University of Southern California Dissertations and Theses

Energy use intensity estimation method based on building façade features by using regression models

(USC Thesis Other)

Energy use intensity estimation method based on building façade features by using regression models

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1

ENERGY USE INTENSITY ESTIMATION METHOD
BASED ON BUILDING FAÇADE FEATURES BY USING REGRESSION MODELS

By

CHAO YANG

A Thesis report presented to the
FACULTY OF SCHOOL OF ARCHITECTURE
UNIVERSITY OF SOUTHERN CALIFORNIA
In partial fulfillment of the
Requirements of the degree
MASTER OF BUILDING SCIENCE

MAY 2015

Copyright 2015 Chao
Yang
2

Dedication
This thesis is dedicated to my family. With the unconditional support from my family, I could
have the chance to pursue my dream and do the academic research that I’m interested in. It
means everything for me to let my family know the completion of this thesis work.
Thank you, Mom! Thank you, Dad!

3

Acknowledgement
Firstly, I would like to express the most sincere appreciation to my thesis committee chair,
Professor Joon-Ho Choi. He was the person who discussed with me for multiple times and
guided me to find out the topic of my thesis. Within the whole process of thesis proposal,
methodology development, data collection, data processing, results representation, and revision
for submission, he gave me the greatest help and guidance when I faced any problem. He always
listened to my idea first and talked about his advice. It is impossible that I could finish this thesis
without those meaningful discussion and his review.
Secondly, I would also like to thank my committee members, Professor Douglas Noble and
Professor Marc Schiler. They always responded to me efficiently about all my questions, even
they had other class and research work to do. Their suggestions come from their rich experience
in this academic field, which are greatly helpful for me to avoid some unnecessary faults and
keep working on the meaningful and practical direction.
In addition, I would like to thank all faculty and classmates from MBS family. I had a great time
for the two years study in this program and learned a lot from each member. This experience of
doing researches and enjoy life is the most precious thing for me.
I also would like to thank Mr. Chris Beers. He gave me many useful suggestions within these
two years to improve my school work and adapt to the new life in a new place.
Last but not least, I appreciate all the encouragement and support from my family. They are the
most important motivation for me to finish my thesis and this MBS program.

4

Contents
Dedication ....................................................................................................................................... 2
Acknowledgement .......................................................................................................................... 3
List of Figures ................................................................................................................................. 9
List of Tables ................................................................................................................................ 14
Abstract ......................................................................................................................................... 16
Hypothesis..................................................................................................................................... 17
Chapter 1 Introduction to the Building Energy Estimation Method ............................................. 18
1.1 Problem ............................................................................................................................... 18
1.2 Explanation of Terms .......................................................................................................... 21
1.2.1 Energy Use Intensity .................................................................................................... 21
1.2.2 Benchmark ................................................................................................................... 22
1.2.3 Architecture 2030 and CBECS .................................................................................... 23
1.2.4 National Median EUI ................................................................................................... 24
1.2.5 Energy Simulation ....................................................................................................... 24
1.2.6 Building Façade ........................................................................................................... 25
1.2.7 Multiple Regression Analysis, Standardized Coefficient and Significance Level ...... 25
1.2.8 Degree Days ................................................................................................................. 26
1.2.9 Partial Least Squares and Cross-validation.................................................................. 27
1.3 Objective ............................................................................................................................. 27
Chapter 2 Background of EUI Benchmarking and Calculation Methods..................................... 29
2.1 Building Energy Reduction Plan and Target ...................................................................... 30
5

2.1.1 International Energy Conservation Code (IECC) ........................................................ 30
2.1.2 ASHRAE 90.1.............................................................................................................. 30
2.1.2 The 2030 Challenge ..................................................................................................... 31
2.2 EUI Baseline ....................................................................................................................... 33
2.2.1 Target Finder ................................................................................................................ 33
2.2.2 CBECS ......................................................................................................................... 34
2.2.3 Meeting the 2030 Challenge through Building Codes ................................................. 35
2.2.4 Simulation Tool ........................................................................................................... 36
2.2.5 Real Recorded Energy Data ......................................................................................... 37
2.3 Building Energy Benchmarking and Disclosure ................................................................. 37
2.4 Façade Features ................................................................................................................... 39
2.4.1 Importance of Façade Features on Building Energy .................................................... 39
2.4.2 Orientation ................................................................................................................... 39
2.4.3 Operable Window ........................................................................................................ 40
2.4.4 Shading ........................................................................................................................ 40
2.4.5 Window-to-wall Ratio ................................................................................................. 41
2.4.6 Volume-to-façade Area Ratio ...................................................................................... 42
2.5 Regression Model for Energy Performance Calculation .................................................... 42
2.5.1 Regression Analysis for Building Energy Performance .............................................. 42
2.5.2 Multiple Linear Regression and Stepwise Regression ................................................ 43
2.5.3 Other Methods for Estimating EUI .............................................................................. 44
Chapter 3 Methodology and Plan of Approach ............................................................................ 46
6

3.1 Introduction of Methodology .............................................................................................. 46
3.2 Energy Use Data Collection................................................................................................ 48
3.2.1 Existing Data ................................................................................................................ 48
3.2.2 Benchmarking Data ..................................................................................................... 48
3.2.3 Monitoring Data ........................................................................................................... 49
3.3 Façade Parameter Determination and Variables Collection ............................................... 49
3.3.1 Façade Feature Determination and Definition ............................................................. 49
3.3.2 Data Reading by Using 3D Model Tool ...................................................................... 51
3.4 Regression Method and Tool .............................................................................................. 52
3.4.1 Regression Tool Introduction ...................................................................................... 52
3.4.2 Multiple Linear Regression, Stepwise Regression, Partial Least Square (PLS).......... 53
3.4.3 Indicators Explanation ................................................................................................. 53
3.4.4 Cross-validation ........................................................................................................... 54
Chapter 4 Regression Models and Results.................................................................................... 55
4.1 Nationwide Annual EUI Model .......................................................................................... 55
4.1.1 Data Sources ................................................................................................................ 55
4.1.2 MLR, Stepwise and PLS .............................................................................................. 61
4.1.3 Validation ..................................................................................................................... 65
4.1.4 Results Comparison ..................................................................................................... 66
4.2 City-based Annual EUI Model – New York City ............................................................... 68
4.2.1 Basic Data Analysis ..................................................................................................... 68
4.2.2 MLR, Stepwise and PLS .............................................................................................. 72
7

4.2.3 Validation ..................................................................................................................... 77
4.2.4 Results Comparison ..................................................................................................... 78
4.3 City-based Annual EUI Model – Los Angeles ................................................................... 80
4.3.1 Basic Data Analysis ..................................................................................................... 80
4.3.2 Regression Models ....................................................................................................... 84
4.3.3 Validation ..................................................................................................................... 85
4.3.4 Results Comparison ..................................................................................................... 86
4.4 Monthly EUI Model for Los Angeles ................................................................................. 88
4.4.1 Basic Data Analysis ..................................................................................................... 88
4.4.2 MLR, Stepwise and PLS .............................................................................................. 90
4.4.3 Validation ..................................................................................................................... 94
4.4.4 Results Comparison ..................................................................................................... 95
4.5 Monthly Cooling and Heating EUI Models for Los Angeles ............................................. 97
4.5.1 Basic Information......................................................................................................... 97
4.5.2 Monthly Heating and Cooling EUI regression models ................................................ 99
4.5.3 Results and Validation ............................................................................................... 100
Chapter 5 Comparison and Model Verification .......................................................................... 103
5.1 Comparison with Current Baseline ................................................................................... 103
5.2 Comparison of Different City’s Annual EUI Model ........................................................ 105
5.3 Model Verification ............................................................................................................ 107
5.3.1 Basic Information and Energy Bills ........................................................................... 108
5.3.2 Simulation .................................................................................................................. 109
8

5.3.3 Comparison and Results ............................................................................................ 115
Chapter 6 Conclusions of Study ................................................................................................. 118
6.1 Current Baseline Problem ................................................................................................. 118
6.2 5 Regression Models at Different Scales and Best Model Selection ................................ 119
6.3 Principle Façade Features and Model Verification ........................................................... 120
6.4 Contribution to Benchmarking and ARCH2030............................................................... 121
6.5 Scope Limit of Regression Models ................................................................................... 121
Chapter 7 Suggestions for Future Work ..................................................................................... 124
7.1 Database Enlargement ...................................................................................................... 124
7.2 Vision-based Façade Features Reading Technique .......................................................... 124
7.3 Regression Models Considering More Factors ................................................................. 125
Reference .................................................................................................................................... 126
Bibliography ............................................................................................................................... 134
Appendix ..................................................................................................................................... 144

9

List of Figures
Figure 1.1 Consumption and Gross Energy Intensity by Census Region for Sum of Major Fuels
for Non-Mall Buildings, CBECS (CBECS 2007)......................................................................... 19
Figure 1.2 IES VE energy simulation tool .................................................................................... 21
Figure 1.3 Energy savings in Portfolio Manager Buildings ......................................................... 23
Figure 1.4 National median reference EUI of selected building types ......................................... 24
Figure 2.1 Improvements from Standard 90-75 to 90.1-2010 by Mark Halverson, PNNL .......... 31
Figure 2.2 The 2030 Challenge Source: 2030. Inc. / Architecture 2030. All Rights Reserved. ... 32
Figure 2.3 EPA-Target Finder ...................................................................................................... 34
Figure 2.4 Benchmarking benefits summary ................................................................................ 37
Figure 2.5 U.S. Building Benchmarking and Transparency Policies by IMT, BuildingRating ... 38
Figure 2.6 Sun path with different building orientation in July .................................................... 40
Figure 3.1 Methodology diagram ................................................................................................. 46
Figure 3.2 Collected existing benchmarking building energy data .............................................. 49
Figure 3.3 Facade features for multiple linear regression............................................................. 50
Figure 3.4 Weather features and built year created by Mark Halverson, PNNL .......................... 50
Figure 3.5 Orientation and adjacent building position.................................................................. 51
Figure 3.6 100 Wall Street SU model ........................................................................................... 52
Figure 3.7 Cross-validation ........................................................................................................... 54
Figure 4.1 Building science-based climate map (U.S. DOE 2013) .............................................. 55
Figure 4.2 Site EUI and construction year .................................................................................... 57
Figure 4.3 Site EUI and renovation .............................................................................................. 57
10

Figure 4.4 Site EUI and WWR ..................................................................................................... 58
Figure 4.5 Site EUI and operable window .................................................................................... 58
Figure 4.6 Site EUI and V/FA ratio .............................................................................................. 59
Figure 4.7 Site EUI and orientation .............................................................................................. 59
Figure 4.8 Site EUI and HDD ....................................................................................................... 60
Figure 4.9 Site EUI and CDD ....................................................................................................... 60
Figure 4.10 Site EUI and climate zone ......................................................................................... 61
Figure 4.11 Standard residual and frequency plot ........................................................................ 63
Figure 4.12 Normal probability plot ............................................................................................. 63
Figure 4.13 Standard residual and fitted value ............................................................................. 64
Figure 4.14 PLS model selection plot ........................................................................................... 64
Figure 4.15 PLS response plot ...................................................................................................... 64
Figure 4.16 PLS standard coefficient plot .................................................................................... 65
Figure 4.17 Validation and error rate ............................................................................................ 66
Figure 4.18 Error rate comparison ................................................................................................ 67
Figure 4.19 Regression results comparison .................................................................................. 67
Figure 4.20 Site EUI and construction year .................................................................................. 69
Figure 4.21 Site EUI and building height ..................................................................................... 70
Figure 4.22 Site EUI and WWR ................................................................................................... 70
Figure 4.23 Site EUI and operable window .................................................................................. 70
Figure 4.24 Site EUI and V/FA ratio ............................................................................................ 70
Figure 4.25 Site EUI and orientation ............................................................................................ 71
11

Figure 4.26 Site EUI and floor area .............................................................................................. 71
Figure 4.27 Site EUI and HDD ..................................................................................................... 72
Figure 4.28 Standard residual and frequency plot ........................................................................ 74
Figure 4.29 Normal probability plot ............................................................................................. 75
Figure 4.30 Standard residual and fitted value ............................................................................. 75
Figure 4.31 PLS model selection plot ........................................................................................... 76
Figure 4.32 PLS response plot ...................................................................................................... 76
Figure 4.33 PLS standard coefficient plot .................................................................................... 76
Figure 4.34 PLS distance plot ....................................................................................................... 77
Figure 4.35 Validation and error rate ............................................................................................ 77
Figure 4.36 Error rate comparison ................................................................................................ 78
Figure 4.37 Regression results comparison .................................................................................. 79
Figure 4.38 Site EUI and construction year .................................................................................. 82
Figure 4.39 Site EUI and WWR ................................................................................................... 82
Figure 4.40 Site EUI and operable window .................................................................................. 82
Figure 4.41 Site EUI and V/FA ratio ............................................................................................ 82
Figure 4.42 Site EUI and orientation ............................................................................................ 83
Figure 4.43 Site EUI and CDD ..................................................................................................... 83
Figure 4.44 Validation and error rate ............................................................................................ 86
Figure 4.45 Error rate comparison ................................................................................................ 86
Figure 4.46 Regression results comparison .................................................................................. 87
Figure 4.47 Monthly site EUI in Building 1 ................................................................................. 88
12

Figure 4.48 Monthly site EUI in Building 2 ................................................................................. 88
Figure 4.49 Monthly site EUI in Building 3 ................................................................................. 89
Figure 4.50 Monthly site EUI in Building 4 ................................................................................. 89
Figure 4.51 Monthly site EUI in Building 5 ................................................................................. 89
Figure 4.52 Standard residual and frequency plot ........................................................................ 91
Figure 4.53 Normal probability plot ............................................................................................. 92
Figure 4.54 Standard residual and fitted value ............................................................................. 92
Figure 4.55 PLS model selection plot ........................................................................................... 93
Figure 4.56 PLS response plot ...................................................................................................... 93
Figure 4.57 PLS standard coefficient plot .................................................................................... 93
Figure 4.58 PLS distance plot ....................................................................................................... 94
Figure 4.59 Validation and error rate ............................................................................................ 94
Figure 4.60 Error rate comparison ................................................................................................ 95
Figure 4.61 Regression results comparison .................................................................................. 96
Figure 4.62 Geometry model of building 2, 3, 4 in DesignBuilder .............................................. 97
Figure 4.63 Annual EUI Calibration for building 2, 3 and 4 ........................................................ 98
Figure 4.64 Estimated monthly heating EUI comparison ........................................................... 100
Figure 4.65 Estimated monthly cooling EUI comparison .......................................................... 101
Figure 4.66 Validation model monthly heating EUI .................................................................. 102
Figure 4.67 Validation model monthly cooling EUI .................................................................. 102
Figure 5.1 U.S. Census regions and divisions (U.S. Department of Commerce 2010) .............. 103
Figure 5.2 Site EUI results comparison in New York City ........................................................ 104
13

Figure 5.3 Site EUI results comparison in Los Angeles ............................................................. 105
Figure 5.4 external view of case study buildings (Martin 2010) ............................................... 108
Figure 5.5 Monthly energy use for 5 years ................................................................................. 109
Figure 5.6 Electricity and natural gas consumption in 2014 ...................................................... 109
Figure 5.7 16th floor plan and zoning ......................................................................................... 110
Figure 5.8 Building geometry ..................................................................................................... 110
Figure 5.9 IES model and surroundings ..................................................................................... 111
Figure 5.10 Annual hourly outdoor dry-bulb temperature.......................................................... 112
Figure 5.11 Annual hourly end-use energy................................................................................. 113
Figure 5.12 Monthly system energy use breakdown .................................................................. 114
Figure 5.13 Monthly electricity and natural gas use ................................................................... 114
Figure 5.14 Monthly carbon emission from system energy use ................................................. 114
Figure 5.15 Annual EUI results comparison............................................................................... 116
Figure 5.16 Monthly EUI results comparison............................................................................. 117
Figure 6.1 Façade features reading illustration. Photo: (Free-wapaperbase 2015) ..................... 120
Figure 7.1 Vision-based façade features reading illustration ...................................................... 124

14

List of Tables
Table 2.1 2030 Challenge baseline and reduction targets (kBtu/sf/yr) ......................................... 34
Table 2.2 The 2030 Challenge Interim Code Equivalents (Marzria and Kershner 2009) ............ 36
Table 3.1 Predictors definition and explanation ........................................................................... 50
Table 3.2 Regression model indicators ......................................................................................... 53
Table 4.1 Total number of buildings in each group ...................................................................... 56
Table 4.2 Coefficients and main indicators .................................................................................. 62
Table 4.3 PLS coefficients and main indicators (continued) ........................................................ 62
Table 4.4 Total number of buildings in each group ...................................................................... 68
Table 4.5 Coefficients and main indicators .................................................................................. 73
Table 4.6 PLS coefficients and main indicators (continued) ........................................................ 74
Table 4.7 Total number of buildings in each group ...................................................................... 80
Table 4.8 Coefficients and main indicators .................................................................................. 84
Table 4.9 Coefficients and main indicators .................................................................................. 90
Table 4.10 PLS coefficients and main indicators (continued) ...................................................... 91
Table 4.11 Basic information of 4 office buildings ...................................................................... 97
Table 4.12 Cases (180 monthly data sets) .................................................................................... 98
Table 4.13 Revised regression models........................................................................................ 100
Table 5.1 Key façade features in NYC and LA .......................................................................... 107
Table 5.2 Building equipment power density, lighting power density and occupancy .............. 111
Table 5.3 Simulation results ....................................................................................................... 113
Table 5.4 Comparison content .................................................................................................... 115
15

Table 5.5 Monthly EUI results and comparison ......................................................................... 116
Table 5.6 Error rates of monthly EUI results .............................................................................. 117
Table 5.7 NMBE and CVRMSE for monthly EUI ..................................................................... 117

16

Abstract
The commercial and residential building sector accounts for about 40% of carbon dioxide (CO
2
)
emissions in the United States per year, more than any other sector (Eddy and Marton 2012). The
most significant factor contributing to CO
2
emissions from buildings is their use of electricity.
Commercial and residential buildings are tremendous users of electricity (Department of Energy
2011), accounting for more than 73% of electricity use in the U.S.
Energy use data from the Commercial Building Energy Consumption Survey (CBECS) is an
average value based on the range of Heating Degree Days (HDD) and Cooling Degree Days
(CDD), which can’t show the specific condition of each building category within one area. In
addition, the average value is too general to evaluate if a specific building case is energy efficient
or not. On the other hand, it is very time consuming to develop a simulation model in software,
which also needs very detailed information about the building itself. The accuracy depends on
how much specific information of envelop thermal conditions, mechanical system performance,
occupancy level and schedule, etc.
Among 3 main factors to influence building energy performance, building façade features are
more easily obtained as opposed to building mechanical systems and schedule information. By
using façade features, certain key attributes could be input to generate a customized baseline
model and to estimate building energy use intensity (EUI).
A simple regression model can be used to calculate the EUI baseline instead of complicated
simulation tools, and the results are accurate and reasonable at an acceptable level. The
calculated baseline can be used for setting a practical baseline for energy reduction target. Due to
its simplicity and quick processing time, the research outcome would also be applicable to the
real-time energy estimation of multiple buildings at an urban scale.
This new method of linear regression analysis is developed to estimate building energy
consumption just based on simple façade attributes and weather conditions. Building façade
features, for example, including shading, window-to-wall ratio, orientation, surface-to-volume
ratio, etc. are easy to obtain. It is meaningful to use a simple way to predict heating and cooling
energy use instead of traditional energy performance simulation tool which is time and resource
17

consuming. Based on collected building physical attribute data, statistical methods could be used
to generate a customized baseline Energy Use Intensity (EUI) estimation model. The proposed
idea will also adopt a simplified building energy performance prediction model as a function of
architectural physical frames and their dynamic ambient environmental condition, such as
monthly cooling/heating degree days. The main goal of this research is to develop a
mathematical method to provide a customized baseline model for buildings, considering specific
façade features and local climate condition. It will provide a direct estimation and prediction of
project building energy performance to provide a reasonable baseline for designer, engineer and
client.
Hypothesis
A simple regression model based on façade features can be used to calculate EUI baselines and
set a more reasonable and accurate energy conservation target.
18

Chapter 1 Introduction to the Building Energy Estimation Method
In 2010, the U.S. consumed 97.8 quads of energy, which represented 19% of global energy
consumption (Program and Efficiency 2012). In the United States, the building sector, including
residential and commercial buildings, accounted for about 41% of primary energy consumption
in 2010. Space cooling, space heating and lighting are the dominant end uses, which accounted
for about 52% of total energy consumed by the building sector. Façade features, such as exterior
wall type, glazing type, shading type, window-to-wall ratio etc. have a great influence on space
heating, cooling and even lighting demand (Shan 2014). A good building façade design will be
greatly useful to reduce energy demand by selecting appropriate façade features according to
local climate characteristics. Conversely, building energy performance could be predicted by
using façade features and climate conditions.
In addition, Energy Use Intensity (EUI) is a very important indicator (Andrews and Krogmann
2009), which can be used to evaluate building energy performance and energy saving potential,
improve operation and maintenance practice, and determine passive design strategies, etc.
Annual EUI could be a baseline for building owners and designers to set a reasonable energy
reduction goal for the following years. In reality, building energy simulation is very time-
consuming and the accuracy depends on available building information for input. Especially on
the urban scale, it’s hard to generate a simulation model for each building to estimate actual
building energy use.
Demands from urban planners and building designers require a new method to set a baseline by
using a simple way at the beginning of design stage, which could be based on easily readable
information like building façade features.
1.1 Problem
An energy use intensity (EUI) baseline currently relies on a national or local energy usage
average. However, the average/median energy consumption can’t be accurate and reasonable to
measure a local specific building case performance, since the building energy consumption rate
varies greatly based on different design condition and climate. The national/regional average
performance criteria that have been popularly adopted to assess building energy performance,
19

have serious limitations since they could not provide a realistic baseline for comparison with
actual energy consumption rates and set a reasonable energy reduction goal for energy efficiency
improvement.

Figure 1.1 Consumption and Gross Energy Intensity by Census Region for Sum of Major Fuels for Non-Mall Buildings, CBECS
(CBECS 2007)
The Commercial Buildings Energy Consumption Survey (CBECS) is a national sample survey
(“About Commercial Buildings Energy Consumption Survey” 2012) that collects information on
the stock of U.S. commercial buildings, including their energy-related building characteristics
and energy usage data (consumption and expenditure). In consumption and expenditures tables
for non-mall buildings, energy use intensity is an average value based on census division, climate
zone, building size or year constructed, which can’t represent the specific physical condition of
each building, since it doesn’t consider any individual building feature. Besides, the average
value, based on census division, climate zone or HDD/CDD range, is too general to categorize
20

weather conditions. The energy use estimation for a building case is not accurate enough to
evaluate energy efficiency. For example, from CBECS consumption and expenditures tables for
non-mall buildings 2003 table C5 (CBECS 2007), an average EUI could be obtained by
specifying census region as well as another factor, like building floorspace area, principal
building type, year constructed, climate zone, number of floors, etc. Office buildings in the west
census region consumed 72.1 kBtu/sf on average, which is a general number for offices in this
region, not a specific baseline to reflect a project’s real energy use.
To more accurately estimate building energy use intensity, many simulation tools could be
utilized for detailed calculation. However, the accuracy significantly depends on how much
building information (envelope, mechanical system performance, occupancy schedule, etc.) can
be obtained for modelling and how proficient energy analysts are for using different programs
(Zhai and McNeill 2013). Even the selection of program will influence the result greatly, since
each program has different calculation methods, capabilities and features

(Crawley et al. 2008).
In addition, considering the time cost of building energy modelling, a new solution to predict
energy use intensity is necessary.

21

Figure 1.2 IES VE energy simulation tool
Figure 1.2 shows building energy modeling by using IES VE 2013. More accurate results depend
on more detailed information input, but that information could hardly be obtained from designers
or owners. Especially in the beginning of design stage, many assumptions should be made to
generate an energy model based on local standards and experience.
1.2 Explanation of Terms
1.2.1 Energy Use Intensity
Energy Use Intensity (EUI) shows a building’s energy use as a function of its size or other
characteristics, which is calculated by dividing annual building energy consumption in one year
by the total gross floor area as kBtu/sf. There are two kinds of EUI: source EUI and site EUI.
Normally, source EUI reflects the total amount of raw fuel consumed for the operation of a
building, which is more comparable and easy to evaluate since all losses from generation,
transmission, delivery and production are taken into consideration. The Environmental
Protection Agency (EPA) suggests using source EUI for equitable evaluation of building energy
performance and Energy Star Score (Energy Star 2014a). On the other hand, site EUI represents
the amount of energy consumed on site, normally electricity and natural gas, which could be
obtained from energy bills.
Generally, a low EUI implies low energy use and high energy performance, but EUI varies
significantly according to building function (Energy Star 2012). From the report (Energy Star
2011), grocery store and hospital functions consume more energy on average since they are more
equipment intensive and their operation period is longer than other building types.
22

The collaboration of ASHRAE, the U.S. Department of Energy, the National Institute of
Standards and Technology and the U.S. Environmental Protection Agency define the importance
of EUI(“What Is Energy Use Intensity? ASHRAE Seeks to Define, Educate” 2011):
1. Establish a single objective definition of energy use intensity (EUI) for the design of
commercial buildings;
2. Determine a single objective baseline EUI for design of commercial buildings from
which to measure relative energy use reductions;
3. Create a performance environment that will support reduction in energy consumption
associated with all loads in commercial buildings;
4. Identify a single objective set of commercial building types and simulation models for
establishment of target design EUIs;
5. Produce one set of design target EUIs for the commercial building sector to guide the
development of future energy codes and standards and building energy codes adopted by
state and local government.
1.2.2 Benchmark
Benchmarking indicates tracking and recording each building’s energy use to establish a baseline
of energy performance. By using an equitable metric to compare each building energy use with
its past performance as well as equivalent and similar buildings, building owners and managers
will be capable of understanding their building energy use more deeply and the potential of
improving efficiency and making the most cost effective decisions.
Benchmarking is significant for energy saving by tracking energy use. Over 35,000 buildings
used U.S. Environmental Protection Agency’s (EPA’s) ENERGY STAR Portfolio Manager to
benchmark energy use (EPA 2011). Fig 1.4 shows the average source EUI reduction since the
baseline year, 2008. The average energy saving in 2011 is about 7% compared with the energy
use in 2008.
23

Figure 1.3 Energy savings in Portfolio Manager Buildings
There are 10 cities, 2 states, and 1 county that have adopted building energy benchmarking and
disclosure laws (IMT 2014a), including Austin, New York City, Seattle, San Francisco,
Philadelphia, Boston, Minneapolis, Chicago, etc. The policies are used to reduce energy
consumption and carbon emission based on improved transparency of building energy use.
1.2.3 Architecture 2030 and CBECS
The non-profit independent organization Architecture 2030 was established to change buildings
from a major greenhouse gas (GHG) emission source into a solution of global climate change
and energy crisis. The 2030 challenge proposed the following targets: for all new buildings,
energy consumption needs to be 60% below the regional (or country) average/median for that
building type. The fossil fuel reduction standard for all new buildings and major renovations
shall be Carbon-neutral in 2030 (Architecture 2030 2011).
Architecture 2030 uses the Commercial Buildings Energy Consumption Survey (CBECS) 2003
data, which provides national and regional medians. The baseline starting point for the common
targets is defined as the national average/median energy consumption of existing U.S.
commercial buildings as reported by the 2003 CBECS. CBECS data is a set of whole-building
energy use measurements gathered by the DOE’s Energy Information Administration, which can
be used to determine a national energy use intensity using kBtu/sf per year as the metric (“About
Commercial Buildings Energy Consumption Survey” 2012).
24

1.2.4 National Median EUI

Figure 1.4 National median reference EUI of selected building types
For benchmarking, the national median source EUI is recommended as a reasonable metric. The
median value means half of the buildings use more energy than this value while half use less
energy. Compared with mean value (arithmetic average), median value represents the mid-point
of energy use for each property type, which is more accurate and comparable because it is less
prone to being skewed by building anomalies with very large individual usage. Figure 1.5 lists
national median reference source EUI and site EUI for selected Portfolio Manager Property types,
which mostly are from CBECS survey data (Energy Star 2014).
1.2.5 Energy Simulation
Computer simulations can be used to estimate building energy performance, which also can be
incorporated into sustainable design. Sustainable building design consists of methods to develop
the built environment while meeting the intent of sustainable development (Zhai and McNeill
2013). Simulation tools have been largely used to assist green building design with quantitative
data provided for decision support. Building energy simulation is a process of building modelling
that focuses on energy consumption and relevant items, like HVAC system, renewable energy,
life-cycle analysis, etc. Building energy simulation can be used in different design phase by
generating different indicators and results. For example, in early design phases, it can help to
make general decisions based on the impact analysis related to alternatives, like orientation,
25

construction fabrics, etc. Other usages include determining energy end use, evaluating specific
technology, conducting saving measurement and verification (M&V) and diagnosing building
operation.
Based on Whole Building Analysis: Energy simulation from Building Energy Software Tools
Directory (U.S. DOE 2011), there are more than 100 kinds of energy simulation tools in market.
Some popular software, like IES, DesignBuilder, eQUEST, Trace, etc. are widely utilized in
building design by architects, engineers and researchers.
1.2.6 Building Façade
The façade, or the exterior walls of a building, is the physical separation between outside
environment and inside conditioned space, which could resist air, heat, water, light, noise
transfer (Cleveland 2009). Many studies focus on the impact of façade design on building energy
performance. A study (Thalfeldt et al. 2013) shows that in heating dominated climate window
properties, external wall insulation, window-to-wall ratio (WWR) and external shading could be
optimized for the lowest life cycle cost as well as higher energy efficiency. Another study (Soto
Francés et al. 2013) presents a mathematical model of ventilated façade with improved
description of thermal responses.
Façade design needs to be suitable for each climate condition by adopting different façade design
strategies. Basic methods for designing high performance façades (Aksamija 2013) include
orienting and developing geometry and massing as response to solar position, providing shading
to reduce cooling load and improve thermal comfort, using natural ventilation, optimizing
exterior insulation and using daylighting.
1.2.7 Multiple Regression Analysis, Standardized Coefficient and Significance Level
In statistics, multiple regression analysis (MRA) is an approach for modelling the relationship
between a scalar dependent variable and multiple explanatory variables by fitting a linear
equation to observe data. Multiple regression analysis (Catalina, Virgone, and Blanco 2008) is
used to predict the single dependent variable (EUI, etc.) by a set of independent variables (façade
features, etc.). Besides to model the relationship between the response variable and the
corresponding explanatory variables, another objective of using MRA (Lam, Hui, and Chan 1997)
26

is to study the effects of various predictors on building energy performance. Other methods,
Artificial Neural Networks (Karatasou, Santamouris, and Geros 2006) is also used to estimate
energy prediction when modeling non-linear process (control schemes for HVAC systems for
example), but MRA is easier and more practical than neural networks method to estimate general
monthly or annual EUI, especially for a large datasets. The predicted model is constructed as the
following form:

(1)
Where is the constant while

are the regression coefficients,

are the significant
predictors and is the random error.
Regression coefficients could be standardized by re-expressing coefficients as the effect of a
one-Standard Deviation (SD) change in

as opposed to a unit change in

. The unit of
predictors and outcome responses could be removed by standardized regression coefficients. The
benefits are not only standardizing statistics but also making predictors comparable of relative
effects on the response. The greater the coefficient is, the more significant this corresponding
predictor could influence the response.
The level of significance (normally called alpha) indicates the probability of how extreme
observed results must be to reject the null hypothesis of a significance test (Taylor 2014). Alpha
ranges from 0 to 1, and it relates to the Confidence Interval (CI) which presents the range of
likely values for a population parameters. A 95% level of confidence with the alpha value of 0.05
is adopted indicates when a p-value of responding predictor is less than 0.05 the result could be
considered statistically significant since the null hypothesis would be rejected. Alpha level of
0.05 is reasonably selected because of the field of building energy estimation in this research.
1.2.8 Degree Days
Degree Days is a method created by engineers to relate each day’s temperature to the demand for
fuel to heat or cool buildings. Heating degree days (HDD) indicate average daily temperatures
(Gronadzik et al. 2010) below the base temperature (e.g. 65°F) while cooling degree days (CDD)
indicate average daily temperatures above the base temperature (e.g. 65°F).
27

To calculate HDD65 for example, each day’s average temperature which is the halfway between
high and low) is subtracted from the base temperature (65°F) and the result is the number of
HDD65 for that day. If the average temperature is higher than the base temperature, there is no
recorded HDD65 for this day. The total HDD65 is the addition of daily HDD65 for the whole
year.
As an outside condition, Degree Days (BizEE 2014) is used to represent how much (in degrees)
and how long (in days) heating or cooling needs to be used and other energy-related calculations.
Normally the heating degree season begins from July 1
st
and the cooling degree day season
begins from January 1
st
. Degree days may change from one area to another area as well as from
one year to another year.
1.2.9 Partial Least Squares and Cross-validation
Partial least squares (PLS) is a biased, non-least squares regression procedure that relates a set of
predictor variables to multiple response variables, which is particularly useful when the
predictors are highly collinear or the numbers of predictors are greater than observations (Roush
1982). PLS could reduce the number of predictors to a set of uncorrelated components and then
perform the least squares regression on these components (Minitab 2014a).
Cross-validation is a method to select a regression model based on the predictive ability of the
models (Shao 1993). The data sets are divided into 2 groups: the first group is used to fit a new
model while the second group is used to assess the predictive ability of the model. To validate
the results of PLS, the cross-validation method should be used to calculate the predictive ability
of potential models by determining the minimum number of components retained in the model.
The optimal number of components and responding model could be selected with appropriate
indicators.
1.3 Objective
A regression model based on basic visualized building façade attributes input is a quite simple
and reliable way to estimate building energy consumption instead of using average data from a
survey or running simulation in software. It will provide a direct estimation and prediction of
project building energy performance to set a reasonable baseline for designer, engineer and client.
28

The main goal of this research is to develop a customized baseline model by using mathematical
methods for buildings, considering specific façade features and local climate condition. Due to
its simplicity and quick processing time, the research outcome would be applicable to set a
reasonable EUI reduction baseline for building performance management and improvement
instead of using national average numbers from a survey.
The first objective is to explore the impact of basic façade features on energy performance in
different climate zones. By using the EUI estimation model, a sensitivity analysis will provide an
overview of how façade features could influence building energy consumption based on real
energy database.
The second objective is to estimate energy use of multiple buildings at an urban scale. For a city,
or a region, it is extremely time consuming to generate an energy model by simulation tool for
each individual building. It is hard to make an energy consumption baseline or limit at an urban
scale, which is useful for urban planning. The outcome tool could make it possible to quickly
process energy use estimation for different buildings based on simple input. The tool would be
beneficial for city planners to be more strategic about setting priorities and allocating resources.
In addition, with the enlargement of the measured building EUI database, multiple estimation
models could be developed for different building function, cities, climate zones or even global-
wide area. Estimation model sets could be at different accuracy level based on the availability of
building façade information and the precision of original data.
Last but not least, this research would draw more attention to the significance of building energy
use disclosure for public. A customized baseline could be more acceptable for building owners to
know building energy saving potential and adopt measures to improve energy efficiency, which
in turn will benefit energy conservation for the whole society.

29

Chapter 2 Background of EUI Benchmarking and Calculation Methods
Globally, buildings consumed about one third of final energy use (Höhne et al.) and contributed
to about 20% of global greenhouse gas emissions. Many actions (Architecture 2030 2014) are
proposed to provide initiatives to meet the GHG emissions reduction targets of a high probability
of holding global warming below 2°C.
The following actions or requirements are representative which can be used to improve the
energy efficiency in building industry:
 On September 26, 2014, DOE issued a determination that ASHRAE Standard 90.1-2013
would achieve greater energy efficiency in buildings subject to the code. DOE estimates
national savings in commercial buildings of approximately: 8.8% energy cost savings, 8.5%
source energy savings and 7.6% site energy savings (U.S. DOE 2014a).
 In Europe, Directive 2010/31/EU Article 9 (EPBD 2014) requires that “Member States
shall ensure that by 31 December 2020 all new buildings are nearly zero-energy buildings;
and after 31 December 2018, new buildings occupied and owned by public authorities are
nearly zero-energy buildings”.
 California building energy code Title 24 requires all new commercial buildings achieving
zero net energy (ZNE) by 2030 while all new residential building achieving ZNE by 2020
(Roth 2013).
 In China, Minimum energy performance standards (MEPS) require a minimum energy
saving level (Bigee 2012) of 50% to 65% in different climate zones.
To determine building energy baseline and reduction targets, currently Target Finder (EPA
2014a) developed by U.S. Environmental Protection Agency is the predominant tool which could
be used to set a simple EUI baseline by inputting a minimum amount of building information.
When the building type is not listed in Target Finder, Commercial Buildings Energy
Consumption Survey (CBECS) 2003 data could be used instead, which a national median of
energy use.
On the other hand, building energy benchmarking policies (Milliken and Jones) are adopted by
many cities and states to drive up demand for energy efficiency. Benchmarking programs can be
30

used to improve building energy performance as well as to increase the public awareness on
energy saving in building sectors. In 2010, New York City was the first city to adopt a
mandatory rating and disclosure program. In 2012, there were about 35,000 benchmarked
building around the U.S. which get an average savings of 2.4% of energy use annually (Milliken
and Jones). In this research, building benchmarking data are used as one of EUI data sets for
regression model development as well as certain amount of samples for validation.
Among building envelope attributes, the most significant façade features are selected for the EUI
estimation model, which could be simply read without digging detailed information about
buildings. Other EUI estimation methods are compared with the regression model and the latter
is considered as the simplest method at a reasonable and acceptable level.
2.1 Building Energy Reduction Plan and Target
2.1.1 International Energy Conservation Code (IECC)
The Internal Energy Conservation Code (ICC 2014) is designed to meet the requirement of
energy-efficient building performance, including building envelopes and systems through code
regulations. The minimum building energy performance for new commercial and residential
buildings is regulated through the improved requirement (IECC 2012) for each part of building
energy use, including heating, cooling, ventilation, lighting, water heating, and other systems.
A study conducted by Pacific Northwest National Laboratory (PNNL) compared the 3 editions
of the IECC, 2006, 2009 and 2012 to analyze the energy reduction by meeting each code. The
results (Zhang, Athalye, and Hart 2013)show that energy savings could be 8.7% by meeting
2009 IECC and 18.6% by meeting 2012 IECC compared with 2006 IECC on a weighted national
basis.
2.1.2 ASHRAE 90.1
The American Society of Heating, Refrigerating and Air Conditioning Engineers’ standard 90.1
is a national guideline (ASHRAE 2013) for designing energy-efficient buildings (except low-rise
residential buildings). It provides detailed requirements for design and construction of new
buildings and systems, new sections of buildings and systems, as well as new systems and
upgraded equipment in existing buildings. ASHRAE 90.1 provides the requirements for all
31

aspects of building (Schwedler et al. 2010), including building construction materials thermal
performance, HVAC system efficiency, load, insulation and controls, lighting, etc. In addition, 3
methods are presented for system and equipment compliance as well as the regulation to use
simulation tool in Appendix G.
The first standard ASHRAE 90-75 was published in 1975, which was the guideline of new
building energy conservation for all buildings (Hunn 2010). The latest version is ASHRAE 90.1-
2013 which determined a higher energy efficiency requirement for buildings. The energy saving
by using this version of standard was estimated by DOE (U.S. DOE 2014a), which predicted 7.6%
site energy savings and 8.7% energy cost savings. The following figure shows the energy saving
improvement from the first version to ASHRAE 90.1-2010 (Hunn 2010).

Figure 2.1 Improvements from Standard 90-75 to 90.1-2010 by Mark Halverson, PNNL
2.1.2 The 2030 Challenge
Residential and commercial buildings (EPA 2014b) are the major source of greenhouse gases
(GHG) emissions, which come directly from fossil fuel combustion for heating, cooling, cooking,
management of waste and wastewater, as well as indirectly from electricity generation consumed
by building appliances and equipment. GHG emission reduction is becoming a serious topic in
the building and energy industry.
32

Architecture 2030 issued The 2030 Challenge to reduce GHG emission and improve the global
and industrial collaboration by setting the following targets (2030 Inc. 2011) :
1. All new buildings, developments and major renovations shall be designed to meet a fossil
fuel, GHG-emitting, energy consumption performance standard of 60% below the
regional (or country) average/median for that building type.
2. At a minimum, an equal amount of existing building area shall be renovated annually to
meet a fossil fuel, GHG-emitting, energy consumption performance standard of 60% of
the regional (or country) average/median for that building type.
3. The fossil fuel reduction standard for all new buildings and major renovations shall be
increased to:
 70% in 2015
 80% in 2020
 90% in 2025
 Carbon-neutral in 2030 (using no fossil fuel GHG emitting energy to
operate).
To accomplish these targets in each time period, innovative sustainable design strategies are
encouraged, as well as generating on-site renewable power and/or purchasing (20% maximum)
renewable energy.

Figure 2.2 The 2030 Challenge Source: 2030. Inc. / Architecture 2030. All Rights Reserved.
Major renovations (2030 Inc. 2011) refer to the energy reduction by improving building
envelope or the technical building systems. The 2030 Challenge requires either more than 25%
of total cost is related to envelope or systems or more than 25% of the envelope surface needs
33

renovation, which also emphasized the significance of building envelope or façade on energy
consumption or the following GHG reduction.
2.2 EUI Baseline
On May 4, 2007, the EUI baseline starting point (2030 Inc. 2011) for their common target goals
as the national average/median energy consumption of existing U.S. commercial buildings was
defined by the following organizations:
 American Institute of Architects (AIA)
 American Society of Heating, Refrigeration and Air-Conditioning Engineers (ASHRAE)
 Architecture 2030
 Illuminating Engineering of North America (IESNA)
 U.S. Green Building Council (USGBC)
 U.S. Department of Energy
The metric for the 2030 Challenge is site EUI in kBtu/sq. ft.-yr, which is from 2030 CBECS data.
CBECS data is collected by Energy Information Administration of U.S. Department of Energy,
which could be adopted to decide a national average EUI baseline. By 2030, the target is not
“Net-zero energy” but “Carbon-neutral” buildings, which uses no fossil fuel and greenhouse gas
emitting energy to operate without the requirement of on-site energy.
2.2.1 Target Finder
Target Finder (EPA 2014a) is an online tool created by Environmental Protection Agency (EPA)
which enables architects, engineers, property owners and managers to assess the average energy
consumption as well as to determine energy reduction baseline and targets regarding to the 2030
Challenge. Basic input includes building location, primary function, gross floor area, energy
estimates (optional), operating hours, number of computers and workers, percentage of
heated/cooled space, etc.
34

Figure 2.3 EPA-Target Finder
When it comes to the determination of a project’s energy consumption baseline and target,
Target Finder is the first tool which is recommended by Architecture 2030. In addition, to ensure
the baseline is an average value with an average fuel mix, Estimated Design Energy is not
encouraged to be input into Section 4.
2.2.2 CBECS
The Commercial Buildings Energy Consumption Survey (CBECS) is a national sample survey
(“About Commercial Buildings Energy Consumption Survey” 2012) that collects information on
the stock of U.S. commercial buildings, including their energy-related building characteristics
and energy usage data (consumption and expenditures).
When the building type is not available in Target Finder, CBECS 2003 data should be used to
obtain the national and/or regional medians and the following reduction targets.
Table 1 (Description and Site 2012) presents EUI values from CBECS 2003 data by using
“Table 1: 2003 CBECS National Median Source Energy Use and Performance Comparison by
Building Types” expect for the building types which could be found in Target Finder marked by
“x”.
Table 2.1 2030 Challenge baseline and reduction targets (kBtu/sf/yr)
Building Type Available in
TargetFinder
Median
Site EUI
50% Target 60% Target 70% Target 80% Target 90% Target
Education 58 29.0 23.2 17.4 11.6 5.8
K-12 School x
College/Univ. 104 52.0 41.6 31.2 20.8 10.4
Food Sales 193 96.5 77.2 57.9 38.6 19.3
Grocery Store x
Convenience 228 114.0 91.2 68.4 45.6 22.8
Food Service 267 133.5 106.8 80.1 53.4 26.7
Restaurant 207 103.5 82.8 62.1 41.4 20.7
35

Fast Food 418 209.0 167.2 125.4 83.6 41.8
Hospital x
Lodging 72 36.0 28.8 21.6 14.4 7.2
Dormitory x
Hotel/motel x
Mall 94 47.0 37.6 28.2 18.8 9.4
Nursing x
Office x
Outpatient 62 31.0 24.8 18.6 12.4 6.2
Clinic 67 33.5 26.8 20.1 13.4 6.7
Medical Office x
Public assembly 42 21.0 16.8 12.6 8.4 4.2
Entertainment 46 23.0 18.4 13.8 9.2 4.6
Library 92 46.0 36.8 27.6 18.4 9.2
Recreation 39 19.5 15.6 11.7 7.8 3.9
Social/Meeting 43 21.5 17.2 12.9 8.6 4.3
Public Order 82 41.0 32.8 24.6 16.4 8.2
Police Station 82 41.0 32.8 24.6 16.4 8.2
Service 45 22.5 18.0 13.5 9.0 4.5
Storage 10 5.0 4.0 3.0 2.0 1.0
Warehouse x
Religious x
Retail Store x
Other 70 35.0 28.0 21.0 14.0 7.0

2.2.3 Meeting the 2030 Challenge through Building Codes
The baseline for the 2030 Challenge currently is from the 2003 Commercial Building Energy
Consumption Survey (CBECS) data for commercial buildings and the Residential Energy
Consumption survey (RECS) for residential buildings. However, the new building standards and
rating systems are still in development. Therefore, for cities and states which are looking to meet
the 2030 Challenge targets, an improvement of requirements (Marzria and Kershner 2009) from
current building energy codes and standards is needed to set up the baseline.
Architecture 2030 developed an interim system of ‘code equivalents’ to take the additional
reduction target from the current code requirements into consideration. The ‘code equivalents’
require the actual energy reduction of a particular code, standard or rating system to meet or
exceed the initial 50% target from the average annual energy use by building type. These code
equivalents can be used to help local governments to amend their existing building codes.
36

The following table provided by Architecture 2030 (Marzria and Kershner 2009) is developed to
specify the additional reduction requirements for each commonly used energy codes and
standards and rating systems. The required additional reductions could be used to meet or exceed
the initial 50% reduction target of the 2030 Challenge.
Table 2.2 The 2030 Challenge Interim Code Equivalents (Marzria and Kershner 2009)
CODE/STANDARD COMMERCIAL RESIDENTIAL
ASHRAE 90.1-2004 30% below -
ASHRAE 90.1-2007 25% below -
ASHRAE 189 (in process) - -
IECC 2006 30% below 30% below
California Title 24 2005 - 15%-20% below
California Title 24 2008 10% below -
Oregon Energy Code 25% below 30% below
Washington Energy Code 25% below 25%-30% below
HERS Index - 65 or less
LEED V4 EA Prerequisite 2: Minimum Energy Performance 5% for new, 3% for renovation
EA Credit 2: Optimize Energy Performance -
GBI Standard Path A, 8.1.1.1: 150 pts -
EECC Option (prescriptive path) - EC-154
NBI Option (prescriptive path) New-Core performance w/ enhanced measures -
Note: Table A, Meeting the 2030 Challenge Through Building Codes (Marzria and Kershner 2009)

2.2.4 Simulation Tool
When detailed building information is available, including building geometry, construction
materials, internal gain (people, equipment, lighting, etc.), mechanical systems, operation
schedule and profile, etc., a building energy simulation tool could be used to estimate building
load as well as energy use for each system. However, for new construction, each variable is still
under evaluation and assumptions based on building codes and standards have to be made to
develop the simulation model, which is not as accurate as expected. The main function of
simulation is to assist in making a decision on some basic building features in the predesign stage.
On the other hand, for an existing building, the accuracy of simulation results depends on how
much detailed building information could be obtained from the owner and facility management,
which brings about another problem of accurate on-site building information accessibility.
37

2.2.5 Real Recorded Energy Data
All of the methods and practices mentioned above would be replaced by real energy use data to
set the energy reduction baseline and target, because the real energy use is the most direct and
reasonable data that could represent the real-time energy consumption of a dedicated building.
Another fact is that building energy use varies every year, even when there is no retrofit or
upgrade of building envelope or systems. The major reason is weather conditions change each
year, which requires building systems to operate for different length of time. It results in energy
usage for meeting different requirements varies. The baseline should be representative to reflect
the building energy consumption under the most common weather condition.
2.3 Building Energy Benchmarking and Disclosure
Building energy benchmarking is a method to get building energy data as a baseline. It will give
owners a better understanding of how much energy their buildings exactly consume for a time
period. To accomplish the task of benchmarking, the energy monitoring and the measured
records from utilities measurement are needed, and the data should be submitted by using a
common format to be available to put into a database. The most commonly used tool is Portfolio
Manager developed by EPA (Energy Star 2014c), which could be used to track and evaluate
energy use for commercial buildings. The benefits of using benchmarking (Milliken and Jones)
to keep track of building energy use are listed in the following figure.

Figure 2.4 Benchmarking benefits summary
38

In the U.S. there are 9 cities (IMT 2014b) which committed to implementing energy
benchmarking and disclosure programs for commercial buildings (Cox, Brown, and Sun 2013),
which include Seattle, San Francisco, Austin, Minneapolis, Cambridge, Boston, New York City,
Philadelphia, Washington, DC, etc. In New York City, the benchmarking policy of Local Law 84
(LL84), part of Greener, Greater Buildings Plan (GGBP) was adopted in 2009 (GGBP 2013),
which requires all non-residential buildings with floor area over 50,000 square feet to submit and
disclose their building energy and water data to the city. The results show that the median source
EUI for office properties in 2010 and 2011 are 213.3 kBtu/sf and 207.3 kBtu/sf and the median
Energy Star score increased from 64 to 67. In 2013, San Francisco Environmental Code Chapter
20 (Environment 2014) requires the non-residential buildings with total floor area over 10,000
square feet must benchmark and report an Annual Energy Benchmark Summary (AEBS) to
government and existing tenants to reduce energy consumption and GHG emissions, and to
improve the commercial building energy efficiency. Even though San Francisco hasn’t released
the energy use metrics to public yet, 2012 report (Sewer 2012) stated that the overall EUI of
benchmarked facilities improved 3.6% from 2011 and the average 2012 carbon footprint
improved 5.1% from 2011.

Figure 2.5 U.S. Building Benchmarking and Transparency Policies by IMT, BuildingRating
39

2.4 Façade Features
2.4.1 Importance of Façade Features on Building Energy
The building façade is the most significant interface between indoor and outdoor environment.
Generally, building walls, roofs, windows, doors, floors, etc. are included in building façade or
envelope. The importance of a façade is not only about the physical function as a separation or a
shelter but also the vital influence of thermal performance on building energy use and the
occupant’s comfort. In addition to the temperature and humidity control (Bolin Rob 2014),
façades also provide building users security, access to daylight and outside view, fire resistance,
indoor environment quality control, aesthetics, etc.
There are so many façade factors having a great impact on building energy use, like R value
which reflects thermal resistance to the heat flow is an importance indicator of insulation and
Solar Heat Gain Coefficient (SHGC) which is a key factor to determine heat gain through
glazing. Thus, a sustainable façade design is an integrated and synergistic method considering
the comprehensive effects and optimizing each indicator in order to obtain the best practice of
building function.
2.4.2 Orientation
Orientation refers to the position of a building relative to the geometric condition or compass.
Good orientation indicates the indoor space could get sufficient sunlight while extra energy to
cool the building is not required because of overheated surfaces. For example, north and south
façades easily receive natural daylight while excessive heat gain could be minimized by well-
designed west and east facades, where there is significant heat transfer resulting from direct solar
heat gain in summer sunrise and sunset time period (Al-tamimi et al. 2009). In addition to
daylighting, local prevailing wind direction is another factor which is supposed to be considered
to make full use of natural ventilation strategy, which could save energy from mechanical
ventilation greatly. As a result, the optimized decision of building orientation results in energy
conservation by incorporating passive strategies and good thermal comfort.
40

Figure 2.6 Sun path with different building orientation in July
2.4.3 Operable Window
Operable windows provide building users controllability of natural ventilation and fresh air as
well as psychological comfort (Daly 2002), which can not only save energy but also decrease
Sick Building Syndrome (SBS) complaints. Mixed-mode ventilation which combined both
mechanical and natural ventilation is an efficient and compromising alternative to enhance
energy efficiency without sacrificing occupant satisfaction. Research stated that in Minnesota
(Guzowski 2003), 10% to 25% of the energy for fan operation could be saved by adopting
operable windows for natural ventilation. The cost of operable windows is about 50% to 75%
more than fixed windows (Carpenter 2014), however, calculating the cost-benefit of using
operable windows should include the consideration of the health and productivity improvement
by providing workers enough controllability on their work space in addition to energy savings.
2.4.4 Shading
To design an effective shading device, both heat gain and daylight accessibility should be taken
into consideration. Exterior shading has a better performance, such as overhang, vertical fins, etc.
because it blocks a significant portion of heat outside the building as a passive strategy. However,
41

in perimeter areas, daylighting is also important for both visual comfort and energy conservation.
A compromise needs to be made considering both sides with daylight control measures. For
interior areas, light shelves could be adopted to make use of reflected daylight and extend
daylighting availability.
2.4.5 Window-to-wall Ratio
Window-to-wall ratio (WWR) is defined as the ratio of total glazed area in the exterior envelope
to the total area of the façade. WWR will influence building energy consumption greatly since
solar radiation is directly introduced into indoor environment through fenestration which is the
most important external heat gain in cooling systems. Other than thermal performance of
windows, like U value, SHGC, Visible Light Transmission (VLT), etc. the simple window area
is generally the most direct factor to obtain solar radiation. WWR as a basic façade indicator, is
strictly limited by building standards and codes. For instance, ASHRAE 90.1-2010 requires the
vertical fenestration area is at most 40% of the gross above-grade wall area as the prescriptive
option (Schwedler et al. 2010). In the LEED rating system for new construction, Energy and
Atmosphere (EA) credit EAp2 Minimum Energy Performance requires achieving higher energy
efficiency by improving building envelope performance by meeting or exceeding the code
compliance (BuildingGreen 2014).
A study (Violeta and Egidijus Saulius 2010) showed the optimal WWR for office buildings in
Lithuania is 20% to 40%. Another research (S 2007) presented that for residential and
commercial buildings in Egypt, the preferred WWR is up to 17% and 20% respectively.
Although lower WWR is suggested on the energy saving side, but with decreasing window area,
the benefit of natural daylighting is reduced. This is why a proper WWR limit is still in debate,
for example, in 2013, ASHRAE proposed a change to standard 189.1, “Standard for the Design
of High-Performance, Green Buildings except Low-rose Residential Buildings”. The proposed
WWR is 30% which is 10% lower than the original 40% for buildings less than 25,000 square
feet. This change was voted down in 2014 (TGP 2014) since it may greatly influence the
accessibility of natural daylight, which is directly relevant to occupants’ health and productivity.
In conclusion, WWR is a fundamental factor in façade features which should be taken into
consideration seriously with respect to proper building design, orientation, high performance
glazing, daylight control, etc.
42

2.4.6 Volume-to-façade Area Ratio
Normally the compactness of a building is presented as surface area to volume (A/V) ratio,
which in this research is replaced by Volume-to-Area ratio instead for consistency. This indicator
is significantly important to the building heating load and potentially energy use for meeting the
load. A compact building form has more advantage to reduce heating energy demand while
limiting the possibility to design an innovative and complicated geometries, because larger
surface area results in more heat gain and loss through the facade. A preferred compactness ratio
should be less than 0.7 square meter to cubic meter (Mcleod, Mead, and Standen 2014). A study
(Behsh 2002) proved that a compact building (low A/V ratio) is less easily affected by outside
weather change than a high A/V ratio building in hot and dry area. On the contrary, normally it is
more effective to use a high A/V ratio building in an area where passive heating or cooling
should be made use of as primary design strategy. In conclusion, Volume-to-area ration should
be well designed with consideration of other factors, solar radiation and direction of wind on the
building for example (ARchitecture 2014).
2.5 Regression Model for Energy Performance Calculation
2.5.1 Regression Analysis for Building Energy Performance
Generally, regression analysis is a statistical method to determine the correlation between
multiple variables. The former term is called the target, response, or dependent variable (Samprit
and Jeffrey S 2013), which is usually presented by “ ”, in addition, other variables could be used
to predict the target, which are called the predictor or independent variables and usually
presented by

,

, etc. The 3 major function of regression analysis are: defining the
relationship between response and predictors, predicting the response by using the model and
testing of hypothesis.
In this research, the response is energy use intensity (EUI) and the predictors are basic façade
features with other factors that take local climate and building codes into account,
Heating/Cooling Degree Day and built year for example. The main purpose of using regression
analysis is to find out the relationship between façade features and EUI locally and nationally,
then the developed model could be used to predict other buildings’ EUI within the same
limitation, depending on EUI type, location, and time period.
43

For estimating building energy consumption, many researchers attempted to use regression
method by considering different predictors. Edward et al. (H. Borgstein and Lamberts 2014) used
a simple linear regression model to analyze 1890 bank branches in 57 different climate zones and
to develop a benchmarking method. The linear regression was carried out by using only one
predictor of CDH which is the number of cooling degree hours of the location. The result
suggested climate has to be incorporated as a correction factor to validate the results.
Another regression model (Chung, Hui, and Lam 2006) was conducted to estimate EUI for
supermarket with central air-conditioning in Hong Kong, and the predictors include 4 categories:
age (building age), occupancy (internal floor area, operational schedule, number of
customers/year), people (occupants’ behavior and maintenance factor, indoor temperature set-
point in summer) and energy system (chiller type of equipment, lighting equipment and control).
The most important factors were selected by backward elimination. The multiple regression is
used to develop a benchmarking process in policy analysis.
Somayeh et al (Asadi, Shams, and Mottahedi 2014) presented a new model to predict and
quantify energy use for commercial buildings especially at the predesign stage. Multiple building
configurations were simulated by using simulation software and Monte Carlo techniques. Totally
10,000 datasets were used to develop a set of regression equations and the maximum error was
less than 5% compared with the simulation results. In this practice, most variables are about
thermal properties of construction materials (with different occupant schedules), which are hard
to obtained from real project and iterated for analysis without using simulation tool.
2.5.2 Multiple Linear Regression and Stepwise Regression
Rather than only using one independent variable as predictor in regression, multiple linear
regression (MLR) has multiple independent variables. The same purpose as simple linear
regression is to develop the relationship between response and predictors and predict the new
response with a new set of predictors at an acceptable confidence level.
The multiple linear regression is presented as the following form:

(1)
Where
44

is the constant while

are the regression coefficients,

are the significant
predictors and is the random error.
The difference between multiple linear regression (MLR) and several simple linear regressions
(Rudolf Jakob and William J 2006) is that in MLR each coefficient indicates the average change
in the response associated with changes in that independent variable while others remain fixed.
This advantage could help to explain what happens when a single variable varies without
changing other predictor variables. In this research, multiple indicators of façade features could
be analyzed and used to develop a multiple linear regression model to predict the response: EUI.
Stepwise regression refers to the process of developing a regression model by adding or
removing variables and recomputing the coefficient (Castree, Kitchin, and Rogers 2013). It could
be used to deal with a large amount of datasets and determine the most significant predictors
after dropping out less important variables, a step at a time based on the statistical significance.
In the case of energy consumption estimation, it is of importance to filter out the least significant
predictors as a practical way to gather information and predict EUI. The best combination of
certain indicators are not only meaningful to collect less amount of data for regression but also
indicate the most important parameters in a certain region, hot and dry area for example. A
research (Filippín, Ricard, and Flores Larsen 2013) selected representative variables by using
stepwise regression for annual energy consumption in the central region of Argentina, which are
volume, availability of the solar resource estimated on the envelope’s surface, volumetric heat
loss coefficient and weighted mean internal surface temperature. The results could be used to
reach a more suitable certification of real energy consumption which is helpful to provide a more
practical criteria of building code.
2.5.3 Other Methods for Estimating EUI
Other than dynamic simulation which is stated in Chapter 1, other mathematical approaches are
adopted to estimate EUI. Rajesh et al. (Kumar, Aggarwal, and Sharma 2013) used Artificial
Neural Network (ANN) to estimate total energy use for heating and carbon emissions. Since
ANN has advantages when making better, quicker and more practical predictions, it is suggested
by the author to use instead of traditional regression method and simulation tool. The results
presented the total load for a six stories building by using ANN method which collected data
45

representing the past history and performance of the real system. Betul et al. (Ekici and Aksoy
2009) selected orientation, insulation and transparency ratio for estimating energy needs by using
ANN, which is tested that this ANN model could provide satisfactory results with deviation of
3.43% and successful prediction rate is 94.8% to 98.5%. An alternative method, hybrid genetic
algorithm-adaptive network-based fuzzy inference system (GA-ANFIS) is used by Kangji et al.
(Li, Su, and Chu 2011) to predict building energy and the results showed it’s more accurate than
the ANN method. GA-ANFIS is a systematic method (Karatasou, Santamouris, and Geros 2006)
based on least squares estimation and statistical tests. Another adaptive ANN model (Yang,
Rivard, and Zmeureanu 2005) is developed to predict dynamic real-time building energy
consumption when considering the unexpected incoming data change. The adaptive ANN model
could be capable of revising itself to update weather and other condition changes.
Decision tree method is another approach to predict building energy use in practice. This
predictive method is featured by the interpretable flowchart-like tree structures which could be
easily used to obtain useful predictors. Zhun et al. (Yu et al. 2010) demonstrated that a decision
tree method can predict building energy demand by 93% accuracy for training data and 92%
accuracy for test data. Compared with other 2 mathematical methods which require complicated
statistical analysis knowledge, decision tree has advantages could be simply utilized to predict
energy, provide significant predictors and show the threshold values which is of importance to
result in high building energy performance.
Case-based reasoning (CBR) was used by Danielle et al. (Monfet et al. 2014) to forecast
building energy demand and the model was validated by real monitored data. CBR is one of the
machine-learning artificial intelligence techniques, which in the same research is compared with
ANN method. The advantages of using CBR include easily updating feature, simple
understanding of reasoning, ability to deal with missing information and large amounts of
predictors. The results showed several indicators are significant for energy prediction: during
occupancy time period, the coefficient of variance of the root-mean-square-error (CV-RMSE),
the normalized mean bias error (NMBE) the root-mean-square-error (RMSE) are below 13.2%,
5.8% and 14 kW.
46

Chapter 3 Methodology and Plan of Approach
3.1 Introduction of Methodology
The purpose of this research is to develop multiple regression models which could be used to
determine the significant façade features relevant to energy aspect as well as to predict energy
performance by entering a minimum number of façade data. Instead of using details of building
information, like construction thermal properties, mechanical system, operation schedule, etc.,
which is used as basic input for simulation and other estimation methods, multiple linear
regression is adopted in this research with easily readable façade features, which include building
height, orientation, volume, floor area, façade area, site area, window-to-wall ratio, volume-to-
façade area ratio, etc.

Figure 3.1 Methodology diagram
There are mainly three parts of the methodology: data collection (DC), data processing (DP), and
model development (MD) as represented in the figure above.
 For data collection, generally 2 types of data should be collected. One is real energy use
data, another is corresponding façade feature. Energy use data which in this research is
47

presented by Energy Use Intensity (EUI) as target metric are from many resources,
including benchmarking and disclosure data by local government, direct energy bill from
building users as well as monitored data for case study and validation. On the other hand,
façade features are collected by using different methods which contain manual estimation
(visual reading and physical model rebuilding), existing building model (SketchUp etc.)
and direct information collection from design drawing or specification. Other potential
factors like built year and HDD/CDD could also be easily obtained from open resource.
 The data processing section serves as data preparation for the following model
development. For annual EUI model development, this step could be skipped since yearly
EUI data is the basic data provided by different building energy resources, the same
processing for available monthly data. However, monthly data are not available in most
cases, in which simulation tool could be adopted to estimate monthly energy performance
first. As a result, monthly data could be estimated accurately after calibrating the
simulation model by real energy bills or annual EUI data.
 Finally, multiple linear regression is used to develop the EUI estimation model package
based on collected façade information and EUI data (monthly and annually). In this
section, the significance of each parameter and correlation between predictors and
response could also be analyzed with the consideration of local code requirements, design
strategies and best practices. Other regression methods would be used for comparison,
which include stepwise regression, partial least square (PLS), etc. In the end, all
regression models should be validated by appropriate method, for example, cross
validation for PLS.
The predicted outcome of this research is a new EUI estimation package, which could provide
the following results for different purpose at different scale:
1. National-wide EUI model
2. City-based EUI model
3. Annual EUI model
4. Monthly EUI model
5. Cooling EUI model
6. Heating EUI model
48

In addition to assisting with EUI benchmarking for improving building energy efficiency, the
research potential outcomes could be applied for new construction to provide a more accurate
baseline and energy reduction target at the predesign stage as well as to evaluate basic façade
design decisions, while for existing buildings, it can help to estimate EUI when there is no
detailed building information available for deep simulation and get a reasonably correct energy
consumption rate by inputting a minimum amount of data.
3.2 Energy Use Data Collection
3.2.1 Existing Data
With the help of the internet and other researchers, 4 office buildings with reported energy data
in downtown Los Angeles plus another 5 buildings in Los Angeles with monthly energy bills are
available for developing an EUI estimation model. In addition, another 188 nation-wide
buildings with annual EUI data are collected from the internet and other researcher’s collection
to be used for annual nation-wide EUI estimation model development.
3.2.2 Benchmarking Data
Among all the cities which committed to disclose building benchmarking data after 2009, only
the following 4 cities have already published the energy data which could be directly used for
analysis. They are New York City (2011 and 2012), Washington, D.C. (2011 and 2012),
Minneapolis (2012) and Boston (2012). Other cities are still in the process of collecting
benchmarking data or unable to disclose energy data from private buildings due to legislative
limitations.
49

Figure 3.2 Collected existing benchmarking building energy data
3.2.3 Monitoring Data
A 16 floor office building in the financial district, downtown Los Angeles, is used for a case
study and a verification sample. With more than 40 sensors installed on the electricity panel,
energy consumption for each system could be measure accurately. These break-down data could
be used to validate the exact relationship between studied predictors with energy consumption
from directly relevant systems, like heating, cooling, ventilation, lighting, etc. In addition,
occupancy satisfaction level could be added with a Post Occupancy Evaluation (POE) study to
take the specific façade features that could affect occupants’ thermal or visual comfort into
consideration as part of health and productivity benefits relevant to façade design.
3.3 Façade Parameter Determination and Variables Collection
3.3.1 Façade Feature Determination and Definition
In this research, all considered building façade features (mainly geometric attributes) could be
easily read without detailed information. The original 17 assumed predictors (including weather
condition, surrounding context and built year) are showed in the following figures and table.
50

Figure 3.3 Facade features for multiple linear regression
A basic assumption of this research is that EUI could be estimated based on only simple façade
features as well as a few other factors, like HDD/CDD which represents dynamic local weather
condition. Another factor, built year, is used to incorporate all the requirements by code in each
time period as showed in the figure above. It assumed that after the first national/local building
energy code was established, a building had to meet the requirements of corresponding codes or
standards, including system efficiency, ventilation rate requirements, etc. The built year is easy
to obtain instead of considering complicated building details. Figure 3.4 shows the energy use
index requirements change created by Mark Halverson, PNNL (Hunn 2010).

Figure 3.4 Weather features and built year created by Mark Halverson, PNNL
Table 3.1 Predictors definition and explanation
No. Façade feature Definition Unit Category
1 Height From open air pedestrian entrance to highest occupied floor
1
Feet Basic
2 Floors Total occupied stories or levels
2
- Basic
3 Orientation Positing of a building with respect to the North
3
- Basic
4 Operable window Window could be open or close based ventilation need
4
- Basic
5 Volume Inner space volume enclosed by external envelope CF Basic
6 WWR Window-to-wall ratio (total window area/total exterior wall area) - Basic
7 Window Area Total glazing area SF Area
8 Façade Area Total area of all parts of the structure’s façade SF Area
9 Site Area Total site area within fixed boundaries SF Area
10 Floor Area Total floor area inside the building envelope SF Area
51

11 V/FA (Compactness) Ratio of volume to façade area - Ratio
12 V/SA Ratio of volume to site area - Ratio
13 FA/SA Ratio of façade area to site area - Ratio
14 HDD Heating degree day (the demand for energy to heat a building) DegreeDays Weather
15 CDD Cooling degree day (the demand for energy to cool a building) DegreeDays Weather
16 Adjacent Building If adjacent building exists to cast shading on objective building
5
- Additional
17 Built Year Year of construction complete year Additional
Note: 1. Height is measure from the level of the lowest, significant, open-air, pedestrian entrance to the finished floor level of the
highest occupied floor within the building (Council on Tall Buildings and Urban Habitat).
2. Floors refer to the total levels of a building which could be used by occupants.
3. Long axis along with North-South is quantified as 1, NE-SW is 2, E-W is 3, SE-NW is 4.
4. With operable window is quantified as 1, without operable window is quantified as 0.
5. No adjacent building is quantified as 0, while adjacent building on the north side is 1, others are clockwise defined by 2 to 8.

Figure 3.5 Orientation and adjacent building position

3.3.2 Data Reading by Using 3D Model Tool
After determining which façade features should be collected for further analysis, multiple tools
are used to read the data through different ways. One method is to make use of SketchUp model
from 3D Warehouse (Trimble 2014) which is an open model resources with real geographic
information. This can be used especially in big cities like New York City where there are a
considerable amount of building models available. The following figure of SketchUp model is
100 Wall Street (ilesoft82 2014).
In SketchUp, each surface could be selected and the corresponding area could be calculated
directly. Window area could be drawn by using the Shape tool which also can define the surface.
All dimensions could be measured by using the Tape Measure tool for height and volume
calculation.
52

Figure 3.6 100 Wall Street SU model
When the existing SketchUp model is not available, other tools would be used to read façade
features, like AutoCAD, Google Earth, etc. In this case the scale tool is significant to duplicate
the original building dimensions.
In some cases, even there is no Google Earth 3D model to be used for data reading. Certain
features could be estimated manually, especially for window area and window-to-wall ratio. To
validate the result, another person’s perception should be added to evaluate the accuracy of the
first person estimation.
3.4 Regression Method and Tool
3.4.1 Regression Tool Introduction
To develop a regression model, many tools could be considered for analysis, like SPSS Statistics
(IBM 2014), MATLAB (MAthWorks 2014), etc. In this research, another statistical analysis tool,
Minitab
®
17 (Minitab 2014b) is adopted for data analysis and regression model development. In
1972, Minitab was developed by Barbara F. Ryan, Thomas A. Ryan., and Brian L. Joiner at the
Pennsylvania State University (NIST 2012). By using Minitab, a large amount of data can be
processed (Minitab 2013a) for basic statistical analysis, regression and correlation analysis,
53

hypothesis tests, model validation, prediction, and graphs making, etc. All façade features, other
factors and EUI data can be input as basic training samples. The correlation between each factor
and EUI could be analyzed by calculating Pearson’s correlation coefficient. Then different
regression models could be used to determine the most accurate model which is sufficient to
predict response values for new observations.
3.4.2 Multiple Linear Regression, Stepwise Regression, Partial Least Square (PLS)
Three regression models are examined to develop the EUI estimation model. Multiple Linear
Regression (MLR) provides the linear relationship between one response with more than one
predictors. The equation form is presented in Chapter 2.
When there are a large number of predictors to be used in regression, stepwise regression should
be used for removing the least significant predictor at each step. The order of removed predictors
also indicate the significance which could be analyzed to determine which façade feature is the
most important in a certain area. This is also called backward elimination (Support Minitab
2014). This automatic process is useful to identify the most significant predictors.
Partial least squares (PLS) regression is a process to limit the original predictors into a smaller
set which is not correlated to each other. PLS regression is extremely useful when collinearity of
original predictors is high, or when the predictor’s number are more than observations. Minitab
uses the nonlinear iterative partial least squares (NIPALS) algorithm (Rosipal) developed by
Herman Wold (Roush 1982). By using this technique, Minitab uses least-squares regression on
the left uncorrelated predictors. In the end, cross-validation is adopted to enhance the model’s
accuracy and predictive ability (Minitab 2013b).
3.4.3 Indicators Explanation
To analyze the results of regression models, multiple indicators could be calculated to evaluate
the characteristics of the corresponding models. The main indicators are listed in the following
table.
Table 3.2 Regression model indicators
No. Indicator Explanation Range/Acceptable
1 Pearson Correlation Whether 2 continuous variables are linearly related (-1,1)/closer to 1
2 P-value The probability of obtaining a test statistic (0,1)/closer to 0
3 VIF Multicollinearity (correlation between predictors) NA/Less than 10
54

4 R
2
The percentage of response variable variation that can be explained (0,100%)/closer to 100%
5 R
2
(adj) R
2
adjusted for the number of predictors in the model (0,100%)/closer to 100%
6 R
2
(pred) Models predictive ability (0,100%)/closer to 100%
7 Durbin-Watson statistic whether or not the correlation between adjacent error terms is zero (1,3)/closer to 2
8 Error rate discrepancy between the estimated values NA/closer to 0

3.4.4 Cross-validation
Partial Least Square (PLS) adopts cross-validation to validate the results, which could calculate
the predictive ability of potential models to determine the optimal number of components that
should be kept in the model (Minitab 2014b). It’s significantly useful when the optimal number
of components is unknown. For each potential model to be validated, the process excludes one
observation at a time and repeats the following procedures:
 Omit one observation
 Recalculate the model without the observation.
 Predict the cross-validated fitted value, for the omitted observation using the recalculated
model and calculates the cross-validated residual value.
 Repeat steps 1 - 3 until all observations are omitted and fit.
 Calculate the prediction sum of squares (PRESS) and predicted R
2
values.
After doing steps 1 - 5 for each model, the optimal model is selected with the number of
components that produces the highest predicted R
2
and lowest PRESS.

Figure 3.7 Cross-validation
55

Chapter 4 Regression Models and Results
4.1 Nationwide Annual EUI Model
4.1.1 Data Sources
In this section, energy use and façade features were collected from 167 nationwide buildings.
The raw data was collected by Andrea Martinez’s unpublished research for her PhD dissertation.
The data resources include U.S. Green Building Council (USGBC) registered projects,
EnergyStar rated buildings, private buildings with disclosed energy record and publicly available
technical reports. Among 167 original datasets, 11 buildings are LEED certified, 26 buildings are
LEED Silver rated, 47 buildings are LEED Gold rated and 37 buildings are LEED Platinum rated.
As a result, 122 buildings are at least LEED certified green buildings, which account for 72.62%
of total datasets. Most cases are mid-rise commercial buildings with a minimum of 3 floors.

Figure 4.1 Building science-based climate map (U.S. DOE 2013)
Original data sources incorporated three more predictors, including renovation, climate zone and
LEED certification level. These three additional predictors were not used in city-based
regression model. Renovation refers to whether or not there was a recorded building retrofit in
recent years. This information was collected from relevant documents and official reports. The
building science-based climate map (U.S. DOE 2013) was based on the IECC (International
Energy conservation Code) climate zone map and was used to divide all buildings into 7
different climate zones, which is illustrated in Figure. 4.1 (U.S. DOE 2013). LEED certification
56

level was also recorded in the original datasets, including uncertified, Certified, Silver, Gold and
Platinum. The higher rated buildings indicated the more efficient buildings on energy
consumption and the supposedly healthier to live in. This level also related directly to building
energy use intensity as a significant factor.
Table 4.1 Total number of buildings in each group
Total # of
Buildings
Built
Year
Renovation WWR Operable
Window
V/FA
Ratio
Axis
Orientation
HDD CDD Climate
Zone
Before 1975 64 - - - - - - - -
After 1975 103 - - - - - - - -
With - 64 - 69 - - - - -
Without - 103 - 98 - - - - -
Below 40% - - 54 - - - - - -
Over 40% - - 113 - - - - - -
Below 40 - - - - 104 - - - -
Over 40 - - - - 63 - - - -
N-S - - - - - 45 - - -
NE-SW - - - - - 37 - - -
W-E - - - - - 48 - - -
NW-SE - - - - - 37 - - -
Below 1000 - - - - - - 22 34 -
1000-1999 - - - - - - 28 63 -
2000-2999 - - - - - - 23 44 -
3000-3999 - - - - - - 41 13 -
4000-4999 - - - - - - 37 10 -
Over 5000 - - - - - - 16 3 -
CZ1 - - - - - - - - 1
CZ2 - - - - - - - - 9
CZ3 - - - - - - - - 41
CZ4 - - - - - - - - 58
CZ5 - - - - - - - - 43
CZ6 - - - - - - - - 12
CZ7 - - - - - - - - 3
Raw datasets were firstly analyzed by dividing into different groups, including built year,
renovation, window-to-wall ratio range, operable window, V/FA ratio range, axis orientation,
HDD and CDD range, as well as climate zone. This process was designed to illustrate if there is
simple correlation between individual façade features with building site energy use before
regression analysis. However, the results in this section will not draw any conclusion about the
linear regression relation of multiple variables and site EUI. Figure 4.2 to 4.10 show the general
correlation and the mean value in each group. Table 4.1 lists the total number of buildings in
different group for analysis.
Figure 4.2 indicates the correlation between site EUI and construction year, which divided all
buildings into 2 groups: before and after 1975, since the first standard ASHRAE 90-75 was
published in 1975 as discussed in Chapter 2. It was the guideline of new building energy
conservation for all buildings (Hunn 2010). Office buildings that were built before 1975 have a
57

higher mean EUI of 66.96 kBtu/sf (64 datasets) compared to 59.22 kBtu/sf after 1975 (103
datasets). The difference is 11.56%, which indicates that construction year (before or after 1975)
is not significantly related to annual EUI at an α-level of 0.05. In fact, with more strict
requirements on building performance from improved energy code, buildings are expected to
consume lower energy.
In addition, figure 4.3 also presents that the mean EUI for buildings with and without renovation
are very close. “0” stands for buildings without renovation (103 datasets) while “1” means
buildings with renovation (64 datasets). Even the fact that buildings after renovation will
consume less energy is expected, in this limited nationwide database 72.62% of buildings are
already LEED certified which means the renovation of these buildings are less needed. Then the
predictor of renovation is not significantly related to annual energy use of nationwide buildings.

Figure 4.2 Site EUI and construction year

Figure 4.3 Site EUI and renovation
Figure 4.4 represents the relation between annual site EUI and current WWR limit for
nonresidential buildings. The overall 40% of window-to-wall ratio (WWR) was used to divide all
datasets into 2 groups. The mean value of buildings with over 40% WWR (113 datasets) was
65.19 kBtu/sf compared to 55.74 kBtu/sf of lower WWR buildings (54 datasets). The difference
Before 1975 After 1975
75
70
65
60
55
50
Construction Year
Site EUI
Interval Plot of Site EUI vs Construction Year
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
1 0
75
70
65
60
55
50
Renovation
Site EUI
Interval Plot of Site EUI vs Renovation
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
58

is 15% and the P value of WWR is 0.099, which means WWR is not significantly related to EUI
at the α-level of 0.05 but close to 0.1, and in reality WWR is an important factor to influence
office building energy use in terms of heating and cooling load by solar heat gain.
Figure 4.5 presents that buildings with the operable windows consumed less energy since the
mixed mode of natural ventilation and mechanical ventilation is more energy efficient. The mean
EUI is 56.33 kBtu/sf for buildings with operable window (69 datasets) and it is 66.35 kBtu/sf for
buildings without operable window (98 datasets). The difference is 15.1% and the P value of
operable window is 0.064% which is slightly higher than the significant α-level of 0.05. The
option of operable windows is a very significant factor to influence building energy use,
especially for reducing the cooling demand in hot and dry area.

Figure 4.4 Site EUI and WWR

Figure 4.5 Site EUI and operable window
V/FA ratio stands for the compactness which has significant impact on heating load. Figure 4.6
illustrates that buildings with V/FA less than 40 (104 datasets) had the lower mean EUI of 62.96
kBtu/sf while buildings with V/FA higher than 40 (63 datasets) had the higher mean EUI of
71.01 kBtu/sf. The difference is 11.33%. Normally compact buildings are consuming less energy
than greater façade area buildings, but the result shows that it also depends on other factors, like
Very High (Over 40%) Prescriptive Limit (Below 40%)
75
70
65
60
55
50
45
WWR
Site EUI
Interval Plot of Site EUI vs WWR
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
1 0
75
70
65
60
55
50
Operable Window
Site EUI
Interval Plot of Site EUI vs Operable Window
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
59

glazing and exterior wall thermal properties, so V/FA ration is not an individual predictor to
influence building energy use in this database.

Figure 4.6 Site EUI and V/FA ratio

Figure 4.7 Site EUI and orientation
Figure 4.7 shows that buildings in N-S (45 datasets) and NW-SE (37 datasets) axis orientations
consumed less energy than the W-E (48 datasets) and NW-SE (37 datasets) axis orientations.
This trend is different from common idea of the impact of building orientation on energy use.
Generally, building with higher façade area on the west and east side have more heat gain and
cooling demand, but the raw data showed there are many buildings with a regular square shape
which resulted in the less sensitivity of axis orientation on energy use. In addition, there are
many other factor are influencing building energy use too, like adjacency and local shades also
have great impact on the heat gain through façade which should be taken into consideration
jointly.
Climate zones in U.S. are defined by the range of heating degree days (HDD) and cooling degree
days (CDD). HDD and CDD stand for the heating and cooling demand in terms of how much the
difference between a base temperature and daily average temperature is. When no façade retrofit
was implemented, the only changing predictors in the regression model are HDD and CDD
Over 40 Below 40
85
80
75
70
65
60
55
V/FA Ratio
Site EUI
Interval Plot of Site EUI vs V/FA Ratio
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
4 3 2 1
90
80
70
60
50
40
Orientation
Site EUI
Interval Plot of Site EUI vs Orientation
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
60

through years which stand for the change in weather conditions. Figure 4.8 and 4.9 illustrate the
correlation between the site energy use and annual HDD and CDD in each building. Buildings
were divided into different groups based on 6 ranges of HDD/CDD, including below 1000, 1000
to 1999, 2000 to 2999, 3000 to 3999, 4000 to 4999, and over 5000. Generally, the mean EUI
increased when the annual HDD increased except the group of 2000 to 2999. With the increase
of yearly heating demand buildings consumed more energy since the more need of fuel for
heating. Compared to HDD, CDD is less sensitive to influence building energy use since most
confidence intervals are overlapped. Generally, the mean EUI of CDD decreases when the CDD
increases, while only buildings with CDD lower than 1000 have a mean EUI slightly higher than
4000 to 4999 group. Considering most confidence intervals are overlapped, CDD can’t be
considered as an individual factor influencing building energy use. It is because for modern
commercial buildings cooling energy use is more constantly required even in winter due to good
performance of air tightness and higher thermal comfort demand, which made the relation more
complicated.

Figure 4.8 Site EUI and HDD

Figure 4.9 Site EUI and CDD
5000+ 4000-4999 3000-3999 2000-2999 1000-1999 1000-
110
100
90
80
70
60
50
40
30
20
HDD
Site EUI
Interval Plot of Site EUI vs HDD
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
5000+ 4000-4999 3000-3999 2000-2999 1000-1999 1000-
100
80
60
40
20
0
CDD
Site EUI
Interval Plot of Site EUI vs CDD
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
61

Furthermore, climate zone is defined by the integrated consideration of HDD and CDD range.
The climate zone division is very significantly relevant to building energy use, which is proved
by the short confidence interval and less EUI interval overlap of each zone which are showed in
Figure 4.10. The result shows that buildings in warm-humid and hot zones consumed less energy
than buildings in dry and cold areas. In the original datasets, the only building of case 1 (LEED
Silver certification) in Honolulu, Climate Zone 1 consumed 27.25 kBtu/sf while only 3 cases in
Climate 7 consumed 200.67 kBtu/sf, 98.01 kBtu/sf and 80.57 kBtu/sf and only the last building
is LEED Gold certified. The limited number of buildings in these two zones resulted in the larger
interval range to cover more possible predictive observations within the groups.

Figure 4.10 Site EUI and climate zone
4.1.2 MLR, Stepwise and PLS
In this research, site EUI are predicted by the façade features by 3 regression methods: Multiple
Linear Regression (MLR), Stepwise Regression and Partial Least Square PLS). The results are
showed in Table 4.2 and 4.3. In MLR, all 22 predictors were included in the every model. Only
east façade area was automatically removed for the less significance. The R
2
value indicates that
all predictors could only explain 58.49% of the variance in EUI while the adjusted R
2
value
means only 49.63% of EUI variable variation that is explained by its relationship with predictor
variables, adjusted for the number of predictors in the model. D-W statistic is 2.21, which means
there is no significant autocorrelation since it’s close to 2. Only operable window, HDD, total
façade area, north and west façade area, climate zone and LEED certification are significantly
related to annual EUI at an α-level of 0.05 since P-values are close to 0.
By comparison, R
2
from stepwise regression means 53.73% of the variance in EUI and it is
slightly lower than MLR. The adjusted R
2
has been improved when compared to MLR. The
7 6 5 4 3 2 1
200
150
100
50
0
-50
Climate Code
Site EUI
Interval Plot of Site EUI vs Climate Code
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
62

predicted R
2
value has been significantly improved to 48.3% which indicates the model might be.
D-W statistic is 2.04 which is also closer to 2. All P-values of corresponding predictors are less
than 0.05 while VIFs are lower than 12. It indicates the regression coefficients are well estimated
without severe multicollinearity. By adding the most significant variable or removing the least
significant variable during each step, the advantage of using stepwise regression is that it can
leave the most important predictors or façade features in the final model, and these predictors are
most significantly related to energy use rather than others. As a result, predictors including floor,
operable window, HDD, CDD, climate zone as well as LEED certification are the most
important factors which have greater impact on energy use for buildings throughout the country.
Table 4.2 Coefficients and main indicators
Indicator Multiple Linear Regression Stepwise Regression Partial Least Squares
R
2
/ R
2
(Adj)/ R
2
(pre) 58.49% 49.63% 33.49% 53.73% 51.39% 48.30% 46.42% - 37.34%
D-W 2.21 2.04 -
Predictor Coef P VIF Coef P VIF Coef P X-Variance
Constant 134 0.20 - 86.60 0.00 - 39.15
0.00 59.72%
Height -0.014 0.95 593.93 - - - -0.0019
Floors -0.36 0.91 649.44 -0.50 0.001 1.28 -0.025
Built year -0.022 0.68 2.39 - - - -0.0027
WWR -0.001 1.00 3.22 - - - 0.015
Orientation 0.15 0.90 1.34 - - - 1.25
Operable Window -21.19 0.00 1.78 -21.69 0.00 1.46 -9.07
Volume -0.00000 0.81 1063.04 - - - 0.00
Window Area 0.000046 0.54 21.33 - - - 0.00
Façade Area 0.00062 0.03 750.28 - - - -0.00
Site Area -0.00006 0.54 9.32 - - - 0.00
Floor Area 0.000007 0.95 1066.73 - - - 0.00
V/FA 0.023 0.83 3.55 - - - 0.077
V/SA 0.043 0.32 18.25 - - - -0.028
FA/SA -2.5 0.11 14.30 - - - -0.34
HDD 0.0092 0.00 13.50 0.0086 0.00 10.8 0.0017
CDD -0.0019 0.16 1.97 -0.0022 0.074 1.72 -0.0002
N Façade Area -0.0011 0.02 137.46 - - - 0.00
S Façade Area -0.00000 0.99 46.53 - - - 0.00
W Façade Area -0.00121 0.02 199.17 - - - -0.00
E Façade Area - - - - - - -0.00
Renovation 0.06 0.99 2.35 - - - -1.20
Climate Zone -9.48 0.02 14.34 -8.56 0.015 11.41 2.37
LEED Certification -3.91 0.00 1.31 -4.07 0.00 1.16 -2.16
Table 4.3 PLS coefficients and main indicators (continued)
Predictors Coef P X-Variance
Height*Floor -0.00
0.00 59.72%
Height*Built Year -0.00
Height*WWR (%) 0.00
Height*Orientation 0.0014
Height*Operable Window -0.074
Height*Volume -0.00
Height*Window Area 0.00
Height*Facade Area -0.00
Height*Site Area -0.00
Height*Floor Area -0.00
Height*V/FA 0.00
63

Height*V/SA -0.00
Height*FA/SA -0.0003
Height*HDD 0.00
Height*CDD -0.00
Height*N -0.00
Height*S -0.00
Height*W -0.00
Height*E -0.00
Height*Renovation -0.0007
Height*Climate Zone 0.0008
Height*LEED Certification -0.0054
PLS is another option of regression analysis. The optimal model is defined as the model with the
highest predicted R
2
in PLS regression. The predicted R
2
is 37.34% with 2 components in the
selected optimal model, which is lower than MLR and stepwise. The P value is 0.00, which is
less than an alpha of 0.05, providing sufficient evidence that the 2-component model is
significant. 2-component model is selected by cross-validation, which has the R
2
value of
46.42%. The X-variance indicates the amount of variance in the predictors that is explained by
the model. The selected model can explain 59.72% of the variance in the predictors.

Figure 4.11 Standard residual and frequency plot

Figure 4.12 Normal probability plot
64

Figure 4.13 Standard residual and fitted value
Figure 4.11 is the histogram of standardized residual and frequency and among 126 observations
3 outliers may exist in the data. Normal probability plot (Figure 4.12) shows an approximately
linear pattern consistent with a normal distribution. The plot of residuals versus the fitted values
(Figure 4.13) shows that the variance of the residuals are constant with a mean of zero.

Figure 4.14 PLS model selection plot

Figure 4.15 PLS response plot
65

Figure 4.16 PLS standard coefficient plot
PLS result plots are also represented by the figures above:
 Figure 4.14 shows that through model selection and validation, optimal model has 2
components. After 2 components, R
2
decrease significantly from cross-validation.
 Figure 4.15 indicates that the model fits the data poorly because the points are not in a
good linear pattern (from the bottom left-hand corner to the top right-hand corner).
 Interpretation of the magnitude and sign of the coefficients (Figure 4.16) shows that
predictor 15 (HDD), 22 (Climate zone) and 12 (V/FA) have the greatest positive impact
on EUI, while predictor 6 (Operable window) and 23 (LEED Certification) have the least
impact on EUI increasing. The results indicate the more heating demand, the colder wand
the higher Volume-to-façade ratio (compactness), the more energy were consumed by
buildings nationwide. On the contrary, buildings with operable windows and higher
LEED Certification levels consumed less energy.
4.1.3 Validation
The main indicators of the regression model showed a lower accuracy of the model’s predictive
ability. To validate the stepwise regression model, the original datasets were divided into two
samples for training and testing. In this section, 10% of original datasets were randomly selected
as validation samples and other datasets were used to develop the regression model. The R
2
of
the training samples model is 47.84% while the D-W statistic is 2.16. The greatest error rate is
60.3% of case 6 while the lowest error rate is 0.0% of case 7. The large range of error also means
an unstable ability to predict new observations.
66

Figure 4.17 Validation and error rate
For PLS, cross-validation is used to validate the optimal model. The result shows the optimal
number of components is 2 when the highest predictive R
2
is 37.34% and the lowest Predictive
Residual Sum of Square (PRESS) is 26529.9 which indicates the sum of the squares of all the
resulting prediction errors. However, PLS still showed less accuracy and predictive ability than
other two methods.
4.1.4 Results Comparison
The error rates for each case calculated by the 3 regression models are compared in Figure 4. 18.
The average error rates for 3 models are 22.28%, 23.26%, 25.34%, however, all of 3 models
show unstable results when EUI is lower than 40. This is not only because 126 original datasets
are not efficient enough to predict all buildings in the whole country, but also energy-efficient
buildings with low EUI can hardly be estimated only by façade features and a few other factors.
To estimate more accurately for nationwide buildings by using only one model, more datasets
have to be collected and more factors have to be taken into consideration. From Figure 4.19, all
estimated EUI from 3 regression models are compared to the reported original EUI value, and
the results show the tabulated estimation by using all 3 regression models, especially when EUI
is lower than 40 kBtu/sf.
67

Based on the result comparison and indicators interpretation, stepwise regression model is
selected as the final EUI estimation model because its relatively high explanation rate of variable
variance as well as better predictive ability. In addition, stepwise regression can also determine
the most significant predictors which are relevant to building energy use across the country.

Figure 4.18 Error rate comparison

Figure 4.19 Regression results comparison

68

4.2 City-based Annual EUI Model – New York City
4.2.1 Basic Data Analysis
Basic building information and energy use data are collected from New York City benchmarking
and disclosure database, which includes building address, reported building square footage and
2011 and 2002 two years’ site Energy Use Intensity (EUI) data that were recorded by using
TargetFinder tool. To limit the scope from the original raw database which has over 14000
buildings in New York City, only office buildings in Manhattan are sorted out as the first
selection. Then the geometry model of each selected building is looked up in 3D Warehouse in
order to read all of the façade dimensions. As the result, there are in total 45 cases (with either
2011 or 2012 EUI data) for the next regression analysis.
All datasets with façade features are firstly analyzed by dividing into different groups. The
results represent the correlation between reported site EUI with each predictor through interval
plotting. The confidence interval is 95% which indicates 95% probability from the future
experiment within this interval. Table 4.4 lists the total number of buildings in each group.
Table 4.4 Total number of buildings in each group
Total # of
Buildings
Built Year Building
Height
WWR Operable
Window
V/FA Ratio Axis
Orientation
Floor Area HDD
Before 1980 31 - - - - - - -
After 1980 14 - - - - - - -
Megatall - 18 - - - - - -
Supertall - 22 - - - - - -
Tall - 5 - - - - - -
Below 40% - - 17 - - - - -
Over 40% - - 28 - - - - -
With - - - 12 - - - -
Without - - - 33 - - - -
Below 40 - - - - 20 - - -
Over 40 - - - - 25 - - -
N-S - - - - - 4 - -
NE-SW - - - - - 16 - -
W-E - - - - - 0 - -
NW-SE - - - - - 25 - -
< 1Million sf - - - - - - 20 -
> 1Million sf - - - - - - 25 -
HDD (2011) - - - - - - - 22
HDD (2012) - - - - - - - 23
Figure 4.20 indicates the correlation between site EUI and construction year, which is divided
into 2 groups: before and after 1980, since the first New York state energy code was established
in 1979 (U.S. DOE 2014b). Office buildings that were built before 1980 (31 datasets) have
higher mean EUI of 102.06 kBtu/sf than 92.74 kBtu/sf after 1980 (14 datasets) and the difference
69

is 9.13%. Even the confidence intervals are slightly overlapped, with more strict requirements of
building performance from improved energy code, buildings consume lower energy as expected.
Tall buildings are grouped into megatall (more than 600 ft), supertall (300 to 600 ft) and tall (165
to 300 ft) for the analysis of height (CTBUH 2013). Figure 4.21 shows the significant difference
of energy use for different height tall buildings. Megatall buildings consume the highest total
energy (including vertical transportation system), followed by super tall and then tall buildings.
The mean site EUI for megatall (18 datasets), supertall (22 datasets) and tall buildings (5 datasets)
are 116.89 kBtu/sf, 90.48 kBtu/sf and 73.52 kBtu/sf. National median site EUI of 67.3 kBtu/sf is
only within the tall building EUI range.
The overall 40% of WWR for prescriptive fenestration requirement (NYCECC 2011) is used to
divide all datasets into 2 groups and the results present that WWR is an important factor to
influence office building energy use in terms of heating and cooling load by solar heat gain. The
mean value of buildings with over 40% WWR (28 datasets) is 107.88 kBtu/sf compared to 84.81
kBtu/sf of lower WWR buildings (17datasets). The difference percentage is 23.38%, which is
significantly different. Figure 4.22 presents that it is more clear that averagely buildings consume
more energy with the increasing WWR when dividing datasets into 4 groups: 20% to 40%, 40%
to 60%, 60% to 80% and over 80% window-to-wall ratio. Normally, buildings will consume
more energy for cooling with large window area since the heat gain through less insulated
glazing.

Figure 4.20 Site EUI and construction year
Before 1980 After 1980
115
110
105
100
95
90
85
80
Construction Year
Site EUI
Interval Plot of Site EUI vs Construction Year
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
70

Figure 4.21 Site EUI and building height

Figure 4.22 Site EUI and WWR

Figure 4.23 Site EUI and operable window

Figure 4.24 Site EUI and V/FA ratio
Tall Supertall Megatall
130
120
110
100
90
80
70
60
50
Tall building
Site EUI
Interval Plot of Site EUI vs Tall building
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
Over 80% 60%-80% 40%-60% 20%-40%
160
150
140
130
120
110
100
90
80
70
WWR
Site EUI
Interval Plot of Site EUI vs WWR
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
1 0
120
110
100
90
80
70
Operable Window (Y/N)
Site EUI
Interval Plot of Site EUI vs Operable Window (Y/N)
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
Over 40 Below 40
120
110
100
90
80
V/FA Ratio
Site EUI
Interval Plot of Site EUI vs V/FA Ratio
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
71

Figure 4.25 Site EUI and orientation
Figure 4.23 indicates that buildings with operable windows consumes less energy since the
mixed mode of natural ventilation and mechanical ventilation is more energy efficient, which is
proved by the fact that the mean value of 84.9 kBtu/sf for buildings with operable windows (12
datasets) is significantly lower than 104.25 kBtu/sf for buildings without operable windows (33
datasets). The difference is 18.56%.
V/FA ratio stands for the compactness which has a significant impact on heating load. Figure
4.24 illustrates that buildings with V/FA less than 40 (20 datasets) have the lower mean EUI of
89.03 kBtu/sf, which means in this heating dominated area, compact buildings are not necessary
consuming less energy than greater façade area buildings. It also depends on glazing and exterior
wall thermal properties and other factors.
Figure 4.25 shows there is no significant difference of EUI between N-S axis orientation (4
datasets) and NE-SW axis orientation (16 datasets) while buildings with NW-SE axis orientation
(25 datasets) have the highest mean EUI value of 111.01 kBtu/sf. It is because when the main
façade faces south west, there is more heat gain through direct sun exposure. In this database,
there is no recorded building facing W-E axis orientation.

Figure 4.26 Site EUI and floor area
Over 1000000 SF Below 1000000 SF
120
110
100
90
80
70
Floor Area
Site EUI
Interval Plot of Site EUI vs Floor Area
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
72

Figure 4.27 Site EUI and HDD
To consider the relation between building floor area and energy use, original datasets were also
divided into 2 groups: below 1000000 sf and over 10000000 sf. It is because in New York City
building database, all datasets are high-rise office buildings and 1000000 sf could be used as a
median line. Figure 4.26 indicates that buildings floor area with over 1,000,000 square feet (25)
have significantly higher energy use than smaller area buildings (20). Larger area buildings have
more complicated systems and operation schedules, which may result in higher energy use.
Another important predictor is heating-degree day, which is extremely important for heating
demand of a building in New York City. Figure 4.27 shows that in total there are only 2 years of
energy data used in this regression research but it is clear that most buildings consumed more
energy in 2011 than in 2012, since the HDD of 3272 in 2011 is higher than 2988 in 2012, while
other façade features didn’t change within these 2 years. It is very possible that weather change
as well as air-conditioning demand have a great and direct impact on building energy use change.
4.2.2 MLR, Stepwise and PLS
In this research, EUI are predicted by the façade features through 3 regression methods by using
Minitab 17: Multiple Linear Regression (MLR), Stepwise Regression and Partial Least Square
PLS). The results are showed in the following tables. Total façade area is replaced by 8 different
directions of façade area. In MLR, all 25 predictors are included in every model. The R
2
value
indicates that all predictors could explain 77.64% of the variance in EUI while the adjusted R
2

value means only 56.18% of EUI variable variation is explained by its relationship with predictor
variables. Adjusted for the number of predictors in the model. The D-W statistic is closer to 2,
which means there is no significant autocorrelation. Only orientation and floor area are
significantly related to annual EUI at an α-level of 0.05 since P-values are close to 0. VIF values
3272 2988
115
110
105
100
95
90
85
HDD
Site EUI
Interval Plot of Site EUI vs HDD
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
73

for all coefficients are greater than 10 which means the regression coefficients are poorly
estimated due to severe multicollinearity.
Table 4.5 Coefficients and main indicators
Indicator Multiple Linear Regression Stepwise Regression Partial Least Squares
R
2
/ R
2
(Adj)/ R
2
(pre) 77.64% 56.18% - 88.15% 84.66% 77.72% 83.73% - 52.15%
D-W 2.02 1.99 -
Predictor Coef P VIF Coef P VIF Coef P X-Variance
Constant 27302 0.17 - -75.30 0.05 - 285.50
0.00 98.47%
Height 0.09 0.59 83.84 0.16 0.00 3.85 0.021
Floors 0.06 0.98 78.14 - - - -0.17
Built year -0.34 0.59 17.67 - - - -0.21
WWR 0.54 0.51 25.16 0.72 0.00 2.03 0.30
Orientation 26.00 0.03 25.61 18.77 0.00 4.53 -5.42
Operable Window -29.9 0.15 12.20 -19.65 0.00 2.11 -37.39
Volume 0.00 0.99 605.78 - - - -
Window Area 0.00014 0.55 100.77 - - - -
Site Area 0.00035 0.73 54.20 - - - -
Floor Area -0.00007 0.03 29.78 -0.00005 0.00 8.55 -
V/FA -0.84 0.81 127.38 - - - -
V/SA 0.19 0.52 132.69 0.14 0.00 4.52 -0.098
FA/SA -10.29 0.11 77.31 -9.47 0.00 8.61 -0.007
Adjacency -1.85 0.50 12.44 - - - -1.097
HDD 5.86 0.18 53879.80 0.0324 0.01 1.02 3.99
CDD -22.7 0.18 53885.99 - - - 0.019
N Façade Area -0.01 0.20 6298.99 - - - 0.065
S Façade Area 0.13 0.23 1023529 0.0013 0.00 11.46 -
W Façade Area -0.0025 0.2 598.28 -0.00063 0.01 13.83 0.001
E Façade Area -0.089 0.243 862326.3 - - - -0.001
NW Façade Area -0.00015 0.806 49.89 - - - 0.001
NE Façade Area -0.00017 0.892 553.60 - - - -
SW Façade Area -0.00012 0.849 148.17 - - - -
SE Façade Area 0.000571 0.471 101.53 - - - -
By comparison, R
2
from stepwise regression means 88.15 % of the variance in EUI. The adjusted
R
2
is also improved when compared to MLR, which is 84.66%. The predicted R
2
value is 77.72%
which indicates the model does not appear to be overfit and has adequate predictive ability since
it’s close to R
2
and adjusted R
2
. D-W statistic is 1.989 which is also closer to 2. All P-values of
corresponding predictors are less than 0.05 while VIFs are less than 10 except south and west
façade areas are slightly higher than 10. The results show that the advantage by using stepwise
regression is not only improved accuracy of each indicator but also the identification a useful
subset of original predictors. The stepwise process systematically adds the most significant
variable or removes the least significant variable during each step. As a result, predictors
including height, WWR, orientation, operable window, floor area, V/SA ratio, HDD as well as
south and west façade area are the most important factors which have greater impact on energy
use for office buildings in New York City.
74

Table 4.6 PLS coefficients and main indicators (continued)
Predictors Coef P X-Variance
Height*Floor -0.002
0.00 98.47%
Height*Built Year -
Height*WWR (%) -
Height*Orientation 0.043
Height*Operable Win 0.060
Height*Volume -
Height*Window Area -
Height*Facade Area -
Height*Site Area -
Height*Floor Area -
Height*V/FA -
Height*V/SA -
Height*FA/SA -0.001
Height*Adjacent Building -0.007
Height*HDD -
Height*CDD -
PLS results are also listed as another experimental option. The optimal model is defined as the
model with the highest predicted R
2
in PLS regression. The predicted R
2
is 52.15% with 13
components in the selected optimal model. The P value is 0.00, which is less than an alpha of
0.05, providing sufficient evidence that the 13-component model is significant. 13-component
model is selected by cross-validation, which has the R
2
value of 83.73%. The X-variance
indicates the amount of variance in the predictors that is explained by the model. The selected
model can explain 98.47% of the variance in the predictors.

Figure 4.28 Standard residual and frequency plot
75

Figure 4.29 Normal probability plot

Figure 4.30 Standard residual and fitted value
Figure 4.28 is the histogram of the standardized residual and frequency. One outlier may exist in
the data, which needs to be proved in other another analysis. A normal probability plot (Figure
4.29) shows an approximately linear pattern consistent with a normal distribution. The point in
the upper-right corner is an outlier (row 33), which could be read from the plot. The plot of
residuals versus the fitted values (Figure 4.30) shows that the variance of the residuals are
constant with a mean of zero.
PLS result plots are represented by the following figures:
 Figure 4.31 shows that through model selection and validation, the optimal model has 13
components. After 13 components, R
2
decreases significantly.
 Figure 4.32 presents the model fits the data adequately because the points are in a linear
pattern, from the bottom left-hand corner to the top right-hand corner.
 Figure 4.33 indicates the interpretation of the magnitude and sign of the coefficients
shows that predictor 29, 34, 26, 20, 21, 11, 19, 6 have the biggest impact on EUI, either
increasing or decreasing energy use.
76

 Distances from the y-model in figure 4,34 identify how well observations are described
by the y-scores. An observation with a large distance value might also be an outlier.
Distances from the x-model identify how well observations are described by the x-scores.
An observation with a large distance value might also be a leverage point.

Figure 4.31 PLS model selection plot

Figure 4.32 PLS response plot

Figure 4.33 PLS standard coefficient plot
175 150 125 100 75 50
175
150
125
100
75
50
Actual Response
Calculated Response
Fitted
Crossval
Variable
PLS Response Plot
(response is Site EUI)
13 components
40 35 30 25 20 15 10 5 1
1.5
1.0
0.5
0.0
-0.5
-1.0
Predictors
Standardized Coefficients
PLS Std Coefficient Plot
(response is Site EUI)
13 components
77

Figure 4.34 PLS distance plot
4.2.3 Validation
To validate the stepwise regression model, original 45 datasets are divided into two samples for
training and testing. The testing samples number is 10% of original datasets and the left samples
are used to develop the regression model. In this case, 5 dataset are randomly selected as testing
samples. The R
2
of training samples model is 91.02% while D-W statistic is 2.04. Error rates for
the 5 validation samples are 5.94%, 9.12%, 4.57%, 6.36% and 8.74%, which are at the accepted
level.

Figure 4.35 Validation and error rate
For PLS, cross-validation is used to validate the optimal model. The result shows the optimal
number of components is 13 when the highest predictive R
2
is 52.15% and the lowest Predictive
Residual Sum of Square (PRESS) is 17895.5, which indicates the sum of the squares of all the
resulting prediction errors.
2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50
35
30
25
20
15
10
5
0
Distance From X
Distance From Y
PLS Distance Plot
13 components
78

4.2.4 Results Comparison
The error rates for each case calculated by the 3 regression models are compared in Figure 4. 36.
It is clear that there are more large errors from MLR than Stepwise and PLS, which also is
proved by the fact that the average error rate of MLR is 16.21% while the average error rates of
Stepwise and PLS are 9.74% and 7.86%. The error rates from lower EUI buildings are larger
than buildings with higher energy use. From Figure 4.41, all estimated EUI from 3 regression
models are compared to the reported original EUI value, and the EUI calculated by Stepwise and
PLS are significantly closer than result from MLR. There are more outliers from PLS rather than
Stepwise.
Based on the result comparison and indicators interpretation, the Stepwise regression model is
selected as the final EUI estimation model because its high explanation rate of variable variance
as well as better predictive ability. In addition, stepwise regression can also determine the most
significant predictors that are relevant to building energy use in certain areas. The selected
predictors can be used to determine the most important target façade features that need to be
collected for EUI estimation. Last but not least, the egression model is easier than PLS since in
PLS a new model is required to precede calculation.

Figure 4.36 Error rate comparison
79

Figure 4.37 Regression results comparison
80

4.3 City-based Annual EUI Model – Los Angeles
4.3.1 Basic Data Analysis
To estimate annual building energy use in Los Angeles, 5 buildings were selected from the
original nationwide datasets. In addition, another 6 buildings with multiple years of site EUI data
were added into the basic datasets. Within the 11 buildings, 10 of them are commercial offices
while another one is a bank building. The floors number ranges from 3 to 12 while the window-
to-wall ratio ranges from 33% to 72.5%. 4 buildings have operable windows as reported or
clearly distinguished.
Buildings with multiple years of energy use record means in the original datasets, only HDD and
CDD varied through different years, in which standards for external weather changes but no
physical façade retrofit or significant envelope change was recorded.
In total, there are 26 cases from 11 buildings that were used in regression analysis for predicting
office building energy use in Los Angeles. All datasets with façade features are firstly analyzed
by dividing them into different groups. The results represent the correlation between reported site
EUI with each individual predictor through interval plotting. The confidence interval is 95%
which indicates 95% probability from the future experiment within this interval. Table 4.7 lists
the total number of buildings in each group.
Table 4.7 Total number of buildings in each group
Total # of
Buildings
Built Year WWR Operable
Window
V/FA Ratio Orientation CDD
Before 1978 21 - - - - -
After 1978 5 - - - - -
Below 40% - 6 - - - -
Over 40% - 20 - - - -
With - - 10 - - -
Without - - 16 - - -
Below 40 - - - 20 - -
Over 40 - - - 6 - -
N-S - - - - 5 -
NE-SW - - - - 10 -
W-E - - - - 7 -
NW-SE - - - - 4 -
Below 2000 - - - - - 14
Over 2000 - - - - - 12
Among 26 original datasets with annual energy use data in Los Angeles area, 5 buildings have
only one year’s energy use data and other 6 buildings have multiple years’ energy use data. The
latest California building energy code, Building Energy Efficiency Standards for Residential and
81

Nonresidential Building (Title 24, Part 6) was released in May. 2012 (California Energy
Commission 2012). It proposed several key factors to improve building energy efficiency,
including envelope insulation and HVAC system testing for residential buildings and lighting
controls, window, unitary HVAC equipment and building commissioning for nonresidential
buildings. In which case, façade improvement is the main area where energy could be saved by
relevant strategies. The first state energy code in California, Building Energy Efficiency
Standard: New Residential and New Nonresidential Buildings (CEC-400-1978-001), was
established in 1978 (California Energy Commission 1978), which initiated the first requirement
of envelope regulation and equipment efficiency. All datasets were firstly divided into two
groups, before and after 1978 and figure 4.38 indicates the correlation between site EUI and
construction year. Office buildings that were built before 1978 (21 datasets) have a significantly
higher mean EUI of 75.91 kBtu/sf than 60.69 kBtu/sf after 1980 (5 datasets). The difference
percentage is 20.17%. The confidence intervals are overlapped because there are only 5 datasets
that were constructed after 1980, which resulted in a wider range of the predictive data
distribution. However, with more strict requirements of building performance from improved
energy code, buildings consumed less energy, as expected.
The overall 40% Window-to-wall ratio is regulated by Title 24 as a prescriptive maximum
percentage for all climate zones in California (California Energy Commission 2012), which was
used to divide all datasets into another two groups. The results present that the mean value of
buildings with over 40% WWR (20 datasets) is 74.4 kBtu/sf compared to 68.2 kBtu/sf of lower
WWR buildings (6 datasets). Even the difference percentage is only 8.33%, WWR is an
important factor to influence office building energy use in terms of heating and cooling load by
solar heat gain. Furthermore, buildings were grouped into 3 parts for detailed analysis: 20% to
40%, 40% to 60% and 60% to 80% WWR. Figure 4.39 presents that the mean EUI of 40% to 80%
is higher than 20% to 40% as expected, but building with 60% to 80% WWR has the lowest
mean EUI. This is contradictory with common sense that larger window will cause more heat
gain through glazing and more energy use for cooling demand. There are only 6 datasets within
60% to 80% group from 2 buildings, so one reason may come from the inefficient numbers of
datasets, which also result in the negative value of the coefficient of WWR in regression model.
More datasets should be added in the future work.
82

Figure 4.38 Site EUI and construction year

Figure 4.39 Site EUI and WWR

Figure 4.40 Site EUI and operable window

Figure 4.41 Site EUI and V/FA ratio
Before 1980 After 1980
90
80
70
60
50
40
30
Construction Year
Site EUI
Interval Plot of Site EUI vs Construction Year
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
60%-80% 40%-60% 20%-40%
110
100
90
80
70
60
50
40
30
20
WWR
Site EUI
Interval Plot of Site EUI vs WWR
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
1 0
100
90
80
70
60
50
40
Operable Window (Y/N)
Site EUI
Interval Plot of Site EUI vs Operable Window (Y/N)
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
Over 40 Below 40
100
90
80
70
60
50
40
V/FA Ratio
Site EUI
Interval Plot of Site EUI vs V/FA Ratio
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
83

Figure 4.42 Site EUI and orientation

Figure 4.43 Site EUI and CDD
Figure 4.40 indicates that buildings with operable windows consumed less energy since the
natural ventilation is an effective way to save energy for cooling in this area, which is also
proved by the fact that the mean value of 67.16 kBtu/sf for buildings with operable windows (10
datasets) is lower than 76.62 kBtu/sf for buildings without operable windows (16 datasets). The
difference percentage is 12.35%. V/FA ratio stands for the compactness which has significant
impact on heating load. Figure 4.41 illustrates that buildings with V/FA less than 40 (20 datasets)
had the similar energy use with higher V/FA ratio buildings (6 datasets). The mean EUI are
73.59 kBtu/sf and 70.94 kBtu/sf separately. It means the V/FA ratio is not significant and clear to
influence energy use only based the current limited database. Figure 4.42 shows there is
significant difference of EUI between different axis orientations. Building in N-S axis orientation
(5 datasets) consumed the highest mean energy of 115.3 kBtu/sf, followed by NE-SW (70.11
kBtu/sf), E-W (64.52 kBtu/sf) and NW-SE (42.08 kBtu/sf). It is because in this cooling
dominated area the main façade facing west and east (N-S axis orientation) has more heat gain
through direct sun exposure. Another important predictor taken into consideration is cooling-
degree day. Figure 4.43 represents the cooling demand of a building and the results showed
4 3 2 1
140
120
100
80
60
40
20
0
Orientation
Site EUI
Interval Plot of Site EUI vs Orientation
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
Over 2000 Below 2000
110
100
90
80
70
60
50
40
CDD
Site EUI
Interval Plot of Site EUI vs CDD
95% CI for the Mean
The pooled standard deviation was used to calculate the intervals.
84

buildings in a year when CDD was higher than 2000 consumed significantly higher energy than
buildings in lower cooling demand year. CDD of 2000 was selected since 2000 is the dividing
line from CBECS climate zone definition, and the range of annual CDD change in Los Angeles
is limited compared to the nationwide building database.
4.3.2 Regression Models
Stepwise regression is selected as the method to predict building energy use in different cities,
because it has the comparatively higher accuracy but also it requires the minimum number of
predictors. The selected predictors also indicate the most significant factors influencing EUI in
the certain area.
In this section, EUIs are predicted by the façade features through 2 regression methods: Multiple
Linear Regression (MLR) and Stepwise Regression. In MLR, floor area, V/FA, V/SA, FA/SA,
adjacent building, north, south, west and east façade area were removed firstly because they
can’t be estimated due to severe multicollinearity. The R
2
value indicates that all predictors could
explain 94.06% of the variance in EUI while the adjusted R
2
value means 88.58% of EUI
variable variation that is explained by its relationship with predictor variables, adjusted for the
number of predictors in the model. The vacant R
2
(predictive) indicated that the MLR model has
less predictable abilities. D-W statistic is 1.73 and closer to 2, which means there is no
significant autocorrelation. Height, floor, orientation, operable window, volume, full façade area,
site area and CDD are significantly related to annual EUI at an α-level of 0.05 since P-values are
close to 0.
Table 4.8 Coefficients and main indicators
Indicator Multiple Linear Regression Stepwise Regression
R
2
/ R
2
(Adj)/ R
2
(pre) 94.06% 88.58% - 93.39% 90.88% 86.39%
D-W 1.73 1.99
Predictor Coef P VIF Coef P VIF
Constant -178 0.85 - -1142 0.121 -
Height 1.46 0.01 66.25 1.645 0.00 48.83
Floors 38.42 0.00 60.57 - - -
Built year 0.06 0.89 4.25 0.554 0.14 3.33
WWR -0.20 0.91 89.57 -0.132 0.525 1.49
Orientation -30.81 0.00 15.05 -31.97 0.00 6.19
Operable Window 43.80 0.03 15.72 - - -
Volume -0.00013 0.00 222.38 -0.00017 0.00 109.01
Window Area 0.0068 0.13 963.42 - - -
Site Area 0.016 0.00 162.69 - - -
Floor Area - - - 0.001198 0.00 42.99
V/FA - - - 6.189 0.00 7.90
V/SA - - - - - -
85

FA/SA - - - - - -
Adjacency - - - - - -
HDD 0.0069 0.89 2.27 - - -
CDD -0.024 0.05 6.63 -0.01273 0.036 2.99
Façade Area -0.0078 0.01 744.48 - - -
N Façade Area - - - - - -
S Façade Area - - - - - -
W Façade Area - - - - - -
E Façade Area - - - - - -
By comparison, R
2
from stepwise regression indicated that 93.39% of the variance in EUI could
be explained by stepwise regression model. The adjusted R
2
was increased when compared to
MLR. In particular, the predicted R
2
value was 86.39%, which indicates the model does not
appear to be overfit and has adequate predictive ability since it’s close to R
2
and adjusted R
2
.
This is a significant improvement of stepwise regression model since the goal is to predict other
buildings in the future, so Stepwise regression is still the best candidate option for EUI
estimation. D-W statistic is 1.99 which is also closer to 2. All P-values of corresponding
predictors (except built year and WWR) are closer to 0 and it presents that those selected
predictors are significantly related to site EUI. The results showed that the advantage by using
stepwise regression is not only improved accuracy of each indicator but also the identification a
useful subset of original predictors. The stepwise process systematically adds the most
significant variable or removes the least significant variable during each step. As a result,
predictors including height, built year, window-to-wall ratio, orientation, volume, floor area,
V/FA and CDD are the most important factors which have greater impact on energy use for
office buildings in Los Angeles.
4.3.3 Validation
To validate the stepwise regression model, the original 25 datasets are divided into two samples
for training and testing. The testing samples number is 10% of original datasets for validation. In
this case, 3 datasets are randomly selected as testing samples while the left datasets are training
samples. The R
2
of training samples model is 90.75% while D-W statistic is 1.86. Error rates for
the 3 validation samples are 5.32%, 6.51% and 3.28%, which are at the accepted level.
86

Figure 4.44 Validation and error rate
4.3.4 Results Comparison
The error rates for each case calculated by 2 regression models are compared in Figure 4.45 and
4.46. For office building in Los Angeles area, the predicted site energy use intensities calculated
by two models are very similar, the average error rate of MLR is 9.00% while the average error
rate of stepwise regression is 9.39%. All estimated EUI from 2 regression models are compared
to the reported original EUI value, and there is significant difference between two results.

Figure 4.45 Error rate comparison
87

Figure 4.46 Regression results comparison
Based on the result comparison and indicators interpretation, the stepwise regression model is
selected as the final EUI estimation model because it keeps relatively high explanation rate of
variable variance but the predictive ability has been significantly improved. In addition, stepwise
regression can also determine the most significant predictors that are relevant to building energy
use in this area. The selected predictors can be used to determine the most important target
façade features that are needed to collect for EUI estimation. The biggest advantage of stepwise
regression model is the minimum input predictors or façade features for the similar accurate
results.

88

4.4 Monthly EUI Model for Los Angeles
4.4.1 Basic Data Analysis
To generate monthly building energy use prediction models in Los Angeles area, buildings with
available monthly energy use data were firstly collected. In this section, 5 buildings from original
annual EUI datasets in Los Angeles were used. Each building had at least 2-year full 12 months
energy use for analysis and all of them are office buildings. The floors number ranges from 4 to
12 while the built year ranges from 35% to 60%. 2 buildings have operable window as reported
or clearly distinguished while other 3 buildings are without operable window.

Figure 4.47 Monthly site EUI in Building 1

Figure 4.48 Monthly site EUI in Building 2
For each building with multiple years of energy use record in the original datasets, only HDD
and CDD varied through different year which standards for external weather changes. There is
no recorded physical façade retrofit or significant envelope change.
89

Figure 4.49 Monthly site EUI in Building 3

Figure 4.50 Monthly site EUI in Building 4

Figure 4.51 Monthly site EUI in Building 5
Figure 4.57 to 4.51 represent monthly heating degree day and cooling degree day change in
different buildings. The greatest heating demand always happened from December to February
while the greatest cooling demand always happened in either July or August. Accordingly, the
greatest monthly energy use also happened in the same month. For building 2, 3 and 4, the peak
energy was used in winter for all the years in this available datasets while for building 1 and 5
there was 1 year in each when the peak energy was used in summer. HDD and CDD are greatly
relevant to the building energy demand, which would be the significant characteristics in
90

monthly building energy regression model development. In total, there are 5 buildings and 180
cases for the analysis to predict the monthly energy use intensity in Los Angeles.
4.4.2 MLR, Stepwise and PLS
In this section, EUI are predicted by using façade features through 3 regression: Multiple Linear
Regression (MLR), Stepwise Regression and Partial Least Square PLS). The results are showed
in following table 4.9. In MLR, the R
2
value indicates that all predictors could explain 91.59% of
the variance in EUI while the adjusted R
2
value means also 91.3% of EUI variable variation that
is explained by its relationship with predictor variables, adjusted for the number of predictors in
the model. Height, floors, built year, WWR and HDD are significantly related to annual EUI at
an α-level of 0.05 since P-values are close to 0. However, the P-value of CDD is relatively high
which means CDD is not as significant as HDD to influence monthly energy use. In addition,
VIF values of height and floors are greater than 10 which means the regression coefficients are
poorly estimated due to severe multicollinearity.
Table 4.9 Coefficients and main indicators
Indicator Multiple Linear Regression Stepwise Regression Partial Least Squares
R
2
/ R
2
(Adj)/ R
2
(pre) 91.59% 91.30% 90.72% 91.59% 91.35% 90.96% 91.82% - 90.89%
D-W 1.05 1.05 -
Predictor Coef P VIF Coef P VIF Coef P X-Variance
Constant 485.7 0.00 - 13.05 0.00 - -37.98
0.00 100%
Height 0.94 0.00 333.06 - - - 0.0032
Floors -13.69 0.00 315.16 - - - 0.025
Built year -0.24 0.00 2.28 - - - 0.026
WWR -0.09 0.00 3.23 - - - 0.026
Orientation - - - - - - -0.45
Operable Window - - - - - - -0.33
Volume - - - -0.00000 0.00 44.20 -0.0000
Window Area - - - 0.00017 0.00 8.68 -0.0000
Site Area - - - - - - -0.0000
Floor Area - - - -0.00007 0.00 35.08 -0.0000
V/FA - - - - - - -0.027
V/SA - - - - - - 0.0032
FA/SA - - - - - - 0.055
HDD 0.0046 0.02 2.06 0.0045 0.02 2.03 0.0013
CDD -0.00005 0.95 2.31 -0.00007 0.92 2.25 -0.0040
N Façade Area - - - - - - -0.0001
S Façade Area - - - - - - -0.0001
W Façade Area - - - - - - -0.0000
E Façade Area - - - - - - -0.0000
By comparison, R
2
from stepwise regression means 91.59% of the variance in EUI and it is same
as multiple linear regression result. The adjusted R
2
is improved when compared to MLR, which
is 91.35%. The predicted R
2
value is 90.96% which indicates the model does not appear to be
overfit and has adequate predictive ability since it’s close to R
2
and adjusted R
2
. All P-values of
91

corresponding predictors are less than 0.05, except CDD, while VIFs are less than 10 except
volume and floor area are higher than 10. The results still show that using stepwise regression
could slightly improve accuracy of each indicator but also identify a useful subset of original
predictors. As a result, to predict monthly EUI, predictors including volume, window area, floor
area, HDD as well as CDD are the most important factors which have greater impact on monthly
energy use for office buildings in Los Angeles.
Table 4.10 PLS coefficients and main indicators (continued)
Predictors Coef P X-Variance
Height*Floor 0.0000
0.00 100%
Height*Built Year 0.0000
Height*WWR (%) 0.0000
Height*Orientation -0.0008
Height*Operable Window -0.0020
Height*Volume -0.0000
Height*Window Area -0.0000
Height*Facade Area -0.0000
Height*Site Area -0.0000
Height*Floor Area -0.0000
Height*V/FA -0.0001
Height*V/SA -0.0000
Height*FA/SA -0.0001
Height*Adjacent Building -0.0000
Height*HDD -0.0000
Height*CDD -0.0000
PLS as another option is listed in the table. The optimal model is defined as the model with the
highest predicted R
2
in PLS regression. The predicted R
2
is 90.89% with 8 components in the
selected optimal model. The P value is 0.00, which is less than an alpha of 0.05, providing
sufficient evidence that the 8-component model is significant. 8-component model is selected by
cross-validation, which has the R
2
value of 91.82%. The X-variance indicates the amount of
variance in the predictors that is explained by the model. The selected model can explain 100%
of the variance in the predictors based on the PLS analysis results.

Figure 4.52 Standard residual and frequency plot
92

Figure 4.53 Normal probability plot

Figure 4.54 Standard residual and fitted value
Figure 4.52 is the histogram of standardized residual and frequency, and standardized residuals
greater than 2 and less than -2 are considered potential outliers by definition. Within 180
observations, there are 8 observations are unusual. Normal probability plot (Figure 4.53) shows
an approximately linear pattern consistent with a normal distribution. The plot of residuals versus
the fitted values (Figure 4.54) shows that the variance of the residuals are constant with a mean
of zero.
PLS result plots are represented by following figures:
 Figure 4.55 shows that through model selection and validation, the optimal model has 8
components.
 Figure 4.56 indicates that the model doesn’t fit the data adequately because the points are
not in a linear pattern.
 Interpretation of the magnitude and sign of the coefficients (Figure 4.57) shows that
predictor 36 (height*façade area) has the greatest positive impact on EUI while 10 (floor
area), 16 (north façade area) and 5 (orientation) have the least impact on EUI incresing.
93

 Figure 4.58 of distances from the y-model identifies how well observations are described
by the y-scores. An observation with a large distance value might also be an outlier.
Distances from the x-model identify how well observations are described by the x-scores.

Figure 4.55 PLS model selection plot

Figure 4.56 PLS response plot

Figure 4.57 PLS standard coefficient plot
94

Figure 4.58 PLS distance plot
4.4.3 Validation
To validate the stepwise regression model, the original 180 datasets are divided into two samples
for training and testing. The number of testing samples is 10% of the original datasets. In this
case, 18 datasets from original 180 valid datasets are randomly selected as testing samples while
other are used to generate the regression model. The R
2
of training samples model is 91.28%.
Error rates for the 14 validation samples are less than 10%, which are at the accepted level, but
the other 4 samples’ recalculated fits have about 20% error rate. It indicates that due to limited
raw datasets for regression, the model is not constantly steady and accurate to predict EUI and
more datasets are still needed for further improvement.
For PLS, cross-validation is used to validate the optimal model. The result shows the optimal
number of components is 8 when the highest predictive R
2
is 90.89% and the lowest Predictive
Residual Sum of Square (PRESS) is 149.54 which indicates the sum of the squares of all the
resulting prediction errors.

Figure 4.59 Validation and error rate
95

4.4.4 Results Comparison
The error rate for each case calculated by 3 regression models are compared in Figure 4. 64. It is
clear that the error rates from MLR and Stepwise are similar, which also is proved by the average
error rate of MLR, which is 9.62%, while the average error rates of Stepwise is 9.63%, while
PLS error rate is slightly lower than MLR and stepwise. From Figure 4.65, all estimated EUI
from 3 regression models are compared to the reported original EUI value, and the EUI
calculated by PLS is slightly closer than result from MLR and stepwise results.
Considering that the PLS model requires more detailed façade information as input, stepwise
regression model is selected as the final EUI estimation model because its simplicity and high
explanation rate of variable variance as well as better predictive ability. In addition, stepwise
regression can also determine the most significant predictors which are relevant to building
energy use in certain area. Stepwise regression can be used to determine the most important
target façade feature that are supposed to be collected for EUI estimation.

Figure 4.60 Error rate comparison

96

Figure 4.61 Regression results comparison
0
2
4
6
8
10
12
14
16
1
7
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
103
109
115
121
127
133
139
145
151
157
163
169
175
Site EUI (kBtu/sf)
Regression Results Comparison
Site EUI MLR STEPWISE PLS
97

4.5 Monthly Cooling and Heating EUI Models for Los Angeles
4.5.1 Basic Information
Four office buildings in downtown Los Angeles with annual measured EUI data (only building 1
has monthly electricity and gas consumption data) are used as basic datasets. Annual EUI in 4
buildings are 53.0 kBtu/sf, 48.1 kBtu/sf, 44.9 kBtu/sf and 56.1 kBtu/sf. Basic building
information are listed in following table.

Figure 4.62 Geometry model of building 2, 3, 4 in DesignBuilder
To get the monthly data from building 2, 3 and 4, DesignBuilder (Tronchin and Fabbri 2008)
was used as a simulation tool to calculate monthly energy use. It is based on the building energy
simulation engine EnergyPlus with a 3D interface and meteorological database.
Four building models were developed in DesignBuilder with the basic input of thermal
conditions and mechanical system information which are gathered from client specifications and
code requirements. All models are office buildings with interior and perimeter zones.
Table 4.11 Basic information of 4 office buildings
No. Built Year Stories Floor area Wall Roof WWR Glazing Cooling Heating
1 1974 55 26500 sf R-15 R-30 37% U-1.22, SHGC 0.22 74 ºF 72 ºF
2 1985 41 23000 sf R-15 R-30 39% U-1.22, SHGC 0.22 74 ºF 72 ºF
3 1982 54 26000 sf R-10 R-18 39% U-1.22, SHGC 0.22 74 ºF 72 ºF
4 1985 42 24000 sf R-12 R-18 35% U-1.22, SHGC 0.22 74 ºF 72 ºF
After developing building models in DesignBuilder and running simulations, the accuracy and
reliability should be validated. Calibration (ASHRAE 2002) is the process of comparing the
simulation outputs to the results of a measurement, determining the deviation and relevant
uncertainty and adjusting the model accordingly. The revised model will accurately describe the
real building energy use after calibration. Models are adjusted after comparing the real annual
energy data with the original simulation results. The typical adjustments include infiltration rate,
98

ventilation rate, occupancy schedule and indoor set point. The final results of the revised model
are shown in the following figure.

Figure 4.63 Annual EUI Calibration for building 2, 3 and 4
The variation between simulated and measured results are 4.94%, 1.44% and 4.67% respectively,
which meet the requirement of overall allowed amount of variation: ±10% mean bias deviation
(MBD). It means that the DesignBuilder models can convey an accurate representation of
building conditions and simulate monthly EUI for heating and cooling in the next step. In
addition, it will also be used in generating more cases by reasonably changing certain façade
features.
Multiple dynamic simulations were generated in order to determine the values of monthly
heating and cooling EUI data with different input. As the result, besides four existing monthly
data sets, 11 more data sets are generated by changing WWR and orientation. In total 15 cases
with full 12 monthly EUI data could generate 180 datasets for regression.
Table 4.12 Cases (180 monthly data sets)
Case Building number Key feature Source
1 1 As built Measured
2 2 As built Estimated
3 3 As built Estimated
4 4 As built Estimated
5 4 WWR 20% Simulated
6 4 WWR 30% Simulated
7 4 WWR 40% Simulated
8 4 WWR 50% Simulated
9 4 WWR 60% Simulated
10 2 Orientation N-S Simulated
11 2 Orientation E-W Simulated
12 3 Orientation NW-SE Simulated
13 3 Orientation N-S Simulated
14 4 Orientation E-W Simulated
15 4 Orientation NW-SE Simulated
99

4.5.2 Monthly Heating and Cooling EUI regression models
Heating and cooling monthly EUI are estimated firstly by Multiple Linear Regression (MLR) as
responses. The first results are shown as following equations:

(4-1)

(4-2)
In these two equations, site area, floor area, V/FA, V/SA and FA/SA are removed automatically
in both equations. A predictor that has a low p-value is likely to be a meaningful addition to the
model. Conversely, a greater (insignificant) p-value suggests that changes in the predictor are not
associated with changes in the response. In the heating EUI equation (4-1), p-value of orientation
and façade area are greater than the common alpha level of 0.05, which indicates that it is not
statistically significant. The first equation of cooling monthly EUI shows the same insignificancy
and multi-collinearity as well. Two models should be revised by removing high p-value and
highly correlated predictors. The revised equations are as follows:

(4-3)

(4-4)
In revised model results as showed in equation 4-3 and 4-4, all the predictor variables are
significantly related to monthly EUI at an α-level of 0.05 since their p-values are 0.000 (except
for HDD in cooling). The VIFs are all less than 10, which suggests that the regression
100

coefficients are well estimated without great multicollinearity. The R
2
value indicates that the
predictors explain 90.08% and 85.44% of the variance in monthly heating and cooling EUI. The
adjusted R
2
are 89.8% and 85.02%, which account for the number of predictors in the model.
Both values indicate that the model fits the data well. The predicted R
2
value are 89.3% and
84.4%, which indicate they have good predictive ability. Since the predicted R
2
value is close to
the R
2
and the adjusted R
2
values, the model does not appear to be overfit.
Table 4.13 Revised regression models
Revised heating monthly EUI Revised cooling monthly EUI
Predictor P-value VIF R
2
D-W P-value VIF R
2
D-W
WWR 0.000 1.00 R
2
: 90.08%
R
2
(adj):
89.8%
R
2
(pred):
89.3%
1.43 0.000 1.00 R
2
: 85.44%
R
2
(adj):
85.02%
R
2
(pred):
84.4%
1.05
V/FA 0.000 5.35 0.000 5.35
FA/SA 0.000 5.36 0.000 5.36
HDD 0.000 2.37 0.048 2.37
CDD 0.000 2.38 0.000 2.38
4.5.3 Results and Validation
The estimated results from the regression models and the comparison with the original data are
illustrated in the following 2 figures and it shows good fitness. Both monthly heating and cooling
EUI calculated by stepwise regression are more accurate and closer to the original EUI.

Figure 4.64 Estimated monthly heating EUI comparison
101

Figure 4.65 Estimated monthly cooling EUI comparison
To validate the accuracy of regression model, a new simulation model is developed by using
DesignBuilder. Since only façade feature inputs are changed in the new model while other
thermal and mechanical system condition keep the same as the calibrated model, the new
simulation model result could be compared with regression model result for validation. In
validation model, building is developed with 45% WWR, east-west orientation, 324,625 sf
façade area, and using HDD and CDD data from 2013.
From the two comparison charts, it is clear that regression model results are similar to the testing
monthly EUI. The revised model is better than original regression model since it is closer to the
testing monthly EUI. It is more accurate for the stepwise model to estimate cooling EUI in
summer and winter than transition seasons. In conclusion, to estimate heating and cooling EUI
more accurately in the future, more datasets should be collected from real buildings especially
from buildings with sub-metering systems, which could record heating and cooling energy use
separately.
102

Figure 4.66 Validation model monthly heating EUI

Figure 4.67 Validation model monthly cooling EUI
103

Chapter 5 Comparison and Model Verification
5.1 Comparison with Current Baseline
The current widely used building energy baseline is from Commercial Buildings Energy
Consumption Survey (CBECS), which provides the total energy consumption, expenditures,
energy use intensity by major fuel, census region, climate zone, building size and year
constructed. Figure 5.1 presents U.S. Census regions and divisions (U.S. Department of
Commerce 2010). Other basic building statistics were also categorized and recorded in the
survey’s tables. In addition, TargetFinder developed by Environmental Protection Agency (EPA)
is also used to determine energy use intensity baseline, which is especially useful when certain
building information are available, like building location, primary function, gross floor area,
energy estimates (optional), operating hours, number of computers and workers, percentage of
heated/cooled space, etc. To get the most accurate metrics, each energy type is recommended to
input into the tool. However, it is difficult to have real energy use information.

Figure 5.1 U.S. Census regions and divisions (U.S. Department of Commerce 2010)
The baseline for buildings in a certain area from CBECS is a relatively constant and stable value
which represents a group of buildings based on only one simple division within a large range. It
could not reflect a specific individual building’s characteristics or a particular year’s weather
condition. Even by using TargetFinder which is highly recommended by Architecture 2030, the
similar “flat” baseline is also generated for a group of buildings in one location.
On the contrary, figure 5.2 and 5.3 indicate the advantage of using regression model to estimated
building energy use baseline. Figure 5.2 and 5.3 present the direct comparison of reported site
104

EUI with baselines from CBECS, TargetFinder and optimal regression model for office
buildings in New York City and Los Angeles. In each case, regression model states the dynamic
change of EUI based on each individual building condition. By collecting easily readable façade
features, it is more accurate to estimate building energy use and set a reasonable baseline and
target for energy conservation.

Figure 5.2 Site EUI results comparison in New York City
105

Figure 5.3 Site EUI results comparison in Los Angeles
5.2 Comparison of Different City’s Annual EUI Model
By using stepwise regression, two models were developed to estimate office building energy use
in New York City and Los Angeles. By removing the least important predictor while keeping the
accuracy in each step, the stepwise regression models could determine the most significantly key
façade features which have the greatest impact on building energy use in each location. The key
façade features in both cities are listed in table 5.1 and the results are showed as follows:
 Height is significant for both east and west cities. Higher buildings consume more energy
than lower buildings. Compared to large scale high-rise buildings, low-rise buildings are
more flexible to utilize passive strategies and renewable energies, like natural ventilation
and onsite PV generation. In addition, the operation schedule is more simple and stable to
save energy from system side. Energy efficient measures (EEM) are more about system
efficiency for high-rise buildings, like adequate mechanical systems sizing and operation
schedule, higher efficiency HVAC systems, high quality lighting alternatives, etc.
 Built year is another important factor in Los Angeles. The built year range is from 1967
to 2010, which results in more related to energy use since most buildings were
constructed with the requirements of improving energy code. By comparison, most
106

buildings in Manhattan have been retrofitted in recent years, so even the built year range
is from 1917 to 2003 built year is not significant. Retrofit time should be also taken into
consideration in the future work.
 Window-to-wall ratio is a key factor of façade which has a great influence on building
energy use. The regression models show that in New York City, office buildings
consume more energy when the ratio of window to wall is greater, as expected. However,
the coefficient of WWR in regression model in Los Angeles is negative, which means
buildings consume less energy when WWR decreases and other factors stay the same.
Normally, there is more sensible heat gain through glazing more than wall and it results
in buildings consume more energy for cooling with large windows, but in this case, due
to the limited number of original datasets for regression, random discrepancy is more
likely to happen. In addition, multiple linear regression represents a complicated relation
between EUI with multiple façade predictors, so other selected factors would also be
taken into consideration. In fact, large window area will also make it possible to use more
daylight to reduce lighting energy use. It is also related to building users’ visual and
mental health. As a result, WWR is a complicated factor which is required to analyze
seriously and weighed to decide the optimal design.
 In New York City, buildings in NW-SE axis orientation have the higher energy use. It is
because that the main façade faces south west has more heat gain through direct sun
exposure. The same result is also showed in Los Angeles, building in N-S axis orientation
consumed the highest energy, followed by NE-SW, E-W and NW-SE. It is because in
this cooling dominated area the main façade facing south has more heat gain through
direct sun exposure.
 Operable window is a basic strategy for natural ventilation and cooling demand reducing.
Buildings with operable window consume significantly lower energy than buildings with
fixed window. In Los Angeles, this factor is not presented because less cases with
operable feature.
 Buildings with larger floor areas consume more energy in Los Angeles while it is
opposite in New York City. This is because collected datasets in New York City are
mostly high-rise office buildings with a more regular square shape, while in Los Angeles,
107

buildings are mostly low-rise office buildings with relatively larger roof area when floor
area is larger.
 In New York City, buildings with greater volume-to-site area ratio consume more energy
while buildings with greater façade area-to-site area ratio consume less energy. These
two factors represent buildings’ compactness and relative envelope area. In addition,
buildings with greater south façade area consume more energy for cooling while
buildings with higher west area consume less energy relatively.
 HDD is an important energy-related factor in New York City while CDD, on the other
side, contributes more to energy use in Los Angeles.
Table 5.1 Key façade features in NYC and LA
Façade Features New York City Los Angeles
Height
✓ ✓
Floors
Built year
✓
WWR
✓ ✓
Orientation
✓ ✓
Operable Window
✓

Volume
✓
Window Area
Site Area
Floor Area
✓ ✓
V/FA
✓
V/SA
✓

FA/SA
✓

Adjacency
HDD
✓

CDD
✓
N Façade Area
S Façade Area
✓

W Façade Area
✓

E Façade Area

5.3 Model Verification
In this section, a typical office building in downtown Los Angeles was selected to testify 3
regression models from chapter 4, including annual EUI model for nationwide buildings, annual
EUI model for office buildings in Los Angeles and Monthly EUI model for office buildings in
Los Angeles. The case study building has 5 years recorded electricity and natural gas
consumption bills and each year has 12 monthly data available. In addition, energy simulation
software IES-VE was used to predict annual, monthly, monthly cooling and heating EUI for the
108

same building as comparison. The intention is to show the accuracy and convenience to estimate
building energy use by using regression models.
5.3.1 Basic Information and Energy Bills
The building is a class-A office situated in downtown Los Angeles. Its total rentable floor area is
about 213,000 square feet with 16 floors. The 16
th
floor has two personal balconies where people
could overlook downtown views. The buildings was built in 1972 and renovated in 1992. The
external view of the buildings is showed in figure 5.4 (Martin 2010).

Figure 5.4 external view of case study buildings (Martin 2010)
Energy consumption data comes from the facility record, including month electricity (kWh) and
gas (therm) use from 2009 to 2014. The building facility management doesn’t charge utility fees
for each tenant based on the real bills but a constant rate from the original contract. The available
information are included in figure 5.5. The energy use unit was converted to MMBtu for
consistency. Figure 5.6 presents the monthly electricity and natural gas consumption charges in
2014 for example. In summer building used more electricity for cooling while in winter building
used more gas for heating and the peak use appeared in September.
109

Figure 5.5 Monthly energy use for 5 years

Figure 5.6 Electricity and natural gas consumption in 2014
5.3.2 Simulation
IES-VE 2013 was used for detailed simulation. Based on the existing floor plan, building
geometry was developed in the ModelIT. Office space was separated into perimeter and interior
areas while other zones were created by space type and orientation. Surrounding buildings were
also developed to take adjacent shading into consideration. Figure 5.7 shows the 16
th
floor plan
and zoning in ModelIT. Weather data comes from TMY2 file from ASHRAE design weather
database v4.0 for Los Angeles International Airport (Climate zone 3B), California.
110

Figure 5.7 16th floor plan and zoning

Figure 5.8 Building geometry
111

Figure 5.9 IES model and surroundings
Construction materials’ thermal properties including opaque envelope thermal conductivity,
glazing U-value, Solar Heat Gain Coefficient (SHGC), and Visual Light Transmission (VLT)
were input based on onsite check and Title 24 corresponding requirements.
Other important inputs are space equipment power density (EPD), lighting power density (LPD)
and occupancy rate as internal gains. A research of typical office equipment power density was
conducted to collect real equipment use data in the 16
th
floor. Investigated spaces included open
office, meeting room, kitchen, plotter and printer room, lobby, etc. Equipment numbers,
power/daily energy use were collected for EPD calculation. In open office for example, each
work station has 1 task light, 1 desktop and 2 monitors. In total, there are 67 task lamps, 120
monitors, 65 desktops, 3 laptops and 2 printers in 5790 square feet floor area. The occupancy
rate for each space was also counted accordingly. Only LPD was assumed based on ASHRAE
90.1 space type template.
Table 5.2 Building equipment power density, lighting power density and occupancy
Space Type EPD (w/sf) LPD (w/sf) Occupancy (sf/occ)
Open office 1.29 0.98 70
Kitchen 9.66 0.65 N/A
Lobby 0.25 0.90 300
Meeting room 0.33 1.23 35
Plotter and printer room 15.00 0.60 N/A
112

For HVAC systems, the majority of floors used typical overhead variable air volume (VAV)
systems while in 2 floors, underfloor air distribution (UFAD) was used in open office area. The
heating source is natural gas boilers and the cooling source is typical electrical chillers. The
seasonal efficiency for heating is 0.89 and the COP for cooling is 3.125. The heating, cooling,
lighting, equipment and occupant profile was used assuming a default office schedule based on
weekdays, weekends and holidays. The results are showed in the following tables and figures.

Figure 5.10 Annual hourly outdoor dry-bulb temperature
Figure 5.10 shows the annual hourly outdoor dry-bulb temperature change. The weather file
Typical Meteorological Year 2 (TMY2) for simulation in IES comes from the National
Renewable Energy Laboratory (NREL). TMY2 are hourly datasets of solar radiation and
meteorological elements for a year (U.S. DOE, 2014). There are 238 locations of TMY2 file
available for building simulation and system comparison. Since TMY2 data are derived from
1961 to 1990, it doesn’t represent a specific year of weather condition, nor the worst scenario.
113

Figure 5.11 Annual hourly end-use energy
Figure 5.11 presents the hourly energy consumption for the main system, including boiler, chiller,
fans and pumps, equipment and lighting. More energy is used for boiler than chiller especially
from December to March. By comparison, energy used for lighting and equipment is constantly
steady though the whole year.
Table 5.3 Simulation results
Month Heating
(MBtu)
Cooling
(MBtu)
Fan/Pump
(MBtu)
Lighting
(MBtu)
Equipment
(MBtu)
Total
(MBtu)
EUI
(kBtu/sf)
Jan 238.70 35.50 183.20 149.10 233.90 840.40 3.64
Feb 191.90 30.90 167.00 140.10 219.60 749.50 3.24
Mar 149.00 46.40 190.70 159.60 250.20 795.90 3.44
Apr 65.70 60.40 189.50 153.10 240.00 708.70 3.07
May 29.50 87.20 201.30 149.10 233.90 701.00 3.03
Jun 10.90 117.90 209.60 153.10 240.00 731.50 3.16
Jul 1.20 179.30 236.00 154.90 242.80 814.20 3.52
Aug 1.70 200.70 242.30 153.90 241.20 839.80 3.63
Sep 3.70 163.50 225.60 153.10 240.00 785.90 3.40
Oct 21.40 108.30 208.70 149.10 233.90 721.40 3.12
Nov 71.40 63.80 190.70 153.10 240.00 719.00 3.11
Dec 176.10 50.60 192.10 159.60 250.20 828.60 3.58
Total 961.40 1144.40 2436.70 1828.10 2865.50 9236.10 39.96
Figure 5.12 indicates the monthly energy use in each system. Equipment and fan/pump are the
biggest energy end-uses in this office building, followed by lighting, cooling and heating. The
highest heating demand happens in December and the highest cooling demand happens in
August. Since electricity is a kind of secondary energy, more carbon is emitted from electricity
use rather than natural gas use, which is showed in figure 5.14.
114

Figure 5.12 Monthly system energy use breakdown

Figure 5.13 Monthly electricity and natural gas use

Figure 5.14 Monthly carbon emission from system energy use

115

5.3.3 Comparison and Results
In this section, 5 years monthly EUI data calculated from recorded energy bills are used to verify
the accuracy of 3 regression models developed in Chapter 4, including nationwide annual EUI
model, annual EUI model and monthly EUI model in Los Angeles. In addition, the simulation
results are contrasted and analyzed.
Table 5.4 Comparison content
Measured
Data
Simulated
Data
Estimated
(Nationwide)
Estimated
(LA Annual)
Estimated
(LA Monthly)
Annual EUI
✓ ✓ ✓ ✓

Monthly EUI
✓ ✓

✓
For annual EUI, both nationwide annual model and city-based annual model for Los Angeles
were used to estimate EUI. Figure 5.15 shows the results and comparison between measured site
EUI and simulated EUI. The building consumed 47.78 kBtu/sf energy in 2010 which was the
highest within 5 years, and 40.71kBtu/sf energy in 2013 which was the lowest. All 5 annual EUI
estimates based on a nationwide model are greater than the results from the local model. The
error rates of 2010 and 2014 are 4.84% and 10.72% from the nationwide model, which are lower
than the error rates from the local model. However, the average error rate of local model is 10.96%
and it is lower than 15.99% of nationwide model. Compared to regression results, EUI from
simulation is based on a time period historic weather condition since it uses TMY2 weather data,
then simulated EUI doesn’t change with different years or weather conditions. Even though the
error rates for the 5 years are 6.22%, 1.85%, 5.20%, 13.42% and 16.37%, and the average error
rate is lower than two regression models, the constant energy consumption results by simulation
are more suitable to size a building’s mechanical system based on a time period weather
condition rather than to set am energy use baseline in a specific year. For further analysis,
normalized mean bias error (NMBE) and coefficient of variation of the root mean square error
(CVRMSE) were calculated based on the requirements from ASHRAE guideline for
measurement of energy and demand saving (ASHRAE 2002). For the annual results, the NMBE
values for simulation, nationwide regression and Los Angeles regression are 11.17%, -19.56%
and 1.7%, while the corresponding CVRMSE values are 12.05%, 19.11% and 13.57%. The
results indicate that Los Angeles regression have the lowest error and stability to estimate annual
energy use. Nationwide regression model is the relatively least accurate method by comparison.
116

Figure 5.15 Annual EUI results comparison
Monthly EUI was calculated through both simulation method and local EUI estimation model in
the Los Angeles area. The results are listed in table 5.5. The simulated result is still constant no
matter in which specific year while the regression model estimates 5-year monthly EUI data
corresponding to measured site EUI. From 2010 to 2013, the highest monthly energy was used
either in December or January while in 2014 the highest energy was used in September. This is
not only about heating and cooling demand for each year’s weather condition, but also about the
percentage of space occupied and used by people. By comparison, simulated monthly EUI is
steadier for each month and the highest value happened in August. Similar to real data, estimated
EUI from the regression model also indicates that the highest energy is used in winter.
Table 5.5 Monthly EUI results and comparison
Source Site EUI (kBtu/sf) Regression Estimated EUI (kBtu/sf) Simulation
Month 2010 2011 2012 2013 2014 2010 2011 2012 2013 2014 Constant
JAN 4.91 4.35 3.61 3.95 3.48 3.87 3.94 4.05 4.21 3.64 3.64
FEB 4.06 4.18 3.48 3.95 3.28 3.79 4.21 4.07 4.06 3.73 3.24
MAR 4.44 3.72 3.89 3.48 3.61 3.68 3.89 4.10 3.67 3.61 3.44
APR 4.63 3.96 3.76 3.36 3.39 3.68 3.68 3.74 3.60 3.64 3.07
MAY 3.69 3.88 3.36 2.77 3.54 3.59 3.61 3.60 3.58 3.58 3.03
JUN 3.38 3.38 3.06 2.65 3.20 3.58 3.59 3.59 3.58 3.58 3.16
JUL 3.44 3.77 3.40 3.23 3.92 3.58 3.58 3.60 3.58 3.57 3.52
AUG 3.51 3.01 3.40 2.98 3.27 3.57 3.58 3.57 3.58 3.57 3.63
SEP 3.88 3.85 3.37 3.72 4.04 3.57 3.59 3.57 3.57 3.57 3.40
OCT 3.58 4.06 3.22 2.96 3.62 3.58 3.59 3.58 3.59 3.57 3.12
NOV 4.05 4.14 3.79 3.81 3.75 3.92 3.89 3.65 3.61 3.59 3.11
DEC 4.21 3.86 3.82 3.85 3.51 3.97 4.46 4.13 4.01 3.96 3.58
Figure 5.16 presents the comparison of monthly EUI and average error rate from each
calculation method. The average monthly error rate is 11.91% of the simulation and 8.88% of the
regression value. Table 5.6 also lists all the error rates for 60 months with measured EUI. The
yearly average error rates are 8.05%, 8.3%, 8.22%, 12.63% and 7.22% for the regression model
and 16.3%, 15.91%, 7.41%, 11.18% and 8.77% for the simulation. Regression is still more
reliable to predict monthly energy use.
117

Figure 5.16 Monthly EUI results comparison
Table 5.6 Error rates of monthly EUI results
Error Rate Regression Simulation
Month 2010 2011 2012 2013 2014 2010 2011 2012 2013 2014
Jan 21.21% 9.55% 12.32% 6.50% 4.59% 25.97% 16.49% 0.85% 7.96% 4.41%
Feb 6.59% 0.72% 17.04% 2.90% 13.65% 20.17% 22.40% 6.83% 17.87% 1.19%
Mar 17.00% 4.61% 5.44% 5.72% 0.20% 22.36% 7.47% 11.44% 0.91% 4.49%
Apr 20.55% 7.04% 0.35% 7.31% 7.23% 33.78% 22.52% 18.39% 8.69% 9.60%
May 2.76% 7.02% 7.13% 29.40% 1.11% 17.87% 21.80% 9.71% 9.49% 14.35%
Jun 5.99% 6.31% 17.25% 35.06% 11.87% 6.36% 6.25% 3.38% 19.35% 1.21%
Jul 3.93% 5.14% 5.70% 10.55% 8.93% 2.35% 6.57% 3.47% 8.91% 10.17%
Aug 1.82% 18.99% 5.01% 19.82% 9.29% 3.53% 20.72% 6.91% 21.76% 11.25%
Sep 7.83% 6.84% 5.81% 3.95% 11.61% 12.30% 11.66% 0.82% 8.63% 15.76%
Oct 0.08% 11.61% 10.96% 21.13% 1.28% 12.81% 23.19% 3.19% 5.39% 13.81%
Nov 3.15% 6.09% 3.47% 5.09% 4.21% 23.17% 24.82% 17.82% 18.25% 16.99%
Dec 5.68% 15.67% 8.17% 4.11% 12.63% 14.88% 7.10% 6.05% 6.95% 2.04%
Average 8.05% 8.30% 8.22% 12.63% 7.22% 16.30% 15.91% 7.41% 11.18% 8.77%
Table 5.7 NMBE and CVRMSE for monthly EUI
NMBE (Simulation) NMBE (Regression) CVRMSE (Simulation) CVRMSE (Regression)
2010 17.86% 7.72% 21.34% 12.83%
2011 14.64% 1.31% 18.7% 9.55%
2012 5.68% -8.01% 10.29% 9.69%
2013 2.02% -10.55% 13.19% 14.46%
2014 6.79% -2.57% 11.32% 8.87%
Average 9.39% 6.03% 14.97% 11.08%
Further analysis was done by calculating NMBE and CVRMSE for monthly EUI results. The
NMBE from regression results are 7.72%, 1.31%, -8.01%, -10.55% and -2.57% in 2010, 2011,
2012, 2013 and 2014. It is better than the simulation results’ errors, which is also proven by
CVRMSE index too. More importantly, the constant EUI calculated by simulation can’t
represent the annual/monthly energy use change in reality, which is the biggest advance of
regression models.

118

Chapter 6 Conclusions of Study
This study was aiming at setting a more specific and accurate energy use baseline for buildings
which pursue energy reduction in the future. Instead of detailed building energy simulation, three
regression models using building façade features were used to estimate building Energy Use
Intensity (EUI) at different scales, including nationwide annual model, city-based annual models
for New York City and Los Angeles, monthly model for Los Angeles, and monthly
heating/cooling model for Los Angeles. The three regression models are Multiple Linear
Regression (MLR), Stepwise Regression and Partial Least Square (PLS). The best model was
selected by results and impact comparison. After model validation, the advantage by using a
regression model was presented through the analysis of current baseline and the optimal stepwise
model. In addition, the most significant façade features, which have the greatest impact on
building energy use, were determined and compared for each city. Lastly, a typical office
building in downtown Los Angeles with 5 years of measured monthly energy data was used to
verify the accuracy and feasibility of the 3 different regression models, by comparing the EUI
results from measured data, simulated data and estimated data.
6.1 Current Baseline Problem
Current building energy conservation action, like Architecture 2030, uses national/regional
median energy use as the baseline to promote energy reduction year by year until 2030 to get to a
carbon-neutral target. The baseline is determined by the national building energy survey
(CBECS) and expressed as kBtu/sf per year as the key metric. However, these average or median
baselines could not represent the real energy use for a specific building, because it doesn’t
consider the yearly local weather condition and particular building characteristics. In addition,
the current baseline is from CBECS 2003 survey data which is also not accurate enough to
represent newly constructed buildings. A median indicates that 50% of total reported buildings
have higher energy use than the single baseline, which will result in a vague target for building
owners or designers. On the other hand, computer simulation is another alternative to estimate
building energy use, but the simulation method is very time consuming and the accuracy highly
depends on the software users’ proficiency and available details of building construction,
operation schedule and system information. It is difficult to estimate the real building energy use,
119

especially for new constructions when there is no detailed information available for complicated
modeling. Besides, it is not necessary to estimate the perfect energy use to set the baseline and
reasonable error rate is acceptable.
6.2 5 Regression Models at Different Scales and Best Model Selection
To develop a package of building EUI estimation models by using façade features, 2 types of
data are required and need to be collected: energy use data and building façade data. Firstly, 10
office buildings with reported annual energy use data (7 of them have monthly energy use data)
were collected from other researchers. 188 nationwide buildings with annual energy information
from internet and open resources were gathered for the nationwide model development.
Secondly, for cities which committed to disclose their building energy benchmarking data after
2009, 25 office buildings in Manhattan, New York City were selected from the benchmarking
database. Lastly, a 16-floor office building in down Los Angeles with 5-year detailed monthly
electricity and natural gas use data was used as a verification case. All of existing, benchmarking
and monitoring data resources above constituted the whole energy database for developing
regression models.
Another type of gathered data is about building façade features, including height, floors,
orientation, operable window, volume, window-to-wall ratio, window area, façade area, site area,
floor area, volume-to-façade area ratio, volume-to-site area ratio, façade area-to-site area ratio,
etc. SketchUp, Google Earth and manual reading were used to collect geometric data. In addition,
HDD and CDD were collected as weather conditions. Built year was added as the condition of
code impact, and adjacency was used to consider the surrounding context factor.
120

Figure 6.1 Façade features reading illustration. Photo: (Free-wapaperbase 2015)
3 regression models were used to estimate building energy use, including Multiple Linear
Regression (MLR), Stepwise Regression and Partial Least Square (PLS). After comparison and
analysis, Stepwise regression was selected to be the optimal estimation model, because it could
not only predict EUI accurately by a minimum number of required data points, but also
determine the most important façade features in different regions or at different scales. As a
result, 5 regression models were developed to estimate nationwide building annual energy use,
annual energy use in New York City and Los Angeles, monthly energy use in Los Angeles,
monthly heating and cooling energy use. The R
2
indexes which indicate how well the regression
line approximates the real data for each model are: 53.73% (nationwide), 88.15% (NYC annual),
93.39% (LA annual), 91.59% (LA monthly), 90.8% and 85.44% (heating and cooling). Due to
less inefficient data numbers, nationwide annual EUI model had the lowest accuracy while others
showed preferable predictive ability.
6.3 Principle Façade Features and Model Verification
By using stepwise regression, two models were developed to estimate office building energy use
in New York City and Los Angeles. By removing the least important predictor while keeping the
accuracy in each step, the stepwise regression models could determine the most significantly key
façade features which have the greatest impact on building energy use in each location.
The common significant façade features in both cities are height, window-to-wall ratio,
orientation and floor area. Other most important façade features for regression in New York City
are operable windows, volume-to-site area, façade area-to-site area, heating degree day, south
and west façade area. Other most important façade features for regression in Los Angeles are
121

built year, volume, volume-to-façade area and cooling degree day. For office building annual
energy use estimation in the future, these façade features should be the priority to collect and in
pre-design phase they also should be taken into serious consideration.
An office building in downtown Los Angeles was used to verify nationwide annual energy
estimation models, annual and monthly energy use models in Los Angeles. The outcome
indicated that EUI estimated by regression models are dynamic according to annual weather
conditions and the error rates were lower on average, when compared to simulation results.
6.4 Contribution to Benchmarking and ARCH2030
Energy Use Intensity (EUI) is an important metric to benchmark building energy use for
comparison with existing counterpart buildings and newly constructed buildings. It is greatly
helpful to set a reasonable baseline to reduce energy use in both pre-design phase and retrofit.
Specifically, for new construction, EUI estimation models could be used to assess basic façade
design decisions as well as determine a more “customized” baseline. For existing buildings, it is
useful to estimate EUI efficiently and quickly only by inputting minimum amounts of data,
which will result in cost and time saving, especially at an upper urban planning level it is
difficult to simulate every single building or obtain the permission of energy use historic data
entry. In addition, the most important façade features in a certain area could be determined.
6.5 Scope Limit of Regression Models
The limitation of the simplified EUI estimation model is the limited range of application. Firstly,
the assumption that other important factors could be incorporated into “built year”, was not
sufficient enough to take every code requirements into consideration. This is reflected by the fact
that among 5 developed regression models, 4 of them have eliminated “built year” after stepwise
selection. Other factors also have great influence on building energy use, such as envelope
thermal properties, HVAC system efficiency, lighting fixtures, and building system use profiles,
etc. Secondly, all buildings in this research are office or bank buildings which have a similar
operation schedule. Five developed models can’t be used to estimate other types of buildings.
For green buildings, since they use more high efficiency systems and renewable energy, it has
less accuracy to estimate building energy use only based on façade features. More information
122

about systems and renewable energy might be considered to incorporate into regression models.
Lastly, except for the nationwide annual EUI model, the other 4 models didn’t consider the
façade and mechanical retrofit into potential predictors. It resulted in the less accuracy by only
considering “built year” without “retrofit year” and “retrofit contents”.
The analysis of correlation between individual façade feature and building energy use illustrated
that most façade features had no simple linear impact on building energy use since most data
distributions were overlapped. The exceptions are weather condition for nationwide annual EUI
model; WWR, operable window, floor area and building height for annual EUI model in New
York City; orientation and CDD for annual EUI model in Los Angeles. The mean value annual
site EUI can generally tell the basic trend of the correlation between individual façade feature
and building energy use for the predictive datasets, but more importantly, multiple linear
regression analysis was more helpful to summarize the complicated impact of multiple façade
features on the building energy use.
For nationwide annual EUI model, the lower accuracy resulted from the limited raw datasets, the
large range of buildings’ height/floors and the large proportion of LEED certified buildings. The
regression results showed the site EUI is more sensitive to floors, operable window, HDD/CDD,
climate zone and LEED certification level.
For annual EUI model in New York City, all buildings are high-rise offices that limited the
model application range. It is more accurately useful for EUI estimation of tall office buildings in
urban financial area. On the other hand, annual EUI model in Los Angeles is less accurate since
the database is smaller and part of it included a single building with multiple years’ energy use
information, which can’t reflect the impact of façade features on local building energy use
(especially when there was no façade retrofit recorded).
For monthly EUI model in Los Angeles, only 5 buildings with at least 2 years’ energy use data
were used to develop the regression model. Even though the most indicator R
2
(predictive) is
90.96%, more raw data are still needed to improve the accuracy and range of application. Both
HDD and CDD are sensitive to monthly energy use because compare to annual weather
condition change, monthly weather condition varies more significantly year by year, which has a
great impact of cooling/heating demand and building energy use.
123

For monthly heating/cooling EUI model, more detailed submetering data are needed to improve
the estimation model, because it’s more about detailed mechanical system choice, operation
schedule and efficiency curve.
124

Chapter 7 Suggestions for Future Work
7.1 Database Enlargement
The productivity of regression models depends highly on the building database size. It would be
more accurate to estimate building EUI by a regression model developed from a larger database,
in order to avoid unnecessary contingency. Other than encouraging more building owners to
disclose their current energy use data, more awareness should be gained at the government level
too. More cities should be encouraged to adopt building energy benchmarking law and disclose
energy use data to public. To develop more detailed energy estimation models (like for heating
and cooling), more datasets should be collected from real buildings especially from buildings
with sub-metering systems, which could record heating and cooling energy use separately.
7.2 Vision-based Façade Features Reading Technique
A building’s façade features, including geometric dimensions and other factors (like Window-to-
wall ratio), are key components of the energy performance, and potential factors which could be
improved in the early-design stage to reduce building energy use. Currently, for the collection of
data of façade features, researchers rely heavily on manual processes, including physical
measurements and visual reading of accessible construction drawing files. Considering the
numerous buildings potentially for planning energy enhancement strategies in the form of retro-
commissioning strategies, it is crucial to effectively collect data of facade features that cover
whole-scale energy monitoring and assessment of each building. One observation is that, in the
scalable collection of façade features, we can use computational modeling technologies for 3D
building structures in an urban areas using a detailed 3D structure of each façade derived from
inhomogeneous imagery data sources.

Figure 7.1 Vision-based façade features reading illustration
125

In the domain of Computer Vision, an aerial view approach has the advantage of providing
building textures without any extra cost over the Light Detection and Ranging sensor (LIDAR)-
based methods. However, the textures of facades obtained from aerial views are of low-
resolution or invisible while those of building roofs are clearly visible (Lee, Jung, and Nevatia
2002). To get accurate façade textures, we need to utilize images taken from ground level or
rooftop heights. The ground view images often require a camera calibration process. This new
technique could highly improve the efficiency of data collection for regression models
development at different levels.
7.3 Regression Models Considering More Factors
To increase the accuracy of regression models for different purposes, more factors should be
added into the basic predictors. This step could also consider regression models at different
levels. For example, a particular EUI estimation model should be developed only for buildings
with renewable energy and high performance mechanical system. Other potential factors include
building retrofit information, basic system efficiency, occupancy rate, operation hours, etc.

126

Reference
2030 Inc. 2011. “The 2030 Challenge.” http://architecture2030.org/about/design_faq.
“About Commercial Buildings Energy Consumption Survey.” 2012.
http://www.eia.gov/consumption/commercial/about.cfm.
Aksamija, Ajla. 2013. Sustainable Facades Design Methods for High-Performance Building
Envelopes.
Al-tamimi, Nedhal Ahmed M, Sharifah Fairuz, Syed Fadzil, and Adel Abdullah. 2009. “THE
EFFECT OF ORIENTATION AND GLAZED AREA TO THE INDOOR AIR
TEMPERATURE IN UNVENTILATED BUILDINGS IN HOT-HUMID” (December):
424–433.
Andrews, Clinton J., and Uta Krogmann. 2009. “Technology Diffusion and Energy Intensity in
US Commercial Buildings.” Energy Policy 37 (2) (February): 541–553.
doi:10.1016/j.enpol.2008.09.085.
http://linkinghub.elsevier.com/retrieve/pii/S0301421508005636.
Architecture 2030. 2011. “The 2030 Challenge.”
http://architecture2030.org/2030_challenge/the_2030_challenge.
———. 2014. “Road Map to Zero Emissions.”
ARchitecture, CLEAR comfortable Low Energy. 2014. “Building Shape-Surface Area to
Volume Ratio.” http://www.new-
learn.info/packages/clear/thermal/buildings/configuration/surcafeareato_vol_ratio.html.
Asadi, Somayeh, Shideh Shams, and Mohammad Mottahedi. 2014. “On the Development of
Multi-Linear Regression Analysis to Assess Energy Consumption in the Early Stages of
Building Design.” Energy & Buildings 85: 246–255. doi:10.1016/j.enbuild.2014.07.096.
http://dx.doi.org/10.1016/j.enbuild.2014.07.096.
ASHRAE. 2002. “Measurement of Energy and Demand Savings.” Vol. 8400.
———. 2013. “About Standard 90.1.” https://www.ashrae.org/resources--
publications/bookstore/standard-90-1.
Behsh, Basam. 2002. “BUILDING FORM AS AN OPTION FOR ENHANCING THE.”
Building Physicas 18: 759–766.
Bigee. 2012. “Regulations for Energy Efficiency of Building Design and Operation in New and
Refurbished Buildings (MEPS China).”
http://www.bigee.net/en/policy/guide/buildings/policy_examples/26/.
127

BizEE. 2014. “BizEE Degree Days Weather Data for Energy Professionals.”
http://www.degreedays.net/.
Bolin Rob. 2014. “Sustainability of the Building Envelope.”
http://www.wbdg.org/resources/env_sustainability.php.
BuildingGreen. 2014. “LEED NC-v2.2 EAp2: Minimum Energy Performance.” LEEDuser.
http://www.leeduser.com/credit/NC-v2.2/EAp2.
California Energy Commision. 2012. 2013 Building Energy Efficiency Standards For Residential
and Nonresidential Buildings.
California Energy Commission. 1978. “1978 Regulations Establishing Energy Conservation
Standards for New Residential and New Nonresidential Buildings.”
Carpenter, Stephen. 2014. “Operable Windows for Offices.”
http://www.greenglobes.com/advancedbuildings/_frames/fr_t_building_operable_windows.
htm.
Castree, Noel, Rob Kitchin, and Alisdair Rogers. 2013. A Dictonary of Human Geography.
Oxford University Press. doi:10.1093/acref/9780199599868.001.0001.
Catalina, Tiberiu, Joseph Virgone, and Eric Blanco. 2008. “Development and Validation of
Regression Models to Predict Monthly Heating Demand for Residential Buildings.” Energy
and Buildings 40 (10) (January): 1825–1832. doi:10.1016/j.enbuild.2008.04.001.
http://linkinghub.elsevier.com/retrieve/pii/S0378778808000844.
CBECS. 2007. “Total Energy Consumption by Major Fuel for Non-Mall Buildings , 2003”
(October 2006): 1–269.
Chung, William, Y.V. Hui, and Y. Miu Lam. 2006. “Benchmarking the Energy Efficiency of
Commercial Buildings.” Applied Energy 83 (1) (January): 1–14.
doi:10.1016/j.apenergy.2004.11.003.
http://linkinghub.elsevier.com/retrieve/pii/S0306261904002028.
Cleveland, Cutler J. 2009. “‘Building Envelope (HVAC).’” Dictionary of Energy.
Council on Tall Buildings and Urban Habitat. “CTBUH Height Criteria.”
http://www.ctbuh.org/TallBuildings/HeightStatistics/Criteria/tabid/446/language/en-
US/Default.aspx.
Cox, Matt, Marilyn a Brown, and Xiaojing Sun. 2013. “Energy Benchmarking of Commercial
Buildings: A Low-Cost Pathway toward Urban Sustainability.” Environmental Research
Letters 8 (3) (September 1): 035018. doi:10.1088/1748-9326/8/3/035018.
http://stacks.iop.org/1748-
9326/8/i=3/a=035018?key=crossref.91d32ac6323854cae5b985bf4390e268.
128

Crawley, Drury B., Jon W. Hand, Michaël Kummert, and Brent T. Griffith. 2008. “Contrasting
the Capabilities of Building Energy Performance Simulation Programs.” Building and
Environment 43 (4) (April): 661–673. doi:10.1016/j.buildenv.2006.10.027.
http://linkinghub.elsevier.com/retrieve/pii/S0360132306003234.
CTBUH. 2013. “Criteria for the Defining and Measuring of Tall Buildings.”
Daly, Alan. 2002. “Operable Windows and HVAC Systems.” Heating/Piping/Air Conditioning
Engineering 74 (12): 22–30.
Department of Energy. 2011. “Buildings Share of Electricity Consumption/Sales.” Buildings
Energy Data Book.
Description, Building Use, and Median Site. 2012. “2030 CHALLENGE Targets : U . S .
National Medians”: 2–3.
Eddy, William F, and Krisztina Marton. 2012. Effective Tracking of Building Energy Use :
Improving the Commercial Buildings and Residential Energy Consumption Surveys.
National Academies Press.
Ekici, Betul Bektas, and U. Teoman Aksoy. 2009. “Prediction of Building Energy Consumption
by Using Artificial Neural Networks.” Advances in Engineering Software 40 (5) (May):
356–362. doi:10.1016/j.advengsoft.2008.05.003.
http://linkinghub.elsevier.com/retrieve/pii/S0965997808001105.
Energy Star. 2011. “DataTrends-Energy Use Benchmarking.”
———. 2014a. “Energy Use Intensity (EUI).” Energy Star.
http://www.energystar.gov/buildings/facility-owners-and-managers/existing-buildings/use-
portfolio-manager/understand-metrics/what-energy.
———. 2014b. “U . S . Energy Use Intensity by Property Type.”
———. 2014c. “The New ENERGY STAR Portfolio Manager.”
http://www.energystar.gov/buildings/facility-owners-and-managers/existing-buildings/use-
portfolio-manager/new-energy-star-portfolio-manager.
Environment, San Francisco Department of. 2014. “Benchmarking Overview.”
http://www.sfenvironment.org/article/benchmarking/benchmarking-overview.
EPA. 2011. “DataTrends Benchmarking and Energy Savings” (2008).
———. 2014a. “EPA’s Target Finder Calculator.” http://www.energystar.gov/buildings/service-
providers/design/step-step-process/evaluate-target/epa’s-target-finder-calculator.
129

———. 2014b. “Climate Change-Commercial and Residential Sector Emissions.”
http://www.epa.gov/climatechange/ghgemissions/sources/commercialresidential.html.
EPBD. 2014. “Towards 2020-Nearly Zero-Energy Buildings.” http://www.epbd-
ca.eu/themes/nearly-zero-energy.
Filippín, C., F. Ricard, and S. Flores Larsen. 2013. “Evaluation of Heating Energy Consumption
Patterns in the Residential Building Sector Using Stepwise Selection and Multivariate
Analysis.” Energy and Buildings 66 (November): 571–581.
doi:10.1016/j.enbuild.2013.07.054.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813004453.
GGBP. 2013. “2013 NEW YORK CITY LOCAL LAW 84 BENCHMARKING REPORT”
(September).
Gronadzik, Walter, Alison Kwok, Benjamin Stein, and John Reynolds. 2010. Mechanical and
Electrical Equipment for Buildings. 11th ed. WILEY.
Guzowski, Mary. 2003. “The ‘ Costs ’ of Operable Windows : On the Question of Operable
Windows in Cold Climate Design.” Proceesing of the Environmental Design Research
Association Conference (June).
H. Borgstein, Edward, and Roberto Lamberts. 2014. “Developing Energy Consumption
Benchmarks for Buildings: Bank Branches in Brazil.” Energy and Buildings 82 (October):
82–91. doi:10.1016/j.enbuild.2014.07.028.
http://linkinghub.elsevier.com/retrieve/pii/S0378778814005659.
Höhne, By Niklas, Pieter Van Breevoort, Yvonne Deng, Julia Larkin, and Gesine Hänsel.
“Feasibility of GHG Emissions Phase-out by Mid-Century Feasibility of GHG Emissions
Phase-out by Mid-Century.”
Hunn, Bruce. 2010. “35 Years of Standard 90.1” (March): 36–46.
IBM. 2014. “SPSS Statistics.” http://www-
01.ibm.com/software/analytics/spss/products/statistics/.
ICC. 2014. “INTERNATIONAL CODE COUNCIL.” http://www.iccsafe.org/Pages/default.aspx.
IECC. 2012. “The 2012 International Energy Conservation Code.”
http://publicecodes.cyberregs.com/icod/iecc/2012/icod_iecc_2012_effectiveuse.htm.
ilesoft82. 2014. “100 Wall Street.”
https://3dwarehouse.sketchup.com/model.html?id=4a3716090fb927e24b98c52c576676c7.
IMT. 2014a. “Building Energy Performance Policy.” http://www.imt.org/policy/building-energy-
performance-policy.
130

———. 2014b. “U.S. Benchmarking Policy Landscape.”
http://www.buildingrating.org/graphic/us-benchmarking-policy-landscape.
Karatasou, S., M. Santamouris, and V. Geros. 2006. “Modeling and Predicting Building’s
Energy Use with Artificial Neural Networks: Methods and Results.” Energy and Buildings
38 (8) (August): 949–958. doi:10.1016/j.enbuild.2005.11.005.
http://linkinghub.elsevier.com/retrieve/pii/S0378778805002161.
Kumar, Rajesh, R K Aggarwal, and J D Sharma. 2013. “Estimation of Total Energy Load of
Building Using Artificial Neural Network.” Energy and Environmental Engineering 1 (2):
25–35. doi:10.13189/eee.2013.010201.
Lam, Joseph C., Sam C.M. Hui, and Apple L.S. Chan. 1997. “Regression Analysis of High-Rise
Fully Air-Conditioned Office Buildings.” Energy and Buildings 26 (2) (January): 189–197.
doi:10.1016/S0378-7788(96)01034-1.
http://linkinghub.elsevier.com/retrieve/pii/S0378778896010341.
Lee, Sung Chun Lee Sung Chun, Soon Ki Jung Soon Ki Jung, and R. Nevatia. 2002. “Integrating
Ground and Aerial Views for Urban Site Modeling.” Object Recognition Supported by User
Interaction for Service Robots 4: 107–112. doi:10.1109/ICPR.2002.1047411.
Li, Kangji, Hongye Su, and Jian Chu. 2011. “Forecasting Building Energy Consumption Using
Neural Networks and Hybrid Neuro-Fuzzy System: A Comparative Study.” Energy and
Buildings 43 (10) (October): 2893–2899. doi:10.1016/j.enbuild.2011.07.010.
http://linkinghub.elsevier.com/retrieve/pii/S0378778811003124.
Martin. 2010. “Architecture Photography.” YOU-ARE-HERE. http://www.you-are-
here.com/los_angeles/modern.html.
Marzria, Edward, and Kristina Kershner. 2009. “Meeting the 2030 Challenge Through Building
Codes.”
MAthWorks. 2014. “MATLAB.” http://www.mathworks.com/products/matlab/.
Mcleod, Rob, Kym Mead, and Mark Standen. 2014. “Passivhaus Primer : Designer ’ S Guide A
Guide for the Design Team and Local Authorities Passivhaus Primer – Designer ’ S Guide :
A Guide for the Design Team and Local Authorities.” BRE.
Milliken, Rebecca, and Betony Jones. “Office Building Benchmarking Guide Engaging the
Hard-to-Reach.”
Minitab. 2013a. Getting Started with Minitab 17.
———. 2013b. “What Is Partial Least Squares Regression.” http://support.minitab.com/en-
us/minitab/17/topic-library/modeling-statistics/regression-and-correlation/partial-least-
squares-regression/what-is-partial-least-squares-regression/.
131

———. 2014a. “Partial Least Squares Overview.”
———. 2014b. “Minitab 17.” http://www.minitab.com/en-us/.
Monfet, Danielle, Maria Corsi, Daniel Choinière, and Elena Arkhipova. 2014. “Development of
an Energy Prediction Tool for Commercial Buildings Using Case-Based Reasoning.”
Energy and Buildings 81 (October): 152–160. doi:10.1016/j.enbuild.2014.06.017.
http://linkinghub.elsevier.com/retrieve/pii/S037877881400499X.
NIST. 2012. “OMNITAB 80.” http://www.nist.gov/itl/sed/omnitab-80.cfm.
NYCECC. 2011. “LOCAL LAWS OF THE CITY OF NEW YORK A LOCAL LAW.” Vol. No.
1.
Program, Buildings Technologies, and Energy Efficiency. 2012. “2011 Buildings Energy Data
Book.”
Rosipal, Roman. “Nonlinear Partial Least Squares : An Overview”: 1–20.
Roth, Bill. 2013. “Building Code Revision Launches California Toward to Zero Net Energy
Buildings.” http://www.triplepundit.com/2013/11/title-24-launches-california-net-energy-
buildings/.
Roush, F.W. 1982. “Applied Linear Regression.” Mathematical Social Sciences 3 (1).
doi:10.1016/0165-4896(82)90010-5.
Rudolf Jakob, Freund, and Wilson William J. 2006. Regression Analysis: Statistical Modeling of
a Response Variable. 2nd ed. Academic Press.
S, Shebl. 2007. ““Effect of Window to Wall Ratio and Different Climate Conditions on Energy
Consumption for Residential and Commercial Buildings in Egypt.” HBRC Journal 3 (2):
115–126.
Samprit, Chatterjee, and Simonoff Jeffrey S. 2013. Wiley Handbooks in Applied Statistics:
Handbook of Regression Analysis. John Wiley & Sons.
Schwedler, Michael C A, Mark M Hydeman, Susanna S Hanson, Stephen V Skalko, Richard V
Heinisch, Ned B Heminger, John F Hogan, et al. 2010. ASHRAE STANDARD Energy
Standard for Buildings Except Low-Rise Residential Buildings. I-P.
Sewer, San Francisco Water Power. 2012. “2012 ENERGY BENCHMARKING REPORT San
Francisco Municipal Buildings” (September).
Shan, Rudai. 2014. “Optimization for Whole Building Energy Simulation Method in Façade
Design”: 1–9.
132

Shao, Jun. 1993. “Linear Model Selection by Cross-Validation.” Journal of the American
Statistical Association 88 (422): 486. doi:10.2307/2290328.
Soto Francés, Víctor Manuel, Emilio José Sarabia Escriva, José Manuel Pinazo Ojer, Emilie
Bannier, Vicente Cantavella Soler, and Gonzalo Silva Moreno. 2013. “Modeling of
Ventilated Façades for Energy Building Simulation Software.” Energy and Buildings 65
(October): 419–428. doi:10.1016/j.enbuild.2013.06.015.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813003538.
Star, E P A Energy. 2012. “DataTrends Energy Use Benchmarking”: 2011–2012.
Support Minitab. 2014. “Basics of Stepwise Regression.” http://support.minitab.com/en-
us/minitab/17/topic-library/modeling-statistics/regression-and-correlation/basics/basics-of-
stepwise-regression/.
Taylor, Courtney. 2014. “What Is the Difference between Alpha and P-Values.” Inferential
Statistics.
TGP. 2014. “Proposed Changes to ASHRAE 189.1 Denied.”
http://www.tgpamerica.com/understanding-wwr/.
Thalfeldt, Martin, Ergo Pikas, Jarek Kurnitski, and Hendrik Voll. 2013. “Facade Design
Principles for Nearly Zero Energy Buildings in a Cold Climate.” Energy and Buildings 67
(December): 309–321. doi:10.1016/j.enbuild.2013.08.027.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813005240.
Trimble. 2014. “3D Warehouse.” https://3dwarehouse.sketchup.com/.
Tronchin, Lamberto, and Kristian Fabbri. 2008. “Energy Performance Building Evaluation in
Mediterranean Countries: Comparison between Software Simulations and Operating Rating
Simulation.” Energy and Buildings 40 (7) (January): 1176–1187.
doi:10.1016/j.enbuild.2007.10.012.
http://linkinghub.elsevier.com/retrieve/pii/S0378778807002447.
U.S. Department of Commerce. 2010. “Census Regions and Divisions of the United States.”
U.S. DOE. 2011. “Whole Building Analysis: Energy Simulation.”
http://apps1.eere.energy.gov/buildings/tools_directory/subjects.cfm/pagename=subjects/pag
ename_menu=whole_building_analysis/pagename_submenu=energy_simulation.
———. 2013. “Building Science-Based Climate Maps.”
———. 2014a. “ANSI/ASHRAE/IES Standard 90.1-2013.”
http://www.energycodes.gov/regulations/determinations.
133

———. 2014b. “Building Energy Codes Program.”
https://www.energycodes.gov/adoption/states/new-york.
———. 2014c. “Weather Data Sources.” U.S. DOE.
http://apps1.eere.energy.gov/buildings/energyplus/weatherdata_sources.cfm#TMY2.
Violeta, Motuziene, and Juodis Egidijus Saulius. 2010. “Simulation Based Complex Energy
Assessment of Office Building Fenestration.” Journal of Civil Engineering and
Management 16 (3): 345–351.
“What Is Energy Use Intensity? ASHRAE Seeks to Define, Educate.” 2011. Targeted News
Service.
Yang, Jin, Hugues Rivard, and Radu Zmeureanu. 2005. “On-Line Building Energy Prediction
Using Adaptive Artificial Neural Networks.” Energy and Buildings 37 (12) (December):
1250–1259. doi:10.1016/j.enbuild.2005.02.005.
http://linkinghub.elsevier.com/retrieve/pii/S0378778805000502.
Yu, Zhun, Fariborz Haghighat, Benjamin C.M. Fung, and Hiroshi Yoshino. 2010. “A Decision
Tree Method for Building Energy Demand Modeling.” Energy and Buildings 42 (10)
(October): 1637–1646. doi:10.1016/j.enbuild.2010.04.006.
http://linkinghub.elsevier.com/retrieve/pii/S0378778810001350.
Zhai, Zhiqiang John, and James S. McNeill. 2013. “Roles of Building Simulation Tools in
Sustainable Building Design.” Building Simulation 7 (2) (December 21): 107–109.
doi:10.1007/s12273-013-0169-9. http://link.springer.com/10.1007/s12273-013-0169-9.
Zhang, J, R Athalye, and R Hart. 2013. “Energy and Energy Cost Savings Analysis of the IECC
for Commercial Buildings.” PNNL (22760).

134

Bibliography
2030 Inc. 2011. “The 2030 Challenge.” http://architecture2030.org/about/design_faq.
“About Commercial Buildings Energy Consumption Survey.” 2012.
http://www.eia.gov/consumption/commercial/about.cfm.
Aksamija, Ajla. 2013. Sustainable Facades Design Methods for High-Performance Building
Envelopes.
Al-tamimi, Nedhal Ahmed M, Sharifah Fairuz, Syed Fadzil, and Adel Abdullah. 2009. “THE
EFFECT OF ORIENTATION AND GLAZED AREA TO THE INDOOR AIR
TEMPERATURE IN UNVENTILATED BUILDINGS IN HOT-HUMID” (December):
424–433.
Andrews, Clinton J., and Uta Krogmann. 2009. “Technology Diffusion and Energy Intensity in
US Commercial Buildings.” Energy Policy 37 (2) (February): 541–553.
doi:10.1016/j.enpol.2008.09.085.
http://linkinghub.elsevier.com/retrieve/pii/S0301421508005636.
Architecture 2030. 2011. “The 2030 Challenge.”
http://architecture2030.org/2030_challenge/the_2030_challenge.
———. 2014. “Road Map to Zero Emissions.”
ARchitecture, CLEAR comfortable Low Energy. 2014. “Building Shape-Surface Area to
Volume Ratio.” http://www.new-
learn.info/packages/clear/thermal/buildings/configuration/surcafeareato_vol_ratio.html.
Asadi, Somayeh, Shideh Shams, and Mohammad Mottahedi. 2014. “On the Development of
Multi-Linear Regression Analysis to Assess Energy Consumption in the Early Stages of
Building Design.” Energy & Buildings 85: 246–255. doi:10.1016/j.enbuild.2014.07.096.
http://dx.doi.org/10.1016/j.enbuild.2014.07.096.
ASHRAE. 2002. “Measurement of Energy and Demand Savings.” Vol. 8400.
———. 2013. “About Standard 90.1.” https://www.ashrae.org/resources--
publications/bookstore/standard-90-1.
Barley, D, M Deru, S Pless, and P Torcellini. 2005. “Procedure for Measuring and Reporting
Commercial Building Energy Performance.” NREL Technical Report (October).
Behsh, Basam. 2002. “BUILDING FORM AS AN OPTION FOR ENHANCING THE.”
Building Physicas 18: 759–766.
135

Bigee. 2012. “Regulations for Energy Efficiency of Building Design and Operation in New and
Refurbished Buildings (MEPS China).”
http://www.bigee.net/en/policy/guide/buildings/policy_examples/26/.
BizEE. 2014. “BizEE Degree Days Weather Data for Energy Professionals.”
http://www.degreedays.net/.
Bolin Rob. 2014. “Sustainability of the Building Envelope.”
http://www.wbdg.org/resources/env_sustainability.php.
Bouyer, Julien, Christian Inard, and Marjorie Musy. 2011. “Microclimatic Coupling as a
Solution to Improve Building Energy Simulation in an Urban Context.” Energy and
Buildings 43 (7): 1549–1559. doi:10.1016/j.enbuild.2011.02.010.
http://dx.doi.org/10.1016/j.enbuild.2011.02.010.
BuildingGreen. 2014. “LEED NC-v2.2 EAp2: Minimum Energy Performance.” LEEDuser.
http://www.leeduser.com/credit/NC-v2.2/EAp2.
California Energy Commision. 2012. 2013 Building Energy Efficiency Standards For Residential
and Nonresidential Buildings.
California Energy Commission. 1978. “1978 Regulations Establishing Energy Conservation
Standards for New Residential and New Nonresidential Buildings.”
Calleja Rodríguez, Gloria, Antonio Carrillo Andrés, Fernando Domínguez Muñoz, José Manuel
Cejudo López, and Yi Zhang. 2013. “Uncertainties and Sensitivity Analysis in Building
Energy Simulation Using Macroparameters.” Energy and Buildings 67: 79–87.
doi:10.1016/j.enbuild.2013.08.009.
Carpenter, Stephen. 2014. “Operable Windows for Offices.”
http://www.greenglobes.com/advancedbuildings/_frames/fr_t_building_operable_windows.
htm.
Castree, Noel, Rob Kitchin, and Alisdair Rogers. 2013. A Dictonary of Human Geography.
Oxford University Press. doi:10.1093/acref/9780199599868.001.0001.
Catalina, Tiberiu, Joseph Virgone, and Eric Blanco. 2008. “Development and Validation of
Regression Models to Predict Monthly Heating Demand for Residential Buildings.” Energy
and Buildings 40 (10) (January): 1825–1832. doi:10.1016/j.enbuild.2008.04.001.
http://linkinghub.elsevier.com/retrieve/pii/S0378778808000844.
CBECS. 2007. “Total Energy Consumption by Major Fuel for Non-Mall Buildings , 2003”
(October 2006): 1–269.
Chung, William, Y.V. Hui, and Y. Miu Lam. 2006. “Benchmarking the Energy Efficiency of
Commercial Buildings.” Applied Energy 83 (1) (January): 1–14.
136

doi:10.1016/j.apenergy.2004.11.003.
http://linkinghub.elsevier.com/retrieve/pii/S0306261904002028.
Cianfrini, Claudio, Massimo Corcione, Emanuele Habib, and Alessandro Quintino. 2012.
“Energy Performance of a Lightweight Opaque Ventilated Facade Integrated with the
HVAC System Using Saturated Exhaust Indoor Air.” Energy and Buildings 50: 26–34.
doi:10.1016/j.enbuild.2012.03.014. http://dx.doi.org/10.1016/j.enbuild.2012.03.014.
Cleveland, Cutler J. 2009. “‘Building Envelope (HVAC).’” Dictionary of Energy.
Council on Tall Buildings and Urban Habitat. “CTBUH Height Criteria.”
http://www.ctbuh.org/TallBuildings/HeightStatistics/Criteria/tabid/446/language/en-
US/Default.aspx.
Cox, Matt, Marilyn a Brown, and Xiaojing Sun. 2013. “Energy Benchmarking of Commercial
Buildings: A Low-Cost Pathway toward Urban Sustainability.” Environmental Research
Letters 8 (3) (September 1): 035018. doi:10.1088/1748-9326/8/3/035018.
http://stacks.iop.org/1748-
9326/8/i=3/a=035018?key=crossref.91d32ac6323854cae5b985bf4390e268.
Crawley, Drury B., Jon W. Hand, Michaël Kummert, and Brent T. Griffith. 2008. “Contrasting
the Capabilities of Building Energy Performance Simulation Programs.” Building and
Environment 43 (4) (April): 661–673. doi:10.1016/j.buildenv.2006.10.027.
http://linkinghub.elsevier.com/retrieve/pii/S0360132306003234.
CTBUH. 2013. “Criteria for the Defining and Measuring of Tall Buildings.”
Daly, Alan. 2002. “Operable Windows and HVAC Systems.” Heating/Piping/Air Conditioning
Engineering 74 (12): 22–30.
Department of Energy. 2011. “Buildings Share of Electricity Consumption/Sales.” Buildings
Energy Data Book.
Description, Building Use, and Median Site. 2012. “2030 CHALLENGE Targets : U . S .
National Medians”: 2–3.
Eddy, William F, and Krisztina Marton. 2012. Effective Tracking of Building Energy Use :
Improving the Commercial Buildings and Residential Energy Consumption Surveys.
National Academies Press.
Ekici, Betul Bektas, and U. Teoman Aksoy. 2009. “Prediction of Building Energy Consumption
by Using Artificial Neural Networks.” Advances in Engineering Software 40 (5) (May):
356–362. doi:10.1016/j.advengsoft.2008.05.003.
http://linkinghub.elsevier.com/retrieve/pii/S0965997808001105.
Energy Star. 2011. “DataTrends-Energy Use Benchmarking.”
137

———. 2014a. “Energy Use Intensity (EUI).” Energy Star.
http://www.energystar.gov/buildings/facility-owners-and-managers/existing-buildings/use-
portfolio-manager/understand-metrics/what-energy.
———. 2014b. “U . S . Energy Use Intensity by Property Type.”
———. 2014c. “The New ENERGY STAR Portfolio Manager.”
http://www.energystar.gov/buildings/facility-owners-and-managers/existing-buildings/use-
portfolio-manager/new-energy-star-portfolio-manager.
Environment, San Francisco Department of. 2014. “Benchmarking Overview.”
http://www.sfenvironment.org/article/benchmarking/benchmarking-overview.
EPA. 2011. “DataTrends Benchmarking and Energy Savings” (2008).
———. 2014a. “EPA’s Target Finder Calculator.” http://www.energystar.gov/buildings/service-
providers/design/step-step-process/evaluate-target/epa’s-target-finder-calculator.
———. 2014b. “Climate Change-Commercial and Residential Sector Emissions.”
http://www.epa.gov/climatechange/ghgemissions/sources/commercialresidential.html.
EPBD. 2014. “Towards 2020-Nearly Zero-Energy Buildings.” http://www.epbd-
ca.eu/themes/nearly-zero-energy.
Farzana, Shazia, Meng Liu, Andrew Baldwin, and Md Uzzal Hossain. 2014. “Multi-Model
Prediction and Simulation of Residential Building Energy in Urban Areas of Chongqing,
South West China.” Energy and Buildings 81: 161–169. doi:10.1016/j.enbuild.2014.06.007.
http://dx.doi.org/10.1016/j.enbuild.2014.06.007.
Filippín, C., F. Ricard, and S. Flores Larsen. 2013. “Evaluation of Heating Energy Consumption
Patterns in the Residential Building Sector Using Stepwise Selection and Multivariate
Analysis.” Energy and Buildings 66 (November): 571–581.
doi:10.1016/j.enbuild.2013.07.054.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813004453.
GGBP. 2013. “2013 NEW YORK CITY LOCAL LAW 84 BENCHMARKING REPORT”
(September).
Ghadamian, Hossein, Mohammad Ghadimi, Mahdi Shakouri, Meisam Moghadasi, and
Mohammad Moghadasi. 2012. “Analytical Solution for Energy Modeling of Double Skin
Façades Building.” Energy and Buildings 50: 158–165. doi:10.1016/j.enbuild.2012.03.034.
http://dx.doi.org/10.1016/j.enbuild.2012.03.034.
Goia, Francesco, Lorenza Bianco, Marco Perino, and Valentina Serra. 2013. “Energy
Performance Assessment of and Advanced Integrated Façade through Experimental Data
Analysis.” SHC 2013, International Conference on Solar Heating and Cooling for
138

Buildings and Industry 00: 1262–1271. doi:10.1016/j.egypro.2014.02.143.
http://dx.doi.org/10.1016/j.egypro.2014.02.143.
Gowri, K, Ma Halverson, and Ee Richman. 2007. “Analysis of Energy Saving Impacts of
ASHRAE 90.1-2004 for the State of New York.” Richland, WA: Pacific Northwest …
(August). http://www.pnl.gov/main/publications/external/technical_reports/PNNL-
16770.pdf.
Gronadzik, Walter, Alison Kwok, Benjamin Stein, and John Reynolds. 2010. Mechanical and
Electrical Equipment for Buildings. 11th ed. WILEY.
Guzowski, Mary. 2003. “The ‘ Costs ’ of Operable Windows : On the Question of Operable
Windows in Cold Climate Design.” Proceesing of the Environmental Design Research
Association Conference (June).
H. Borgstein, Edward, and Roberto Lamberts. 2014. “Developing Energy Consumption
Benchmarks for Buildings: Bank Branches in Brazil.” Energy and Buildings 82 (October):
82–91. doi:10.1016/j.enbuild.2014.07.028.
http://linkinghub.elsevier.com/retrieve/pii/S0378778814005659.
Höhne, By Niklas, Pieter Van Breevoort, Yvonne Deng, Julia Larkin, and Gesine Hänsel.
“Feasibility of GHG Emissions Phase-out by Mid-Century Feasibility of GHG Emissions
Phase-out by Mid-Century.”
Huang, Jianen, Henglin Lv, Tao Gao, Wei Feng, Yanxia Chen, and Tai Zhou. 2014. “Thermal
Properties Optimization of Envelope in Energy-Saving Renovation of Existing Public
Buildings.” Energy and Buildings 75: 504–510. doi:10.1016/j.enbuild.2014.02.040.
http://dx.doi.org/10.1016/j.enbuild.2014.02.040.
Hunn, Bruce. 2010. “35 Years of Standard 90.1” (March): 36–46.
IBM. 2014. “SPSS Statistics.” http://www-
01.ibm.com/software/analytics/spss/products/statistics/.
ICC. 2014. “INTERNATIONAL CODE COUNCIL.” http://www.iccsafe.org/Pages/default.aspx.
IECC. 2012. “The 2012 International Energy Conservation Code.”
http://publicecodes.cyberregs.com/icod/iecc/2012/icod_iecc_2012_effectiveuse.htm.
ilesoft82. 2014. “100 Wall Street.”
https://3dwarehouse.sketchup.com/model.html?id=4a3716090fb927e24b98c52c576676c7.
IMT. 2014a. “Building Energy Performance Policy.” http://www.imt.org/policy/building-energy-
performance-policy.
139

———. 2014b. “U.S. Benchmarking Policy Landscape.”
http://www.buildingrating.org/graphic/us-benchmarking-policy-landscape.
Karatasou, S., M. Santamouris, and V. Geros. 2006. “Modeling and Predicting Building’s
Energy Use with Artificial Neural Networks: Methods and Results.” Energy and Buildings
38 (8) (August): 949–958. doi:10.1016/j.enbuild.2005.11.005.
http://linkinghub.elsevier.com/retrieve/pii/S0378778805002161.
Kazanasmaz, Tuǧçe, Ilknur Erlalelitepe Uygun, Gülden Gökçen Akkurt, Cihan Turhan, and
Kenan Evren Ekmen. 2014. “On the Relation between Architectural Considerations and
Heating Energy Performance of Turkish Residential Buildings in Izmir.” Energy and
Buildings 72: 38–50. doi:10.1016/j.enbuild.2013.12.036.
Kumar, Rajesh, R K Aggarwal, and J D Sharma. 2013. “Estimation of Total Energy Load of
Building Using Artificial Neural Network.” Energy and Environmental Engineering 1 (2):
25–35. doi:10.13189/eee.2013.010201.
Lam, Joseph C., Sam C.M. Hui, and Apple L.S. Chan. 1997. “Regression Analysis of High-Rise
Fully Air-Conditioned Office Buildings.” Energy and Buildings 26 (2) (January): 189–197.
doi:10.1016/S0378-7788(96)01034-1.
http://linkinghub.elsevier.com/retrieve/pii/S0378778896010341.
Lee, Sung Chun Lee Sung Chun, Soon Ki Jung Soon Ki Jung, and R. Nevatia. 2002. “Integrating
Ground and Aerial Views for Urban Site Modeling.” Object Recognition Supported by User
Interaction for Service Robots 4: 107–112. doi:10.1109/ICPR.2002.1047411.
Li, Kangji, Hongye Su, and Jian Chu. 2011. “Forecasting Building Energy Consumption Using
Neural Networks and Hybrid Neuro-Fuzzy System: A Comparative Study.” Energy and
Buildings 43 (10) (October): 2893–2899. doi:10.1016/j.enbuild.2011.07.010.
http://linkinghub.elsevier.com/retrieve/pii/S0378778811003124.
Manfren, Massimiliano, Niccolò Aste, and Reza Moshksar. 2013. “Calibration and Uncertainty
Analysis for Computer Models - A Meta-Model Based Approach for Integrated Building
Energy Simulation.” Applied Energy 103: 627–641. doi:10.1016/j.apenergy.2012.10.031.
http://dx.doi.org/10.1016/j.apenergy.2012.10.031.
Martin. 2010. “Architecture Photography.” YOU-ARE-HERE. http://www.you-are-
here.com/los_angeles/modern.html.
Marzria, Edward, and Kristina Kershner. 2009. “Meeting the 2030 Challenge Through Building
Codes.”
MAthWorks. 2014. “MATLAB.” http://www.mathworks.com/products/matlab/.
140

Mcleod, Rob, Kym Mead, and Mark Standen. 2014. “Passivhaus Primer : Designer ’ S Guide A
Guide for the Design Team and Local Authorities Passivhaus Primer – Designer ’ S Guide :
A Guide for the Design Team and Local Authorities.” BRE.
Milliken, Rebecca, and Betony Jones. “Office Building Benchmarking Guide Engaging the
Hard-to-Reach.”
Minitab. 2013a. Getting Started with Minitab 17.
———. 2013b. “What Is Partial Least Squares Regression.” http://support.minitab.com/en-
us/minitab/17/topic-library/modeling-statistics/regression-and-correlation/partial-least-
squares-regression/what-is-partial-least-squares-regression/.
———. 2014a. “Partial Least Squares Overview.”
———. 2014b. “Minitab 17.” http://www.minitab.com/en-us/.
Monfet, Danielle, Maria Corsi, Daniel Choinière, and Elena Arkhipova. 2014. “Development of
an Energy Prediction Tool for Commercial Buildings Using Case-Based Reasoning.”
Energy and Buildings 81 (October): 152–160. doi:10.1016/j.enbuild.2014.06.017.
http://linkinghub.elsevier.com/retrieve/pii/S037877881400499X.
Mukhopadhyay, Jaya, Juan-Carlos Baltazar, Hyojin Kim, and Jeff S. Haberl. 2011.
“COMPARISON OF ASHRAE STANDARD 90.1, 189.1 AND IECC CODES FOR
LARGE OFFICE BUILDINGS IN TEXAS.” Energy Systems Laboratory: 1–14.
Mustafaraj, Giorgio, Dashamir Marini, Andrea Costa, and Marcus Keane. 2014. “Model
Calibration for Building Energy Efficiency Simulation.” Applied Energy 130: 72–85.
doi:10.1016/j.apenergy.2014.05.019. http://dx.doi.org/10.1016/j.apenergy.2014.05.019.
Neto, Alberto Hernandez, and Flávio Augusto Sanzovo Fiorelli. 2008. “Comparison between
Detailed Model Simulation and Artificial Neural Network for Forecasting Building Energy
Consumption.” Energy and Buildings 40: 2169–2176. doi:10.1016/j.enbuild.2008.06.013.
NIST. 2012. “OMNITAB 80.” http://www.nist.gov/itl/sed/omnitab-80.cfm.
NYCECC. 2011. “LOCAL LAWS OF THE CITY OF NEW YORK A LOCAL LAW.” Vol. No.
1.
Program, Buildings Technologies, and Energy Efficiency. 2012. “2011 Buildings Energy Data
Book.”
Rosipal, Roman. “Nonlinear Partial Least Squares : An Overview”: 1–20.
141

Roth, Bill. 2013. “Building Code Revision Launches California Toward to Zero Net Energy
Buildings.” http://www.triplepundit.com/2013/11/title-24-launches-california-net-energy-
buildings/.
Roush, F.W. 1982. “Applied Linear Regression.” Mathematical Social Sciences 3 (1).
doi:10.1016/0165-4896(82)90010-5.
Rudolf Jakob, Freund, and Wilson William J. 2006. Regression Analysis: Statistical Modeling of
a Response Variable. 2nd ed. Academic Press.
S, Shebl. 2007. ““Effect of Window to Wall Ratio and Different Climate Conditions on Energy
Consumption for Residential and Commercial Buildings in Egypt.” HBRC Journal 3 (2):
115–126.
Samprit, Chatterjee, and Simonoff Jeffrey S. 2013. Wiley Handbooks in Applied Statistics:
Handbook of Regression Analysis. John Wiley & Sons.
Schwedler, Michael C A, Mark M Hydeman, Susanna S Hanson, Stephen V Skalko, Richard V
Heinisch, Ned B Heminger, John F Hogan, et al. 2010. ASHRAE STANDARD Energy
Standard for Buildings Except Low-Rise Residential Buildings. I-P.
Sewer, San Francisco Water Power. 2012. “2012 ENERGY BENCHMARKING REPORT San
Francisco Municipal Buildings” (September).
Shan, Rudai. 2014. “Optimization for Whole Building Energy Simulation Method in Façade
Design”: 1–9.
Shao, Jun. 1993. “Linear Model Selection by Cross-Validation.” Journal of the American
Statistical Association 88 (422): 486. doi:10.2307/2290328.
Soto Francés, Víctor Manuel, Emilio José Sarabia Escriva, José Manuel Pinazo Ojer, Emilie
Bannier, Vicente Cantavella Soler, and Gonzalo Silva Moreno. 2013. “Modeling of
Ventilated Façades for Energy Building Simulation Software.” Energy and Buildings 65
(October): 419–428. doi:10.1016/j.enbuild.2013.06.015.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813003538.
Star, E P A Energy. 2012. “DataTrends Energy Use Benchmarking”: 2011–2012.
Support Minitab. 2014. “Basics of Stepwise Regression.” http://support.minitab.com/en-
us/minitab/17/topic-library/modeling-statistics/regression-and-correlation/basics/basics-of-
stepwise-regression/.
Susorova, Irina, Meysam Tabibzadeh, Anisur Rahman, Herek L. Clack, and Mahjoub Elnimeiri.
2013. “The Effect of Geometry Factors on Fenestration Energy Performance and Energy
Savings in Office Buildings.” Energy and Buildings 57: 6–13.
doi:10.1016/j.enbuild.2012.10.035. http://dx.doi.org/10.1016/j.enbuild.2012.10.035.
142

Taylor, Courtney. 2014. “What Is the Difference between Alpha and P-Values.” Inferential
Statistics.
TGP. 2014. “Proposed Changes to ASHRAE 189.1 Denied.”
http://www.tgpamerica.com/understanding-wwr/.
Thalfeldt, Martin, Ergo Pikas, Jarek Kurnitski, and Hendrik Voll. 2013. “Facade Design
Principles for Nearly Zero Energy Buildings in a Cold Climate.” Energy and Buildings 67
(December): 309–321. doi:10.1016/j.enbuild.2013.08.027.
http://linkinghub.elsevier.com/retrieve/pii/S0378778813005240.
Trimble. 2014. “3D Warehouse.” https://3dwarehouse.sketchup.com/.
Tronchin, Lamberto, and Kristian Fabbri. 2008. “Energy Performance Building Evaluation in
Mediterranean Countries: Comparison between Software Simulations and Operating Rating
Simulation.” Energy and Buildings 40 (7) (January): 1176–1187.
doi:10.1016/j.enbuild.2007.10.012.
http://linkinghub.elsevier.com/retrieve/pii/S0378778807002447.
U.S. Department of Commerce. 2010. “Census Regions and Divisions of the United States.”
U.S. DOE. 2011. “Whole Building Analysis: Energy Simulation.”
http://apps1.eere.energy.gov/buildings/tools_directory/subjects.cfm/pagename=subjects/pag
ename_menu=whole_building_analysis/pagename_submenu=energy_simulation.
———. 2013. “Building Science-Based Climate Maps.”
———. 2014a. “ANSI/ASHRAE/IES Standard 90.1-2013.”
http://www.energycodes.gov/regulations/determinations.
———. 2014b. “Building Energy Codes Program.”
https://www.energycodes.gov/adoption/states/new-york.
———. 2014c. “Weather Data Sources.” U.S. DOE.
http://apps1.eere.energy.gov/buildings/energyplus/weatherdata_sources.cfm#TMY2.
U.S. DOE. 2011. “Metering Best Practices A Guide to Achieving Utility Resource Efficiency.”
Vesterberg, Jimmy, Staffan Andersson, and Thomas Olofsson. 2014. “Robustness of a
Regression Approach , Aimed for Calibration of Whole Building Energy Simulation Tools.”
Energy & Buildings 81: 430–434. doi:10.1016/j.enbuild.2014.06.035.
http://dx.doi.org/10.1016/j.enbuild.2014.06.035.
Violeta, Motuziene, and Juodis Egidijus Saulius. 2010. “Simulation Based Complex Energy
Assessment of Office Building Fenestration.” Journal of Civil Engineering and
Management 16 (3): 345–351.
143

“What Is Energy Use Intensity? ASHRAE Seeks to Define, Educate.” 2011. Targeted News
Service.
Yang, Jin, Hugues Rivard, and Radu Zmeureanu. 2005. “On-Line Building Energy Prediction
Using Adaptive Artificial Neural Networks.” Energy and Buildings 37 (12) (December):
1250–1259. doi:10.1016/j.enbuild.2005.02.005.
http://linkinghub.elsevier.com/retrieve/pii/S0378778805000502.
Yu, Jinghua, Liwei Tian, Changzhi Yang, Xinhua Xu, and Jinbo Wang. 2013. “Sensitivity
Analysis of Energy Performance for High-Rise Residential Envelope in Hot Summer and
Cold Winter Zone of China.” Energy and Buildings 64: 264–274.
doi:10.1016/j.enbuild.2013.05.018. http://dx.doi.org/10.1016/j.enbuild.2013.05.018.
Yu, Zhun, Fariborz Haghighat, Benjamin C.M. Fung, and Hiroshi Yoshino. 2010. “A Decision
Tree Method for Building Energy Demand Modeling.” Energy and Buildings 42 (10)
(October): 1637–1646. doi:10.1016/j.enbuild.2010.04.006.
http://linkinghub.elsevier.com/retrieve/pii/S0378778810001350.
Zhai, Zhiqiang John, and James S. McNeill. 2013. “Roles of Building Simulation Tools in
Sustainable Building Design.” Building Simulation 7 (2) (December 21): 107–109.
doi:10.1007/s12273-013-0169-9. http://link.springer.com/10.1007/s12273-013-0169-9.
Zhang, J, R Athalye, and R Hart. 2013. “Energy and Energy Cost Savings Analysis of the IECC
for Commercial Buildings.” PNNL (22760).
Zhou, Zhi Hua, Shan Hu, and Tao Du. 2012. “Study on Determination of Best Window-Wall
Ratio of Office Building in Cold Area.” Applied Mechanics and Materials 260-261: 209–
216. doi:10.4028/www.scientific.net/AMM.260-261.209.
http://www.scientific.net/AMM.260-261.209.

144

Appendix
Table A. Database for annual EUI estimation model in New York City
Case Site EUI
(kBtu/sf)
Height
(ft)
Floor Built Year WWR
(%)
Orientation Operable
Window
Volume
(cf)
Window Area
(sf)
1 109 369 28 1965 45 2 0 4011417 55423
2 89 499 36 1979 33.9 4 0 5449080 64278
3 68 451 40 1989 47.5 4 0 3668882 87790
4 82 570 50 1983 24.5 2 0 6543600 62285
5 76 318 26 1970 32.88 2 0 7552182 63636
6 100 628 49 1972 73.11 2 0 9029766 221864
7 67 328 26 1955 55.49 2 0 7846672 129480
8 80 291 28 1983 30.79 2 0 10170450 61237
9 90 647 37 1984 30.54 2 0 15421892 109140
10 90 653 50 1986 64.94 2 1 14499477 245009
11 89 503 39 1961 33.27 1 1 17868477 128018
12 125 575 38 2003 48.08 4 0 21283342 246648
13 72 411 29 1916 36.46 4 1 15682527 113651
14 155 688 52 1967 46.36 4 0 15824000 186482
15 67 256 16 1934 25.84 4 1 13718634 70521
16 123 574 42 1967 38.54 4 0 16391978 152292
17 155 615 47 1971 60 4 0 18318254 256427
18 120 689 50 1974 83.47 4 1 20345923 420702
19 81 743 52 2006 76.33 1 0 31711983 469845
20 133 545 42 1960 63.28 4 0 22330536 251915
21 124 752 54 1985 59.39 4 0 18663766 270071
22 144 634 44 1969 60.89 4 0 22141465 306783
23 96 369 28 1965 45 2 0 4011417 55423
24 92 300 25 1970 48.36 1 0 3252462 44628
25 79 499 36 1979 33.9 4 0 5449080 64278
26 65 451 40 1989 47.5 4 0 3668882 87790
27 62 365 29 1969 71.92 2 1 5886720 121266
28 74 318 26 1970 32.88 2 0 7552182 63636
29 93 628 49 1972 73.11 2 0 9029766 221864
30 63 328 26 1955 55.49 2 0 7846672 129480
31 72 291 28 1983 30.79 2 0 10170450 61237
32 106 628 46 1971 70 4 1 11795724 277803
33 112 647 37 1984 30.54 2 0 15421892 109140
34 77 653 50 1986 64.94 2 1 14499477 245009
35 83 503 39 1961 33.27 1 1 17868477 128018
36 119 575 38 2003 48.08 4 0 21283342 246648
37 87 411 29 1916 36.46 4 1 15682527 113651
38 137 688 52 1967 46.36 4 0 15824000 186482
39 57 256 16 1934 25.84 4 1 13718634 70521
40 109 574 42 1967 38.54 4 0 16391978 152292
41 147 615 47 1971 60 4 0 18318254 256427
42 113 689 50 1974 83.47 4 1 20345923 420702
43 121 545 42 1960 63.28 4 0 22330536 251915
44 114 752 54 1985 59.39 4 0 18663766 270071
45 146 634 44 1969 60.89 4 0 22141465 306783
Table B Database for annual EUI estimation model in New York City (Continued)
Case Façade
Area (sf)
Site Area
(sf)
Floor Area
(sf)
V/FA V/SA FA/SA Adjacent
Building
HDD CDD
1 123162 17017 292627 32.57 235.73 7.24 4 3272 2018
2 189610 10920 358522 28.74 499.00 17.36 6 3272 2018
3 184822 11098 387406 19.85 330.59 16.65 0 3272 2018
4 254224 22083 487501 25.74 296.32 11.51 0 3272 2018
5 193540 25310 669211 39.02 298.39 7.65 0 3272 2018
6 303466 23100 762051 29.76 390.90 13.14 8 3272 2018
7 233339 29299 777658 33.63 267.81 7.96 0 3272 2018
8 198885 34000 802690 51.14 299.13 5.85 0 3272 2018
145

9 357368 35778 871353 43.15 431.04 9.99 0 3272 2018
10 377285 33347 1002327 38.43 434.81 11.31 0 3272 2018
11 384785 44757 1067151 46.44 399.23 8.60 1 3272 2018
12 512994 45839 1071012 41.49 464.31 11.19 8 3272 2018
13 311714 40363 1166100 50.31 388.54 7.72 8 3272 2018
14 402248 52636 1178137 39.34 300.63 7.64 0 3272 2018
15 272913 75460 1234704 50.27 181.80 3.62 0 3272 2018
16 395153 40166 1278463 41.48 408.11 9.84 8 3272 2018
17 427378 71874 1319496 42.86 254.87 5.95 0 3272 2018
18 504016 40992 1440000 40.37 496.34 12.30 0 3272 2018
19 615544 41224 1715800 51.52 769.26 14.93 0 3272 2018
20 398096 67344 1724226 56.09 331.59 5.91 8 3272 2018
21 454742 49802 1798779 41.04 374.76 9.13 0 3272 2018
22 503832 70471 1842494 43.95 314.19 7.15 0 3272 2018
23 123162 17017 292627 32.57 235.73 7.24 4 2988 1945
24 92283 12919 325000 35.24 251.76 7.14 7 2988 1945
25 189610 10920 358522 28.74 499.00 17.36 6 2988 1945
26 184822 11098 387406 19.85 330.59 16.65 0 2988 1945
27 168612 16128 426531 34.91 365.00 10.45 0 2988 1945
28 193540 25310 669211 39.02 298.39 7.65 0 2988 1945
29 303466 23100 762051 29.76 390.90 13.14 8 2988 1945
30 233339 29299 777658 33.63 267.81 7.96 0 2988 1945
31 198885 34000 802690 51.14 299.13 5.85 0 2988 1945
32 396861 21782 833901 29.72 541.54 18.22 2 2988 1945
33 357368 35778 871353 43.15 431.04 9.99 0 2988 1945
34 377285 33347 1002327 38.43 434.81 11.31 0 2988 1945
35 384785 44757 1067151 46.44 399.23 8.60 1 2988 1945
36 512994 45839 1071012 41.49 464.31 11.19 8 2988 1945
37 311714 40363 1166100 50.31 388.54 7.72 8 2988 1945
38 402248 52636 1178137 39.34 300.63 7.64 0 2988 1945
39 272913 75460 1234704 50.27 181.80 3.62 0 2988 1945
40 395153 40166 1278463 41.48 408.11 9.84 8 2988 1945
41 427378 71874 1319496 42.86 254.87 5.95 0 2988 1945
42 504016 40992 1440000 40.37 496.34 12.30 0 2988 1945
43 398096 67344 1724226 56.09 331.59 5.91 8 2988 1945
44 454742 49802 1798779 41.04 374.76 9.13 0 2988 1945
45 503832 70471 1842494 43.95 314.19 7.15 0 2988 1945
Note: 1. Height is measure from the level of the lowest, significant, open-air, pedestrian entrance to the finished floor level of the
highest occupied floor within the building (Council on Tall Buildings and Urban Habitat).
2. Floors refer to the total levels of a building which could be used by occupants.
3. Long axis along with North-South is quantified as 1, NE-SW is 2, E-W is 3, SE-NW is 4.
4. With operable window is quantified as 1, without operable window is quantified as 0.
5. No adjacent building is quantified as 0, while adjacent building on the north side is 1, others are clockwise defined by 2 to 8.
Other original database was from unpublished research and collected by the collaboration of
multiple researchers. Raw data will be disclosed when all different analysis is done and the final
data disclosure is agreed by each data owner.

Abstract (if available)

Abstract The commercial and residential building sector accounts for about 40% of carbon dioxide (CO₂) emissions in the United States per year, more than any other sector (Eddy and Marton 2012). The most significant factor contributing to CO₂ emissions from buildings is their use of electricity. Commercial and residential buildings are tremendous users of electricity (Department of Energy 2011), accounting for more than 73% of electricity use in the U.S. ❧ Energy use data from the Commercial Building Energy Consumption Survey (CBECS) is an average value based on the range of Heating Degree Days (HDD) and Cooling Degree Days (CDD), which can’t show the specific condition of each building category within one area. In addition, the average value is too general to evaluate if a specific building case is energy efficient or not. On the other hand, it is very time consuming to develop a simulation model in software, which also needs very detailed information about the building itself. The accuracy depends on how much specific information of envelop thermal conditions, mechanical system performance, occupancy level and schedule, etc. ❧ Among 3 main factors to influence building energy performance, building façade features are more easily obtained as opposed to building mechanical systems and schedule information. By using façade features, certain key attributes could be input to generate a customized baseline model and to estimate building energy use intensity (EUI). ❧ A simple regression model can be used to calculate the EUI baseline instead of complicated simulation tools, and the results are accurate and reasonable at an acceptable level. The calculated baseline can be used for setting a practical baseline for energy reduction target. Due to its simplicity and quick processing time, the research outcome would also be applicable to the real-time energy estimation of multiple buildings at an urban scale. ❧ This new method of linear regression analysis is developed to estimate building energy consumption just based on simple façade attributes and weather conditions. Building façade features, for example, including shading, window-to-wall ratio, orientation, surface-to-volume ratio, etc. are easy to obtain. It is meaningful to use a simple way to predict heating and cooling energy use instead of traditional energy performance simulation tool which is time and resource consuming. Based on collected building physical attribute data, statistical methods could be used to generate a customized baseline Energy Use Intensity (EUI) estimation model. The proposed idea will also adopt a simplified building energy performance prediction model as a function of architectural physical frames and their dynamic ambient environmental condition, such as monthly cooling/heating degree days. The main goal of this research is to develop a mathematical method to provide a customized baseline model for buildings, considering specific façade features and local climate condition. It will provide a direct estimation and prediction of project building energy performance to provide a reasonable baseline for designer, engineer and client.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Building energy performance estimation approach: facade visual information-driven benchmark performance model

PDF

Energy efficient buildings: a method of probabilistic risk assessment using building energy simulation

PDF

Energy performance of different building forms: HEED simulations of equivalent massing models in diverse building surface aspect ratios and locations in the US

PDF

A simplified building energy simulation tool: material and environmental properties effects on HVAC performance

PDF

Double skin façade in hot arid climates: computer simulations to find optimized energy and thermal performance of double skin façades

PDF

Mitigating thermal bridging in ventilated rainscreen envelope construction: Methods to reduce thermal transfer in net-zero envelope optimization

PDF

Facade retrofit: enhancing energy performance in existing buildings

PDF

Energy savings by using dynamic environmental controls in the cavity of double skin facades

PDF

A proposal for building envelope retrofit on the Bonaventure Hotel: a case study examining energy and carbon

PDF

The intelligent control strategy of kinetic façades for daylight and energy performance: evaluating the daylight effect of adaptive systems based on parametric workflow

PDF

Multi-domain assessment of a kinetic facade: determining the control strategy of a kinetic façade using BIM based on energy performance, daylighting, and occupants’ preferences; Multi-domain asse...

PDF

A parametric study of the thermal performance of green roofs in different climates through energy modeling

PDF

Developing a data-driven model of overall thermal sensation based on the use of human physiological information in a built environment

PDF

Kinetic facades as environmental control systems: using kinetic facades to increase energy efficiency and building performance in office buildings

PDF

Energy simulation in existing buildings: calibrating the model for retrofit studies

PDF

Occupant-aware energy management: energy saving and comfort outcomes achievable through application of cooling setpoint adjustments

PDF

Real-time simulation-based feedback on carbon impacts for user-engaged temperature management

PDF

An analysis of building component energy usage: a data driven approach to formulate a guideline

PDF

Bridging performance gaps by occupancy and weather data-driven energy prediction modeling using neural networks

PDF

District energy systems: Studying building types at an urban scale to understand building energy consumption and waste energy generation

Asset Metadata

Creator Yang, Chao (author)

Core Title Energy use intensity estimation method based on building façade features by using regression models

School School of Architecture

Degree Master of Building Science

Degree Program Building Science

Publication Date 04/23/2017

Defense Date 03/23/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag baseline model,energy use intensity,facade,OAI-PMH Harvest,regression

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Choi, Joon-Ho (committee chair), Noble, Douglas (committee member), Schiler, Marc (committee member)

Creator Email chaoyang@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-560624

Unique identifier UC11298574

Identifier etd-YangChao-3393.pdf (filename),usctheses-c3-560624 (legacy record id)

Legacy Identifier etd-YangChao-3393.pdf

Dmrecord 560624

Document Type Thesis

Format application/pdf (imt)

Rights Yang, Chao

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA