homes are not s ufficient to statistically validate the model but does provide an overall sense of its accuracy. The sample size restrictions. The validation involved installing data loggers into the 24 homes for a period of two weeks and recording the energy consumption of the plug loads. This through of the home, and literature review. The information was compiled and extrapolated to determine the estimated yearly MEL. The yearly MEL of the 24 homes was used to validate and calibrate the new model.
89 Validation of the Model Survey The first step in the validation process was the creation of a survey. The survey was u sed to select homes that w ere good candidates for the study and to assist with the MEL data collection. Because of the limited resources of the study, the survey was initially sent to personal acquaintances of the researchers who were then asked to forwar d it to 10 additional people. The personal acquaintances were initially targeted because they had diverse characteristics that were used in the new MEL model. When asked to forward the survey, the survey respondent s w ere requested to forward it to homes with less common characteristics but ones that were used in the new model For example, lower income households (less than 40K) living in single family homes represent 16% of the population but were good homes to test the model. This method resulted in 35 homes filling out the survey. The two primary reasons that homes were not selected to participate was because either sufficient number of homes with similar conditions had already been selected or the home had unusual living conditions such as college student or renting a single room in a house. The characteristics included in the model (which will be elaborated on in more detail later) were home size, income, number of household members and if a home business was present. The survey was able to coll ect information on homes with a wide range of these characteristics. Homes sizes ranged from 1,040sqft to 3,697sqft of conditioned space. Income was another important characteristic and ranged from 12 thousand to 144 thousand dollars of annual income. H omes observed had from one to seven household members living year round in the home. Three households with home businesses were surveyed. The survey was structured so that it described the goals of the study in detail and asked householders about the appl iances they owned, the characteristics of their family,
90 and other energy influencing factors. The survey can be found in Appendix A. The survey had two primary objectives. First, it was a means of calculating energy consumption for appliances and hardwi red loads that could not be monitored in the study. For example, some plug loads were located in wet locations and would damage the logger if monitored. Other loads such as door bells and security systems were hardwired and unable to monitor their energy use. Additional ly some loads are seasonal, such as electric blankets, and cannot be accurately extrapolated with a two week monitoring period. The survey helped identify these loads, which with the aid of typical energy consumption data from past studi es would allow this study to estimate the energy use. The second objective of the survey was to identify which homes were best fitted for this study and would be good candidates to be monitored. The stepwise regression analysis identified the most energy influencing factors (income, number of household members, size of home and presence of home business). The survey asked householders about those factors and was used to target homes with a good range of responses of those factors. Those factors would al so be used to input into the new model to validate the results. The survey was checked for accuracy in three ways. First, the survey answers were reviewed with the homeowners of homes selected to be monitored in person to make sure they fully understood the questions. When possible, other members of the household who did not complete the survey were asked to participate in the verification. It was found that other members may have knowledge about appliances that were in the home and have a better feel f or their use than the person filling out the survey. A second means of verifying the survey was with the through. During the walk through if answers to the survey questions
91 were not consistent with what was observed in the house then th e survey was adjusted. An example is appliances that were not indicated in the survey but were observed during the walk through. The third verification procedure was that the size of the home ze of the home is critical information as it is used in the new model and is the only characteristic considered in the HERS MEL model. Validation of the Model Data Loggers The purpose of this study was to develop a new method of modeling residual MELs an d like any new model the results needed to be validated. The details for how the model was validated will be discussed in detail in C hapter 4 but a critical piece of the process was to monitor the actual MELs in real world homes. The device selected need ed to be able to record electrical consumption data for multi week periods of time, have a high degree of accuracy, be cost effective for the study and preferably have been used in previous similar studies without technical challenges with the hardware. S everal devices were investigates such as the TED 500 C, ONSET Data Loggers, Current Cost EnviR and Dent Instruments PowerScout 3+ but none of these system selected for the st udy. See Figure 3 1. variety of energy consumption data at user adjusted intervals that range from 1 second to 1 week and has a large data storage capacity (120,000 records). The type of information that can be collected is Wattage, Volts, Amps, Power Factor, Duty Cycle and Frequency. The information is collected in the device and then downloaded to a computer via a USB plug. It is UL rated for 120v/15 amp and the accuracy is wit hin
92 1.5%. The data is date stamped and transferable to an excel file for easy evaluation. The memory is non volatile so the information is not lost in the event of power outages. This device has a reliable track record and was the data logger used in th e Moorefield study. The estimated cost of each device is approximately $200 without any bulk or educational discounts. The study would like to thank Jamie Bullivant and his staff at ThinkTank Energy Products Inc. for providing the data loggers at a subst antially reduced cost. Validation of the Model Monitoring of the Test Homes Based on the responses from the survey, 24 homes were selected as good candidates for monitoring. These households were contacted to set up a convenient time to install the dat a loggers. The householder was told to budget at least an hour as several other key steps of the study were performed during the initial visit. The first step of the visit was the verification of the survey. The survey was reviewed with the householder to verify that they fully understood the questions, and then it was revised if the plug and hard wired loads. Again, the survey responses were adjusted if what was indicated in the survey deviated from what was observed during the walkthrough. All plug loads that had a static draw, such as clocks, had their energy use measured l A entire year. Plug loads with varying energy use, such as microwaves and computers, were plugged into a data logger and monitored for two weeks. Y early energy consumption was extrapolated from the two week period. The householders were asked about their vacation and travel habits so that the time they are away from the
93 house was included in the annual energy use calculation. All plug loads were monitored provided they did not unreasonably inc onvenience the householder or were expressly asked not to be logged (due to temporary power interruption). The householder was instructed not to unplug the data logger, adjust the settings, or add/remove appliances to it. After the two week monitoring per iod the data loggers were collected. The householder was asked about factors that could affect the results such as power outages, tampering of the data loggers, unexpected trips, or any other unusual occurrences. Once the data was analyzed the energy co nsumption and energy cost were emailed to the householder. Validation of the Model Compiling of Information Collected from Test Homes Test home monitoring templates were created to rec ord the collected data. For MELs that could not be monitored, a commo n library of unit energy consumption (UEC) values from literature review was used. All test homes and the new model used the same UEC values. The information from the data loggers was downloaded from the device into a spreadsheet. Snapshots of the energ y use was collected every 30 seconds so that the average energy use was averaged over the two week period and then extrapolated for a full year. The information from the data loggers were analyzed by the research er to see if any of the values were beyond what was expected as a means to check for corruption in the data. When corruption of the data occurred, the appliances were either logged again when possible or replaced with UEC data from literature review. Validation of the Model Comparing Actual MEL with New Model Prediction The final step in the validation process is to compare the actual MEL of the 24 test houses with what the new model predicted. The occupant characteristics of the 24
94 householders were entered into the new model to calculate the p redicted energy use. Actual energy use was reviewed to see if they are consistently higher or lower than what was predicted by the model. A pattern was observed and the model was calibrated accordingly. ASHRAE Guidelines 14 (2002) limits the range of ca librating a new model with actual observations to +/ 15% .This limit was imposed on this research as well (Srinivasan et al., 2011c). One of the independent variables that was used to model MELs is householder income. Income is a sensitive subject that th e potential test householders may have been uncomfortable answering. To overcome this challenge a regression formula using the RECS data was created to estimate income. Independent variables such as education, age, size of home, and several other variabl es were used to estimate the actual income. The survey that the householders were asked to complete did ask about income but had larger ranges to limit the invasiveness of the questions. The survey asked the respondents to classify their income from 0 40K, 40 80K, 80K 120K, and 120K or more. Their response was included in the regression equation. Validation of the Model Defining Success For this study, success was defined using three characteristics of the model. The first measure of success was that the new model predicted the MEL of the test houses better than current practices (HERS index MEL model). There were 24 test houses so the new model needed to predict the actual MEL better than the HERS index for 13 or more times. The second measure that was used to determine the success of
95 The third and perhaps most important mea sure of success was how closely the new the average MEL for all homes In other words, how tight the individual observations are from the was predicted using the new model and the HERS index model for each of the 12,083 RECS respondents. The standard deviation of the new model was significantly smaller than the HERS index. If the new model met these t hree conditions it was considered Chapter Summary This study provided a detailed methodology for how this research was conducted. This study used the large sample size of the RECS to calculate a MEL value of over 12 thousand households. A stepwise regression analysis was then performed to create a new MEL model that is based on occupant characteristics as opposed to just the building characteristics as has been done in the past. The MEL of 24 test homes was determined by monitoring them for a two week period. The new model was then compared and calibrated with the actual MEL of the test homes.
96 Figure 3 1. Watts Up? Pro data logging device ( Photo courtesy of ThinkTank Energy Products, I nc.)
97 CHAPTER 4 CALCULATION OF THE MEL S FROM THE 2009 RECS Calculation is to us e only its physical characteristics such as size and number of bedrooms. The hypothesis of this research is that the addition of occupant characteristics will improve the prediction of a MEL A principal means to test this hypothesis was using the information provided in (RECS). The EIA provid es a 12,083 survey micro sampling of the data to perform custom analysis. Table 4 1 provides a sample of the data provided in the RECS. Of the 12,083 units included with the RECS, 8,700 were single family homes, 2,800 were apartments and 500 were mobile homes. Based on the data provided in the RECS, the MEL for each of the survey respondents was calculated The RECS did not ask the householders about every MEL appliances but does include information on approximately two thirds of the total residual load The remaining third of MELs was included in the calculation using the same methodology as the Building America Program and the HERS index. This c hapter provides a detailed description of how the MEL of each of the households was calculated The infer ences made with this information will be addressed in C hapter 5 Residual MEL Defined that does not fall into the major end use categories of heating, cooling, water heating and lighting. Some organizations have created additional groupings such as major appliances, pools and televisions. As the HERS index is the predominate energy rating
98 system for residential housing and considered the baseline for this study, its definiti on of loads, hard wired loads and hot tubs. There are exceptions to this general definition. For example, portable fans, air purifiers/cleaners, humidifiers and dehumidifie rs are related to climate control but are included in the residual MEL category. Portable space heaters, HVAC thermostats, air handler stand by power and ceiling fans are also related L. Well pumps are associated with water heaters but are included with the HERS definition. Hardwired loads such as GFCI receptacles, intercom systems, and security systems are all included as a residual MEL. Major appliances such as ranges, clothes wash ers, clothes dryers, primary refrigerator and secondary refrigerator/freezers are excluded from the HERS definition. The HERS index breaks from most organizations in its categorization of televisions. The HERS includes television peripherals such as DVD players and cable boxes in its definition but considers televisions a major appliance that is address separately. Developing of the MEL Calculation To calculate energy consumption a unit energy consumption (UEC) value needs to be determined. The UEC is the estimated total annual energy used by the average person with a typical model appliance. It averages usage patterns over a range of people and weighs the energy consumption of the different appliance models based on their market saturation. For this study, the Roth study (Roth et al., 2008b) was the primary source of UEC data. This is the same source used by the HERS index and many of the appliances in the Building America analysis spreadsheets (Hendron and Engebrecht, 2009). Table 4 2 has a list of the UEC used in the calculation and will be
99 elaborated further in this c hapter. Market saturation is also an important factor. The HERS index uses the market saturation data provided by the Roth study which gathered its information largely from the 2001 RECS. The MEL calculation described in this c hapter uses the more current 2009 RECS data to derive its market saturation. Resident usage data, such as number of hours per day a computer is used, is included in the calculation if provided in the RECS dat a. Kitchen Appliances Microwaves The first major grouping of appliances within the RECS is small kitchen appliances, which include microwaves, toasters, and coffee makers. Energy use for most appliances was calculated by multiplying the assumed energy i ntensity of the appliance by mode, by length of time in each mode, and by the market saturation of the appliance. Based on the Roth study (Roth et al., 2008b), the assumed energy intensity of a typical microwave was 1,500W in active mode and 3W in stand b y mode. The survey broke down how often the resident uses the microwave to cook meals and snacks into five categories: 1) Most, 2) About Half, 3) A Few, 4) Rarely, and 5) Never. Unfortunately, the survey did not qualify a precise length of time to corres pond to these 37). The s were used for this category. The remaining categories were extrapolated from this value. minutes per all. The market saturation for microwaves is 96% for all households included in the
100 RECS. Using the calculation described above, the average home uses 100kW h per year as compared with the HERS index estimate of 126kWh per year. Kitchen Appliances Coffee Maker The second kitchen appliance included with the RECS is the coffee maker. The energy consumption from coffee makers was calculated by multiplying the energy draw from the various modes (Active, Warming, Idle) by the duration in each mode and then multiplying it by the market saturation. Using data from the Roth Study (2008b), homes with a coffee maker have it in the active mode (the mode actually brewi ng coffee) for 6.3 minutes per day, warming mode for 38 minutes per day, and in the idle mode the other 97% of the day. Although there are various types of coffee makers, over 90% of the types used are the automatic drip style. This type of coffee maker on average uses 1,100W in active mode, 70W in warming mode, and .4W in idle mode (Roth et al., 2008b). The idle mode is primarily used to power a clock or timer. The market saturation of coffee makers surveyed in the RECS is 63%. Using the calculation d escribed above, the average coffee maker uses 59kWh. When market saturation is considered, the average energy use per household is 37kWh/year, which is consistent with the HERS index. Kitchen Appliances Toaster The third small kitchen appliance included with the RECS is the toaster. This was a more difficult appliance to calculate as the RECS combined toasters and toaster ovens into the same category. The toaster and toaster oven do have similar usage patterns and energy draws, but they have significan t differences that need to be addressed. The calculation used the market saturation of each appliance to weigh the specific characteristics together. There are 104 million toasters nationwide with an
101 average power draw of 1,050W. The toaster is used for approximately 6 minutes a day (Roth et al., 2008b). The Roth Study also indicates that there are 64 million toaster ovens with an average power draw of 1,300W (Roth et al., 2008b). These appliances are used for approximately 4 minutes a day. These dura tions seem anecdotally high, but with the lack of alternative consumption data they were used in the calculation. Using a weighted average approach for the two appliances, the toaster assumed in the calculation had a power draw of 1,145W and was used 4.9 minutes/day ( Table 4 3 ) The average toaster appliance using this calculation has an annual energy consumption of 13kWh, where the HERS index used 35kWh. Home Entertainment Television Peripheral Devices Although televisions are not considered a residu al MEL in the HERS index, there with televisions, the overall diversity of these appliances is high and the market saturation is low. To calculate the energy use, the esti mated UEC of each appliance was multiplied by the market saturation. The specific peripheral appliances included are cable boxes, DVRs (built into cable box and standalone unit), digital converter boxes, VCRs, DVDs, VCR/DVD combination units, and home the ater units. For these appliances the UEC was provided from the Roth Study. Video game consoles such as Xbox systems, Nintendo GameCubes, and Playstations were also included with the calculation, but their UEC was provided from the Roth and McKenney study (2007). Of all of the TV peripheral devices, the set top box (STB) uses by far the most amount of energy. STB includes satellite receivers, cable receivers, and personal video recorders (PVR). However, there are some significant differences between the energy use in each of these types of units. The RECS provided survey data on standalone
102 satellite and cable boxes, satellite and cable boxes with integrated PVRs, and standalone PVRs. Using the Roth Study data, active and standby power draw for each of t hese groups was calculated using a weighted average approach. Table 4 4 demonstrates the active and stand by draws used in the calculation. With the exception of VCRs, all television peripherals calculated with the RECS data increased from the estimates used in the HERS model. For example, average DVD energy consumption increased from 27kWh/year to 46kWh/year. Home theater systems increased from 15kWh/year to 22kWh/year. The most significant increase in energy consumption was from set top boxes which i ncreased from 92kWh/year to 226kWh/year. This trend seems reasonable as the RECS data is more current and appears to reflect the increased proliferation of home entertainment equipment. Home Computing Similar to home entertainment, home computing is anoth er area of high energy use and where the RECS provided detailed information. The RECS asked the survey respondents about the type of computer they use, length per day they use it, monitor type, and about the peripheral equipment of the three most used com puters in the home. The RECS questioned the survey respondents on the estimated duration they used their computer per day. The survey questionnaire asked the respondents to choose if they used their computer : 1) never, 2) less than 1 hour/day, 3) betwee n 1 and 3 hours/day, 4) 3 and 6 hours/day, 5) 6 and 10 hours/day, or 6) more than 10 hours/day. For the calculations a fixed value needed to be used instead of ranges, so the actual value used for the choices were: 1) = 0, 2) = 1 hr, 3) = 2 hrs, 4) = 4.5 hrs, 5) = 8hrs, and 6) = 12 hrs. The energy calculation was performed by taking the energy performance of the type of computer (desktop or laptop) and multiplying it by the length per day it was
103 used. This calculation was done for the three most used com puters. The RECS also asks the survey respondents if they own more than three computers. Where this is the case, a laptop computer with the UEC as provided in the Roth study, was assumed for each computer over three. The type of computer has a significa nt impact on home computer energy use. A desktop (excluding monitor) uses over three times the energy as a laptop computer (Roth et al., 2008b). Monitors are also a significant source of energy use. Monitors are used with desktop computers and also with laptops equipped with docking stations. The RECS question the survey respondents about the number and type of monitors they used, and their energy use is included with the calculation. Flat screen monitors were estimated to use 31W per hour and the olde r CRT monitors were assumed to use 61W per hour (Roth et al., 2008b). Length of time in use was assumed to be the same as the computer. There are a number of peripheral appliances associated with home computing. The appliances specifically addressed with the RECS are high speed internet modems, wireless routers, printers, faxes, and copiers. For each of these devices the UEC, as provided by the Roth study, was multiplied by the market saturation to account for their energy use in the MEL calculation. Es timated h ome c omputing energy use has decreased from 261kWh/year using the HERS data to 195kWh/year with the RECS data. The RECS data is more current than the information used in the HERS model and seems to track the trends of more efficient home computer equipment. Increased proliferation of laptops over desk tops, flat screen monitors over CRT monitors and more energy efficient printers contribute to this reduced energy use. Energy used for m odems and wireless routers (not included with HERS data) have nearly tripled.
104 Rechargeable Portable Appliances and Electronic Devices Rechargeable consumer products are a category that has experienced significant growth. The RECS breaks this category into two groupings. The first grouping is called digital cameras, and electric shavers. The RECS asked the respondents how many devices they owned and also about how the device is charged. The energy consumption is dramatically affected by if the appliance is continually plugged in, if the charger is plugged in when no t being charged, or if the charger is removed when not actively being charged after use. Similar to other questions in the survey, ranges were provided to the respondent to choose from so reasonable extrapolations needed to be made. When asking the surve y respondent how many devices they owned, they were given these choices: 1) 1 3, 2) 4 8, 3) more than 8, or 4) none. To quantify these ranges, if the recipient answered 1 then 2 devices were assumed, 4 then 6 devices were assumed, or 3 then 10 devices were assumed. To determine the energy draw based on the charging habits of the occupants the Roth study data was used. Table s 4 5 and 4 6 demonstrate the list of representative devices used to create the average energy use of both categories of recharge able consumer products by charging habit. Data was not available for length of time devices were charging for respondents who only plugged in the device when actively being charged. For those households it was assumed that each device was actively charge d for 1 hour per day.
105 Well Water Pump The Roth study provides UEC information for well water pumps, which is based on a study conducted by Author D. Little, Inc. (ADL) commissioned by the US DOE ( Zogg and Alberino, 1998) The ADL study assumed that a res idence would require 10 gpm flow rate at a pressure between 30 and 50 psi. No information is provided on the assumed well depth or pipe size. The pump was assumed to have an average energy consumption of 725W and used 115hr/year (19 minutes/day). The wa ttage and time on suggest (Zogg and Alberino, 1998). As discussed in Chapter 2 the assumed pump size was not provided in the ADL study, but based on the 10 gpm flow and the 725 wattage the pump would be approximately a horse power pump (Xylem, 2012). This size pump would be undersized for most homes, and the energy consumption was revised in the MEL calculation. The ADL study does not indicate the well depth assumed; howe ver, The Water Systems Council indicates that most wells are between 100 and 500 feet in depth (2003). The well pipe size is also not provided for a 10gpm system (indicated in the ADL study), but a 1 1979). Assuming a well depth of the same average service pressure of 40psi (service pressure = ), ), Using the Red 1/2 horse power (Red Jacket, 2012). Common 1 1/2 horse power residential deep submersible pumps have an average energy draw of 1,500W (Xylem, 2012). Assuming the sam e run time from the ADL study of 115hr/year, the UEC for a more appropriately sized
106 pump is 172.5kWh/year. As discussed in Chapter 2 the Parker study sites the Roth study (Roth et al., 2008b) as the source of the well pump UEC data but uses 862kWh instead of the 83.4kWh actually published. It is believed that the UEC published in the Parker study actually came from the California Residential Appliance Saturation Study (RASS Volume 2), which also estimates the UEC for single family houses to be 862kWh (K EMA, 2010). The RASS study does not provide the methodology for how the UEC was calculated but quantitatively seems very high for a national average. This research will replace the HERS UEC of 862kWh with 172.5kWh in its MEL calculation. es Included with Calculation The RECS also provide market saturation data for several other electrical devices that are not associated with any appliance category. Specifically these appliances are engine block warmers, large aquariums, cordless phones, a nswering machines, and spas. These appliances were included with the calculation by multiplying the market saturation as provided by the RECS by the UEC as found in the Roth study. The energy use calculated with this study for engine block warmers did no t change significantly from the HERS data. Cordless phones were not included in the HERS calculation but were only 3kWh/year with the RECS data. Large aquarium energy use estimates decreased from the HERS data of 27kWh/year to 8kWh/year The most dramat Average spa energy use increased from 61kWh/year to 117kWh/year This increases was due solely to market saturation as the UEC for both models was the same. not Included with the RECS The HERS index calculates MELS by taking a list of the most common appliances and multiplying their UEC by the market saturation. Table 4 7 demonstrates
107 a complete list of the appliances included. The total energy consumption of their MEL multiplier of .91kWh/year/sqft. Within the HERS list is a category titled such as electric blankets, air purifiers, and door bells and by multiplying their UEC by the market saturation. The UEC and market saturation data was collected from a wide range of publications (Floyd and Webber, 1998; KEMA, 2010; Sanchez et al., 1998; U S EIA, 2001e). This research expanded this list by adding 17 appliances not included with the original HERS list such as water beds, irons and, hairdryers. The UEC and market saturation were compiled from other MEL studies (Hendron and Engebrecht, 2010; KEMA, 2010; McMahon and Nordman, 2004; Porter et al., 2006; Roth et al., 2008b; US DOE, 2011c). Irons, water bed heaters, and hair dryers were three appliances with the highest average energy use that were added. Table 4 8 demonstrates a list of all appl appliances t Chapter Summary In this c hapter it was discussed how the information provided in the RECS was used to calculate a MEL value for over 12 thousand households. UEC values in combination with market saturation and usage patterns for the individual households were used in the calculation. This c hapter showed how the energy consumption of 60+ individual product types were calculated and used in the MEL calculation. Most of the
108 UEC values came from the Roth study; however, these values were checked for reasonableness and adjusted if need ed such as with the well pump.
109 Table 4 1. Information a vailable from RECS Occupant Characteristic Appliance Location Microwave Urban or Ural Toaster Number of B edrooms Coffee Maker Education TV peripherals (VCR, DVD, Cable Cox, DVR) Household Members Home Theater Household Age Computer, Monitor and Printer for 3 most used Computers If Home During Day Cable Internet Modem and Wireless Router Retirement Statu s Fax Machine Household Income Copy Machine Age of Home Well Water Pump Engine Block Heater Aquariums Stereo Equipment Cordless phone and Answering Machine Spa
110 Table 4 2. UEC used in MEL calculation Appliance UEC (kWh/year) Information Source Microwave Most Meals (18min/day) 190 Roth et al., 2008 b Half Meals (12min/day) 135 Roth et al., 2008 b A Few Meals (2min/day 44 Roth et al., 2008 b Rarely (30sec/day) 31 Roth et al., 2008 b Toaster 36 Roth et al., 2008 b Coffee Maker 59 Roth e t al., 2008 b Set Top Box 124 Roth et al., 2008 b Set Top Box w / DVR Calculated ( 33 74) Roth et al., 2008 b Stand Alone DVR 237 Roth et al., 2008b Digital Converter Box Calculated (6 20) Based on SIIG Converter CE CV0111 S Video Game System 41 Roth et al., 2008a VCR/DVD Player 50 Roth et al., 2008b VCR Player 47 Roth et al., 2008b DVD Player 30 Roth et al., 2008b Home Theater System 89 Roth et al., 2008b "Other" peripheral TV Calculated ( 46 66) Roth et al., 2008b 3 Most Used Computers Calcula ted ( 9 329) Roth et al., 2008b Secondary Computer (4+) 72 Roth et al., 2008b Monitory Calculated (11 267) Roth et al., 2008b DSL / Fiber Opt ic Internet 53 Roth et al., 2008b Cable Internet Access 53 Roth et al., 2008b Satellite Internet Access 53 R oth et al., 2008b Wireless Router 53 Roth et al., 2008b Printers 27 Roth et al., 2008b Fax 70 Roth et al., 2008b Copier 340 Hendron & Engebrecht, 2010 Well Pump 173 Roth et al., 2008b Auto Block Warmer 250 Roth et al., 2008b Heated Aquarium 210 Roth et al., 2008b Stereo Equipment 122 Roth et al., 2008b Cordless Telephone 27 Roth et al., 2008b Answering Machine 34 Roth et al., 2008b Recharge Portable Appliances Calculated (12 420) Roth et al., 2008b Recharge Electronic Devices Calculated ( 3 169) Roth et al., 2008b Humidifier 100 Roth et al., 2008b Spa 2040 Roth et al., 2008b
111 Table 4 3. Toaster oven UEC values Market Saturation (millions) Energy Draw (W) Duration of Use (min/day) UEC (kWh/year) Toaster 104 1,050 6 38.2 Toaster Oven 64 1,300 4 31.6 Weighted Average 1,145 4.9 36.4 Table 4 4. Weighted average of energy use by mode for set top boxes. Cable Satellite Weighted Ave of Cable and Satellite Stand Alone STB Installed (millions) A ctive (W) Off (W) Installed (millions) Active (W) Off (W) Active (W) Off (W) Analog Set Top Box 28 16 16 n/a n/a n/a 3.4 3.4 Digital Set Top Box 42 14 14 61 13 13 10.4 10.4 HD Set Top Box 1 22 21 1.4 21 18 0.4 0.3 Weighted Ave STB 14.1 14.1 STB with I nt e gra t ed PVR/DVR Personal Video Recorder Set Top Box 4 26 21 6 25 25 20.5 12.7 HD DVR Set Top Box 1 29 24 1.4 42 40 7.1 4.7 Weighted Ave STB with PVR/DVR 27.6 17.4
112 Table 4 5. Weighted average recharge able portable appliance energy use Rechargeable Portable Appliances Installed (millions) No Load (W) Charge Maintenance Actively Charging Cordless Phone 122 2.3 3.1 4 Cordless Phone w/ TAD 57 2.8 3.8 4.4 Cell Phone 200 0.3 0.52 2.6 Camcorder 64 0.37 0.39 9.6 Digital Camera 3.7 0.4 0.4 3 PDA 21 0.58 0.61 0.47 Rechargeable Toy 0.2 1 4.9 6 MP3 Player 23 0.26 0.62 3.7 Rechargeable Toothbrush 9 1.7 1.6 1.7 Shaver 28 0.3 0.68 2.3 Trimmer / Clipper 7.7 0.3 0.68 2.3 Weighted Average 1.1 1.5 3.9 Ta ble 4 6. Weighted average rechargeable electronic devices energy use Rechargeable Electronic Devices Installed (millions) No Load (W) Charge Maintenance Actively Charging Cordless Power Tools 51 0.9 3.6 15.9 Standalone Batter Charger 8.6 1.1 3.1 11.8 Cordless vacuum 21 0.8 3.7 4.7 Rechargeable Lawnmower 0.005 10 6.5 40 Weighted Average 1.2 4.8 16.7
113 Table 4 7. Appliance e nergy c onsumption in HERS Index as compared with RECS b ased m odel. End Use HERS MEL (kWh/year) RECS Calculated MEL (kWh/year) Difference (kWh) Difference (%) Desk top & Notebook Computer 168 140 28 17% Well Pump 147 20 127 86% Microwave 126 99 27 21% Recharg Electronics 69 57 12 18% Spa 61 117 56 92% Set top Box 92 226 134 146% Computer Monitor 54 35 19 36% Co mponent Stereo 49 53 4 7% Clothes Iron* 49 49 0 0% Vacuum Cleaner* 41 41 0 0% Printer/MFD 39 20 19 49% Coffee Maker 37 37 0 1% VCR Player 37 15 22 60% Hair Dryer* 36 36 0 0% Toaster/ Toaster Oven 35 13 22 61% Water Bed* 33 33 0 0% Component A udio* 32 32 0 0% Aquarium 27 8 19 70% DVD Player 27 46 19 72% Cable/DSL Modem 21 35 14 67% Home Theater in a Box 15 22 7 44% Security System* 15 15 0 0% Clock Radio* 14 14 0 0% Portable Audio* 5 5 0 0% Other Miscellaneous 329 455 126 68% 10% Est imate for Peripheral Electronics 156 59 97 62% Wireless Router not include d 25 n/a n/a Fax not include d 6 n/a n/a Copier not include d 29 n/a n/a Auto Block Warmer In "Other Misc" 3 n/a n/a Cordless Phone not include d 17 n/a n/a Answering Machine I n "Other Misc" 16 n/a n/a Humidifier In "Other Misc" 15 n/a n/a Total 1,714 1,795 81 4.7% RECS data not available so HERS data used
114 Table 4 8. Appliance inc ategory Appliance Original Inclusion UEC (kWh) Market S aturation Energy (kWh) Source of Information Dehumidifier HERS 400 0.15 60.0 Sanchez et al., 1998; Saturation from 2001 RECS Electric Blankets HERS 120 0.28 33.5 Sanchez et al., 1998 Portable Air Cleaner (Attached Fixture) HERS 500 0.05 24.2 Sanchez e t al., 1998 Ground Fault Interrupt HERS 7 3.00 21.0 Floyd & Webber, 1998 Electric Clock HERS 15 0.94 14.0 Sanchez et al., 1998; UEC from Roth et al., 2008 b Doorbell HERS 18 0.65 11.5 Sanchez et al., 1998 Slow Cooker HERS 16 0.56 9.0 Sanchez et al., 19 98 Digital Picture frame HERS 88 0.02 1.8 Not cited in HERS Calculation Waffle Iron/ Sandwich Grill HERS 25 0.31 7.8 Sanchez et al., 1998 Garage Door Opener HERS 30 0.26 7.8 Sanchez et al., 1998 Hot Plate HERS 30 0.23 6.9 Sanchez et al., 1998 Blender HERS 7 0.77 5.6 Sanchez et al., 1998 Exhaust Fan HERS 15 0.35 5.2 Sanchez et al., 1998 Men's Shaver HERS 13 0.36 4.6 Sanchez et al., 1998 Timer HERS 18 0.23 4.3 Sanchez et al., 1998 Garbage Disposer HERS 10 0.39 3.9 Sanchez et al., 1998 Grow Lights HE RS 800 0.00 3.7 Sanchez et al., 1998 Sump/Sewage Pump HERS 40 0.09 3.7 Sanchez et al., 1998 Deep Fryer HERS 20 0.15 2.9 Sanchez et al., 1998 Window Fan HERS 20 0.14 2.9 Sanchez et al., 1998 Floor Fan HERS 8 0.35 2.8 Sanchez et al., 1998 Bottled Water Dispenser HERS 300 0.01 2.8 Sanchez et al., 1998 Heat Tape HERS 100 0.03 2.8 Sanchez et al., 1998
115 Table 4 8. Continued Appliance Original Inclusion UEC (kWh) Market Saturation Energy (kWh) Source of Information Air Cleaner Electric (not mounted) H ERS 55 0.05 2.7 Sanchez et al., 1998; Saturation from KEMA, 2010 Hair Setter HERS 10 0.26 2.7 Sanchez et al., 1998 Desk Fan HERS 8 0.31 2.5 Sanchez et al., 1998 Power Strip/ Surge Protector HERS 3 0.94 2.5 Sanchez et al., 1998 Heating Pads HERS 3 0.66 2.2 Sanchez et al., 1998 Stand Fan HERS 8 0.27 2.2 Sanchez et al., 1998 Can Opener HERS 3 0.64 2.1 Sanchez et al., 1998 Home Medical Equipment HERS 400 0.00 1.9 Sanchez et al., 1998 Hot Oil Corn Popper HERS 2 0.10 0.3 Sanchez et al., 1998 Automatic G riddles HERS 5 0.25 1.4 Sanchez et al., 1998 Hand Mixers HERS 2 0.85 1.3 Sanchez et al., 1998 Electric grill HERS 180 0.01 1.2 Sanchez et al., 1998 Women's Shaver HERS 12 0.10 1.2 Sanchez et al., 1998 Air Corn Popper HERS 6 0.19 1.2 Sanchez et al., 199 8 Electric Kettle HERS 75 0.01 0.8 Sanchez et al., 1998 Hand Held Electric Vacuum HERS 4 0.19 0.8 Sanchez et al., 1998 Instant Hot Water HERS 160 0.00 0.7 Sanchez et al., 1998 Compactor HERS 50 0.01 0.6 Sanchez et al., 1998 Curling Iron/ Flat Iron HE RS 1 0.52 0.5 Sanchez et al., 1998 Food Slicer HERS 1 0.41 0.4 Sanchez et al., 1998 Stand Mixers HERS 1 0.21 0.3 Sanchez et al., 1998 Electric Knife HERS 1 0.36 0.3 Sanchez et al., 1998 Juicer HERS 0 0.04 0.0 Sanchez et al., 1998 Home Security System NEW 61 0.24 14.3 Roth et al., 2008 b Carbon Monoxide Detector NEW 18 0.26 4.7 Hendron & Engebrecht, 2009
116 Table 4 8. Continued Appliance Original Inclusion UEC (kWh) Market Saturation Energy (kWh) Source of Information Smoke Detector NEW 4 0.84 3.4 Hendron & Engebrecht, 2009 Trash Compactor NEW 50 0.01 0.5 Hendron & Engebrecht, 2009 Lawn Mower (electric) NEW 100 0.05 5.0 Roth et al., 2008b ; Market saturation from 2009 California RASS Kiln NEW 50 0.02 1.0 Roth et al., 2008b Pipe and Gutter Hea ters NEW 53 0.01 0.5 Roth et al., 2008b Water Bed NEW 1096 0.02 25.2 Roth et al., 2008b Iron NEW 53 0.92 48.9 Roth et al., 2008b Baby Monitor NEW 18 0.10 1.8 Roth et al., 2008b ; Saturation from Hendron & Engebrecht, 2009 Hair Dryer NEW 42 0.86 36.1 R oth et al., 2008b Night Light NEW 8 2.00 16.0 McMahon & Nordman, 2004; Saturation from Porter et al., 2006 Electric Tooth Brush NEW 12 0.06 0.7 Roth et al., 2008b Image Scanner NEW 138 0.14 19.3 Saturation from 2009 California R ASS; UEC from Roth et al., 2008b Standby from Hendron & Engebrecht, 2009 Irrigation Time NEW 45 0.05 2.3 Hendron & Engebrecht, 2009 Small Freshwater Aquarium (5 20 gal) NEW 105 0.02 2.5 Roth et al., 2008b Small Marine Aquarium (5 20 gal) NEW 245 0.00 0.6 Roth et al., 2 008b Total kWh 455.2
117 CHAPTER 5 RESULTS Overview of Findings for All Households As discussed in detail in C hapter 4 a MEL was calculated for all 12,083 RECS respondents. The HERS index is the most common energy rating system for residential home s and was the baseline in which the new model was compared. The way the HERS index calculates their residual MEL is with a square foot multiplier of .91. This multiplier was derived by listing the UEC of typical appliances found in a home, multiplying it by the market saturation of the appliance, and dividing it by the average home size as provided by the 2004 RECS. The home size used by the HERS index is 1,900sqft. However, the 2009 RECS indicate that the average home size has increased to 2,200sqft, wh ich is an increase of nearly 16%. Using the calculated MEL for each of the 12,083 RECS respondents, the study found that the average household uses 1,795kWh (including hot tub use). Hot tubs are categorized as a MEL by the HERS index but are difficult to include in a model as they have a low saturation but high energy use. This section will be comparing the RECS calculated MEL with the HERS MEL, so hot tubs were included to be consistent. More will be discussed on the impact of hot tubs later in this c h apter. The 1,795kWh is approximately 16% of the averaged total electrical load of the homes (11,300kWh/year). This is consistent with the MEL percentage range found in literature ( Barley et al., 2008; Ecos Consulting, 2004; Porter et al., 2006; Roth et a l., 2006; Sanchez et al., 1998; US DOE, 2012; Wentzel et al., 1997). A breakdown of which appliances make up the 1,795kWh can be found in Figure 5 category. Other Miscellaneous MELs are made up of 60+ small appliances such as
118 irons, crock pots, and waterbeds. The complete list can be found in Table 4 8 in C hapter 4 large number of them they make up a subs tantial portion of the total MEL. Set top boxes (13%), computers (8%), and microwaves (6%) have the next highest draw. However they still make up relatively small portions of the total energy use. An important take away from this graph is that no one ap pliance or group of appliances drives the total MEL. The MEL is made up of many small loads, which is a leading contributor to why MELs are so difficult to predict. As mentioned above, the HERS index derived their residual MEL multiplier by creating a lis t of UEC of the most common appliances, multiplying it by its market saturation and dividing it by the square foot of the home. The UEC and market saturation were provided from literature review. The RECS calculated MEL of this study uses almost all of t he same UEC values as the HERS index but uses the RECS 2009 for market saturation for the appliances that make up approximately two thirds of the load. Table 4 7 provides a comparison between the HERS Index and the RECS calculated residual MEL breakdown. Notice that the total HERS index residual load for 1,900sqft. The MEL load of 1,714kWh divided by 1,900sqft gives you the HERS index residual MEL multiplier of .91kWh/sqft. Ultimately the difference between the two total residual MEL is very small with only a 4.7% difference. Although not conclusive, the fact that the calculated MEL for all households is close to the established MEL model helps to validate that the MELs we re estimated accurately.
119 Some interesting findings can be seen by comparing the HERS and RECS MELs as shown in Table 4 7 MELs. The calculated MEL from set top boxes has over doubled from th e HERS value to 134kWh annually. This is explained largely from market saturation. The market saturation use of cable set top boxes has increased from 40% to nearly 80% for the primary television. Additionally, the proliferation of DVRs (digital video r ecorders) contributed to this increase. The energy intensity used for well pumps significantly decreased from 147kWh to 20kWh/year. This decrease was caused by two factors. First, the market saturation from the HERS to the 2009 RECS decreased from 17% t o 12%. The much larger impact was due to the change in UEC. As described in C hapter 4 it was felt that the UEC used by the HERS grossly overestimated the energy consumed from well pumps. The HERS index had a UEC of 862kWh, whereas this research felt th at 173kWh closer estimated the actual averaged load. See C hapter 4 for reviewing Table 4 7 that auto block warmers, answering machines, and humidifiers are appliances w as included in the RECS so their market saturation could be used to calculate their load. T 34kWh) should be added to the RECS category. The true comparison is between Wh (455kWh+34kWh = 489). This addition furthers the spread between the two models. RECS market saturation data was used
120 for auto block warmers, answering machines, and humidifiers so this accounts for some of the variance; but the main cause was that th is study added 17 appliances not included with the HERS typical house. These appliances with their market saturation can be found in Table 5 1 Combined, the added 17 appliances add an additional he Building America saturations. Analysis of Independent Variables Ultimately this research created a new model for predicting MELs by using a stepwise multilinear regression. However, before a multilinear regression model could be created a review of the independent variables (occupant characteristics) was completed. This was done to determine the overall impacts that the occupant characteristics had on MELs, which character istics had the most impact, and which were statistically significant. A review of statistical terms and definitions has been provided in Appendix B if clarification is needed. Trending of Independent Variables The first step in answering the research qu estion is to identify what occupant characteristics to include in our hypothesis. There are obviously a near infinite number of characteristics that could be used. However, to be able to test the significance of any particular characteristic it must be o ne that was included in the RECS. The occupant characteristics and the category of independent variable included in the survey can be found in Table 5 2 The next step in the analysis was a simple test in trending. This was performed by averaging the ME L of groups of households with occupant characteristics of interest. For example, the study would want to determine if there is a
121 correlation between when the home was built and the MEL. Figure 5 2 shows that there is a strong trend that the newer the ho me the higher the MEL. The trending analysis just provides an overall sense if an occupant characteristic is influencing the MEL. Firm conclusions cannot be made from this kind of review, but knowing the individual trending of the occupant characteristic s is of interest and worth identifying. The trending analysis also can show if the relationship between the variables is linear or nonlinear. Descriptive Statistics sta statistical characteristics of an independent variable. The traditional descriptive statistics include the sample size, minimum value, maximum value, range of values, mea n, standard deviation, and variance. Table 5 3 includes the descriptive statistics bet ween the minimum and the maximum. Some of the variables are categorical, meaning that the responses are grouped in only a few options. For example, the married (0). For t hat variable, the minimum is zero while the maximum is 1. (The standard deviation and variance do not provide any insights for categorical variables.) Part of what makes predicting MELs so difficult is not only are there many factors that influence it bu t also a large range of values in the factors. The energy use for Miscellaneous Electrical Load (first line) ranges significantly from 680kWh to 4,885kWh (range of 4,205) highlights this point. Size of home, number of household members, number of childre n, age of householder, and income all have notably large ranges. Of
122 particular interest to this study is the standard deviation and the variance. The standard deviation i ndicates that the data points are close to the mean. Conversely, a large standard deviation indicates a large dispersion of the data points from the mean. The variance is the standard deviation squared. If the observations are normally distributed (whic h they are in this sample), approximately 68% of the observations fall within +/ one standard deviation. By dividing the standard deviation by the mean you can get a sense for how dispersed the factors are. Miscellaneous Electrical Loads (excluding hot tubs) has a standard deviation that is +/ 31% of the mean. To put the dispersion into for all RECS respondents is $58,319 and has a standard deviation of $43,251. This mean s that for approximately 68% of the population the incomes range from $15,068 to $101,570. This is a very large range of incomes that could greatly influence the number house 2,172sqft but the spread of one standard deviation ranges from 717sqft to 3,626sqft. This home size range includes small apartments to very large single family houses. The average num ber of household members is 2.7 but approximately 68% (one standard deviation) of the sample has between 1.2 and 4.2. The lifestyle of a single person living alone varies significantly to a home with two adults with two teenage children. Because of the m any characteristics that influence MEL use and the large dispersion within the characteristics it is clear that no one characteristic can effectively predict MEL usage. Income, size of home, and household members were singled out
123 as they were the characte ristics used in the new model to predict MELs. Why those characteristics were used over others will be explained in more detail later in this c hapter. Testing for Significance Another step in the review of the individual occupant characteristics was con ducting a test for statistical significance. For this an Analysis of Variance (ANOVA) test was used. An ANOVA test is a statistical method used to compare the means of multiple independent variables (occupant characteristics) with a prediction model. As this study is using a 95% confidence level, the ANOVA test will determine if the 5% or less of the variance is caused by chance or sampling error and that the results would be the same 95% of the time with a different sample from the same population. In this test the hypothesis (Ha) is that occupant characteristics influence MELs. By contrast the null hypothesis (Ho) is that occupant characteristics do not influence MELs. The ANOVA tests the probabilit y of getting these observed results or results more extreme if the null hypothesis was true. Stated another way, the occupant characteristics are considered significant if the probability that their explanatory power is due to sampling error is less than 5%. An ANOVA table for all of the occupant characteristics is found in Table 5 4 using the independent variable to describe the dependent variable. Here each of the independent variables is being tested for significance, so the regression model only has
12 4 one predictor (single linear regression). When the new model was developed a similar ANOVA test was performed to test the significance using multiple independent variables. That will be discussed later in t his c difference between what was observed and what the model predicted squared. The ratio of the residual sum of squares to the total sum of squares provides some insights into how well the model predicts the dependent variable. The total degrees of freedom (df) are the sample size minus one (N 1). The regression degrees of freedom are the number of categories in an independent variable min us one (k 1). Residual degrees of freedom are the difference between the total and regression degrees of freedom. Degrees of freedom do not provide any insights into the data aside from determining the sample size and number of categories in the indepe ndent variable. They do help in the variable is statistically significant. For this study, all of the independent variables reviewed have a very low p value (<.0005) so all are considered statistically significant at 95% confidence. There is a diff erence between statistical significance and practical significance. The practical significance requires a mind behind the numbers to see which variables should be included in a regression model to function in the real world
125 el produced in this study does limit the number of occupant characteristics it includes, which increases its practical functionality at the very slight expense of some statistical significance. More will be discussed on this when the step wise regression model is explained. Before the ANOVA study can be performed there are three assumptions that must be satisfied. The first is that the residuals from the sample are normally distributed or has the familiar bell shaped curve. A residual is the difference b etween what was observed (each of the 12,083 MEL values from the RECS) and what the model predicted it should be. Figure 5 3 graphs the residuals for all 12,083 observations. The vertical axis is the frequency of the residual and the horizontal axis is how many residual standard deviations there are from the predicted value. Figure 5 3 shows a very near normal distribution with a slight right tail skew. Having an exactly normal distribution is extraordinarily unlikely but the data provided in the RECS is very close. With a lower limit of zero kWh and no upper limit also lends itself to a right tail skew. Regardless, the central limit theorem states that because of the large sample, a near normalized distribution can be treated as if it were normal. T he second assumption is that any errors in the individual cases are independent from one another. In this study each of the 12,083 survey respondents were randomly selected and independent of each other. Additionally, the calculation of the MEL is on an individual case by case basis without any inferences made from the sample as a whole. The third assumption is the homogeneity of the variance, which is sometimes called the equality of variance. This means that the variances in the observations of an ind ependent variable are similar throughout the data set. Figure 5 4 shows the predicted values from the model on the x
126 axis and the standard deviations of the residuals on the y axis. When looking for homogeneity of the residuals, it is expected that the g raph will be without pattern or form. Figure 5 4 shows that the variances are homogenously spread across the data set. Correlation of Independent Variables When reviewing the occupant characteristics with an ANOVA study it is important to understand the correlation between not just the MEL (dependent variable) but also the other occupant characteristics (independent variables). The Pearson Correlation test, developed by Karl Pearson in the early 20 th century, is widely accepted in the academic community and was used in this study. The Pearson Correlation test from 1 +1. The sign indicates the slope or direction of the correlation. The closer the r value is to eit her 1 or +1 the higher the correlation is between the two variables. Appendix C provides a summary of the correlations of all of the variables being considered. All of the variables are listed along the left side and top of the tables. The intersection of the variables on the left side with the variables along the top indicates the correlation between the two. You will notice that there is a diagonal line from the upper left to the lower right where all of the r values are 1.000. At these intersection s the variable on the left and the top are the same. The table is showing that there is a 100% correlation between a variable and itself. The first column in Appendix C is the dependent variable Miscellaneous Electrical Load and is the most interesting f or this study. The table shows that the income (r = .424), sqft of Home (r = .410), and size of garage (r = .349) have the highest explanatory power of miscellaneous electrical loads. Marital status (r = .310), type of house (r = .284), and education (r = 2 60) also have
127 high r values. The table also shows that there can be a high inter correlation between the occupant characteristics. For example there is a 59.9% correlation between the square footage of the home and its garage size. The phenomenon i s called multicollinearity highly correlated with themselves you cannot just take the characteristics with the highest r values to construct a model. To provide the best fit equation that describes discussed in more detail later in th is c hapter. Before stepwise regression is discussed, the following paragraphs will review each of the occupant characteris tics and how they relate to MEL usage. Housing type One of the independent variables considered in this study was housing type. This includes mobile homes, attached and detached single family homes, and apartments. Housing type is not an occupant charac teristic but worth reviewing for insights into MEL usage. Figure 5 5 shows single family homes standing out from the other categories as having a significantly larger MEL. Intuitively, other factors such as home size and income play a role in the energy consumption. The average detached single family home in the RECS sample is 2,700sqft. The group with the next largest square footage is attached single family homes that have an average size of 1,900sqft. There is a 30% difference in size between attache d and detached homes. The difference in size is even greater in the other housing types with apartments with 2 4 units averaging 1,100sqft, mobile homes 1,000sqft, and apartments with 5+ units at 900sqft. Income has a similar distribution. Households living in detached single family homes have an average income of $67K as compared with attached single family
128 homes, apartments with 2 4 units, mobile homes, and apartments with 5+ units with incomes of $57K, $37K, $32K, and $38K, respectively (Table 5 5 ). The Pearson Correlation table (Appendix C ) confirms this as the size of the home and that householder income has a 43.9% and 22.7% correlation, respectively. What is less .2 64) and .262) also are strongly correlated to housing type. Urban/rural location is less interesting because it is so weakly correlated to MEL usage (r = .086), but marital status is highly correlated (r = .310). AIA cli mate zone and Building America climate regions AIA climate zone and Building America (BA) climate regions were other variables considered. Figure 5 6 shows the association between MEL and the AIA climate zones, while Figure 5 7 shows the energy use by BA climate region. The AIA climate zone (Figure 5 6) does appear to show a pattern with homes in extreme heating or cooling areas tend to use more MEL. The BA climate regions appear to show no pattern of location to MEL usage (Figure 5 7 ). The BA climate r egions use cooling degree days (CCD) and heating degree days (HDD) as the driver to define the boarders of the regions but are more generalized than the AIA climate zones. This looser affiliation with climate conditions is likely the reason why there is no correlation between BA climate regions squared was used to normalize the graph. The AIA climate region squared shows a small correlation between climate zone and the yea r the home was built (r = 21.9% Appendix C ). This is reasonable as different parts of the country experience different growth at different periods of time. This correlation is not of much interest to the study, because the year the home was built has a v ery low correlation with MEL
129 usage (r = 3.6%). No other characteristics have any significant correlation to either the AIA Climate Zones or BA Climate Regions. Home size and number of household members Both the HERS index and the BA program use the size of the house as the principal indicator of MEL intensity. This is well founded as the size of the home is only behind income with the highest correlation with MEL (r = 41.0%) (Appendix C ). As shown in Figure 5 8 there is a very strong trend between home size and MEL intensity. Most of the individual appliances that make up the MEL increase steadily as home size increases. However, TV set top boxes, hot tubs/spas, and desktop/laptop computers rise sharply with the increase in home size ( Figure 5 9 ) Ot her characteristics that were reviewed that are related to home size are total number of rooms and number of bedrooms ( Figures 5 10 and 5 11 ) Total number of r ooms (Figure 5 10) rises steadily and has a similar trend with the square footage of the home ( Figure 5 8). Number of bedrooms also has a similar pattern of energy use, except that once the number of bedrooms exceeds five the energy use starts to decline (Figure 5 11) The weight that the model places on these high bedroom houses is low as they rep resent less than 1% of the homes included with the RECS sample. A point worth highlighting is that the BA program uses number of bedrooms as an independent variable to calculate MELs. Number of bedrooms is used as a surrogate for number of household memb ers, assuming that number of members influences MELs. This assumption is at least partially validated as Figure 5 12 shows that there is a strong association between number of bedrooms and number of household members. The residuals or difference from mea n to trend line is highest with homes with four or more bedrooms. This suggests that the BA model may become increasingly inaccurate for homes with a large
130 number of bedrooms. This is a known limitation by the BA program (Building America, 2007). This can be partially explained by Figure 5 13, which shows the relationship between number of household members and energy use. Intuitively, it would make sense that the more people in the home the higher the MEL; and this holds true for homes between one and four members. However, energy consumptions starts to trend back down for homes with five or more members, suggesting that other factors are influencing energy use. Appendix C shows that marital status (r = 44.1%), householder age (r = 35.3%), number of children (r = 84.0%), and retirement status (r = 26.3%) are all highly correlated to the number of household members in the home. Interestingly education (r = 4.6%) and income (r = 17.3%) have a relatively low correlation to the number of household membe rs Home Business A home business has a correlation of 19.7% with MELs, which is fairly moderate when compared to sqft of the home (r = .410) and income (r = 42.4). The difference between households with a home business and those without is approximately 500kWh (Figure 5 14). 500kWh is nearly 30% of the average household MEL. However, what makes this characteristic interesting is that it is not very correlated to any of the other characteristics. The highest correlation is with income but only 13.8%. W hat this means is that even though it s correlation with the dependent variable (MEL) is 19.7%, very little of it s correlation is explained with other variables. This variable is part of the new model because its correlation with MEL is unique and independ ent from other variables.
131 Income and Education Of all of the characteristics included with the RECS income has the highest correlation with MEL usage and has an r value of 42.4%. As one would expect there is a lot of correlation with other independent va riables. Three specifically worth mentioning are square footage of home (r = .461), education (r = .443), and marital status (r = .351). The association of MEL and square footage has been shown before in Figure 5 8. Figure 5 15 shows the comparisons bet ween MELs and education and marital status. Statistically, income and education are very similar. They both have high explanatory power on MELs but also have high correlations with other independent variables. Because of their similarity, their explanat ory power overlaps. Stated another way, income and education explain the same parts or variance of the MEL. Income has a higher r value and is used in the final model. However, if income is not available education could be used fairly successfully as a s ubstitute. Marital status also has a high r value. Figure 5 16 shows that married households use 400kWh/year more than unmarried households This substantial swing in kWh and the correlation with ME Ls is high (r = .310), marital status is also highly correlated with the square footage of the house (r = .283), number of household members (r = .441), and as stated before income (r = 35.1). Much of its explanatory power is shared with these other chara cteristics. In other words, it does not add much that has not already been explained with other independent variables. T his is the reason marital status was not included in the final new MEL model. Other independent variables There are o ther occupant cha racteristics that were reviewed in this study Because of the large sample size, all of them were statistically significant but do not
132 correlate well with the MEL. T hese independent variables are householder age, number of children, if the householder wa s home throughout the day, and retirement status. The range of MEL usage for each of the se characteristics can be found in Figure s 5 17 through 5 20. One occupant characteristic that surprising ly did not impact MELs significantly was whether the househol der was home throughout the day or not Intuitively it would seem obvious that the longer the householders are in the house the more MEL energy they would use. This study found that the difference between households home during the day and not is less th an 100kWh/year. This small difference in energy use is the driver behind an r value of .082. Stepwise Regression The research question for this study was to determine if the addition of occupant characteristics improve d the predict ion of MELs To do this a new regression model or explanatory equation was created based on the RECS data. The study used a systematically adds and removes independent variables (occupant characteristics) to determine which ones best explain the dependent variable (MEL in kWh). The regression study created a series of models each with an increasingly large number of independent variables. The first model would only use one independent var iable and have the lowest explanatory power expressed as the models R squared (square of residuals). The last model would use all (or most) of the independent variables to explain the dependent variable and have the highest explanatory power. However, th e explanatory power did not increase linearly with the addition of more independent variables. Each additional independent variable would have a diminishing explanatory
133 judg such that it provides a balance between high explanatory power and simplicity to use in application. Errors in Variable Before discussing the specific s of the various stepwise models it should be noted that an assumption w as made about the independent variables used This assumption was that the is zero The error in variable is the potential errors when observing the regression predictors. In this case the predictors are the occupant characteristic s included in the RECS. Essentially, th is assumption is that the data used from the RECS was collected accurately and represents the survey respondents correctly. Although it is very likely that some error occurred in the data collection it will be treat ed as zero for two reasons. First, the sample size is very large so a small number of errors will not significantly influence the regression equation. Second, although there maybe errors in collection of the data it would be random and thus not bias the results in any particular direction Random errors in data collection tend to balance themselves out with large sample sizes. Stepwise Models The calculated RECS MEL (dependent variable) and the entire occupant characteristics (independent variables) we re input into the statistical software package SPSS. The software ran a stepwise regression analysis and created 17 regression models, each with an increasing number of independent variables. The first model has only income as the independent variable to explain the MEL. The last model (model R) has 17 independent variables used to explain the MEL. Table 5 6 provides a list of the independent variables included in model R in order of highest to lowest explanatory power. Table 5 7 summarizes the 17 mode ls. The R and R Square values are
134 6 explain MELs. The R value for this model is .424, and because the model has only one independent variable R is the same as what is provided in Appendix C As the number of variables increases, the R and R square values get larger. This is because for each new model one variable is added to further explain the MEL value. However, each new independent variable does not correlate with the MEL equally. In fact, because the stepwise process selects the independent variables with the highest r values first each new model increases the that shows the increased explanatory power of the model from the previous model. For Each subsequent model has a reduced explanatory power but increases in complexity and difficulty to use in practical application. A balance was struck between explanatory power and ease of use in application. This decision does have some subjectivity change in R square, and it was felt that this minor increase does not outweigh the increase in difficulty to use the model in practice. Multico l linearity of Model need to be verified. First, the new model needs to be tested for statistical significance. All of the independent variables used were statistically s ignificant, so it is extremely likely that their additive explanatory power would also be significant. Table 5 8 is the ANOVA table for models A E and as expected has a p value less than .0005 for all
135 models and is statistically significant. The second characteristic of the model that needs to be verified is if there is too much correlation between the independent variables. This l l linearity makes the model unstable where small changes in the independent variables can have exaggerated effects on the output. There are three tests used to check for multicol l inearity, and the results of these tests can be found in Table 5 9 erance is simply the inverse of the VIF. The rule of 10 (or .1 for tolerance) is widely accepted as the threshold of an unacceptable level of multicol l 9 shows that Models A D all have very small VIFs with values that d is introduced that the VIF increases to 8. 968 This is expected as there is a strong correlation between household members (IV3) and household members squared (IV5). Still, the VIF is under the threshold that strongly suggests that multicol l inearity is not a concern in these models. Another test to determine if there is a multicolinearity problem considered to have an acceptable level of multico l linearity (Manzoor et al., 2011). Here again, multicolinearity is not being flagged as an issue as all of the condition indexes for the independent variables are much lower than 30. Based on these three tests, multicol linearity is not being considered an issue in any of the first five models (A E). Components of Explanatory Equation The MEL models created from this study are explanatory equations and use a constant and coefficients to estimate MELs. A constant (a) is this explanatory equation is predicting MEL, the constant generally represents the
136 minimum energy use for all residential homes. Coefficien ts ( b k ) are sometimes called x ) into the predicted value units as shown in the generic template for a multilinear explanatory equation below: Model E is the model that this study is presenting as the best balance between explanatory power and ease of use. In model E the independent variables used x of the home, number of household members, if the household contains a home business, and the household members squared. These are all quantitative variables and extend from zero to infinity. The x value for income is simply the total annual income of the household in US dollars. Similarly, the x value for square footage of the home is the actual square footage of the condition ed space of the household. The x value for number of household members is the number of people of any age who live in the home. This excludes members not actually living in the home, such as children away at college or serving in the military. One of the variables is household members squared. Number of household members was squared to better model the nonlinear (U shaped) relati onship between household members and MEL usage. Figure 5 13 demonstrates that as the number of people in the home increases beyond four the MEL intensity decreases. The squaring of the household members captures the decreased energy use from large member households. Home business is a dichotomous independent variable, meaning that
137 there are only two options. If there is a home business the x value would be one. If there is not a home business in the household then the x value would be zero. The consta nt and coefficient values for the models A E are shown in Table 5 10 The constant values indicated in Table 5 1 0 above. Model E with the coefficient labels is provided below: Example of E xplanatory E quation To illustrate the explanatory equation, an exampl e of a very common home will of $55,000 ( x 1 ), lives in a 1,400sqft home ( x 2 ), has three members ( x 3 ), and does not operate a home business ( x 4 ). With these independent vari ables the equation would be as follows: Using model E the estimated annual MEL in kWh for household Alpha described above would be 1,635kWh/year. However, with any statistical equation there is a
138 certai as the margin of error. Table 5 1 0 provides the margin of error for models A E. Model E (which was used above) has a margin of error of 857kWh. The confidence interval for household Alpha is 1,635 +/ 857kWh. What this means is that with 95% fairly wide interval but not unexpected considering the large standard deviations of the independent variables (Table 5 3 ) and the relatively weak correlations between the variables (Appendix C ). The wide range does not mean that the model does not have any practical explanatory power. In fact, it will be shown later in this c hapter that model E is a be tter predictor of MELs than current practices. Hot Tubs It should be highlighted that the RECS calculation of the MEL does not include hot tubs. This is a distinction between this model and the HERS model. The HERS model includes hot tubs as part of th e MEL multiplier. The HERS model assumes a UEC of 2,040kWh/year and a market saturation of 3% (Figure 2 6). Considering that the average MEL of all RECS respondents is 1,795, a UEC of 2,040kWh is an extreme value for one appliance. The market saturation of 3% is also an extreme value and shows that the energy use does not reflect the vast majority of the population. The this means is that for 97% of the population, 61kWh is added to the model for an house so it does not skew the model significantly. However, for the 3% of homes with hot tubs the HERS model grossly underestimates the MEL Due to the high UEC and low market saturation this appliance cannot be averaged with the rest of the MELs and
139 is disaggregated out for the new model. Homes with hot tubs should add 2,040kWh/year to the new model to account for the appliance. Although not as extreme, water beds also have a high UEC (1,096kWh/year) and low saturation (3%). A case could be made to disaggregate water beds out from the residual MEL category as so would add another layer of complexity to the model without a sufficiently large return in precision. Improved Standard Deviation Using the New Model The supposition of this study is that the new model is a better predictor of MELs than the HERS model. To test this, the MEL of all of the RECS respondents was calculated using the new model and the HERS model. For the new model this included all of the coefficients shown in the equation above, but the HERS index only uses a multiplier of .91kWh/sqft of t he home. The results of the two models can be found in Table 5 1 1 survey respondents. The average between the two models is relatively close. Using the HERS model the average energy consumption is 1,977kWh, whereas using the new model the energy consumption is 1,792kWh. There is almost no difference between the Calculated derived from the calculated MEL. The diffe rence between the HERS model and the new model is only 185kWh, which is a 9.4% difference from the HERS model. What is more noteworthy is the standard deviation, which is the dispersion from the individual observations from the predicted model. The stand ard deviation of the HERS model is that the calculated MEL accurately represents the true MEL for each of the households,
140 what Table 5 1 1 shows is that the new model and HERS model explanatory power for entire populations is very close. The new model improves the overall population predictive power by 9.4%. However, at the individual home level the new model is more likely to predict the true MEL by 54.8%, because there is less deviation between the model and individual observations. Validation of Model Once the new model was created it was validated with 24 test homes. The MEL for each of the homes was calculated by monitoring plug loads with data loggers, an occupant survey, and literature review of typical UEC values. The homes were monitored with the data loggers for a two week period with the yearly MEL extrapolated from the data collected. The data collected from the data loggers was combined with the information collected from the occupant survey to estimate a yearly MEL for the home. The occupant survey primarily provided information about the occupant such as income, marital status, and size of home and information on appliances used that could not be monitore d. The yearly MEL of the 24 homes was compared with the calculated MEL using the new model The new model was calibrated b ased on actual data collected at the test homes the new model was calibrated Before contin to test the model which is not enough to statistically validate it. To statistically validate the model, a minimum number of 385 households would need to be monitored (using Since this was not a funded research project r esource and time limitations prevented the monitoring of this many homes. The comparison of the new
141 model wi th these test homes does still have value but only provides an overall sense of partial validation and not necessarily a statistical validation. Survey The occupant survey had two primary objectives. First, the survey was to help help identify what homes would be good candidates to validate the model. An important distinction to hi ghlight is that the test homes were selected based on their value to test the model and not necessarily homes that represented the average population. These two criterions are not completely dissimilar but the priority was placed on testing the model. The occupant characteristics used in the model are income, square footage of homes, number of household members (household members squared), and if there was a home business. The test homes selected had a wide diversity of these characteristics. Before disc ussing the diversity of the test homes, the occupant Household s Income Income can be a sensitive subject to some people. As the survey and participation in the study was voluntary without any compensation it was felt that asking about income in the survey should be addressed delicately. Income was grouped into ranges of $40K, and the occupants were asked to indicate what grouping best described their household. Specifically, the question asked if the hou sehold earned 1) less than $40K, $40K $79K, $80K $120K, and $120K+. Using the RECS, the stepwise regression process was used to create a model that predicted income using this response as an independent variable and other variables such as education, home
142 size, and marital status. Using the income group s and the other independent variables listed created a regression model with an R 2 value of 92.7%. The estimated income using the regression model was used in the MEL model (model E) to predict the tes t homes MEL. Income: 1) <40K 2) $40K $79K 3) $80K $120K 4) $120K+ Education Level: 0) No schooling 1) K 12 2) High school diploma 3) Some college , 7) Professional degree 8) Doctorate degree Marital Status: 1) Married 2) Unmarried Surve y Response The survey was of course used specifically for this study but several questions were included that could be useful for future studies. For example, the HERS index does not include televisions as a residual MEL but other organizations such the B uilding America program does include it. With this in mind, television size was one of the questions asked about. Space heaters are another appliance not included as a MEL in the HERS model but asked about in this study. The complete survey can be found in Appendix A The results of the survey have been summarized into two categories: house characteristics and occupant characteristics. Table 5 1 2 shows the house characteristics of the 24 test houses. The Model Validation (MV) houses are listed 1 24. stipulated in the IRB protocol. Of the house characteristics only square footage is being
143 included with the MEL model. The house size ranges from 1,040sqft 3,697sqft with an average of 1,947sqft. The RECS data indicated that the average household size was 2,172sqft with a standard deviation of 1,454sqft. This is slightly above the average test house size but, given the standard deviation, approximately 68% (empirical rule) o f the RECS homes range from 718sqft and 3,626. This home size spread matches very was 1,900sqft so this sample mirrored the HERS model well. Survey R esponse I ncome Th e second type of questions that the survey addressed was occupant characteristics can be used to improve MEL modeling. Income, number of household members, and if there i s a home business were all addressed in the survey and summarized in Table 5 1 3 Table 5 1 3 estimated income using a regression model. Incomes range from $12,323 $143,590 per year with an average of $81,062. Over 85% of all of the households included with the RECS have incomes within this range. Table 5 1 3 shows that the number of household members ranges from 1 7 people and has an average of 3.4 people. Over 95% of the households included in the RECS have from 1 to 7 household members. Of the 24 test houses only 3 contain home businesses. Ideally, more households with home businesses would have been collected but obtaining these types of homes was difficult. Home businesses provide the least explanatory power of all of the occupant characteristics in the new model so it was not felt that this dramatically impaired the test
144 Skewness of the Test Houses As mentioned earlier, the primary purpose of the test homes was to validate the model. However, it is worth highlighting that there is some skewness i n the characteristics of the test homes that do not represent the population well. For examp le, Table 5 1 2 indicates that 83% (20 of 24) of the homes are single family detached as opposed to only 65% indicated in the RECS data. Similarly, of the test homes only 8% (2 of 24) are in a rural area and 25% rent (6 of 24), whereas the RECS data shows this to be 65% and 32% respectively. The test homes also had more members, were better educated, and had higher incomes than the average RECS respondent. Table 5 1 3 shows the average number of people in the test homes was 3.4 members, the median educatio n level was a and had an income of 81K. The average RECS household had 2.7 members, the highest level of education US population was not the goal of the test houses but is worth noting, especially when looking at what homes the new model predicted better than current practices. Data Loggers The actual MEL for 24 test houses was recorded to validate the accuracy of the new model. The majority of the MEL was recorded using ? loggers. The data loggers were provided to the study at a substantial discount by the manufacture ThinkTank Energy Products, Inc. The loggers have a limited amount of seconds. Recording data every 30 seconds allowed the data loggers to record 15 days of information. The data was trimmed to 14 days so the ratio of weekdays and weekends would not be skewed. The number of data loggers installed depended on the number o f plug loads
145 that could be recorded, but they ranged from 4 to 11 with an average of 7. Homes with only four data loggers were homes with either low income, few household members, or both. There was no noticeable pattern in the householders with a higher number of data loggers installed. An unexpected advantage found with using the Watts Up ? Pro E S data logger is that the data recorded cannot be modified or deleted from the unit without plugging it into a computer. This was an advantage because many hou seholders, children in particular, were very interested in the data being recorded. Power information such as watts, amps, volts, and power factor can be shown on the display by pushing buttons on the device. Pushing the buttons and watching the differen t read out change satisfied the curiosity of the children but did not corrupt the data recorded. Comparing Recorded Energy Consumption with Published UEC data The New Model used the RECS survey in combination with published UEC of small appliances to calcu late the MEL of the RECS respondents. Additionally, published UEC data was used to supplement the information recorded with the data loggers when monitoring specific MEL in the test homes was not possible. Because of the weight of importance that this st udy places on the published UEC information, a comparison was made between the actual energy consumption recorded with the data loggers and the published UEC. The studies that created the published UEC data were considerably more involved than the monitor ing of 24 test houses. The UEC studies used much larger sample sizes and incorporated the energy consumption from many different makes and manufacturers of the various appliances. With this in mind it would not be appropriate to use the recorded data fro m the test houses to supersede or invalidate the published UEC. Still, the comparison was used to show a general
146 The summary of the comparison betwe en what the data loggers recorded and what was expected based on the published UEC is provided in Table 5 1 4 Through the course of the study, data loggers were used to record two week periods over 160 times. On several occasions the data recorded was co rrupted and could not be used. Other times, uncommon appliances without published UEC data such as weather stations, foot massagers, and rice cookers were recorded. In both cases, these data logging events were excluded fr om Table 5 1 4 because they are n ot applicable to a UEC comparison. The comparison includes 140 data logging events, which recorded the energy consumption from 2 71 appliances. The energy consumption (extrapolated for the entire year) recorded on the data loggers was 13,851kWh. The publ ished UEC for the 2 71 appliances recorded on the data loggers was 13,933kWh. This is a difference of less than one percent. While this would seem to indicate that the published UEC mirrors very closely to what was observed in the test houses, it is impor tant to look at the variance from what was expected from what was observed at the individual appliance level. For the 140 data logging events the average difference from what was recorded to the published UEC was 119%. Several data logging event deviate d from what was expected significantly; in excess of 2,000% in some cases. These large outliers influenced the average considerably. A better measure of evaluation is the median difference between published UEC and what was recorded. In this study the m edian differences was only 20%. What can be inferred from this data is that while there can be significant deviations from published UEC to what is actually observed,
147 overall the UEC is fairly accurate, As mentioned before, a sample size of 24 households is not enough to make any conclusions but this study does support the findings of past studies. An important question to address is that if any specific appliances are consistently skewed from the published UEC data. Because of the limited number of data loggers available to the study, not every appliance was monitored individually. In fact, the majority of the data loggers had multiple appliances being recorded so that the m ultiple appliances on the same logger had the advantage of being able to include more test houses in the study but the individual energy consumption for each appliance was lost. However, due to various specifics of the test houses, 61appliances in 20 prod uct type groups were individually recorded. A summary of the energy use recorded can be found in Table 5 1 5 appliances recorded with data loggers included with the product type group. The appliances recorded in the appliance group. Similar to Table 5 1 4 columns shows the difference between what was recorded and the published UEC data. Two appliances stand out and are worth highlighting. First, is the energy consumption from a dehumidifier. The energy consumption recorded greatly exceeds the published First the sample size is only one so no conclusions should be drawn. The second factor to consider is that in this house the dehumidifier was used to reduce allergies for a small child. The dehumidifier was on for approximately 9 months a year, which is
148 unus ual especially for colder climates. The second appliance worth highlighting is the microwave. The microwave was recorded individually in 14 households. The average energy use is 62kWh/year and is 112% less than the published UEC. What makes this of par ticular interest is that over 95% of households own a microwave. These two appliances are recommended to be investigated further in future studies. Calculating MEL for Test Houses The first step in validating the model with the test houses is to calculate the MEL using the new model. The variables used in Model E are Income, Size of Home, Household Members, Home Business, and Household Members squared. All of the recorded characteristics of the 24 test houses are provided in Tables 5 1 2 and 5 1 3 but the characteristics used in the model are summarized in Table 5 1 6 As mentioned earlier the HERS index includes hot tubs with their MEL model. This study does not recommend including hot tubs with the other residual MEL due to its high UEC and low market s will be added to Model E so that it will have the same definition of MEL. The hot tub variable is a simple dichotomous variable meaning that when a hot tub is present the UEC of 2,040kWh/year will be added. Model E with the hot tub variable will be referred
149 Using the equation shown above, the annual MEL for all 24 test houses was shown in Table 5 1 6 The topic of calibration will be discussed later in this c hapter. The next step in validating the new model is to compare it with what the HERS index model predicted and what the MEL actually was. The HERS model only uses the squar e footage of a home to predict the MEL. The model multiplies the square footage by .91kWh/year. The HERS predicted MEL for all 24 test houses can be found in Table 5 1 7 Table 5 1 7 also provides the new model predicted MEL and the actual MEL monitored i indicates which of the two models more closely estimated the true MEL. Of the 24 homes monitored, the HERS model more accurately predicted the actual MEL 7 times whereas the new model was more accurate 17 times. What is perhaps a better indicator of the predictive power o f the model is the standard deviation of the difference between what the models predicted and the actual MEL. As mentioned earlier, the words, how tight the observation Of the 24 test houses, the standard deviation for the new model was 610kWh, which is significantly less than the HERS models standard deviation of 1,060kWh. Similarity in Test Homes Where the HERS Model was the Better Predictor As mentioned above the HERS model predicted the MEL value better in 7 out of the 24 test homes Those seven homes were evaluated to determine if there were any commonalities between them No appreciable pattern of occupant chara cteristics was
150 found with the seven homes. Income, size of home, marital status, age and housing type were all well distributed. The one significant pattern that was noticed was that the HERS model tended to predict homes with smaller MEL consumption bet ter than the new model. Table 5 18 lists all of the test houses in order of MEL consumption in kWh. Of the 10 homes with the lowest kWh MEL, the HERS model better predicts 7 of them. This comparison suggests that the new model better predicts homes with larger MEL consumption whereas the HERS model is more accurate for homes with smaller MEL consumption Calibration of the New Model An interesting trend that can be seen in Table 5 1 7 is that the new model tended to overestimate the actual MEL. Of the 24 homes, the New Model overestimated the actual MEL 17 times (71%). Because there is a fairly consistent pattern in the increase its predictive power. The energy modeling community generally limits the percent a model can be calibrated to +/ 15% (ASHRAE, 2002; Srinivasan, 2011c). The average difference between what the new model predicted from the actual was 155kWh (see Table 5 1 7 ). Using this average, 155kWh was reduced constant (y intercept). In other words, the constant from the New Model equation 1,024 would be reduced to 869. The average predicted MEL for the test houses was 2,031, so a calibration of 155kWh is 7.6% of the average and is well wi thin energy modeling
151 The results using the calibrated new model can be seen in Table 5 19 In the table the formula above was used to calculate the predicted MEL and was compared with the HERS model and the actual MEL of the 24 test houses. The calibrated new model predicted the MEL better than the HERS mod el for 19 of the test houses (79%). The calibrated model predicted the energy consumption of two more houses than the uncalibrated model. Statistical Review of Calibrated New Model In addition to testing if the calibrated new model was a better predictor of MEL as the HERS model, the data collected from the test houses was also used to statistically be specific to this study the confidence interval is what the new model predicts the MEL to be plus or minus the margin of error (857kWh) given a 95% confidence level. The data collected from the test houses mirror what was expected very closely ( Table 5 20 ) Of the 24 test houses where the true MEL was known, 23 of them (95.8%) were within the confidence interval. The MEL of test house MV6 was outside of the confidence interval. Having one test house (4.2% of the sample) outside the confidence interva l was very possible given the confidence level of 95%. It is interesting to note that although MV6 was
152 outside the confidence level, the new model was still a better predictor than the HERS model. Test Homes with Highest Deviation from Calibrated New Mode l As discussed earlier, the HERS model tended to predict lower MEL values better than the new model However, it is worth discussing the deviation of the six homes with the highest deviation. Table 5 21 shows the six test homes with the highest deviation The upper portion of the chart indicates the four test houses with the highest houses all had one unusual appliance with a high UEC that skewed their MEL higher than th e average household and what the model predicted. The test house with the highest deviation from the model and the only test house which exceeded the margin of error was test house MV6. MV6 was a single family house with two retired householders. The ho me was fairly unremarkable with the exception that as a hobby one of the householders was an amateur meteorologist. In the home there was a computer server that serviced a website to publish local weather conditions. The weather station computer server w as accompanied by two monitors, a battery backup, two surge protectors, and various other small accessories. These appliances were in the on mode 24 hours a day. The annual energy consumption of the weather system was 1,063kWh annually. If this system w as not in use the new model still would have predicted the energy consumption more closely than the HERS model and the actual MEL would have been within the confidence interval. The next three test houses where the model underestimated the energy use had similar situations where a single uncommon appliance skewed the energy consumption. Test house MV12 had a large turtle tank with a heat lamp, MV09 had a dehumidifier, and MV14 had a computer
153 server. Each of these appliances is fairly uncommon but ha s lar ge energy consumptions. There were two homes that the new model significantly overestimated energy consumption. Although it is difficult to say exactly which appliances these hou seholders were younger than 30 years old, were renters, and had incomes less than 20K. Another interesting characteristic is that both households were energy conscious and unplugged appliances when they were not in use. comparison was made between the estimated MEL of the RECS respondents (n = 12,083) using the HERS model and model E (including hot tubs). A similar comparison was done using the calibrated new model. The only di fference between model E (including hot tubs) and the calibrated new model is the constant number (y intercept). The calibration reduced the constant by 155kWh. The difference in average MELs for all respondents using both models is 341kWh or about 17.2% ( Table 5 22 ) The more significant statistic that this table shows is the standard deviation. The standard deviation of the MEL as predicted by the HERS index is 1,323kWh, whereas with the calibrated new model is only 598kWh. (The standard deviation s f or the new model and the calibrated new model are the same because only the constant was adjusted.) What this reduced standard deviation means is that the calibrated new model can predict the MEL of individual homes better than the HERS model by 54.8%. Ch apter Summary This study takes a unique approach to modeling MELs as square foot area is being supplemented with characteristics of the occupant to more accurately predict MEL a high
154 statistical significance due to the large sample size of the RECS. The new model was then tested on 24 actual homes. Of the 24 homes, the new model predicted the MEL better than the HERS model 19 times. Additionally, the HERS model and the new mo del were used to estimate the MEL of all of the 12,083 RECS respondents. The average energy consumption for all RECS respondents using both models was relatively close. What is significantly more important is the standard deviation of the predicted MELs. The standard deviation of the HERS model was twice that of the new model. What this means is that while the new model and HERS model explanatory power for entire populations is very close, the new model is more likely to predict the true MEL of individua l homes by 54.8%.
155 Table 5 1 Appliances not included in RECS model but added to new model Appliance Original Inclusion UEC Market Saturation Energy (kWh/yr) Source of Information Security System NEW 61 0.24 14.3 Roth et al., 2008 b CO Detector NEW 18 0.26 4.7 Hendron & Engebrecht, 2009 Smoke Detector NEW 4 0.84 3.4 Hendron & Engebrecht, 2009 Trash Compactor NEW 50 0.01 0.5 Hendron & Engebrecht, 2009 Lawn Mower (electric) NEW 100 0.05 5.0 Roth et al., 2008 b ; Market saturation from 2009 Californ ia RASS Kiln NEW 50 0.02 1.0 Roth et al., 2008 b Pipe / Gutter Heaters NEW 53 0.01 0.5 Roth et al., 2008 b Water Bed NEW 1096 0.02 25.2 Roth et al., 2008b Iron NEW 53 0.92 48.9 Roth et al., 2008b Baby Monitor NEW 18 0.10 1.8 Roth et al., 2008b ; Saturatio n from Hendron & Engebrecht, 2009 Hair Dryer NEW 42 0.86 36.1 Roth et al., 2008b Night Light NEW 8 2.00 16.0 McMahon & Nordman, 2004; Saturation from Porter et al., 2006 Tooth Brush NEW 12 0.06 0.7 Roth et a l., 2008b Image Scanner NEW 138 0.14 19.3 Saturation from 2009 California RASS; UEC from Roth et al., 2008 b Standby from Hendron & Engebrecht, 2009 Irrigation Time NEW 45 0.05 2.3 Hendron & Engebrecht, 2009 Aquarium (5 20 gal) NEW 105 0.02 2.5 Roth et al., 2008 b Marine Aquarium (5 20 gal) N EW 245 0.00 .6 Roth et al., 2008 b Total 183
156 Table 5 2 Occupant characteristics reviewed for significance Independent Variable Independent Variable Type Notes Year Home Constructed Quantitative Discrete Type of Housing Unit Categorical No minal AIA Climate Zone Categorical Nominal Average temperatures from 1971 2000 Building America Climate Region Categorical Nominal Urban vs. Rural Classification Categorical Nominal Based on US Census Classification Size of Home Quantitative Discrete Unit of Measure = square footage Number of Bedrooms Quantitative Discrete Number of Total Rooms Quantitative Discrete Level of Education Categorical Ordinal Number of Household Members Quantitative Discrete Marital Status Cate gorical Nominal Age of Householder Quantitative Discrete Number of Children in Household Quantitative Discrete Home Business Categorical Nominal Householder Home During Weekday Categorical Nominal Householder Retirement Status Catego rical Nominal Income Quantitative Discrete Income provided in ranges of 5K
157 Table 5 3 Descriptive s tatistics of i ndependent v ariables Range Min Max Mean Stand Dev Variance SD / Mean MEL 4,205 680 4,885 1,678 516.7 267,017 31% House Type 4 1 5 3 Categorical AIA Zone by CDD and HDD 4 1 5 3 Categorical BA Climate Region 4 1 5 3 Categorical Urban / Rural *** 1 0 1 80% Urban Categorical Year Made 89 1920 2009 1971 24.8 61 6 1% Year Made Range 7 1 8 4.0 Categorical Size of Home** 16,022 100 16,122 2,172 1,453 2.11E+06 67% Size of Garage 3 0 3 0.8 Categorical Education 8 0 8 3.4 Categorical Household Members 13 1 14 2.7 1.5 2.30 57% Marital Status 1 0 1 0.6 Categorical Householder Age 69 16 85 50 16.7 279.63 34% No. of Children 1 1 0 11 1 1.1 1.29 153% Home Business **** 1 0 1 10% Home Bus Categorical At Home During Day **** 1 0 1 60% Home During Day Categorical Retired **** 1 0 1 30% Retired Categorical Income 147 5 K 2 5 K 150 K 58 3 K 43 .3K 1.87E+09 74% Does not include H ot Tubs. ** Size provided in Square Feet ***Urban = 0, Rural = 1 **** Negative Response = 0, Affirmative Response = 1
158 Table 5 4 ANOV A t able for all i ndependent v ariables ANOVA* Sum of Squares df Mean Square F Sig. (p value) Type of Home Regression (model) 364,123,446 4 91,030,862 384 .000 Residual (error) 2,861,984,241 12,078 236,958 Total 3,226,107,687 12,082 AIA Climate Zone Regression (model) 9,881,466 4 2,470,366 9 .000 Residual (error) 3,216,226,221 12,0 78 266,288 Total 3,226,107,687 12,082 AIA Climate Zone (squared) Regression (model) 9,881,466 4 2,470,366 9 .000 Residual (error) 3,216,226,221 12,078 266,288 Total 3,226,107,687 12,082 BA Climate Region Reg ression (model) 9,007,555 4 2,251,889 8 .000 Residual (error) 3,217,100,132 12,078 266,360 Total 3,226,107,687 12,082 Urban / Rural Regression (model) 23,865,500 1 23,865,500 90 .000 Residual (error) 3,202,242,187 12,081 265,064 Total 3,226,107,687 12,082 Year Home Built Regression (model) 140,170,780 89 1,574,953 6 .000 Residual (error) 3,085,936,907 11,993 257,312 Total 3,226,107,687 12,082 Year Home Built (range) Regression (mode l) 65,983,002 7 9,426,143 36 .000 Residual (error) 3,160,124,685 12,075 261,708 Total 3,226,107,687 12,082 Size of Home Regression (model) 1,510,199,261 3,755 402,184 2 .000 Residual (error) 1,715,908,426 8,327 206,066 Total 3 ,226,107,687 12,082
159 Table 5 4. Continue d ANOVA* Sum of Squares df Mean Square F Sig. (p value) Regression (model) 395,176,574 3 131,725,525 562 .000 Residual (error) 2,830,931,113 12,079 234,368 Total 3,226,107,687 12,082 Educati on Regression (model) 266,956,376 8 33,369,547 136 .000 Residual (error) 2,959,151,310 12,074 245,085 Total 3,226,107,687 12,082 Number of Household Members Regression (model) 323,627,758 12 26,968,980 112 .000 Residual (error) 2,902,479,929 12,070 240,471 Total 3,226,107,687 12,082 Number of Household Members (squared) Regression (model) 323,627,758 12 26,968,980 112 .000 Residual (error) 2,902,479,929 12,070 240,471 Total 3,226,107,687 12,082 Marital Status Regression (model) 309,154,029 1 309,154,029 1,280 .000 Residual (error) 2,916,953,658 12,081 241,450 Total 3,226,107,687 12,082 Householder Age Regression (model) 184,105,234 68 2,707,430 11 .00 0 Residual (error) 3,042,002,452 12,014 253,205 Total 3,226,107,687 12,082 Householder Age (squared) Regression (model) 184,105,234 68 2,707,430 11 .000 Residual (error) 3,042,002,452 12,014 253,205 Total 3,226,107,687 12,082 Number of Children Regression (model) 102,273,299 11 9,297,573 36 .000 Residual (error) 3,123,834,388 12,071 258,788 Total 3,226,107,687 12,082
160 Table 5 4 Continue d ANOVA Sum of Squares df Mean Square F Sig. (p value) Num ber of Children (squared) Regression (model) 102,273,299 11 9,297,573 36 .000 Residual (error) 3,123,834,388 12,071 258,788 Total 3,226,107,687 12,082 Home Business Regression (model) 125,836,932 1 125,836,932 490 .000 R esidual (error) 3,100,270,755 12,081 256,624 Total 3,226,107,687 12,082 Home During Day Regression (model) 21,605,041 1 21,605,041 81 .000 Residual (error) 3,204,502,646 12,081 265,251 Total 3,226,107,687 12,082 Reti red Regression (model) 14,240,336 1 14,240,336 54 .000 Residual (error) 3,211,867,350 12,081 265,861 Total 3,226,107,687 12,082 Income (grouped by 40K for income estimation purposes) Regression (model) 570,003,803 3 190,001,268 8 64 .000 Residual (error) 2,656,103,884 12,079 219,894 Total 3,226,107,687 12,082 Income (grouped by 5K) Regression (model) 631,903,761 23 27,474,077 128 .000 Residual (error) 2,594,203,926 12,059 215,126 Total 3,226,107,687 12,082 Dependent Variable Miscellaneous Electrical Load (no hot tub) Table 5 5 Comparison of h ousing t ype with s ize and i ncome Size in Sqft Average Income Mobile Homes 1,000 32K Single Family Detached 2,700 67K Single Family Attached 1,900 57K Apartment (2 4 units) 1,100 37K Apartments (5+ units) 900 38K
161 Table 5 6 Independent v ariables in s tepwise r egression m odel in o rder of h ighest to l owest e xplanatory p ower Independent Variables Description 1 Income 2 Sqft of Home 3 # of H ousehold Members 4 Home Business 5 # of H ousehold Members (squared) 6 Size of Garage 7 Education 8 House Type 9 At Home During Day 10 No. of Children 11 House Holder AGE(squared) 12 Householder Age 13 Year Made 14 Marital Status 15 Retired 16 AIA Zone by CDD and HDD 17 AIA Zone by CDD and HDD (squared)
162 Table 5 7 Model s ummary Model R R Square R Square Change Margin of Error @ 95% (kWh) Independent Variable (see table 5 6 ) A 0.424 0.180 0.180 917 1 B 0.489 0.239 0.059 884 1,2 C 0.509 0.259 0. 020 872 1,2,3 D 0.523 0.274 0.014 863 1,2,3,4 E 0.534 0.285 0.011 857 1,2,3,4,5 F 0.541 0.293 0.008 852 1,2,3,4,5,6 G 0.546 0.298 0.005 849 1,2,3,4,5,6,7 H 0.551 0.304 0.006 845 1,2,3,4,5,6,7,8 I 0.555 0.309 0.005 842 1,2,3,4,5,6,7,8,9 J 0.557 0.310 0.002 842 1,2,3,4,5,6,7,8,9,10 K 0.558 0.312 0.002 841 1,2,3,4,5,6,7,8,9,10,11 L 0.564 0.318 0.006 837 1,2,3,4,5,6,7,8,9,10,11, 12 M 0.565 0.319 0.001 836 1,2,3,4,5,6,7,8,9,10,11, 12,13 N 0.565 0.320 0.000 836 1,2,3,4,5,6,7,8,9,10,11, 12,13,14 O 0.56 6 0.320 0.000 836 1,2,3,4,5,6,7,8,9,10,11, 12,13,14,15 P 0.566 0.320 0.000 836 1,2,3,4,5,6,7,8,9,10,11, 12,13,14,15,16 R 0.566 0.321 0.000 835 1,2,3,4,5,6,7,8,9,10,11, 12,13,14,15,16,17
163 Table 5 8 ANOVA t able for m odels A E ANOVA* Sum of Square s df Mean Square F Sig. (p value) Model A Regression (model) 631,903,761 23 27,474,077 128 .000 Residual (error) 2,594,203,926 12,059 215,126 Total 3,226,107,687 12,082 Model B Regression (model) 580,878,112 1 580,878, 112 2,653 .000 Residual (error) 2,645,229,575 12,081 218,958 Total 3,226,107,687 12,082 Model C Regression (model) 769,927,519 2 384,963,760 1,893 .000 Residual (error) 2,456,180,168 12,080 203,326 Total 3,226,107,687 12,082 Model D Regression (model) 835,852,700 3 278,617,567 1,408 .000 Residual (error) 2,390,254,987 12,079 197,885 Total 3,226,107,687 12,082 Model E Regression (model) 882,536,473 4 220,634,118 1,137 .000 Residua l (error) 2,343,571,214 12,078 194,036 Total 3,226,107,687 12,082 Dependent Variable Miscellaneous Electrical Load (no hot tub)
164 Table 5 9 Multico l linearity d iagnostic Multico l linearity Diagnostic Model Independent Variable / Dimens ions Tolerance VIF Condition Index A Income 1.000 1.000 3.027 B Income 0.787 1.270 3.649 Sqft of Home 0.787 1.270 3.991 C Income 0.779 1.283 3.493 Sqft of Home 0.777 1.287 4.390 Household Members 0.957 1.045 5.376 D Income 0.775 1.290 2.032 Sqft of Home 0.771 1.297 3.573 Household Members 0.957 1.045 4.479 Home Business 0.975 1.025 5.485 E Income 0.764 1.309 2.157 Sqft of Home 0.770 1.298 2.668 Household Members 0.112 8.968 4.592 Home Business 0.975 1.026 4.955 Household Members (squared) 0.114 8.772 15.415
165 Table 5 10 Coefficients of m odels A E Model IV# Constants and Coefficients Values Model Margin of Error @ 95% (kWh) A (Constant) 1382 917 1 Income 0.0051 B (Constant) 1259 884 1 Income 0.0036 2 Sqft of Home 0.0970 C (Constant) 1153 872 1 Income 0.0034 2 Sqft of Home 0.0903 3 Household Members 50 D (Constant) 1152 863 1 Income 0.0033 2 Sqft of Home 0.0860 3 Household Members 49 4 Home Business 226 E (Constant) 1024 857 1 Income 0.0031 2 Sqft of Home 0.0846 3 Household Members 151 4 Home Business 221 5 Household Members (squared) 14 Table 5 11 Estimation of the MEL of the RECS r espondents using the HERS and n ew m odel Sample (n) Calculated RECS MEL (kWh ) HERS Model (kWh) New Model (kWh) Difference (kWh) Difference (%) Model Average* 12,083 1,795 1,977 1,792 185 9.4% Standard Deviation 1,323 598 725 54.8% Hot tubs included in model values
166 Table 5 12 Survey r esults: h ouse c haracteristic s ID# Type of Home* Urban /Rural Size (sqft) Bed rooms Baths Total Rooms Garage Own /Rent MV01 SFD U 2,700 5 2 14 None Rent MV02 SFD U 2,200 4 2 13 2 Car Rent MV03 SFD U 2,489 5 2 14 2 Car Own MV04 SFA U 1,148 2 2 6 1 Car Own MV05 SFD U 2,008 4 2 9 2 Car Own MV06 SFD U 1,524 3 2 9 1 Car Own MV07 SFD U 1,481 3 2 8 1 Car Own MV08 SFD U 1,714 3 2 9 None Own MV09 SFD U 1,118 3 2 7 None Own MV10 Apartment U 1,200 2 2 7 None Rent MV11 SFD U 1,860 3 2 9 2 Car Own MV12 SFD U 1,734 3 2 9 2 Car Own MV 13 SFD R 2,955 5 5 15 2 Car Own MV14 SFD U 3,697 5 5 15 3 Car Own MV15 SFD U 1,152 3 1 6 None Own MV16 Townhouse U 1,420 2 2 6 1 Car Own MV17 SFD U 2,130 3 2 10 2 Car Own MV18 SFD U 1,040 3 2 8 1 Car Own MV19 SFD U 2,834 3 3 11 3 Car Own MV20 Mobile Home R 1,728 3 2 8 None Rent MV21 SFD U 2,707 5 3 12 2 Car Rent MV22 SFD U 1,452 2 1 6 1 Car Own MV23 SFD U 1,902 4 2 10 2 Car Rent MV24 SFD U 2,531 4 3 11 2 Car Own SFD = Single Family Detached; SFA = Single Family Attached
167 Table 5 13 Surve y results: o ccupant c haracteristics ID# HH Mem Married Home Bus in Home During day Education Income Range Estimated Income MV01 5 Married No Yes Masters 80K 119K $101,677 MV02 3 Married Yes Yes Bachel ors 120K+ $141,637 MV03 7 Married No Yes Masters 40 K 79K $59,064 MV04 1 Single No Yes Bachel ors <40K $12,323 MV05 2 Married No No Masters 80K 119K $100,399 MV06 2 Married No Yes Associates 80K 119K $99,114 MV07 4 Married No No Bachel ors 80K 119K $98,546 MV08 3 Married No No Masters 80K 119K $100,122 MV09 4 Married No No Masters 40K 79K $57,775 MV10 2 Married No No Bachel ors <40K $15,336 MV11 2 Married No Yes Masters 80K 119K $99,631 MV12 2 Married No No Bachel ors 80K 119K $99,412 MV13 5 Married Yes Yes Assoc iates 80K 119K $102,34 3 MV14 3 Married Yes Yes Ph.D. 80K 119K $104,700 MV15 5 Married No Yes Bachel ors 40K 79K $57,078 MV16 1 Single No No Ph.D. 40K 79K $57,181 MV17 3 Married No No Profess ional 80K 119K $102,498 MV18 5 Married No Yes Masters <40K $16,543 MV19 4 M arried No No Masters 120K+ $143,590 MV20 1 Single No Yes Masters <40K $14,225 MV21 6 Married No No Masters 120K+ $142,843 MV22 3 Married No Yes Ph.D. 80K 119K $101,962 MV23 3 Married No Yes Bachelors 40K 79K $58,411 MV24 5 Married No Yes Masters 4 0K 79K $59,103
168 Table 5 14 Comparison between a veraged r ecorded e nergy c onsumption and p ublished UEC. Recorded with Data Logger Published UEC Difference (kWh) Difference (%) Cumulative Energy Use from Data Loggers (kWh) 13,851 13,933 82 0.6% Average Difference from Recorded to Published 119 % Median Difference from Recorded to Published 20% Data Logging Events (2 week periods)* 140 Appliances Recorded* 271 Uncommon appliances recorded without published UEC not included T able 5 1 5 Comparison between e nergy u se of i ndividual a ppliances and p ublished UEC. Sample Size (n) Average Recorded Consumption Published UEC Difference (kWh) Difference (%) Toaster Oven 2 34 33 1 3% Noise Maker 3 8 7 1 9% Laptop 2 59 72 13 22% Cable Set Top Box 1 173 133 40 23% Cell Phone Charger 1 3 4 1 25% Garage Door 2 41 30 11 27% Coffee Maker 7 89 59 30 34% DVD/VCR Combo 1 113 50 63 56% Electric Kettle 1 197 75 122 62% Clock Radio 3 24 9 15 62% Dehumidifier 1 1,244 400 844 68% Phone with Integrated Answering Machine 2 18 31 13 72% Radio 5 36 9 27 75% TV Converter Box 1 29 4 24 85% Toaster 8 21 39 18 86% Digital Picture Frame 1 47 88 41 88% Fan 4 82 8 74 90% Microwave 14 62 131 69 112% Phone 1 12 26 14 123% Iron 1 2 53 51 2 339%
169 Table 5 1 6 Calculated MEL using n ew MEL m odel ID# Estimated Income Size (sqft) H H Mem Home Business H H Mem Squared Hot Tub Uncalibrated New Model (kWh) MV01 101,677 2,700 5 No 25 No 1,965 MV02 141,637 2,200 3 Yes 9 No 2,185 MV03 59,064 2,489 7 No 49 No 1,785 MV04 12,323 1,148 1 No 1 No 1,296 MV05 100,399 2,008 2 No 4 No 1,743 MV06 99,114 1,524 2 No 4 No 1,698 MV07 98,546 1,481 4 No 16 No 1,827 MV08 100,122 1,714 3 No 9 No 1,798 MV09 57,775 1,118 4 No 16 No 1,674 MV10 15,336 1,200 2 No 4 No 1,419 MV11 99,631 1,860 2 No 4 No 1,728 MV12 99,412 1,734 2 No 4 Yes 3,757 MV13 102,343 2,955 5 Yes 25 Yes 4,250 MV14 104,700 3,697 3 Yes 9 Yes 4,242 MV15 57,078 1,152 5 No 25 No 1,700 MV16 5 7,181 1,420 1 No 1 No 1,454 MV17 102,498 2,130 3 No 9 No 1,841 MV18 16,543 1,040 5 No 25 No 1,569 MV19 143,590 2,834 4 No 16 No 2,077 MV20 14,225 1,728 1 No 1 No 1,351 MV21 142,843 2,707 6 No 36 No 2,086 MV22 101,962 1,452 3 No 9 No 1,782 MV23 58,41 1 1,902 3 No 9 No 1,689 MV24 59,103 2,531 5 No 25 No 1,823
170 Table 5 1 7 New m odel and HERS m odel c omparison with a ctual MEL from t est h ouses ID# Uncalibrated New Model (kWh) HERS Model (kWh) Actual Test Home MEL (kWh) New Model to Actual (kWh) H ERS to Actual (kWh) Better Predictor MV01 1,965 2,457 1,909 56 548 New Model MV02 2,185 2,002 2,465 279 463 New Model MV03 1,785 2,265 1,663 123 602 New Model MV04 1,296 1,045 1,067 229 22 HERS MV05 1,743 1,827 1,028 715 800 New Model MV06 1,698 1 ,387 3,426 1,728 2,039 New Model MV07 1,827 1,348 2,199 372 852 New Model MV08 1,798 1,560 1,183 615 377 HERS MV09 1,674 1,017 2,166 492 1,148 New Model MV10 1,419 1,092 516 903 576 HERS MV11 1,728 1,693 1,179 549 514 HERS MV12 3,757 1,578 4,39 4 637 2,816 New Model MV13 4,250 2,689 4,567 318 1,878 New Model MV14 4,242 3,364 4,684 443 1,320 New Model MV15 1,700 1,048 1,375 324 327 New Model MV16 1,454 1,292 935 519 357 HERS MV17 1,841 1,938 1,198 643 740 New Model MV18 1,569 946 1,22 3 346 276 HERS MV19 2,077 2,579 2,006 71 573 New Model MV20 1,351 1,572 447 904 1,125 New Model MV21 2,086 2,463 1,336 750 1,128 New Model MV22 1,782 1,321 1,096 685 225 HERS MV23 1,689 1,731 1,442 247 288 New Model MV24 1,823 2,303 1,523 300 780 Ne w Model Averag e 2,031 1,772 1,876 155 105 17NM / 7HERS Standard Deviation 610 1,060
171 Table 5 18. New Model in order of MEL magnitude Order ID# Uncalibrated New Model (kWh) HERS Model (kWh) Actual Test Home MEL (kWh) Better Predictor 1 MV20 1, 351 1,572 447 New Model 2 MV10 1,419 1,092 516 HERS 3 MV16 1,454 1,292 935 HERS 4 MV05 1,743 1,827 1,028 New Model 5 MV04 1,296 1,045 1,067 HERS 6 MV22 1,782 1,321 1,096 HERS 7 MV11 1,728 1,693 1,179 HERS 8 MV08 1,798 1,560 1,183 HERS 9 MV17 1,841 1,938 1,198 New Model 10 MV18 1,569 946 1,223 HERS 11 MV21 2,086 2,463 1,336 New Model 12 MV15 1,700 1,048 1,375 New Model 13 MV23 1,689 1,731 1,442 New Model 14 MV24 1,823 2,303 1,523 New Model 15 MV03 1,785 2,265 1,663 New Model 16 MV01 1,965 2,45 7 1,909 New Model 17 MV19 2,077 2,579 2,006 New Model 18 MV09 1,674 1,017 2,166 New Model 19 MV07 1,827 1,348 2,199 New Model 20 MV02 2,185 2,002 2,465 New Model 21 MV06 1,698 1,387 3,426 New Model 22 MV12 3,757 1,578 4,394 New Model 23 MV13 4,250 2 ,689 4,567 New Model 24 MV14 4,242 3,364 4,684 New Model Average 2,031 1,772 1,876 17NM / 7HERS
172 Table 5 19 Calibrated new m odel and HERS m odel c omparison with a ctual MEL from t est h ouses ID# Calibrated New Model (kWh) HERS Model (kWh) Actual Te st Home MEL (kWh) New Model to Actual (kWh) HERS to Actual (kWh) Better Predictor MV01 1,810 2,457 1,909 99 548 New Model MV02 2,030 2,002 2,465 434 463 New Model MV03 1,630 2,265 1,663 32 602 New Model MV04 1,141 1,045 1,067 74 22 HERS MV05 1 ,588 1,827 1,028 560 800 New Model MV06 1,543 1,387 3,426 1,883 2,039 New Model MV07 1,672 1,348 2,199 527 852 New Model MV08 1,643 1,560 1,183 460 377 HERS MV09 1,519 1,017 2,166 647 1,148 New Model MV10 1,264 1,092 516 748 576 HERS MV11 1,573 1,693 1,179 394 514 New Model MV12 3,602 1,578 4,394 792 2,816 New Model MV13 4,095 2,689 4,567 473 1,878 New Model MV14 4,087 3,364 4,684 598 1,320 New Model MV15 1,545 1,048 1,375 169 327 New Model MV16 1,299 1,292 935 364 357 HERS MV17 1,6 86 1,938 1,198 488 740 New Model MV18 1,414 946 1,223 191 276 New Model MV19 1,922 2,579 2,006 84 573 New Model MV20 1,196 1,572 447 749 1,125 New Model MV21 1,931 2,463 1,336 595 1,128 New Model MV22 1,627 1,321 1,096 530 225 HERS MV23 1,534 1,731 1,442 92 288 New Model MV24 1,668 2,303 1,523 145 780 New Model Average 1,876 1,772 1,876 0 105 19NM / 5HERS Standard Deviation 610 1,060
173 Table 5 20 Confidence i nterval for c alibrated n ew m odel and t est h ouses ID# Calibrated New Model (kWh) Confidence Interval* Actual Test Home MEL (kWh) Actual Within Confidence Interval Better Predictor MV01 1,810 (953 2,667) 1,909 Yes New Model MV02 2,030 (1,173 2,887) 2,465 Yes New Model MV03 1,630 (773 2,487) 1,663 Yes New Model MV04 1,141 (28 4 1,998) 1,067 Yes HERS MV05 1,588 (731 2,445) 1,028 Yes New Model MV06 1,543 (686 2,400) 3,426 No New Model MV07 1,672 (815 2,529) 2,199 Yes New Model MV08 1,643 (786 2,500) 1,183 Yes HERS MV09 1,519 (662 2,376) 2,166 Yes New Model MV10 1,264 (407 2,121) 516 Yes HERS MV11 1,573 (716 2,430) 1,179 Yes New Model MV12 3,602 (2,745 4,459) 4,394 Yes New Model MV13 4,095 (3,238 4,952) 4,567 Yes New Model MV14 4,087 (3,230 4,944) 4,684 Yes New Model MV15 1,545 (688 2,402) 1,375 Y es New Model MV16 1,299 (442 2,156) 935 Yes HERS MV17 1,686 (829 2,543) 1,198 Yes New Model MV18 1,414 (557 2,271) 1,223 Yes New Model MV19 1,922 (1,065 2,779) 2,006 Yes New Model MV20 1,196 (339 2,053) 447 Yes New Model MV21 1,931 (1,074 2,788) 1,336 Yes New Model MV22 1,627 (770 2,484) 1,096 Yes HERS MV23 1,534 (677 2,391) 1,442 Yes New Model MV24 1,668 (811 2,525) 1,523 Yes New Model Margin of Error at 95% Confidence is 857kWh
174 Table 5 21 Six t est h ouse with h ighest d eviation from m odel ID# Calibrated New Model (kWh) Actual Test Home MEL (kWh) New Model to Actual (kWh) Better Predictor Notes Underestimated MV06 1,543 3,426 1,883 New Model Weather Station Server MV12 3,602 4,394 792 New Model Large Turtle Ta nk with Lamp MV09 1,519 2,166 647 New Model Dehumidifier MV14 4,087 4,684 598 New Model Computer Server Overestimated MV10 1,264 516 748 HERS Newly married couple in rental MV20 1,196 447 749 New Model Single, young college grade, income <15K T able 5 22 Estimation of the MEL of the RECS r espondents using the HERS and calibrated new m odel Sample (n) Calculated RECS MEL (kWh) HERS Model (kWh) Calibrated New Model (kWh) Difference (kWh) Difference (%) Model Average* 12,083 1,795 1,977 1,636 341 17.2% Standard Deviation 1,323 598 725 54.8% Hot tubs included in model values
175 Figure 5 1. Breakdown of calculated MEL for all RECS r espondents by p ercentage
176 Figure 5 2. Comparison between the a ge of the h ome and MEL Figure 5 3. Distribution of MEL s tandardized r esiduals 1400 1500 1600 1700 1800 1900 2000 2100 Before 1950 1950 1959 1960 1969 1970 1979 1980 1989 1990 1999 2000 2004 2005 2009 MEL kWh Year Home Built
177 Figure 5 4. Homogeneity of v ariances for all p redicted MEL v alues
178 Figure 5 5. Comparison between the t ype of h ome and MEL Figure 5 6. Comparison between AIA c limate z one and MEL 0 500 1000 1500 2000 2500 Mobile Homes Single Family Detached Single Family Attached Apartment (24 units) Apartments (5+ units) MEL kWh Housing Type 1650 1700 1750 1800 1850 1900 MEL kWh AIA Climate Zone
179 F igure 5 7. Comparison between B uilding A merica c limate r egion and MEL Figure 5 8. Comparison between h ome s ize and MEL 1650 1700 1750 1800 1850 1900 MEL kWh Building America Climate Regions 0 500 1000 1500 2000 2500 <500 500 999 1,000 1,499 1,500 1,999 2,000 2,499 2,500 2,999 3,000+ MEL kWh Home Size (sqft)
180 Figure 5 9. Comparison between h ome s ize and i ndividual MEL a ppliances. <500 500 999 1,000 1,499 1,500 1,999 2,000 2,499 2,500 2,999 3,000+ Computer 62 81 116 135 154 169 201 Well Pump 2 7 12 20 24 26 34 Rechargeable Electronics 23 33 44 56 64 69 82 Spa 0 10 40 100 142 152 265 TV Set Top Boxes 98 156 205 228 252 256 287 Computer Monitor 14 20 28 35 40 44 48 Cable/DSL Modem 17 25 31 36 39 40 44 Home Theater 6 12 15 21 25 27 32 TV Peripherals 28 42 55 63 63 67 74 Wireless Router 12 17 21 25 27 30 34 Copier 11 12 21 29 32 34 49 Humidifier 4 7 11 14 16 21 24 Computer Spa TV Set Top Boxes 0 50 100 150 200 250 300 350 MEL kWh Home Size (sqft) Comparison of Household Size and Individual MEL Appliances Computer Well Pump Rechargeable Electronics Spa TV Set Top Boxes Computer Monitor Cable/DSL Modem Home Theater TV Peripherals Wireless Router Copier Humidifier
181 Figure 5 10. Comparison between the t otal n umber of r ooms in the h ome and MEL Figure 5 11. Comparison between n umber of b edrooms in the h ome and M EL 0 500 1000 1500 2000 2500 3000 3500 <5 5 7 8 10 11 13 14 16' 17+ MEL kWh Total Number of Rooms in Home 0 500 1000 1500 2000 2500 3000 1BR 2BR 3BR 4BR 5BR 6BR 7BR MEL kWh Number of Bedrooms in Home
182 Fi gure 5 12. Comparison between n umber of b edrooms in h ousehold and h ousehold m embers 0 1 2 3 4 5 6 7 Number of Household Members Members in Household Members in Household
183 Fi gure 5 13. Comparison between n umber of p eople in t he household and MEL Figure 5 14. Comparison between h ouseholds with h ome b usiness and MEL 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 MEL kWh Household Members Living in Home 0 500 1000 1500 2000 2500 Home Business No Home Business MEL kWh Households with Home Business
184 Figure 5 15. Comparison between the h e ducation l evel and MEL Figure 5 16 Comparison between m arital s tatus and MEL 0 500 1000 1500 2000 2500 MEL kWh Education Level 0 500 1000 1500 2000 2500 Married Not Married MEL kWh Marital Status
185 Figure 5 17. Comparison between the a ge of the h ouseholder and MEL Figure 5 18. Comparison between the n umber of c hildren in the household and MEL 0 500 1000 1500 2000 2500 <30 30 39 40 49 50 59 60 69 70+ MEL kWh Age of Householder (years) 0 500 1000 1500 2000 2500 0 1 2 3 4 5 6 MEL kWh Number of Children in Home
186 Fi gure 5 19. Comparison between h ouseholds with m embers h ome d uring w eekdays and MEL Figure 5 20. Compa rison between the h ouseholders r etirement s tatus and MEL 1680 1700 1720 1740 1760 1780 1800 1820 1840 1860 At Home During Weekday Not At Home During Weekday MEL kWh Home During Day 1700 1720 1740 1760 1780 1800 1820 1840 Not Retired Retired MEL kWh Retirement Status
187 CHAPTER 6 W HOLE HOUSE SWITCH Overview Over the past 30 years, the intensity of all major energy use categories has decreased in the residential market, with the exception of miscellaneous electr ical loads (MELs). MELs stand alone as the single category in which energy intensity has steadily increased over time (EIA, 2011; Nordman and Sanchez, 2006; Parker et al., 2011; KEMA, 2010). Additionally, according to a report commissioned by the US Depa rtment of Energy, MELs will grow to 36% of the energy used in code compliant homes by 2020 (Roth et al., 2008b). Therefore, the reduction of MELs is a key area of research for national energy reduction and for achieving zero net energy homes. Current prac tices for reducing MELs fall into the two general categories of technical or behavioral (Mohanty, 2001). Perhaps the most obvious of the technical improvements is from increased energy efficiency from advances in technology. Old style CRT computer monito rs for example use twice the energy to operate than modern LCD monitors and four times as much as Energy Star rated units (Roth et al., 2008b). Equipping units with sleep or low power modes is another improvement that many appliances have adopted. Howeve r, having the option of a low power mode is only effective if the owner chooses to use it. Many manufacturers have encouraged reduced energy use by having low power setting enabled as the factory default. Providing energy guide labels that inform consume rs of expected energy cost has been very successful with major appliances. Having smaller appliances list their energy consumption would similarly use market pressure to decrease energy use (International Energy Agency, 2001). Changing the behavior of th e occupant is another way of
188 reducing MELs. Opower is a company that partners with utility providers to analyze households and provide them with comparisons between their energy use and their neighbors. Participation in the Opower program has yielded an average savings of 2.8% of the homes total energy consumption with some municipalities experiences over a 6% decrease (Parker et al., 2011). Home automation through the use of timers and occupancy sensors is another way of reducing MELs. Energy dashboard s and smart meters are devices or networks of devices that provide the householder real time feedback on their energy use. Several nationwide studies have shown that when householders are provided this instant feedback total energy consumption is reduced from 5 15% (Parker et al., 2011). Stand by power is the energy used by an appliance when it is not functioning or is in the off mode. It is sometimes called a phantom load, vampire draw or leaking electricity. Existing literature indicates that stand b y power accounts for 2 8% of a and Huber, 1997). Several measures are available to reduce stand by power. Perhaps the most effective is to simply unplug the appliance when it i s not used. The inconvenience of this measure especially in areas where the plug is not easily accessible has kept this from being a widely used measure. Another option is to use smart power strips where peripheral equipment is controlled by the primary appliance. Technical improvements to appliances such as more efficient power supplies (low voltage transformers) can reduce stand by power by 40% (Mohanty, 2001). To decrease stand by power, one energy efficiency measure (EEM) that households have avail able is the whole house switch (WHS). The WHS uses switches
189 to send wireless signals to various disconnectors that sever the power to appliances. The switches are generally conveniently located at home exits and the master bedroom. The concept is that ho useholders can eliminate stand by power loss when they leave the home or go to sleep. This study uses data from the Residential Energy Consumption Survey (RECS), with a sample size of 12,083, conducted by the US Energy Information Agency (EIA) to estimate the effectiveness of the WHS in reducing home power consumption. The calculated results were then verified by testing the WHS with 24 typical single family homes. Whole House Switch Defined The premise behind the WHS is that it conveniently reduces stan d by power loss with minimal interruption of services to the householder. For this study the WHS has two primary components. First is the disconnector, which is used to sever power to the appliance (Figure 1) A plug type disconnector has the appliance plug into it, and then it plugs into a wall receptacle; in this way, it is similar to a common surge protector. An integrated disconnector works similarly, but it replaces the traditional electrical The integrated disconnector looks identical to a common electrical receptacle with all relay switches lying behind the wall plate (Figure 1) Both the plug type and integrated type disconnectors are typically controlled wirelessly. This measure is idea l for retrofit applications due to the ease of installation and because the disconnectors can be moved with appliances as they are relocated. For new construction, homeowners have the option of installing hardwired switches in area of expected high stand by loss like the primary television, home office and kitchen outlets.
190 The second primary component of the WHS is the controller that commands the disconnectors. A hardwired controller can be set up as a common house switch, or a free standing remote cont rol can be used (Figure 6 2) Controllers can command as many disconnect switches as the householder requires. Controllers commonly use an omni directional radio frequency transmitter with a range radius of 50 100 feet. Each disconnect switch also conta ins a repeater that receives and then rebroadcasts the command. The more disconnectors there are on the network, the larger the web becomes. The technology required for the WHS is available off the shelf. Wireless communication protocols such as Z Wave and Zigbee are being used by manufacturers such as General Electric, Honeywell, Levitron, and Black and Decker. Using a common communication protocol allows components from different manufacturers to be used together in the same home network. Although th is study focuses on the reduction of stand by power, this technology can also be adopted to reduce other energy demands. Thermostats, lighting, and entire circuits can be controlled with these wireless protocol enabled devices. Residential Energy Consum ption Survey Since 1978, the US Energy Information Agency has been conducting the Residential Energy Consumption Survey (RECS). The survey is very detailed and contains over 90 pages of questions. The questions primarily relate to home energy use but al ages, income, and characteristics of the home. The survey also asks about appliances in the home such as coffee makers, toasters, computers, printers, and DVD players. More detai led questions, such as duration of use, type, and size, are asked about more
191 energy intensive appliances, such as televisions, computers, and microwaves. The survey does not rely completely on the responses of the occupant but verifies the information with utility and tax records as well as field measurements, to the extent possible (US EIA, 2011). The RECS is published every four years with the most current survey conducted in 2009. The 2009 RECS is different from previous years as it makes available to the public the individual responses from over 12,000 households interviewed. The current study will make use of this large dataset to estimate the potential savings of the WHS. Assumptions Used to Calculate the WHS Energy Savings Despite the comprehensive nature of the RECS questions, not all of the information needed to calculate the effectiveness of the WHS was provided. Specifically, reasonable assumptions of the length of time that the householder would implement the WHS were made. The three most sig nificant periods of time when the EEM could be implemented are when the homeowner is asleep, away during the day, and away for overnight trips. According to the US Department of Labor, the average American adult sleeps between 8.2 and 9.0 hours per day (2 012). However, of the 12,000 householders who participated in the RECS, only 12% lived by themselves, and there is insufficient data in the literature to determine when all members of a household are asleep. Since that is a period during which the WHS wou ld be activated, an assumed time had to be used for this study: the assumption of 6 hours was made. The RECS asked the respondents whether someone was home throughout the day. For the respondents who indicated that there was someone home throughout the d ay, it was assumed that the home was vacated for 2 hr/weekday to account for miscellaneous errands. For households that did not have someone at home during the day, US
192 Census data (2009) was used to estimate that they were away 9.25 hr/weekday for work (8 hrs), commute to work (45 min), and miscellaneous errands (30 min). All householders were assumed to be away 4 hr per weekend day. Another important consideration is the number of days the householder was away traveling. The US Department of Labor esti mates that the average paid time off for full time workers ranges from 7 18 days per year (1996). In addition, the average full time worker has between 7 and 9 paid holidays per year. Although not intended for recreational use, paid sick days, which typi cally range from 7 11, can often be converted to paid time off. For this study, it was assumed that all householders were away from home for 14 days per year. See Table 6 1 for a summary of the assumed hours that the WHS was implemented. Calculation for Appliances Addressed in the RECS The RECS is very comprehensive and asks the survey respondent about appliances that make up, on average, about two Some of the appliances, such as fax machines and cordless phones, us e much of their energy in stand by mode, but it is unlikely that the householder would want to disconnect the power to these devices when they are asleep or away from home. Table 6 2 provides the list of appliances included in the RECS that have good pote ntial for stand by power savings. For some of the appliances listed, the only information that the RECS provides is whether the appliance was present in the home. For appliances reported to be present, the unit energy consumption (UEC) from the literatur e review was used to calculate the total load (Hendron and Engebrecht, 2010; Roth et al., 2008a; Roth et al., 2008b). The UEC is the estimated total annual energy used by the average person with a typical model appliance. It averages usage patterns over a range of
193 people and weighs the energy consumption of the different appliance models based on their market penetration. Other appliances, such as television peripherals, computers, monitors, microwaves, rechargeable electronics, and rechargeable tools, h ave more detailed information collected by the RECS about the appliance type and usage. When present, this information was used to calculate the UEC for the corresponding survey respondent. Whether the homeowner is home during the day is directly linked to how many hours the WHS would be used. The estimated savings were calculated from using the WHS based on the hours of assumed WHS activation listed in Table 6 1 and based on the wattage values provided in Table 6 2. Calculation for Appliances Not Addres sed in the RECS The RECS did not ask the respondents about all their appliances and those not asked about create approximately one E Program (Hendron and Engebrecht, 2010; RESNET, 2006). A list of the most common appliances not included in the RECS was created. The UEC of each of these appliances as found in a literature review (KEMA, 2010; Roth et al., 2008b; Sanchez et al., 1998) was then multiplied by its market saturation to estimate the typical energy load of the appliance as a national average. Market saturation is defined as the average number of a ppliance per home (total appliance / all homes). A similar method was used to calculate the WHS energy savings. The wattage for each appliance in its lowest power mode was multiplied by its market saturation and by the weighted average of the number of h ours the EEM would be activated (Table 6 3). Specifically, this study
194 compiled the UEC, stand by energy use, and market saturation of 63 appliances resulting in an average savings potential of 20.5kWh/year per home. Effectiveness of WHS The methods descri bed above were applied to all 12,083 RECS respondents in order to calculate the effectiveness of hypothetically utilizing the WHS in these homes. Results indicate that the average WHS savings over all the respondents would be 282 kWh/year, which is 7.9 4). The range is one standard deviation from the mean and captures approximately two thirds of the homes. Factors like varying efficiencies of appliance models, number of appliances, individual usage patter ns and householder preferences cause a high degree of variability in residential MELs. This high variability is reflected in Table 6 5 with a relatively high standard deviation. The RECS provided the yearly energy consumption for all of the survey respon dents. The average respondent used 11,288kWh per year. This equates to a WHS savings of 1.2 cost. Householders with specific characteristics, such as urban vs. rural, married vs. unmarried, and various in one standard deviation) divided by the average total utility for the various household group. The results indicate that the overall savings varies moderately across the different occupant groups, ranging from 1.1 4.7%. The occupant characteristics that had the most influence on the effectiveness of the WHS was whether the householder was home throughout the day or not. The savings potential and payback peri ods will be discussed below. Savings and Simple Payback Period for Common WHS Network: The vast majority of stand by energy loss is from television peripherals, computers, computer
195 peripherals, and kitchen appliances (Table 6 4). These appliances are u sually clustered, thus multiple appliances could be controlled by a single disconnect switch. Every home is configured differently, but if the typical home were to have a WHS system with five disconnectors that controlled the primary television peripheral s (1 disconnector), secondary TV with peripherals (1), primary computer with peripherals (2), and a modem and wireless router (1), an average savings of 209 kWh/year would be achieved. This is approximately $25 per year ($0.118 /kWh) in energy savings, and it is disconnector system with two controllers would cost approximately $170.00 retail (www.Zwaveproducts.com) and would pay for itself in 6.9 years. Testing of WHS Calculations The WHS savings calcu lations using the RECS were tested by simulating the WHS on 24 actual homes. Twenty four homes are not sufficient to statistically validate the model; loads were two week period. Limitations of the study did not allow for an actual WHS to be installed in each of the test homes. The information collected from the data loggers was used to only used periodically were accounted for using the UEC obtained from a literature review (Hendron and Engebrecht, 2010; KEMA, 2010; Roth et al. 2008a; Roth et al., 2008b; Sanchez et al., 1998). The lowest wattage for each appliance, as recorded by the data loggers, was used to calculate the stand by loss savings if a WHS was installed. Table 6 6 provides a summary of the findings. The test hom es were closely divided between households home (14) expected savings, calculated based on the RECS data, if all available appliances are on a WHS
196 network. Applianc es like refrigerators and security systems that are not applicable to this type potential if a WHS was installed in the test houses. This is based on the actual appli ances observed and recorded with data loggers. Each of the 24 householders were questioned about how often they are asleep and away from the home to estimate the likely duration the WHS would be implemented. From the empirical rule it is expected that ap proximately 68 % of the standard deviations. This distribution around the mean was observed with 18 test houses (75%) within one standard deviation and 6 test house s (25%) within two standard deviations. Not all appliances have the same stand by power savings potential. In real world situations, homeowners would purchase enough disconnecting devices to control the most energy consuming appliances and to control appliances that are clustered together. For each of the 24 test homes, reasonable retrofit packages were selected. Each home was assumed to have a stationary switch at the primary exit and one remote control in the master bedroom. The number of disconne ctors varied by the type and clustering of appliances observed in each test home. The retrofit packages and savings are provided in Table 6 6. Households with someone home during the day experienced an average savings of 115 kWh/year ($13.57), so the sys tem would take 13.9 years to pay for itself. As expected, householders away during the day would enjoy a much higher return, with an average savings of 184 kWh/year ($21.76), corresponding to a simple payback period of 7.2 years. The average return on in vestment is not very favorable. However it is important to note that of the 24 houses, 5 of them (21%) had simple payback period of 5 years or less. Test house MV07 was projected to experience savings of 440 kWh ($51.88) with a payback of 2.5 years. If there is one finding that should be taken away common households that are away from the home during the day and with heavy saturations of
197 television peripherals, co mputers and office equipment would have high energy savings and a short payback period. Chapter Summary The energy intensity of miscellaneous electrical loads (MELs) fueled by expansion in the home entertainment, personal electronics, and convenience items markets, has steadily outpaced other energy end uses. The whole house switch (WHS) is an energy efficiency measure that reduces MELs by eliminating much of the stand by power loss. Using data from the RECS the effect of the WHS on energy costs was calcu lated. It was found that the WHS has the potential to save the average household 282 kWh/year or 7.9 total MEL. This corresponds to an approximate savings of 1.2 electrical consumption. The calculations w ere tested by simulating the WHS on 24 actual single family homes. The results from the 24 sample homes were within one standard deviation from what was predicted using the RECS. Sample retrofit WHS packages were simulated and showed that the effectivene ss of the WHS was greatly influenced by whether the householder was home during the day. A principle finding of the study is that the savings that can be enjoyed by the WHS is fairly modest when averaged over a large population; however, they can be subst antial for certain householders. Specifically, householders away from the home during the day and with heavy saturations of television peripherals, computers and office equipment would have high energy savings and a short payback period.
198 Table 6 1 Hour s assumed that the WHS would be activated All Household Members Asleep Away from Home on Weekday Away from Home on Weekend Traveling Total At Home During Day 6hrs/day 2hrs/day 4hrs/day 14 days/yr (336hrs/yr) 3,336hrs/yr Not At Home During Day 6hrs/day 9 .25hrs/day 4hrs/day 14 days/yr (336hrs/yr) 5,149hrs/yr Table 6 2. Power u sage i nformation for a ppliances i ncluded in RECS Appliance UEC (kWh/year) Stand by Wattage Information Source Personal Video Recorders (DVR or TIVO) 237 27.0 Roth et al., 2008 b C ombo Cable Box / DVR 224 24.0 Roth et al., 2008 b Cable Box with no built in DVR 133 15.0 Roth et al., 2008 b Satellite Dish Box 129 14.0 Roth et al., 2008 b Copy Machine 340 10.0 Hendron & Engebrecht, 2010 DSL/Cable Modem 53 6.0 Roth et al., 2008 b VCR 4 7 4.5 Roth et al., 2008 b DVD/VCR Combo 50 4.5 Roth et al., 2008 b Digital TV Converter Box 4 4.0 Based on SIIG Analog to Digital Converter CE CV0111 S Component / Rack Stereo 122 3.0 Roth et al., 2008 b Microwave 131 3.0 Roth et al., 2008 b Wireless Rou ter 53 6.0 Roth et al., 2008 b DVD Player 30 2.3 Roth et al., 2008 b Laptop PC (Plugged In) 72 2.0 Roth et al., 2008 b Desktop PC w/ Speakers 235 2.0 Roth et al., 2008 b Printer 27 2.8 Roth et al., 2008 b Video Gaming System 41 1.0 Roth et al., 2008 a PC M onitor CRT 77 1.0 Roth et al., 2008b PC Monitor LCD 36 1.0 Roth et al., 2008b Home Theater (HTIB) 89 0.6 Roth et al., 2008b Coffee Maker 59 0.4 Roth et al., 2008b Rechargeable Electronics Calculated 1.5 Rot h et al., 2008b Rechargeable Tools Calculated 4.8 Roth et al., 2008b
199 Table 6 3. Power u sage i nformation for a ppliances not i ncluded in RECS Appliance UEC (kWh/year) Market Saturation Stand by Wattage WHS Savings Potential (kWh/year) Information Source Doorbell 18 65% 2 5.7 Sanchez et al., 1998 G arage Door Opener 30 26% 4 4.6 Sanchez et al., 1998 Image Scanner 138 14% 6 3.4 Saturation KEMA, 2010 UEC from Roth et al., 2008a Stand by from Hendron & Engebrecht 2010 Men's Shaver 13 36% 1 2.2 Sanchez et al., 1998 Desk Fan 8 31% 1 1.4 Sanchez et al ., 1998 Power Strip 3 94% 0.3 1.2 Sanchez et al., 1998 Digital Picture frame 88 2% 10 0.9 Smith et al., 1991 Women's Shaver 12 10% 1 0.6 Sanchez et al., 1998 Electric Tooth Brush 12 6% 1 0.4 Roth et al., 2008 b Air Cleaner Electric (not mounted) 55 5% 1 0.2 Sanchez et al., 1998 Saturation from KEMA, 2010. Total 20.5 Table 6 4. Average MEL and WHS s avings p otential from RECS d ata Appliances Total MEL (kWh/year) WHS Savings (kWh/year) WHS Savings (%) Television Peripherals 453 167 36.9% Recharge able Electronics 57 25 43.9% Computer and Office Equipment 284 54 19.0% Small Kitchen Appliances 137 13 9.5% Well Pump 20 0 0% Spa 117 0 0% Other MELs 736 24 3.3% Total 1,795 282 15.7%
200 Table 6 5. Effectiveness of WHS by o ccupant g roup Occupant Gro up N WHS Savings (kWh/yr) Stand Dev Total MEL % of MELs % of Total Utility All Households 12,083 282 141 1,795 7.9 23.6% 1.2 3.7% Households Home During Weekdays 6,881 245 117 1,835 7.0 19.7% 1.1 3.0% Households Away During Weekdays 5,202 331 155 1,743 10.1 27.9% 1.7 4.7% Retired 3,567 254 131 1,743 7.1 22.1% 1.1 3.6% Not Retired 8,516 294 144 1,817 8.3 24.1% 1.3 3.8% Income Less than 40K/year 3,755 208 114 1,444 6.5 22.3% 1.0 3.5% Income 40K/year 80K/year 5,229 285 130 1,784 8.7 23.3% 1.4 3.7% Income 80K/year 120K/year 1,710 356 138 2,125 10.3 23.2% 1.7 3.8% Income More than 120K/year 1,389 381 141 2,382 10.1 21.9% 1.6 3.4%
201 Table 6 6. Validating WHS with t est h ouses ID# RECS Calc d WHS Potential (kW h/yr) Simulated WHS Potential on Test Homes (kWh/yr) Retrofit Package Savings (kWh/yr) Retrofit Package Savings ($/yr)* No. of Disc o n t Retro Cost** Simple Pay back of Retro Package Home During Day MV01 245 155 96 $11.29 4 $160 14.2 MV02 245 185 86 $1 0.20 3 $130 12.8 MV03 245 317 127 $14.99 3 $130 8.7 MV04 245 182 129 $15.25 3 $130 8.5 MV06 245 361 196 $23.18 2 $100 4.3 MV11 245 210 160 $18.82 2 $100 5.3 MV13 245 264 202 $23.84 2 $100 4.2 MV14 245 138 86 $10.12 3 $130 12.8 MV15 245 197 16 2 $19.16 2 $100 5.2 MV18 245 142 52 $6.17 2 $100 16.2 MV20*** 245 18 8 $0.94 1 $70 74.2 MV22 245 193 116 $13.66 2 $100 7.3 MV23 245 237 89 $10.49 2 $100 9.5 MV24 245 163 101 $11.87 3 $130 11.0 Average 245 197 115 $13.57 2.4 $113 13.9 Not Home During Day MV05 331 398 230 $27.19 4 $160 5.9 MV07*** 331 516 440 $51.88 3 $130 2.5 MV08 331 286 178 $20.95 2 $100 4.8 MV09*** 331 159 152 $17.90 2 $100 5.6 MV10*** 331 72 45 $5.30 2 $100 18.9 MV12*** 331 495 207 $24.40 3 $130 5.3 MV16 331 28 5 243 $28.72 2 $100 3.5 MV17 331 205 126 $14.86 2 $100 6.7 MV19 331 309 147 $17.35 3 $130 7.5 MV21*** 331 162 77 $9.09 2 $100 11.0 Average 331 289 184 $21.76 2.5 $115 7.2 Nationwide retail price of $.118 /kWh (EIA, 2012) ** All retrofit pac kages have 2 controllers *** Within 2 standard deviations of expected
202 Figure 6 1. Z Wave enabled disconnectors ( Photo courtesy of Zwave Products, Inc) Figure 6 2. Z Wave enabled control switches ( Photo courtesy of Zwave Prod ucts, Inc)
203 CHAPTER 7 CONCLUSIONS Overview The focus of this dissertation study was on modelin g MELs in the residential section The HERS rating system uses a simple square foot multiplier to model MELs. The hypothesis of this study is that the HERS model can be improved by incorporating occupant characteristics. As described previous ly this study calculated a MEL value for 12,083 homes using the RECS data and regressed a new MEL model. The new MEL model uses size of the home, income of the househ old, number of household members and presence of a home business as variables in the model. In the paragraphs below, the key conclusions that were made from this study are discussed. Key Conclusion #1: Occupant Characteristics are Better Predictors of M ELs The hypothesis of this study is that occupant characteristics are a better way of predicting MELs that current practices (square foot multipliers). The correlations between all the variables ( ) are provided in Appendix C. Ho wever, the correlations between all the occupant characteristics independent variables and MEL dependent variable can be found in table 7 1. Notice that the variable with the highest correlation to MEL s is income and not the square footage of the home. T used to improve current practices. Practical Benefits from Key Conclusion #1: Key conclusion #1 indicates that occupant characteristics are a critical component of understanding M ELs. There are t wo practical applications from this key conclusion First, utility providers can benefit by better understanding the nature of residential MELs. Utility providers are keenly
204 interested in reducing overall load but more specifically peak load where the cost to produce energy is the highest Intuitively, it might seem that homes where the occupant is vacated during the day when utilities are at their highest production would have the lowest MEL. This study found that the correlation betwe en MEL and homes vacated during the day is less than 9%. While MELs can certainly be targeted for overall energy demand reduction the energy saved will not be focused during peak load hours. The second practical application that can be drawn from this ke y conclusion is how residential energy use is perceived. The two most widely recognized MEL models are the HERS index and Building America Program (BA) which b oth use building characteristics to predict MELs. Similarly, when industry and academia offers energy reduction solutions it usually focuse s on technical improvements Th is could be improved technology, building automation or more efficient designs. This study reinforces that a critical f actor in understanding energy is occupant behavior. This st udy e concept This study will help both industry and academia look at energy differently than just the physical components of a building. Looking at energy through the lens of demographics, affluenc e, and behavior could lead to new solutions to improve energy performance. Key Conclusion #2: The New Model This study regressed an equation that significantly improves the current HERS model by incorporating income, household members and presence of a home business. Hot tubs are not recommended to be included in the definition of MEL; however, to be consistent with the HERS index it was included with the model The equation has been
205 Modest Improvement in Accuracy for Modeling Average MEL Th is study found that the average MEL as calculated with the RECS is very similar to whereas the average RECS respondent used 1,795kWh/year ( Table 4 7 ) T he HERS and new model are within 5% of each other. A test performed with this study was to compare the HERS model and the new model with each of the 12,083 RECS households. The calculated average energy consumption for all of the households using both models is relatively close. Using the HERS model the average energy consumption is 1,977kWh, whereas using the new model the energy consumption is 1,792kWh. The difference is only 185kWh or 9.4% from the HERS model. The conclusion drawn from this test is that the new model only modestly imp roves the accuracy of MEL predictions when reviewing a large number of homes averaged together. Significant Improvement in Individual Home Modeling The purpose behind this study is the modeling of individual homes and not averaged MELs over large communi ties. What is significantly more noteworthy in the test described above is the standard deviation. The standard deviation is the dispersion of the individual observations from the predicted model. The standard deviation of the HERS model is 1,323kWh whe
206 this (598kWh) The conclusion that can be drawn from this is that the new model is more likely to predict the true MEL by 54.8%, because there is less dispersion between the model and individual o bservations. MELs are influenced by many factors. Only through the use of large sample sizes can any meaningful predictions be made. However, even with large sample sizes predicting individual homes MEL is difficult without a large margin of error Hot T ubs The published UEC data for hot tubs is 2,040kWh/year and has a market saturation of 3%. Considering that the average MEL of all RECS respondents is 1,795, a UEC of 2,040kWh is an extreme value for one appliance. The market saturation of 3% is also a n extreme value and shows that the energy use does not reflect the vast majority of homes Due to the high UEC and low market saturation it is recommended that this appliance not be defined as a residual MEL and be disaggregated into its own category simi lar to the television Practical Benefits from Key Conclusion #2: The creation of the new model was the heart of this dissertation study. The objective of the study was to create a better way of modeling MEL than current practices. As the HERS index i s the most widely accepted means of modeling MELs it was used as the baseline which this study sought to improve. Perhaps the party that could most benefit from this study is the RESNET certified energy rater. These energy raters go into homes and physic ally measure characteristics to provide a whole house HERS score. The RESNET rater currently uses the .91kWh/sqft HERS model. However, because they are in direct contact with the home owner they have access to all of the coefficients used in the new
207 mode l. The accuracy of their estimation of the MEL value could increase by 54.8% on average with very little additional effort The target of this research was to better model MELs on an individual home bases. However, utility providers, developers and commu nity planners can still benefit from the new model. When new communities are planned they typically attract buyers with similar characteristics. Developers could plan to cater to more affluent householders or low income affordable houses. Similarly, dev elopers could choose to build a retirement community where the average household has only one or two members or build in close proximity to a school where the household size is typically larger. Developers will know the typical buyer they are marketing to and can use those common characteristics in the new model to better predict the MEL for that community. With a better understanding of the MEL at the community level utility providers and community planners can better predict energy use when the project is completed. Key Conclusion # 3 : Published UEC Values Compared to Recorded Consumption A comparison between what the data loggers recorded and what was expected based on the published UEC was performed in this study. The comparison includes 140 data logg ing events that record ed the energy consumption from 2 71 appliances. The total energy consumption recorded on the data loggers was 13,851k Wh. The published UEC for the 2 71 appliances recorded on the data loggers was 13,933kWh. This is a difference of le ss than one percent. While this would seem to indicate that the published UEC mirrors very closely to what was observed in the test houses, it is important to look at the distribution around what was expected or the standard deviation. For the 140 data l ogging events the average difference from what was
208 recorded to the published UEC was 119%. What this means is that there were some large deviations in what was expected at an individual appliance level. There are two appliances that are worth highlighting The first appliance is the dehumidifier. The energy consumption recorded greatly exceeds the published UEC. The second appliance is the microwave. The microwave was recorded individually in 14 households. The average energy use of 62kWh/year is 112% less than the published UEC. These two appliances were not recorded with data loggers in a sufficient number of homes to justify the discrediting the published UEC data. However, because of the large deviation it is something worth making note of for fu ture studies. Practical Benefits from Key Conclusion #3 : This study used published UEC appliance data extensively with the calculation of MEL of the RECS respondents. The validation of the published UEC data with actual consumption was of course self s erving for the study. However, energy modelers can also benefit from these validated UEC values. If a home is being modeled and the appliances are known, these UEC values can be used to calculate energy use. As the published UEC values have been shown t o be consistent with the actual load from 24 test houses, energy modelers can incorporate them into their models with a higher degree of confidence than before this study was conducted. Future Studies This study can be expanded in several ways to help imp rove energy modeling. The following are recommended for a future study: This study was conducted using the 2009 RECS which collected its data from the 16 most populated states. The new model used this information as an aggregate and is generalized for all US homes. For future studies, this same methodology could be applied to each of the 16 states individually. The predictive power of
209 state specific models would likely exceed the nationwide model created with this study. Use the RECS to calculate all of the other traditional energy end uses. A great deal of inferences could be made by having the energy consumption broken down to the appliance level of over 12 thousand households. The RECS provide the total yearly energy consumption of each of the respond ents. A regression explanatory equation using total energy consumption (dependent variable) and characteristics of the household (independent variable) could be created to predicted total energy consumption. The regression equation would likely not be as accurate as if the energy use was calculated using energy modeling software such as eQ UEST BEOpt, or EnergyPlus but would likely be easier and faster for quick estimates. The Building America (BA) program has a similar model as the HERS index for calcula ting MEL. The BA definition of MEL is also similar but with a few distinctions such as including televisions and ceiling fans. This dissertation study could be modified to calculate MELs as defined by BA and improve upon their model. The BA program uses state multipliers for the four most populated states to adjust their MEL model. This study could be used to expand these multipliers for more states as the 2009 RECS collected data from the 16 most populated states. This study has shown that including occ upant characteristics is a better way of modeling MELs than simply a square foot multiplier. However, the RECS data and the calculated MEL could be used to improve the HERS index multiplier. mber of Chapter Summary MELs stands alone as the single energy end use in which energy intensity has steadily increased over time and is not showing any signs o f changing. Occupant behavior is the most critical component to MELs but remains one of the least researched areas of residential energy use. Current models ignore the occupant as a factor who influences energy use. This study looked to improve the curr ent MEL by incorporating characteristics of the occupant as a surrogate for occupant behavior. This
210 Consumption Survey to calculate a MEL and then make inferences about occupant current model by 9.4% and decreased the dispersion from the model (standard deviation) by 54.8%. The model was tested on 24 real world houses in which it predicted th e MEL better 17 out of 24 times. The modeling and reduction of MELs is a key area of research for national energy reduction and goals for zero net energy homes.
211 Table 7 1. Correlation between o ccupant c haracteristics and MEL Occupant Characteristics (I ndependent Variable) Correlation Coefficient (r) Income 0.424 Sqft of Home 0.410 Size of Garage 0.349 Marital Status 0.310 House Type 0.284 Education 0.260 Household Members 0.241 Home Business 0.197 Household Members (squared) 0.166 No. of Chil dren 0.137 Year Made 0.133 Year Made Range 0.132 Urban / Rural 0.086 House Holder AGE(squared) 0.085 At Home During Day 0.082 No. of Children (squared) 0.072 Retired 0.066 Householder Age 0.047 AIA Zone by CDD and HDD 0.043 AIA Zone by CDD an d HDD (squared) 0.036 BA Climate Region 0.022
212 APPENDIX A S URVEY
225 APPENDIX B STATISTICAL ANALYSIS AND TERMINOLOGY This dissertation made statistical inferences of all US households based on the 2009 RECS data. It is important to describe the statistical methods and terminology US households. The large sampl e size provides the advantage of mirroring the whole sample sizing, it can be said with 99% confidence that the RECS sample mean is within 1.5% of the true population me an (Agresti and Finlay, 2009). The calculated MEL value shaped. Figure 2 1 0 shows the MEL intensity in kWh along the x axis and the frequency of the intensity along the y axis. The RECS sample is slightly skewed to the right. This was anticipated as the lowest MEL value is bounded by zero kWh whereas the upper end has no limit. Standard Deviation An important statistical measure that was used in this dissertation is the standard from the mean (average MEL in kWh from the sample). This is important because as measure the success or failure of the model. The empirical rule states that if the sample is normally distributed then approximately 68% of all observations are within one standard deviation of the mean, 95% fall within two standard deviations and nearly all observations are within three standard deviations (Agresti and Finlay, 2009).
226 ANOVA This dissertation created a new MEL model by regressing several independent variables that explain the dependent variable. The dependent variable is the MEL and the independent variables are characteristics of the occupant. Before the regression could be completed the independent variables needed to be checked for statistical significance. Testing for significance is verifying that the pattern in the ob servations is reliable and not cause by random chance. A large sample size is the principle way of increasing statistical significance. The means of testing significance ( Table A 1 below ) The ANOVA table shows the variance from what was predicted in the model from what s Squares The sum of squares is the squared difference between what was observed and wh the number of characteristics describing the dependent variable. In the example ANOVA table provided in Table A 1 is a simple linear regression and only has one independent variable explaining the dependent variable. As such the df is one. However, if this were a multilinear regression model with three independent variables explaining the dependent variable then the df would be three. The total df is the sample size (n) mi nus 1. The residual df is the total df minus the regression df. The mean square is calculated by dividing the sum of squares by the df. In the example provided, the residual sum of square (2,645,229,575) divided by the degrees of freedom (12,081)
227 equals Statistic used to determine the probability of the observed values being seen if the null hypoth esis is true. The null hypothesis is that the independent variables do not describe the dependent variable. The F statistic is converted to a more easily An ANOVA table is used to analyze your da significance. The p value is the probability that your data would be seen if the independent variables do not predict the dependent variable. The widely accepted rule is that a p value less that .05 can be used to claim that an independent variable is statistically significant (Agresti and Finlay, 2009). Table A 2 graphically explains the calculation of the ANOVA table as described above. Besides significance the quality of correl ation is also important in reviewing measure of how correlated two variables are. In other words it indicates the strength of relationship between two variables. 1 < r < 1. The sign The correlation of coefficient shows the correlation between two variables and used for simple linear regression formulas. For multilinear regression formulas where
228 multiple independent variables will be used to describe 2 2 the correlation of coefficient (r). R 2 explains the predictor variables. The new MEL model that was created in this dissertatio n is a multilinear regression and the coefficient of determination was used to analyze the results. There is no hard line or even general rule as to what an acceptable value that the R 2 must be to be considered acceptable. However, it is worth highlighti ng the R 2 values used in the Parker study which is the basis of the HERS model ranged from .0369 to .1696. The new MEL model (Model E) created in this dissertation had a R 2 value of .285. Stepwise Regression The type of regression analysis chosen to creat e the new MEL model was a variable (MEL)] on each explanatory variable separately and keeping the regression with the highest R 2 (Kennedy, 2001). Stated another way the stepwise regression process systematically introduces and then removes each of the independent variables 2 ). The higher the R 2 is the hi gher the explanatory power of the regressed equation. Margin of Error and Confidence Interval Another important statistical measure for a regression equation is its standard error. The standard error is how much the sample mean is expected to deviate fr om the true population mean. It is an estimate of the standard deviation of a regression equation. It is important for this dissertation because it is used to calculate the margin of error and the confidence interval. The margin of error is a measure of the amount of
229 random sampling error given a set confidence level. The larger the confidence level the higher the margin of error. 95% is a very common confidence level and will be used in this study. When poll results are published they generally indic ate their findings with a +/ percentage of uncertainty. The +/ percentage of uncertainty is the margin of error. The range between the findings plus the margin of error and minus the mar g in of error is called the confidence interval. The confidence in terval is a range of values that with a set confidence level contains the true population value. For this dissertation 24 test homes were monitored to record their actual MEL. The new MEL model will be used to predict the MEL and then compared with the actual value. One of the ways that it was determined that the new model was successful was that all of the test houses were within the margin of error of the model. Table A 1. Sample ANOVA table ANOVA Model Sum of Squares df Mean Square F Statistic P Value Regression (model) 580,878,111 1 580,878,112 2,653 .000 Residual (error) 2,645,229,575 12081 218,958 Total 3,226,107,686 12082
230 Table A 2 Calculations of ANOVA table ANOVA Model Sum of Squares df Mean Square F Statistic P Value Regres si on (model) k MSR = SSR/k F obs =MSR/MSE Area to right of F distribut ion Residual (error) n k 1 MSE = SSE/(n k 1) Total n 1
231 APPENDIX C PEARSON CORR ELATION TABLE
232 LIST OF REFERENCES Agresti, A., & Finlay, B. (2009). Describing Variability of the Data. Pearson Hall, Inc 4 th EDS, (pp.49 51). Statistical Methods for the Social Sciences Upper Saddle River, New Jersey. Argyrous, G. (2011). Stati stics for Research: With a Guide to SPSS SAGE Publications, Ltd. London, England. ASHRAE 90.1 1989. (1989). ASHRAE/IESNA Standard for Energy Efficient Design of New Buildings Except Low Rise Residential Atlanta, GA. ASHRAE Standards Committee. ( 2002). ASHRAE Guideline 14, Measurement of Energy and Demand Savings Atlanta GA. Barley, C.D., Haley,C., & Anderson, R. (2008). Building America System Research Plan for Reduction of Miscellaneous Electrical Loads in Zero Energy Homes. (NREL Technica l Report NREL/TP 550 43718). Bonino, D., Corno, C., & De Russis, L. (2012). Home Energy Consumption Feedback: A user Survey. Energy and Buildings 47(2012). Pp. 383 393. Brisken, W.R. & Reque, S.G. (1956). Heat Load Calculations by Thermal Response ASHRAE Transactions, Vol. 62, Pt. 1. Brown, R., Rittleman, W., Parker, D., & Homan, G. (2007). Appliances, Lighting, Electronics, and Miscellaneous Equipment Electricity Use in New Homes Lawrence Berkeley National Laboratory. (LBNL 62440). Buildin g America. (2007). BA2007 Final Report: Potential Savings in Miscellaneous Electric Loads from use of Next Generation Appliances Retrieved from http://www.bira.ws/projects/files/9C3b_Potential_Savings_Appliances.pdf ( July 13, 2011 ) Darby, S. (2006) The effectiveness of feedback on energy consumption: a review for DEFRA of the literature on metering, billing and direct displays. Environmental Change Institute, University of Oxford. Retrieved from http://www.eci.ox.ac.uk/research/energy/electric metering.php ( October 25, 2012 ) Dawson Haggerty, S., Lanzisera, S., Taneja, J., Brown, R., & Culler, D. (2012). @scale: Insights from a Large, Long Lived Appliance Energy WSN. Retrieved from http://www.cs.berkeley.edu/~taneja/publications/ipsn12atsca le.pdf ( April 8, 2012 ) Ecos Consulting. (2004). Power Supply Efficiency: What Have We Learned? Durango, CO, California Energy Commission PIER Program.
233 Engebrecht, C. & Hendron, B. (2010). Overview of Building America House Simulation Protocols. Re trieved from http://apps1.eere.energy.gov/buildings/publications/pdfs/building_america/rpm2010_per f_eval_protocols.pdf ( March 7, 2012 ) Fairey, P., Colon, C., Martin, E., & Chandra, S. (2006). Comparing Apples, Oranges and Grapefruit: An Analysis of Cu rrent Building Energy Analysis Standards for Building America, Home Energy Ratings and the 2006 International Energy Conservation Code. Florida Solar Energy Center. (Report # FSEC CR 1650 06). Fanara, A., Clark, R., Duff, R., & Polad, M. (2006). How S mall Devices are Having a Big Impact on U.S. Utility Bills Retrieved from http://www.energystar.gov/ia/partners/prod_development/downloads/EEDAL 145.pdf?87d3 18e8 ( April 8, 2012 ) Floyd, D. & Webber, C. (1998). Leaking Electricity: Individual Field Measurement of Consumer Electronics. Florida Solar Energy Center. (FSEC PF 344 98). Retrieved from http://www.fsec.ucf.edu/en/publications/html/FSEC PF 344 98/index.htm#Table ( August 29, 2012 ) Fung, A., Aulenback, A., Ferguson, A., & Ismet Ugursal, V. (2003). Standby power requirements of household appliances in Canada. Energy and Buildings 35 (2003). pp. 217 228 Groves, S. (2009). The Desire to Acquire: Forecasting the Evolution of Household Energy Services Simon Fraser University, Burna by Canada. (Report 485). Haberl, J. & Cho, S. (2004). Literature Review of Uncertainty of Analysis Methods. Report to the Texas Commission on Environmental Quality. (ESL TR 04/11 01). Retrieved from http://www esl.tamu.edu/docs/terp/2004/esl tr 04 11 01.pdf ( August 21, 2012 ) Hendron, R., Anderson, R., Judkoff, R., Christensen, C., Eastment, M., Norton, P., Reeves, P., & Hancock, E. (2004). Building America Performance Analysis Procedures. National Renewable Energy Laboratory. (NREL/TP 550 35567) Hendron, R. (2006 ). Building America Performance Analysis Procedures for Existing Homes. National Renewable Energy Laboratory. (NREL/TP 550 38238). Hendron, Robert & M. Eastment. (2006). Development of an Energy Savings Calculation Methodology fo r Residential Miscellaneous Electric Loads. Retrieved from http://www.ntis.gov/products/stoncd/docs/DE2006891591.pdf ( April 11, 2012 ) Hendron, R. & Engebrecht, C. (2009). Building America Research Benchmark Definition. National Renewable Energy Labor atory. (NREL/TP 550 47246).
234 Hendron, R. & Engebrecht, C. (2010). Building America House Simulation Protocols. National Renewable Energy laboratory. Retrieved from http://www.nrel.gov/docs/fy11osti/49246.pdf ( April 11, 2012 ) Hill, P.R. (1957). A Me thod of Computing the Transient Temperature of Thick Walls from Arbitrary Variation of Adiabatic Wall. Temperature and Heat Transfer Coefficient (National Advisory Committee for Aeronautics, Technical Note, 4105). Holden, H.L. (1963). The Calculation of Fluctuating Heat Flow in Building. Australian Computer Conference, Melbourne (DBR Report 236). CSIRO. Huang, J. & Gu, L. (2002). Prototypical Residential Buildings to Represent the US Housing Stock Berkeley, CA: Lawrence Berkeley National Laboratory. I nternational Building Performance Simulation Association. (2012). History of Building Energy Modeling. Retrieved from http://www.bembook.ibpsa.us/index.php?title=History_of_Building_Energy_Modeling Accessed 21 August 2012 ( April 11, 2012 ) Internationa l Energy Agency. (2001). Things That Go Blip in the Night: Standby Power and How to Limit it. Organization for Economic Co operation and Development. Paris France. James, P., Sonne, J., Vieira, R., Parker, D., & Anello, M. (1996). Are Energy Savin gs Due to Ceiling Fans Just Hot Air? Retrieved from http://eec.ucdavis.edu/ACEEE/1994 96/1996/VOL08/089.PDF ( April 10, 2012 ) Katzev, R., Cooper, L., & Fisher, P. (1981) The Effect of Feedback and Social Reinforcement on Residential Electricity Consump tion. Journal of Environmental Systems 10(1981) pp. 215 227. KEMA, Inc. (2010). 2009 California Residential Appliance Saturation Study. California Energy Commission. (Publication number: CEC 200 2010 004 ES). Retrieved from http://www.energy.ca.gov/appliances/rass/ ( August 28, 2012 ) Kennedy, P. (2001). The Classical Linear Regression Model. The MIT Press 4 th EDS, (pp. 52). A Guide to Econometrics. Cambridge, Massachusetts. Kusuda, T (1969). Thermal Response Factors for Multi Layer Structures of Various Heat Conduction Systems. ASHRAE Transactions, Vol. 75, Pt. 1, (pp. 246 271). Kusuda, T. (1970). Algorithms for Psychometric Calculations (Building Science Series 21). National Burea u of Standards, Washington, DC. Kusuda, T., (1971). Use of Computers for Environmental Engineering Related to Buildings (Science Series 39). National Bureau of Standards, Washington, DC.
235 Kusuda, T. (1974). Computer Program for Heating and Cooling Loads in Buildings NBSLD. National Bureau of Standards, Washington, DC. Lawrence Berkeley National Laboratory. (2012). Home Energy Saver: Engineering Documentation Miscellaneous Equipment Energy Consumption Retrieved from https://sites.google.com/a/lbl. gov/hes public/Home Energy Saver/calculation methodology/calculation of energy consumption/miscellaneous equipment energy consumption ( July 12, 2012 ) Lee, Y. S., Yi, Y. K., & Malkawi, A. (2011). Simulating Human Behavior and its Impact on Energy Uses. Paper presented at the Proceedings of Building Simulation 2011: 12 th Conference of International Building Performance Simulation Association. Sydney, 14 16 November. Malin, N. (2008). BIM Companies Acquiring Energy Modeling Capabilities. Green Source Retrieved from http://greensource.construction.com/news/080403BIMModeling.asp ( August 22, 2012 ) Manzoor, S., Ullah, H., Hussain, M. & Ahmad, A. (2011). Effect of Teamwork on Employee Performance International Journal of Learning & Development Vo l 1(1). Retrieved from http://www.google.com/url?sa=t&rct=j&q=&esrc=s&frm=1&source=web&cd=4&cad=rja& ved=0CEsQFjAD&url=http%3A%2F%2Fwww.macrothink.org%2Fjournal%2Findex.php %2Fijld%2Farticle%2Fdownload%2F1110%2F861&ei=UcDAUJbKC4a69gSqhoC4CA& usg=AFQjCNElhiMc yvpVe6idKpW U8alkXMWsQ&sig2=OAG6jmfU2Ainq8B07skdBQ ( December 12, 2012 ) Building Energy Use. Energy and Buildings 42(2010),pp. 173 177. McMahon, J., & Nordman, B. (2004). De veloping and Testing Low Power Mode Measurement Methods Lawrence Berkeley National Laboratory. (LBNL#P500 04 057). Meier, A. (2001). A Worldwide Review of Standby Power Use in Homes Lawrence Berkeley National Laboratory. Retrieved from http://esch olarship.org/uc/item/03m799xz ( September 18, 2012 ) Meier, A & Huber, W. (1997). Results from the investigations on leaking electricity in the USA. Lawrence Berkeley National Laboratory (Report #40909). Meier, A., Nordman, B., Busch, J., Payne, C., Br own, R., Homan, G.K., Sanchez, M. & Webber, C. (2007). Low power mode Energy Consumption in California Homes. California Energy Commission, PIER Buildings End Use Energy Efficiency Program. (CEC 500 2008 035).
236 Meier, A., Rainer, L., & Greenberg, S. (1992 ). Miscellaneous Electrical Energy Use in Homes. Energy 17(5), pp. 509 518. Retrieved from http://energy.lbl.gov/ea/akmeier/pdf/meier rainer misc e use.pdf ( August 6, 2012 ) Meyers, R., Williams, E. & Matthews, H. (2010). Scoping the Potential of Mon itoring and Control Technologies to Reduce Energy Use in Homes. Energy and Buildings 42(2010) pp. 563 569. Midwest Plan Service. (1979). Private Water Systems Handbook (Report MWPS 14). Iowa State University. Ames, Iowa. Mills, E. (2008). The Ho me Energy Saver: Documentation of Calculation Methodology, Input Data, and Infrastructure. Lawrence Berkeley National Laboratory. (LBNL 51938).Mohanty, B. (2001). Standby Power Losses in Household Electrical Appliances and Office Equipment. Presented at the Regional Symposium on Energy Efficiency Standards and Labeling. May 29, 2001. Moorefield, L., Frazer, B., & Bendt, P. (2011). Office Plug Load Field Monitoring Report. California Energy Commission, PIER Energy Related Environmental research Pro gram. (CEC 500 2011 010). Muncey, R.W. (1963). Thermal Response of a Building to Sudden Changes of Temperature or Heat Flow. Australian Journal of Applied Science Vol. 14, Pt. 2, pp. 123 128. Nessi, A., & Nisolle., L. (1925). Regimes Variables de Fon ctionnement dans les Installations de Chauffage Central. Paris, France: Dunod. Nordman, B. & Sanchez, M. (2006). Electronics Come of Age: A Taxonomy for Miscellaneous and Low Power Products Lawrence Berkeley National Laboratory. (LBNL 63559). O'B rien, R.M. (2007). A Caution Regarding Rules of Thumb for Variance Inflation Factors. Quality and Quantity 41(5) pp. 673 690 Retrieved from http://web.unbc.ca/~michael/courses/stats/lectures/VIF%20article.pdf ( December 12, 2012 ) Parker, D., Fairey, P ., & Hendron, R. (2011). Updated Miscellaneous Electricity Loads and Appliances Energy Usage Profiles for Use in Home Energy Ratings, the Building America Benchmark Procedures and Related Calculations. Florida Solar Energy Center. (FSEC CR 1837 10). Pi pes, L.A. (1957). Matrix Analysis of Heat Transfer Problems. Franklin Institute Journal, 263(3), pp. 195 205
237 Porter, S.F., Moorefield, L., & May Ostendorp, P. (2006). Final Field Research Report California Energy Commission, PIER Program. (CEC 500 0 4 030). Pratt, R., C. Richman, C.E., Ritland, K., Sandusky, W., & Taylor, M. (1989). Description of Electric Energy Use In Single Family Residence in The Pacific Northwest. End Use Load and Consumer Assessment Program (ELCAP). DOE/BP 13795 21. Richla nd, WA: Pacific Northwest National Laboratory. Red Jacket Inc. (2012). GRIZZLY Retrieved from http://completewatersystems.com/wp content/uploads/2011/04/BRJGRIZZ.pdf ( August 27, 2012 ) RESNET. (2 006). 2006 Mortgage Industry National Home Energy Rating System Standards Oceanside CA: RESNET. RESNET. (2012). The History of RESNET. Retrieved from http://www.resnet.us/about/our history ( June 27, 2012 ) Ross, J.P. & Meier, A. (2000). Whole Hous e Measurements of Standby Power Consumption. In The Second International Conference on Energy Efficiency in Household Appliances. Naples, Italy; September 27 29, 2000. (LBNL 45967). Roth, K., & Brodrick, J. (2008). Home Energy Displays. ASHRAE Journal 50(7) pp. 136 137 Roth, K.W., Ponoum, R., et al. (2006). U.S. Residential Information Technology Energy Consumption in 2005 and 2010. Cambridge, MA. Roth, K.,& McKenney, K. (2007). Energy Consumption by Consumer Electronics in U.S. Residences. TIAX LLC. (Reference No. D5525). Roth, K., McKenney, K., & Brodrick, J. (2008a). Small Devices, Big Loads. ASHRAE Journal 2008(June) Roth, K., McKenney, K., Ponoum, R., & Paetsch, C. (2008b). Residential Miscellaneous Electric Loads: Energy Co nsumption Characterization and Savings Potential in 2006 and Scenario based Projections for 2020. TIAX LLC. (Reference No. D0370). Samarakoon, E., & Soebarto, V. (2011). Testing the Sensitivity of User Patterns in Building Energy Performance Simulatio n. Paper presented at the Proceedings of Building Simulation 2011: 12 th Conference of International Building Performance Simulation Association Sydney, 14 16 November.
238 Sanchez, M.C., Koomey, J.G., Moezzi, M.M., Meier, A.K., & Huber, W. (1998). Miscellan eous Electricity Use in the U.S. Residential Sector. Lawrence Berkeley National Laboratory. (LBNL 40295). Smith, B.A., Uhlaner, R.T., Cason, T.N., & Courteau, S. (1991). Residential Energy Usage Comparison: Findings. Quantum Consulting. (EPRI CU 7392). E lectric Power Research Institute, Palo Alto, CA. Southern California Edison. (2008). Office of the Future Phase II Report: The 25% Solution (ET 08.01 Final Report). Srinivasan, R.S., Lakshmanan, J., Srivastav, D. & Santosa, E. (2011a). Benchmarkin g Plug Load Densities for K 12 Schools. Paper presented at the Proceedings of Building Simulation 2011: 12 th Conference of International Building Performance Simulation Association, Sydney, 14 16 November Srinivasan, R.S., Lakshmanan, J., Santosa, E., & Srivastav, D. (2011b). Plug Load Densities for Energy Analysis: K 12 Schools. Energy and Buildings 43(11), pp. 3289 3294. DOI: 10.1016/j.enbuild.2011.08.030. Srinivasan, R.S., Lakshmanan, J., & Srivastav, D. (2011c). Calibrated Simulation of a n Existing Convention Center: The Role of Event Calendar and Energy Modeling Software. 12 th Conference of International Building Performance Simulation Association, Sydney, 14 16 November Retrieved from http://www.ibpsa.org/proceedings/BS2011/P_1216.pd f ( November 10, 2012 ) Steemers, K., & Yun, G.Y. (2009). Household energy consumption: a study of the role of occupants. Building Research & Information 37(5 6),pp. 625 637 Stein, L.F. (2004). California Information Display Pilot: Technology Assess ment Final report by Primen to Southern California Edison. Stern, P. (1999). Information, Incentives, and Pro Environmental Consumer Behavior Journal of Consumer Policy 22(1999) pp. 461 478. Stewart, J.P. (1948). Solar Heat Gain through Walls and R oofs for Cooling Load Calculations. ASHVE Transactions, Vol. 54, pp. 361 388. Tupper, K., Franconi, E., Chan, C., Fluhrer, C., Jenkins, M., & Hodgin, S. (2012). Building Energy Modeling Innovation Summit Retrieved from http://rmi.org/content/Files/S ummit_PreRead_FINAL.pdf ( August 21, 2012 ) US Census Bureau. (2009). Commuting in the United States: 2009. Retrieved from http://www.census.gov/prod/2011pubs/acs 15.pdf ( September 11, 2012 )
239 US Department of Energy. (2011a). About Building America. Retrieved from http://www1.eere.energy.gov/buildings/building_america/m/about.html ( March 7, 2012 ) US Department of Energy. (2011b). Program Goals. Retrieved from http://www1.eere.energy.gov/buildings/building_america/m/program_goals.html ( March 7, 2012 ) US Department of Energy. (2011c). Analysis Spreadsheets Retrieved from http://www1.eere.energy.gov/buildings/building_america/m/analysis_spreadsheets.html ( March 8, 2012 ) US Department of Energy. (2012). Building Energy Data Book: Chapter 2 Residential Sector Retrieved from http://buildingsdatabook.eren.doe.gov/ChapterIntro2.aspx ( April 12, 2012 ) US Department of Labor. (1996). Average paid holidays, and days on vacation and sick leave for full time employees. Retrieved from http:/ /www.bls.gov/news.release/ebs.t05.htm ( September 11, 2012 ) US Department of Labor. (2012). Charts from the American Time Use Survey. Retrieved from http://www.bls.gov/tus/charts/chart16.pdf ( September 11, 2012 ) US Energy Information Agency. (1995). Household Energy Consumption and Expenditures DOE/EIA 0321(93). Retrieved from http://www.eia.gov/ftproot/consumption/032193.pdf ( August 27, 2012 ) US Energy Information Agency. (2001). Residential Energy Consumption Survey (RECS). Retrieved from http://www.eia.gov/consumption/residential/data/2001/ ( August 29, 2012 ) US Energy Information Agency. (2011a). Residential Energy Consumption Survey (RECS ). Retrieved from http://www.eia.gov/consumption/residential/index.cfm ( November 28, 2011 ) US E nergy Information Agency. (2011b). EIA household energy use data now includes detail on 16 States. Retrieved from http://www.eia.gov/consumption/residential/methodology/states.cfm ( November 28, 2011 ) US Energy Information Agency. (2011c). How does E IA estimate energy consumption and end uses in U.S. homes? Retrieved from http://www.eia.gov/consumption/residential/methodology/2009/brief.cfm ( November 28 th 2011 )
240 US Energy Information Agency. (2011d). Share of energy used by appliances and consumer electronics increases in U.S. homes. Retrieved from http://www.eia.gov/consumption/residential/reports/electronics.cfm ( July 13, 2011 ) US Energy Information Agency. (2012). Table 5.3. Average Retail Pric e of Electricity to Ultimate Customers: Total by End Use Sector, 2002 June 2012 Retrieved from http://www.eia.gov/electricity/monthly/epm_table_grapher.cfm?t=epmt_5_03 ( September 18, 2012 ) US Environmental Protection Agency. (2013). Carbon Dioxide Emissions. R etrieved from http://www.epa.gov/climatechange/ghgemissions/gases/co2.html (February 8 th 2012) US General Services Administration. (2012). Energy Performance and Operations Retrieved from http://www.gsa.gov/portal/content/102283 ( Augus t 22, 2012 ) Wang, C., Yan, D., & Jiang, Y. (2011). A Novel Approach for Building Occupancy Simulation. Building Simulation, 4, pp. 129 167 DOI: 10.1007/s12273 011 0044 5. Water Systems Council. (2003). Determining the Depth of a Well Retrieved fr om http://www.watersystemscouncil.org/VAiWebDocs/WSCDocs/9522259IN_DEPTH.PDF ( August 27, 2012 ) Wenzel, T., Koomey, J., Rosenquist, R., Sanchez, M., & Hanford, J. (1997). Energy Data Sourcebook for the U.S. Residential Sector. Lawrence Berkley Nationa l Laboratory (Report #40297). Retrieved from http://enduse.lbl.gov/Info/LBNL 40297.PDF ( November 18, 2011 ) Wiggins, M., McKenney, K., & Brodrick, J. (2009). Residential Energy Monitoring. ASHRAE Journal June(2009). Wilson, A., & Morrill, J. (1990 ). Consumer Guide to Home Energy Savings Berkeley, CA: American Council for an Energy Efficient Economy. Wilson, A. (2008). Energy Dashboards: Using Real Time Feedback to Influence Behavior Environmental Building News, 17(12) Retrieved from htt p://www.buildinggreen.com/auth/article.cfm/2008/11/24/Energy Dashboards Using Real Time Feedback to Influence Behavior/ ( March 14, 2012 ) Wood, G. & Newborough, M. (2007). Energy Use Information Transfer for Intelligent Homes: Enabling Energy Conserva tion with Central and Local Displays. Energy and Buildings 39(2007) pp. 495 503. Xylem, Inc. (2012). Water Products: Technical Data and Pump Fundamentals for Goulds Water Technology, Bell & Gossett, Red Jacket Water Products and Centripro
241 (Technical Manual TTECHWP). Retrieved from http://completewatersystems.com/wp content/uploads/2011/06/TTECHWP.pdf ( August 27, 2012 ) Yamaguchi, Y., Fugimoto, T., & Shimoda, Y. (2011). Occupant Behavior Model for Households to Estimate High Temporal Resolution Re sidential Electricity Demand Profile Paper presented at the Proceedings of Building Simulation 2011: 12 th Conference of International Building Performance Simulation Association. Sydney, 14 16 November.Zogg, R. & Alberino, D. (1998). Electricity Consum ption by Small End Uses in Residential Buildings Arthur D. Little Inc. (Reference No. 34732 00).
242 BIOGRAPHICAL SKETCH Joe Burgett was born in 1980 in Saint Petersburg Florida. He is the younger of 2 sons to Dan and Sally Burgett. After graduating hi gh school Joe attended the University of Florida where he received his Bachelor of Science from the M.E. Rinker, Sr. School of Building Construction. After graduating, Joe spent eight years in the construction industry working largely for The Weitz Compan y. While in the industry, Joe worked primarily on state and local government project but has experience with hotels, hospitals, biomedical research, schools, multifamily condos, single family residences and retirement campuses. The majority of his constr uction experience was in operations serving as a superintendent and project manager however he also spent several years in preconstruction. While in preconstruction he estimated over 800 million dollars of work. In 2010, Joe returned to the University of and PhD degrees in construction management. He is currently enrolled in the M.E. Rinker, Sr. School of Building Construction and is expected to graduate with his Ph.D. in the spring of 2013. Joe is married to his beautiful wife Jill and has three daughters with her. His youngest daughter, Brynn, was born in November 2011, Emma, was born in March of 2010 and his older daughter, Kate, was born in September of 2008. Joe and his family currently reside in Gainesville, Florida.