Energy and Buildings 110 (2016) 57–70

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Prediction of cooling energy use in buildings using an enthalpy-based

cooling degree days method in a hot and humid climate
Minjae Shin a,b , Sung Lok Do b,∗
a
Department of Architecture, College of Architecture, Texas A&M University, College Station, TX 77843, United States
b
Energy Systems Laboratory, Texas A&M Engineering Experiment Station, College Station, TX 77845, United States

a r t i c l e i n f o a b s t r a c t

Article history: The cooling degree days (CDD) method has widely been used for predicting the cooling energy consump-
Received 10 September 2015 tion of buildings with air conditioning. The conventional CDD values primarily are based on outdoor
Accepted 15 October 2015 dry-bulb temperatures that neglect the influence of latent heat on the total energy load. This study pro-
Available online 23 October 2015
poses an enthalpy-based CDD method to account for latent heat as well as sensible heat. Two institutional
buildings were used to examine the applicability and accuracy of the proposed enthalpy-based CDD, as
Keywords:
compared to the conventional temperature-based CDD. To achieve this examination, multiple linear
Building energy analysis
regression analyses were first performed to determine the base temperature/enthalpy points for each
Regression model
Cooling degree days
building during the different data periods (weekdays, weekends, and all week). Next, the CDD values
Base temperature/enthalpy were separately calculated using the determined base temperature/enthalpy points, and linear regres-
Latent load sion models were developed to predict the cooling energy use. Each individually-predicted cooling energy
use employing the temperature- and enthalpy-based CDD methods was then compared to the measured
cooling energy use. As a result, the comparison utilizing the enthalpy-based CDD method resulted in a
percent error of approximately 2% less than that of the temperature-based CDD method.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction such as DOE-2, eQUEST, EnergyPlus, TRNSYS, and ESP-r. On the

other hand, statistical methods have also been developed to predict
Energy consumption in buildings has been increasing glob- building energy use that use correlations among influencing vari-
ally, and associated concerns such as the impending exhaustion of ables such as weather data. Correlations are typically determined
energy resources and serious environmental impacts (e.g., ozone using a linear regression model that represents the relationship
layer depletion and climate change) have been raised [1–4]. The between the building’s historical energy use and the corresponding
reduction of building energy use has become one of the primary weather data (e.g., outdoor temperature or cooling/heating degree
topics of interest for most developed/developing countries [2]. days).
Various techniques and methods have been developed to predict When the past measured data of a building system’s perfor-
building energy use. The predicted results facilitate the analysis of mance and climate details are available, statistical methods have
energy use and general energy efficiency in buildings before and widely been used to predict building energy consumption; they
after construction and/or energy retrofits. Furthermore, the results tend to be simple and easily applicable to any type of building.
can be used to make decisions regarding how to minimize building The best known statistical methods used in the field of building
energy consumption, as well as how to improve overall building energy simulation are the regression method [8–13] and degree
energy performance. day method [14–16]. When developing relationships useful to an
Methods for predicting building energy use can be categorized analysis of building energy consumption, outdoor dry-bulb temper-
into two types [5–7]: engineering methods and statistical meth- atures are most often referenced as the input variable typically used
ods. Engineering methods generally use physical calculations as for both the regression and degree day methods. These methods
principles for building energy use. A number of computer simula- appear to be the most useful in temperature-based analyses that
tion programs based on engineering methods have been developed, account only for sensible heat. However, in order to better predict
cooling energy use in buildings, latent heat should be considered
in addition to sensible heat. Consequently, the authors believe that
∗ Corresponding author. Tel.: +1 979 220 7465; fax: +1 979 862 8687. an enthalpy-based analysis that includes latent heat can advanta-
E-mail address: [email protected] (S.L. Do). geously be used to predict the cooling energy use of a building,

http://dx.doi.org/10.1016/j.enbuild.2015.10.035
0378-7788/© 2015 Elsevier B.V. All rights reserved.
58 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

Fig. 1. Overall research methodology

especially buildings located in cooling-dominated areas in hot and degree day method, based on the outdoor dry-bulb temperature
humid climates. and enthalpy data. Both buildings were located on the Texas A&M
This study examines a building’s cooling energy use as it University—College Station campus. Texas is located in a cooling-
is predicted with outdoor dry-bulb temperature versus outdoor dominated climate, and the majority of annual energy consumption
enthalpy. To develop a mathematical linear relationship between stems from cooling energy use. Thus, a determination of the cool-
the cooling energy use and weather data (i.e., either temperature or ing degree days was important to accurately predicting the cooling
enthalpy in this study), the regression method and cooling degree energy use. The selected buildings had a variety of functions:
day method are used. Then, the developed relationships are used laboratory, office, and classroom. The heating, ventilation, and
to predict the building’s cooling energy use at a given condition air-conditioning (HVAC) systems in these buildings consisted of
of temperature or enthalpy. To evaluate the goodness-of-fit of the single-duct variable air volume (SDVAV) air handling units (AHUs).
predicted values against the measured values, this study uses sta- In addition, each building had a direct digital control (DDC) sys-
tistical indices such as the percent error, R-Squared, and Coefficient tem that operated the AHUs and pumps, while the terminal boxes
of Variation of Root Mean Square Error (CV-RMSE) values. were pneumatically controlled. All of the AHUs had pre-heating and
cooling coils controlled through the energy management control
2. Methodology system (EMCS). Further information about the case-study buildings
can be found in Table 1.
The case-study buildings and methodology of this analysis are
described in this section. Fig. 1 shows that this study is based on 2.2. Data collection: Weather and building energy consumption
actual energy consumption data, corresponding outside dry-bulb data
temperature data, and relative humidity data for one year. Enthalpy
values were calculated using dry-bulb temperature and relative A regression analysis is typically performed using the measured
humidity data. All of these measured and calculated data were used weather data available from hundreds of weather stations across
in a multiple linear regression analysis to estimate the base tem- the United States. The weather variables required for this analysis
perature and enthalpy in each case. Thereafter, these values were
used to calculate the cooling degree days for each of the case-study Table 1
buildings. The cooling energy uses of each of the case-study build- General information about the case-study buildings.

ings were predicted using a regression model with the estimated Building A Building B
cooling degree days, based on the temperature and enthalpy data.
Construction year 2011 2005
Finally, the predicted and measured cooling energy use were com- Building function Laboratory, office, Laboratory, office,
pared to each other in order to investigate which cooling degree classroom classroom
day method was more accurate. Conditioned area [m2 ] 13,928 9950
Number of floors 2 3
Occupancy schedule 08:00–17:00 24 h (laboratory)
2.1. The case-study buildings 06:00–22:00 (office)
06:00–21:00
Two case-study buildings were selected for use in evaluat- (classroom)
Building system SDVAV with reheat SDVAV with reheat
ing the cooling energy prediction procedure through the cooling
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 59

Table 2 Table 3
Summary of building energy consumption data. Regression models used in building energy simulations [8].

Building A Building B Regression Equation Examples of Examples of

model dependent independent
Data collection period Mar. 2012 through Feb. Apr. 2013 through Mar.
variables variables
2013 2014
Total number of days 334 days 362 days Mean or 1P E = Eb Whole building –
Missing days 31 days in Aug. 2012 3 days in Mar. 2014 model electric
2P model E = C + B1 (T) Chilled or hot Temperature
water
consumption
were dry-bulb temperature and relative humidity. Data for these
3P cooling/ E = C + B1 Chilled or hot Temperature/
variables were available from a nearby local government weather heating (B2 − T)+ water degree days
station: the National Climatic Data Center (NCDC) [17]. The hourly model E = C + B1 consumption
observations collected were converted to daily data and then used (T − B2 )+
for the analysis. 4P cooling/ E = C + B1 Chilled or hot Temperature
heating (B3 − T)+ − B2 water
For energy consumption data, the daily chilled water energy use
model (T − B3 )+ consumption
data for the case-study buildings were collected from the EMCS E = C − B1
for each building. Descriptions of the building energy use data are (B3 − T)+ + B2
summarized in Table 2. (T − B3 )+
5P model E = C − B1 Chilled and hot Temperature/
(DDTH ) + B2 water degree days
2.3. Enthalpy calculation (DDTC ) consumption
E = C + B1
Enthalpy can be obtained from the dry-bulb temperature and (B3 − T)+ + B2
(T − B4 )+
either the relative humidity or dew point temperature at a cer-
VBDD model E = C + B1 (DDBT ) Hot water Heating degree
tain pressure level. If the relative humidity is known, the moist air consumption days
enthalpy (hair ) can be written as the following equation:
Where E is the energy use or demand estimated by the equation, C is the constant
0.62198 × ∅ × Pws mean in energy units, Bn is the regression coefficient or change-point temperature, T
hair = 0.240t + × (1061 + 0.444t) (1) is the outdoor temperature, DDTH is the heating degree days (or degree hours) based
P − ∅ × Pws on the balance point temperature, DDTC is the cooling degree days (or degree hours)
where t is the air dry-bulb temperature (◦ C), ∅ is the decimal repre- based on the balance point temperature, and DDBT is the heating/cooling degree
days (or degree hours) based on the balance point temperature. The + symbol means
sentation of relative humidity, Pws is the saturation vapor pressure positive values only for the parenthetical expression.
(Pa) over liquid water for a temperature range of 0 to 200 ◦ C, and P is
the barometric pressure of the atmospheric air (kPa). The saturation
vapor pressure (Pws ) can be written as the following equation: information/data, and allow for the prediction of the dependent
variables at the given conditions of the independent variables.
C8
ln Pws = + C9 + C10 T + C11 T 2 + C12 T 3 + C13 ln T (2) In the field of building energy simulation, a regression approach
T
is a simple and powerful technique that can be used in a variety
where C8 through C13 are coefficients that can be found in the of ways (such as estimation, savings calculation, and calibration)
ASHRAE Handbook of Fundamentals (2013), and T is the absolute to analyze a building’s energy use and efficiency [9,12]. For exam-
temperature (i.e., K = ◦ C +273.15). ple, regression analysis is often used to predict how a building’s
All of the enthalpy data used in this study were derived from the energy use or demand for electricity changes as the value of one
dry-bulb temperature and relative humidity data from Eqs. (1) and or more independent variables (e.g., outdoor temperature, degree
(2). Fig. 2 illustrates the daily average outdoor dry-bulb tempera- days, or occupancy) is varied. To quantify the effects of the indepen-
ture data obtained from the NCDC for 2012, and the specific daily dent variables, a regression model (which is a mathematical model
average outdoor enthalpies calculated using the above equations. based on a statistical analysis of some measured data) can be used
[8]. Several regression models, also called change-point models,
2.4. Regression analysis are regularly used for building energy analyses (see Table 3 and
Fig. 3).
Regression analysis is a common statistical approach used to These regression models were implemented in the American
investigate the relationships among dependent and independent Society of Heating, Refrigerating, and Air-Conditioning Engineers
variables. These relationships are developed from past or present (ASHRAE) Inverse Modeling Toolkit (IMT) in 2002, and developed

40 120
Outdoor Dry-bulb Temperature
Outdoor Dry-bulb Temperature [°C]

Outdoor Specific Enthalpy [kJ/kg]

35 Outdoor Specific Enthalpy 105

30 90

25 75

20 60

15 45

10 30

5 15

0 0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Fig. 2. Measured outdoor temperatures and calculated outdoor enthalpies for the site.
60 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

Fig. 3. Regression models used in the building energy simulation [8,9,18]: (a) mean or 1 parameter (1P) model, (b) 2 parameter (2P) model, (c) 3 parameter heating (3PH)
model (similar to a variable based degree-day model (VBDD) for heating), (d) 3 parameter cooling (3PC) model (VBDD for cooling), (e) 4 parameter heating (4PH) model, (f)
4 parameter cooling (4PC) model, and (g) 5 parameter (5P) model.

from the ASHRAE Research Project RP-1050 under the guidance Haberl and associates in 2003 conducted four sets of precision tests
of Technical Committee 4.7, Energy Calculations [18]. The ASHRAE for the IMT [11], which compared the IMT results against results
IMT can derive various regression models of building energy use, from the Statistical Analysis System (SAS) software [19], EModel
including linear, change-point linear, variable-based degree-day, [20], and PRInceton Scorekeeping Method (PRISM) [21]. These com-
multi-linear, and combined regression models, and find the best- parisons showed that the IMT accurately calculated the intended
fit change-point model by using a two-part grid search method [10]. regression models by providing well-matched statistical indices.
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 61

2.6. Model evaluation using statistical index

Using the defined regression models and CDD method, this

study predicts the cooling energy use of a building. To evaluate
the goodness-of-fit of the predicted values against the measured
values, this study uses statistical indices such as the R-Squared and
Coefficient of Variation of Root Mean Square Error (CV-RMSE). The
R-Squared (i.e., R2 ) and CV-RMSE values can be calculated using
Eqs. (4), (5), and (6), respectively.
n  2
i=1
Yi − Ŷi
2
R =1−   2 (4)
n
i=1
Yi − Ȳi

Fig. 4. Diagram of building heat gain/loss factors n  2
i=1
Yi − Ŷi
RMSE = (5)
n
2.5. CDD calculation
RMSE
CV − RMSE = × 100 (6)
The Cooling Degree Days (CDD) method measures the influence Ȳ
of temperature change on a building’s energy use, assuming that where Y represents the measured cooling energy use (◦ C), Ŷ is the
the building use and efficiency of the heating, ventilating, and air corresponding value of the cooling energy use predicted by the
conditioning (HVAC) equipment are constant [15,16]. The CDD indi- model (◦ C), n is the total number of data, and Ȳ is the mean of
cator is determined using the daily average outdoor temperature by the measured cooling energy use (◦ C). In addition, percent error is
comparing it to a defined reference point temperature (also called a simply defined, as follows:
base temperature); in this study, this is called a temperature-based
CDD analysis. A conventional temperature-based CDD is defined as Ymeasured − Ypredicted
% error = × 100 (7)
the sum of the positive differences between the outdoor and ref- Ymeasured
erence temperatures over a given period [16]. See the following
The larger the R-Squared and percent error values and the
equation:
smaller the CV-RMSE value, the closer the predicted values are to

n the actual values.
 +
CDD(tbal ) = t̄o,i − tbal (3)
i=1 3. Results and discussion

where CDD(tbal ) is the cooling degree days for a defined reference 3.1. Determination of base temperature and enthalpy
temperature, n is a number of days over a desired period, t̄o,i is
the daily average outdoor temperature (◦ C), and tbal is a defined Prior to the degree day analysis, the base temperatures (i.e., the
reference point temperature (◦ C). reference point temperatures) and enthalpies of the case-study
In degree-day theory, the reference point temperature of a buildings were determined by a three parameter cooling (3PC)
building is defined as a means of accounting for indoor thermal regression analysis. Fig. 5 shows the results from the regression
comfort, which means it is the outdoor temperature that separates analysis determining both a base temperature and enthalpy for
times when a building requires cooling from times when it does Building A (the results for buildings B can be found in Appendix
not [14]. For example, a set-point of 18.3 ◦ C (65 F) is often used for A). In Fig. 5, the base temperature/enthalpy can be obtained at the
the reference point temperature to determine CDD [22]. If the aver- change point, which indicates if the chilled water consumption is
age outdoor temperature on a certain day is 30 ◦ C, then the value dependent upon or independent of the daily average outdoor tem-
of the CDD becomes 11.7 and thus the mechanical cooling system perature/enthalpy. That is, cooling is not necessary for the building
should be operated to cool down the building by 11.7 ◦ C to reach at or below the change point. The red vertical lines were drawn
18.3 ◦ C. The reference point temperature is not a fixed value, but at the change points corresponding to the different data periods
rather a variable value depending upon the following factors (see (weekdays, weekends, and all week); they illustrate the determined
Fig. 4): outdoor weather conditions (e.g., temperature, humidity, base temperature/enthalpy.
solar radiation, wind, and others), desired indoor temperature (e.g., As a result, it was found that for the all week period, the base
thermostat set-point), the building’s heat gain or loss (e.g., envelope temperature and enthalpy were 13.4 ◦ C and 32.3 kJ/kg, respectively.
thermal resistance, infiltration, and others), and internal heat gain On the other hand, as shown in the plot, it was observed that the
(e.g., lights, equipment, and people). The reference point tempera- base temperatures and enthalpies corresponding to the weekday
ture must be set higher as the outdoors get hotter, the indoor space and weekend periods were significantly different; it appeared that
is kept warmer, the indoor space gets more heat from the outdoors, the case-study building had different energy use patterns for dif-
or the internal heat adds more to the indoor space. A building’s ferent periods. For example, as compared to the results of the all
thermal characteristics and local climate conditions influence the week period, the base temperatures for weekdays and weekends
determination of the reference point temperature, and each build- were 1.9 ◦ C higher and 4.3 ◦ C lower, respectively. This indicates
ing may have its own reference point temperature when using the that the building consumed more baseline energy (i.e., constant
degree day method. Therefore, it is important to pick an appropriate building energy use such as lighting and appliances, and excluding
base temperature for degree-day-based calculations, and degree cooling) for weekdays than for weekends. In addition, the resulting
days in the most appropriate base temperatures are unlikely to be slopes of the regression lines were different: greater for weekdays
those most readily available. As a building’s base temperature typ- and smaller for weekends. This implies that the cooling energy use
ically varies throughout the year, even the most appropriate base for weekdays increased more rapidly as the outdoor temperature
temperature is usually only an approximation. increased than it did for weekends.
62 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

60,000 60,000
Data Points for All Week Data Points for Weekdays
3PC Model for All Week Data Points for Weekends

Chilled Water Energy Use [kWh/day]

50,000 50,000 3PC Model for Weekdays
Chilled Water Energy Use [kWh/day]

3PC Model for Weekends

40,000 40,000

30,000 30,000

20,000 20,000

10,000 10,000

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Outdoor Dry-bulb Temperature [°C] Outdoor Dry-bulb Temperature [°C]
(a) (b)
60,000 60,000
Data Points for All Week Data Points for Weekdays
3PC Model for All Week Data Points for Weekends

Chilled Water Energy Use [kWh/day]

50,000 50,000
Chilled Water Energy Use [kWh/day]

3PC Model for Weekdays

3PC Model for Weekends

40,000 40,000

30,000 30,000

20,000 20,000

10,000 10,000

0 0
0 10 20 30 40 50 60 70 80 90 100 110 0 10 20 30 40 50 60 70 80 90 100 110
Specific Enthalpy [kJ/kg] Outdoor Specific Enthalpy [kJ/kg]
(c) (d)

Fig. 5. Determination of base temperatures and enthalpies using the 3PC regression model: (a) with outdoor temperatures for all weeks, (b) with outdoor temperatures for
weekdays and weekends, (c) with enthalpies for all weeks, and (d) with enthalpies for weekdays and weekends.

Table 4 summarized the results from the regression analyses value may be acceptable with a larger data set (e.g., two or three
for the case-study buildings, with statistical indices (i.e., the R- years of monthly data). Regarding the coefficient of determina-
Squared and CV-RMSE). For the base temperatures and enthalpies, tion, the R-Squared value, in both cases the highest values were
the buildings had significantly different values depending upon the observed in the models using the data for weekdays. A high R-
considered periods (i.e., all week, weekdays, and weekends), even Squared value indicates that there is a strong correlation between
though the buildings were located in same climate conditions. This the x and y values (e.g., outdoor temperature and energy consump-
implies that the differences were caused by the characteristics of tion), which means that the cooling systems of the buildings tended
each building, such as the building system operation schedule and to be controlled and maintained relatively well during the week-
occupancy level and schedule. days. In addition, it was observed that the calculated R-Squared
When the energy use information was compared to the cor- values using the outdoor enthalpy data for each building resulted
responding outdoor dry-bulb temperature, the coefficient of the in a better goodness-of-fit than using the outdoor temperature data
determination, the R-Squared value, was statistically determined. for the regression analyses, as shown in Table 4.
This is a measure of the proportion of variability explained by the Regarding the CV-RMSE values for the day-level comparisons
linear relationship in the sample of paired data. It is a number (i.e., the comparisons between the regression model’s predicted
between zero and one, with a value close to zero suggesting a poor and measured data, on a daily basis), typically those less than 15
model. In general, a value above 0.7 is considered an acceptable to 30% are considered good models. The results showed that the
level indicating confidence in the relationship; however, a lower CV-RMSE from the analysis using the weekday enthalpy data for

Table 4
Base temperatures/enthalpies for the case-study buildings.

Building A Building B

All week Weekdays Weekends All week Weekdays Weekends

Base temperature (◦ C) 13.4 15.3 11.0 14.6 14.3 14.5

R2 0.8630 0.8787 0.8360 0.9121 0.9541 0.9422
CV-RMSE 13.8 13.3 14.4 22.6 16.0 17.8
Base enthalpy(kJ/kg) 32.3 33.6 26.8 35.0 34.5 37.1
R2 0.9299 0.9409 0.9110 0.9285 0.9690 0.9673
CV-RMSE 9.9 9.3 10.6 20.4 13.2 13.3
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 63

Table 5
Monthly cooling degree days and number of data points for Building A.

CDD Number of data points for CDD calculation

Month Temperature-base (◦ C) Enthalpy-base (kJ/kg) CDDT (at 13.4 ◦ C) CDDh (at 32.3 kJ/kg)

Mar. 202 622 27 27

Apr. 284 767 30 30
May 390 1092 31 31
Jun. 475 1363 30 30
Jul. 498 1571 31 31
Aug. Data not available Data not available
Sep. 419 1152 30 30
Oct. 250 702 28 28
Nov. 132 340 21 20
Dec. 98 299 16 16
Jan. 60 209 13 15
Feb. 49 133 15 12
Total 2857 8250 272 270

Building A was 9.3%, which is considered an excellent analysis result A were calculated using Eq. (3). Table 5 presents the calculated
indicating that the energy consumption by regression model was monthly temperature-based CDD (CDDT ) values and enthalpy-
well-matched to the measured data among the case-study build- based CDD (CDDh ) values. The resulting monthly CDDT and CDDh
ings. In addition, it is clear from the overall CV-RMSE results that values showed that the building consumed the largest cooling
the regression model using the outdoor enthalpy data predicted energy during the month of July. However, it should be noted that
the energy consumption more accurately than the model using the the building might have consumed more cooling energy in August
outdoor dry-bulb temperature. than in July. Unfortunately, this study could not identify the cooling
energy use in August because the data for August were not available
3.2. Calculations of cooling degree days (due to technical issues).
This study was not able to conduct a comparison of the CDD
Based on the determined base temperatures and enthalpies for values for the CDDT and CDDh due to the different CDD units
the all week values using the 3PC regression models outlined in (i.e., CDDT in ◦ C, whereas CDDh in kJ/kg). However, this study did
Section 3.1, the monthly cooling degree days (CDD) for Building include a comparison of the number of data points used for the CDD

60,000 60,000
Chilled Water Energy Use [kWh/day]
Chilled Water Energy Use [kWh/day]

50,000 50,000

40,000 40,000

30,000 30,000

20,000 20,000

10,000 10,000

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Cooling Degree Day with Outdoor Temperature [°C] Cooling Degree Day with Outdoor Temperature [°C]
(a) (b)
60,000 60,000
Chilled Water Energy Use [kWh/day]

Chilled Water Energy Use [kWh/day]

50,000 50,000

40,000 40,000

30,000 30,000

20,000 20,000

10,000 10,000

0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Cooling Degree Day with Specific Enthalpy [kJ/kg] Cooling Degree Day with Specific Enthalpy [kJ/kg]
(c) (d)
Fig. 6. Cooling energy predictions with linear regression models for Building A: (a) CDDT for weekdays, (b) CDDT for weekends, (c) CDDh for weekdays, and (d) CDDh for
weekends.
64 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

Table 6
Measured and predicted monthly total cooling energy consumptions for building A.

Cooling energy consumption for weekdays (kW h) Cooling energy consumption for weekends (kW h)

Measured Predicted Differences Measured Predicted Differences

(predicted-measured) (predicted-measured)

CDDT CDDH CDDT CDDH CDDT CDDH CDDT CDDH

Mar. 475,945 511,110 539,158 35,165 63,213 174,494 199,243 201,766 24,749 27,272
Apr. 499,633 566,401 554,809 66,767 55,176 218,568 254,734 246,205 36,167 27,637
May 731,625 740,919 727,986 9,294 −3,639 248,322 259,581 252,825 11,259 4,502
Jun. 825,325 800,319 784,919 −25,006 −40,406 323,675 321,036 319,787 −2,640 −3,888
Jul. 890,082 840,515 880,600 −49,567 −9,482 349,590 329,684 343,090 −19,906 −6,499
Aug. Data not available Data not available
Sep. 648,516 697,682 663,999 49,166 15,482 315,099 323,395 317,016 8,296 1,917
Oct. 587,095 597,475 591,333 10,380 4,238 188,568 189,676 190,547 1,108 1,978
Nov. 420,490 435,417 419,830 14,927 −661 169,039 168,777 166,061 −262 −2978
Dec. 360,243 346,984 341,114 −13,259 −19,129 238,466 210,683 218,037 −27,783 −20,429
Jan. 391,871 355,022 365,239 −36,849 −26,632 150,910 130,149 137,214 −20,761 −13,697
Feb. 353,062 312,065 313,392 −40,996 −39,670 142,244 132,600 124,894 −9644 −17,350

calculation. As shown on the right side of Table 5, marginal differ- approximately 270 days of data points, which represented approx-
ences in the monthly total numbers of data points were observed. imately 80% of one year (excluding the missing data in August). The
To predict the annual cooling energy consumption for a build- total amount of data indicated that the site where the case study
ing, both the temperature- and enthalpy-based CDD methods used building was located in a cooling-dominated climate. In addition,

Fig. 7. Percent errors between the measured and predicted monthly total cooling energy consumptions of Building A: (a) for weekdays, and (b) for weekends.
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 65

10,000
Whole Period Full Cooling Period

9,000

Absolute Difference between Actual and Predicted

8,000

7,000

Energy Use [kWh/day]

6,000

5,000

4,000

3,000

2,000

1,000

0
CDD-Temp. CDD-Enthalpy CDD-Temp. CDD-Enthalpy

(a)
10,000
Whole Period Full Cooling Period
9,000
Absolute Difference between Actual and Predicted

8,000

7,000
Energy Use [kWh/day]

6,000

5,000

4,000

3,000

2,000

1,000

0
CDD-Temp. CDD-Enthalpy CDD-Temp. CDD-Enthalpy

(b)
Fig. 8. Distribution of absolute difference values between the daily measured and predicted cooling energy consumption data of Building A: (a) for weekdays, and (b) for
weekends.

based on the number of data points used for the CDD calculation, it within the acceptable range (R2 > 0.7), but the model using CDDh
was found that the full cooling period (i.e., the period that required showed a better goodness-of-fit for the predicted values against
cooling for every day in the whole month, according to reference the measured values (around 10% higher than the correlation with
point temperature/enthalpy) was April through September. the CDDT ).
Due to the different building energy use patterns for weekdays
3.3. Cooling degree days energy prediction and weekends, different periods were used for the comparison.
The R-Squared values using the CDDT regression models were not
To estimate the cooling energy consumptions (i.e., chilled water noticeably different for weekdays and weekends (0.8142 for week-
energy use), a linear regression model was used. The linear models days and 0.8066 for weekends). On the other hand, the CDDh
were developed based on the relationship between the CDD (i.e., regression models resulted in significantly different R-Squared val-
CDDT and CDDh ) and the corresponding measured cooling energy ues (0.9174 for weekdays and 0.8701 for weekends). As a result,
use (see Fig. 6). As shown in Fig. 6, the linear models used to calcu- the predicted cooling energy uses from the CDDh regression mod-
late the energy consumptions and coefficients of the determination els appear to be more closely related to the measured data of the
(R-Squared) are presented in the top left corner of each plot. Based building than the predicted values from the CDDT regression mod-
on the resultant R-squared values as compared with the measured els.
energy use, CDDh resulted in a better prediction of energy con- The monthly total measured and predicted cooling energy con-
sumption than did CDDT . In both cases where the regression models sumptions for Building A were compared. To accomplish this, the
used CDDT and CDDh , the R-Squared values were observed as being daily measured and predicted data were converted to monthly
66 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

energy consumption. Table 6 presents the results of the compar- cooling-dominated climate. To accomplish the study object, a
ison, including the absolute differences between the measured and three parameter cooling (3PC) regression model was used to
predicted values. In addition, Figs. 7 and 8 illustrate the differences determine the base temperature/enthalpy of each case-study
by comparing the CDDT and CDDh for weekdays and weekends (the building. Measured year-long daily climatic data were used to
results for buildings B can be found in Appendix A). drive the 3PC regression models. Based on the determined base
Fig. 7 illustrates the percent errors between the monthly mea- temperature/enthalpy, the CDDT and CDDh were individually
sured and predicted total cooling energy consumptions for each calculated; they were then used to predict the cooling energy use.
cooling degree day (i.e., CDDT and CDDh ) method. Overall, the CDDh Next, each predicted cooling energy use derived from the CDDT and
method appeared to more accurately predict the energy use of the CDDh was compared to the measured data, and the accuracy of the
building, as it was observed that the annual energy consumption predicted values was evaluated. Based on the results of the analyses
predicted with the CDDh method had a 1% less annual percent- conducted in this study, the following conclusions can be made:
age error than the value predicted with the CDDT method. More
obvious distinctions in percent error between the CDDT and CDDh • Using the 3PC regression analyses, it was observed that both
methods were found during the full cooling season (April through case-study buildings had different base temperatures/enthalpies
September). The smallest percent errors for both CDD methods for depending upon the data period: weekdays, weekends, or all
weekdays (see Fig. 7a) were observed in May: 0.5% by the CDDh week. As compared to the results for the all week period, the
method and 2.3% by the CDDT method; the CDDh method better base temperatures/enthalpies were higher for weekdays and
predicted the cooling energy use (about 1.8% in percentage error) lower for weekends. This indicates that the buildings consumed
than did the CDDT method. For weekends (see Fig. 7b), the small- more baseline energy during weekdays than weekends. In addi-
est percentage errors of the CDDT and CDDh methods were 0.6% in tion, the resultant regression line slopes were different – greater
September and 0.6% in October, respectively. Moreover, the percent for weekdays and smaller for weekends – which implies differ-
errors of the predicted cooling energy consumptions for weekdays ent cooling energy use patterns for different periods. Therefore,
during the full cooling season were 2.0% by the CDDh method and it can be concluded that the data periods used for energy
4.0% by the CDDT method. For weekends, the overall percentage analysis, according to the building’s energy use characteristics,
errors were 1.0% by the CDDh method and 3.0% by the CDDT method. need to be determined to better predict the building’s energy
On the other hand, the largest prediction errors for week- use.
days were observed in March (13.3%) by the CDDh method and • The monthly CDDT and CDDh values were calculated separately.
in April (13.4%) by the CDDT method. Both CDD methods resulted The total number of data points used for the CDDT and CDDh were
in larger percent error values during the partial cooling seasons 272 days and 270 days, respectively, which represented approx-
(from January to April and November to December); the reason for imately 80% of one year (excluding the missing data in August).
this appears to be that the rapid outdoor weather changes affected This indicates that the site where the case study buildings were
the predicted values. As a result, a more accurate prediction of located was in a cooling-dominated climate. In addition, the full
energy consumption was achieved for Building A by using the CDDh cooling season for the buildings was identified as April through
method. September.
Fig. 8 uses box-whisker plots to show the distribution of the • The linear regression analyses using the CDDT and CDDh methods
absolute difference values between the daily measured and pre- indicate that use of the CDDh model resulted in a closer goodness-
dicted cooling energy consumptions using the CDDT and CDDh of-fit for the predicted values (when compared to the measured
methods for weekdays and weekends. The overall differences were values) than did the CDDT models. This implies that the cor-
approximately the same for the predicted values using the CDDT relation between the CDDh and cooling energy consumption is
method, regardless of the data period (whole period or full cooling significantly improved – around 10% higher – than the correla-
period); however, using the CDDh method significantly improved tion with the CDDT (e.g., the R-Squared of 0.8142 for CDDT and
the predictions made, especially for the full cooling period. While 0.9174 for CDDh using the weekday period). Thus, it was expected
the CDDT method showed a similar level of absolute difference on that the CDDh method would predict values more accurately than
the median, the results by the CDDh method were quite different. the CDDT method.
• Compared with the measured cooling energy use for the full cool-
4. Summary and conclusions ing season (e.g., April through September), the average percent
errors using the CDDT and CDDh methods yielded 4% and 2%,
The concept of the CDD is one of a simplified method for pre- respectively. Since the CDDh method reduced the percent error
dicting cooling energy consumption in buildings. However, one of to 2%, this study concludes that the CDDh method provides more
the CDD method’s issues is that latent cooling loads cannot be cap- precise results for cooling energy consumption in buildings in
tured during the CDD calculation process; it generally uses dry-bulb cooling-dominated climates.
temperature. Especially in hot and humid climates, it is essential
to consider not only the sensible cooling load but also the latent In general, this study has identified that cooling energy predic-
cooling load when estimating cooling energy consumption. In this tions using CDD with enthalpy yield accurate results, especially
study, a new CDD method using specific enthalpy values (CDDh ) during the full cooling season. However, application of the pro-
instead of dry-bulb temperature (CDDT ) was proposed to predict posed method in different climates and building types may produce
cooling energy consumption. different results. For instance, the latent heat may not have a sig-
The primary objective of this study was to evaluate the accu- nificant effect in dry or cold climates. Therefore, further research is
racy of the CDDT and CDDh approaches to predict daily total required that applies the proposed method to other existing build-
cooling energy use in two institutional buildings located in a ings in different weather conditions.
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 67

Appendix A. Analysis results for Building B

Figs. A1–A3

30,000 30,000
Chilled Water Energy Use [kWh/day]

Data Points for All Week Data Points for Weekdays

Chilled Water Energy Use [kWh/day]

3PC Model for All Week Data Points for Weekends
25,000 25,000 3PC Model for Weekdays
3PC Model for Weekends

20,000 20,000

15,000 15,000

10,000 10,000

5,000 5,000

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Outdoor Dry-bulb Temperature [°C] Outdoor Dry-bulb Temperature [°C]
(a) (b)

30,000 30,000
Chilled Water Energy Use [kWh/day]

Data Points for All Week Data Points for Weekdays

Chilled Water Energy Use [kWh/day]

3PC Model for All Week Data Points for Weekends
25,000 25,000 3PC Model for Weekdays
3PC Model for Weekends

20,000 20,000

15,000 15,000

10,000 10,000

5,000 5,000

0
0
0 10 20 30 40 50 60 70 80 90 100 110
0 10 20 30 40 50 60 70 80 90 100 110
Outdoor Specific Enthalpy [kJ/kg]
Outdoor Specific Enthalpy [kJ/kg]
(c) (d)

Fig. A1. Determination of base temperatures and enthalpies using the 3PC regression model for Building B: (a) with temperatures for all weeks, (b) with temperatures for
weekdays and weekends, (c) with enthalpies for all weeks, and (d) with enthalpies for weekdays and weekends.
68 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

Fig. A2. Percent errors between the measured and predicted monthly total cooling energy consumptions of Building B: (a) for weekdays, and (b) for weekends.
 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70 69

6,000
Whole Period Full Cooling Period

Absolute Difference between Actual and Predicted

5,000

4,000

Energy Use [kWh/day]

3,000

2,000

1,000

0
CDD-Temp. CDD-Enthalpy CDD-Temp. CDD-Enthalpy

(a)

6,000
Whole Period Full Cooling Period
Absolute Difference between Actual and Predicted

5,000

4,000
Energy Use [kWh/day]

3,000

2,000

1,000

0
CDD-Temp. CDD-Enthalpy CDD-Temp. CDD-Enthalpy

(b)
Fig. A3. Distribution of absolute difference values between the daily measured and predicted cooling energy consumption data of Building B: (a) for weekdays, and (b) for
weekends.

Table A1

Table A1
Measured and predicted monthly total cooling energy consumptions for building B.

Cooling energy consumption for weekdays (kW h) Cooling energy consumption for weekends (kW h)

Measured Predicted Differences Measured Predicted Differences

(predicted-measured) (predicted-measured)

CDDT CDDH CDDT CDDH CDDT CDDH CDDT CDDH

Apr. 177,074 172,136 178,793 −4,937 1,720 43,677 45,904 43,880 2,226 203
May 316,453 296,823 309,204 −19,630 −7,249 70,455 70,050 67,336 −404 −3119
Jun. 391,790 368,727 365,138 −23,063 −26,652 137,114 135,107 127,722 −2,007 −9,392
Jul. 430,334 435,034 417,328 4,700 −13,006 110,975 111,471 111,901 496 926
Aug. 439,116 444,655 432,548 5,539 −6,568 126,687 133,518 124,919 6,831 −1,768
Sep. 365,908 383,480 364,546 17,572 −1,363 111,693 111,473 115,784 −220 4,091
70 M. Shin, S.L. Do / Energy and Buildings 110 (2016) 57–70

Table A1 (Continued)

Cooling energy consumption for weekdays (kW h) Cooling energy consumption for weekends (kW h)

Measured Predicted Differences Measured Predicted Differences

(predicted-measured) (predicted-measured)

CDDT CDDH CDDT CDDH CDDT CDDH CDDT CDDH

Oct. 237,091 240,689 256,704 3,598 19,613 54,969 50,507 55,185 −4,462 216
Nov. 82,154 84,572 91,544 2,418 9,390 34,159 33,328 37,916 −831 3,757
Dec. Data not available Data not available
Jan. 42,886 51,834 58,038 8,948 15,152 14,490 13,917 17,436 −574 2,946
Feb. 61,051 71,507 72,698 10,456 11,647 24,075 24,534 28,705 459 4,630
Mar. 65,077 72,124 71,862 7,047 6,785 39,734 35,932 39,811 −3,803 77

References [11] J.S. Haberl, A. Sreshthaputra, D.E. Claridge, J.K. Kissock, Inverse model toolkit
(1050RP): application and testing, ASHRAE Trans. -Res. 109 (Part 2) (2003)
[1] S. Bouton, J. Creyts, T. Kiely, J. Livingston, T. Nauclér, Energy Efficiency: A 435–448.
Compelling Global Resource, McKinsey Sustainability & Resource [12] BPA., Regression for M&V: Reference Guide, Bonneville Power Administration,
Productivity, 2010. Portland, OR, 2011.
[2] UNEP, Buildings and climate change: summary for decision-makers, in: [13] J.K. Kissock, A Methodology to Measure Retrofit Energy Savings in
United Nations Environment Programme (UNEP) Sustainable Buildings & Commercial Buildings, Department of Mechanical Engineering, Texas A&M
Climate Initiative (SBCI), Paris, France, 2009. University, 1993 (Ph.D. Thesis).
[3] IEA, Key world energy statistics, in: International Energy Agency (IEA), Paris, [14] CIBSE, Degree-days: Theory and Application, The Chartered Institution of
France, 2014. Building Services Engineers, London, England, 2006.
[4] L. Pérez-Lombard, J. Ortiz, C. Pout, A review on buildings energy consumption [15] G. Krese, M. Prek, V. Butala, Analysis of building electric energy consumption
information, Energy Build. 40 (3) (2008) 394–398. data using an improved cooling degree day method, Strojniski Vestnik/J.
[5] H.-x. Zhao, F. Magoulès, A review on the prediction of building energy Mech. Eng. 58 (2) (2012) 107–114.
consumption, Renewable Sustainable Energy Rev. 16 (6) (2012) [16] ASHRAE, 2009 ASHRAE Handbook—Fundamentals (SI Edition), American
3586–3592. Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc, Atlanta,
[6] N. Li, J.-y. Kwak, B. Becerik-Gerber, M. Tambe, H.V.A.C. Predicting, energy GA, 2009.
consumption in commercial buildings using multiagent systems, in: [17] National Climatic Data Center, Quality Controlled Local Climatological Data,
Proceedings of the 30th International Symposium on Automation and 2015, http://www.ncdc.noaa.gov/qclcd/QCLCD.
Robotics in Construction and Mining, ISARC, 2013. [18] J.K. Kissock, J.S. Haberl, D.E. Claridge, Development of a Toolkit for Calculating
[7] L.G. Swan, V.I. Ugursal, Modeling of end-use energy consumption in the Linear, Change-Point Linear and Multiple-Linear Inverse Building Energy
residential sector: a review of modeling techniques, Renewable Sustainable Analysis, Models Final Report for ASHRAE Research Project 1050-RP,
Energy Rev. 13 (8) (2009) 1819–1835. American Society of Heating, Refrigerating and Air-Conditioning Engineers,
[8] ASHRAE, ASHRAE Guideline 14: Measurement of Energy and Demand Savings, Inc., Atlanta, GA, 2002.
American Society of Heating, Refrigerating and Air-Conditioning Engineers, [19] SAS, SAS User Manual, SAS Institute Inc., Metaire, LA, 2001.
Inc., Atlanta, GA, 2002. [20] J.K. Kissock, W. Xun, R. Sparks, D. Claridge, J. Mahoney, J. Haberl, EModel
[9] J.S. Haberl, C. Culp, D.E. Claridge, ASHRAE’s Guideline 14-2002 for Version 1.4de, Energy Systems Laboratory, Department of Mechanical
Measurement of Energy and Demand Savings: How to Determine What Was Engineering, Texas A&M University, 1994.
Really Saved by the Retrofit, Energy Systems Laboratory, College Station, TX, [21] M. Fels, J.K. Kissock, M. Marean, C. Reynolds, PRISM (Advanced Version 1.0)
2005. Users Guide, Center for Energy and Environmental Studies, Princeton
[10] J.K. Kissock, J.S. Haberl, D.E. Claridge, Inverse modeling toolkit (1050RP): University, 1995.
numerical algorithms for best-fit variable-base degree-day and change-point [22] EPA, Climate Change Indicators in the United States, 2014, EPA, Washington,
models, ASHRAE Trans. -Res. 109 (Part 2) (2003) 425–434. DC, 2014.