Professional Documents
Culture Documents
A Holistic Utility Bill Analysis Method For Baselining Whole Commercial Building Energy Consumption in Singapore
A Holistic Utility Bill Analysis Method For Baselining Whole Commercial Building Energy Consumption in Singapore
www.elsevier.com/locate/enbuild
Abstract
The methodology for baseline building energy consumption is well established for energy saving calculation in the temperate zone both for
performance-based energy retrofitting contracts and measurement and verification (M&V) projects. In most cases, statistical regression
models based on utility bills and outdoor dry-bulb temperature have been applied to baseline monthly and annual whole building energy use.
This paper presents a holistic utility bills analysis method for baseline whole building energy consumption in the tropical region. Six
commercial buildings in Singapore were selected for case studies. Correlationships between the climate data, which are monthly mean
outdoor dry-bulb temperature (T0), relative humidity (RH) and global solar radiation (GSR), and whole building energy consumption are
derived. A deep prediction study based monthly mean outdoor dry-bulb temperature (T0) and whole building energy consumption is stated.
The result shows that variations of the energy consumption in most of these buildings are contributed by T0 and can be well predicted at 90%
confidence level only with it. The analysis of such kind of model is especially useful for building managers, owners and ESCOs to track and
baseline energy use during pre-retrofit and post-retrofit periods in the tropical condition.
# 2004 Elsevier B.V. All rights reserved.
0378-7788/$ – see front matter # 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2004.06.011
168 B. Dong et al. / Energy and Buildings 37 (2005) 167–174
PIannual ð%DyÞ
ymeasured
¼ 100
ybaselineprojected
(
a 2 VARModel
t 1 ;n p
2
2 ^
Y baselineporjected
)1=2
þ ðmeasurmment error fractionÞ2 (9)
3. Methodology
Table 1
Size and energy use in six buildings
Building Baseline year Total building Air conditioned Total building energy
area (m2) area (m2) consumption (MWh/year)
A 2000 20,165 12,268 3,470
B 2000 32,368 24,825 6,034
C 2000 42,026 25,822 9,998
D 2001 60,894 36,688 11,551
E 2001 42,060 25,833 8,031
F October 2000– September 2001 43,187 34,753 7,880
nearest to six buildings among four stations because the b. The utility bill period is one month later than the
onsite weather data are not available. The monthly data is weather data period. It means that the present utility
found by averaging the hourly data of the whole month. bill actually represents the previous month’s electricity
Actually, there is little difference among these four weather use. This situation always happens
points in terms of monthly mean temperature. Fig. 2 shows c. The utility bill period is 15 days later than the weather
the monthly mean outdoor dry-bulb temperature from year data period. It means the readings of utility bills begin
2000 to 2001. The highest temperature appeared in May around the middle of the month.
2000 as 28.6 8C, while the lowest temperature appeared in
January 2000 as 26.4 8C. Hence, the overall range of tem- For each of the baseline model, a multiple linear regres-
perature is only 2.2 8C, which determines the difference of sion (MLR) analysis between climates parameters and
energy consumption among 24 months should be with little whole building energy consumption in the baseline year
change. period was conducted by SPSS based on all these three
In addition, there always exists the problem that whether possibilities. Then, the model which has the best-fit regres-
the utility bill was recorded beginning with the first day of sion is selected. The regression results are shown in Table 2.
the month and finishing at the last day. The uncertainty in Due to the obvious difference among three R2, we did not
reading dates should also be considered. calculate CV-RMSE for each model. If the three R2 are very
close, then an additional CV value is needed. From the Table
3.2. Baseline model identification 2, it shows that building D and F have the highest R2 value
when the utility bill period is the same month as the weather
Following the methodology flowing charts, before any data, while the R2 value appears highest in building B, C and
regression analysis began, we should first verify the bill E when the utility bill period is one month later than the
reading dates. The building owners did not provide the temperature data file. Building A has the highest R2 value
detailed information on utility bills’ reading dates. Every when the bill period is 15 days later than weather data
building has its own policy for recording electricity use. period. The results in Table 2 also indicate that the way that
Hence, It is difficult to verify the specific utility reading building owner records the utility bill cannot be arbitrarily
period. Reddy et al. [5] pointed out that there were three decided because any way exits the possibility to have the
possibilities in recording building energy use. These three highest R2 value. Hence, all three possibilities should be
methods are also considered in this study: considered at the same time to establish the best relationship
between utility bills and weather data variations.
a. The utility bill period is correspondent with weather After the bill reading dates are settled down, following
data period. It means that the utility bill reading dates the methodology flowing chart, the MLR regression results
are the same dates as weather data dates. of the best fit model were selected and shown in Table 3.
Table 3
Summary of multiple linear regression results
Variables A B C D E F
Correlation t Correlation t Correlation t Correlation t Correlation t Correlation t
Partial Model Partial Model Partial Model Partial Model Partial Model Partial Model
R R2 R R2 R R2 R R2 R2 R2 R2 R2
T0 0.93 0.92 7.10 0.86 0.79 4.77 0.82 0.76 4.6 0.87 0.78 5.06 0.86 0.81 4.8 0.89 0.82 5.6
RH 0.35 1.01 0.36 1.07 0.15 0.43 0.12 0.31 0.28 0.84 0.44 1.37
GSR 0.66 2.49 0.35 1.06 0.15 0.43 0.13 0.37 0.08 0.23 0.18 0.51
172 B. Dong et al. / Energy and Buildings 37 (2005) 167–174
a suddenly high data, and we can speculate that there should a prediction interval (PI), which shows the error band of
be certain unexpected reasons making building energy forecast value. For building A, the monthly electricity use in
consumption abruptly go down. Therefore, the energy con- year 2001 is predicted to be 24.11 (1.71) (kWh/m2/month),
sumption predicted by the baseline model can be used by an increase of about 0.92% (12.51), while the measured
building owner to check whether the energy consumption energy use is 23.52 (kWh/m2/month), an increase of about
follows a normal development. 2.45%. Although the difference between modeled and mea-
Table 5 gives a summary of percentage changes of energy sured energy use is only 0.59 (kWh/m2/month), the error
consumption per square meter on monthly basis and statis- band is wide. The worst prediction for is building D, with a
tics analysis from baseline year to final year for all six modeled energy consumption of 24.33 (5.46) (kWh/m2/
buildings. For building A, changes are clearly negative, month) and decreases its energy use by 16.58 (48.15),
which means the energy consumption shows an increasing while the measured energy use is 28.59 (kWh/m2/month)
trend. However, in terms of measured energy consumption,
building D and E have negative percentage change, which
means that the energy use is lower than the previous year.
Such kind of behavior is contrary to the normal electricity
use trends.
In addition, building B has a positive value on the
modeled percentage change, which indicates the opposite
trends of the measured value. Building E is in such condition
as well. The other four correctly predict the energy con-
sumption trends. All of the models’ VARs, which indicate
the variance of the forecast error, are small, except for
building D. It increases according to the building energy
consumption. It appears highest in building D, which is
29.79 (kWh/m2/month) and lowest in building B, 1.35
(kWh/m2/month). This means in terms of accuracy, the
predicted data in building B is the best one.
Fig. 4. Predicted vs. measured monthly energy use with 90% PI band.
Furthermore, direct deviations from modeled and mea-
sured energy use are all below 5%, except for building D.
The deviation is from the difference of predicted and
measured on based on measured one. The lower the devia-
tion, the more accurate the predicted energy consumption is.
At 90% confidence level, following equation (4), one can get
Table 4
Summary of single variant regression analysis
Building A B C D E F
2
R 0.76 0.72 0.75 0.76 0.78 0.77
CV-RMSE 3.48 3.8 2.78 3.49 3.76 3.32
Fig. 5. Residues of estimated whole building energy consumption.
B. Dong et al. / Energy and Buildings 37 (2005) 167–174 173
Table 5
Summary of identified baseline models and relevant statistical measures for six buildings
Building Baseline year Final year Annual energy use (kWh/m2/ VAR (Model) (kWh/m2/month)
month)
Measured Modeled
A 2000 2001 23.52 24.11 2.92
B 2000 2001 20.26 20.59 1.35
C 2000 2001 32.87 32.09 2.60
D 2001 2000 28.59 24.22 29.79
E 2001 2000 25.55 25.76 2.05
F October 2000–September 2001 2000 18.81 18.34 2.83
Deviation (%) 90% PI (kWh/month/m2) Percentage change (from 90% PI error (without measurement error)
baseline year to final year)
Measureda Modeled
A 2.51% 3.10 0.92%b 2.45% 12.51%
B 1.64 10.06 1.61 0.04 10.06
C 2.42 2.9 1.09 2.36 8.73
D 15.33 48.15 16.58 18.06 48.15
E 0.8 2.59 0.81 1.14 10.24
F 2.5 3.04 2.55 4.22 15.77
a
‘‘Measured’’ percentage change is the result of direct bill comparing.
b
Negative value means the predicted energy use is higher than the measured one.
and it decreases by 18.06. It shows a large uncertainty in prediction interval. Obviously, the former value has more
estimates. statistical meaning.
Overall, Table 5 shows that all the error bands of per-
centage changes are wider than percentage changes, what-
ever measured or modeled. Hence, the percentage changes 4. Conclusion
are not very significant statistically. The reason could be as
follows: This paper presents a holistic statistical regression pro-
cedure to identify baseline models of energy use in tropical
(a) Low R2 value. Although, following Reddy (1997a), it region when the utility bill reading dates are not clearly
shows that the models are included among the excellent known. A multiple linear regression analysis was performed
model, all R2 are lower than 0.8, which means more in the process of verifying model parameters. A deep study
than 20% of the variation in Y of its mean value can not on the model prediction ability was carried out after setting
be explained by the model. This study indicates that, in up the final baseline model. The results of applying to six
tropical region, the building energy consumption has buildings showed that most of the projected baseline models
just a certain relationship with temperature, but not very successfully tracked and detected changes in energy use.
strong. Although judging from Tables 4 and 5 and previous arbitrary
(b) Little changes in monthly mean temperature in tropical standards [5], the baseline models are all excellent and have
region. This leads to small percentage changes both in good prediction ability, the PI for Dyð%Þ is high and the
measured and modeled annual energy use. Further- prediction results are therefore not statistically significant.
more, together with reason (a) and the goal of the However, the deviation directly between measured and
prediction interval of baseline-projected energy use molded energy use is small. Hence, comparing measured
to meet 90% confidence level, the width of PI becomes and modeled energy use directly is much clearer and mean-
large. ingful.
In conclusion, baseline building energy consumption on
Finally, although all the measured and modeled percen- the whole building level solely using monthly utility bills
tages are within 90% uncertainty interval, one shall not only and outdoor dry-bulb temperature as model variants is
look at such fraction intervals to judge its prediction ability. practical and achievable in the tropical region. Furthermore,
As discussed before, 90% error band for percentage changes comparative low R2 indicates that there are other weather or
has little statistical meaning and the actual value of PI for non-weather variables to influence building energy con-
modeled energy use is more comparative and objective. For sumption. The relative low confidence prediction indicates
example, in building B, the modeled value is 20.59 (kWh/ that further study based on daily and hourly whole building
m2/month) with 10.06 prediction interval at 90% confidence consumption data are needed. The study also indicates that
level, while the modeled percentage is 0.04% with 10.06% the weather data obtained from the weather station is
174 B. Dong et al. / Energy and Buildings 37 (2005) 167–174
sufficient and appropriate for developing tracking models [4] M. Krarti, J. Kreider, D. Cohen, P. Curtiss, Prediction of Energy
for changes in building energy use. Saving for Building Retrofits Using Neural Networks, ASME J. Solar
Energy Eng. 120 (3) 1998.
[5] T.A. Reddy, N.F. Saman, D.E. Claridge, J.S. Haberl, W.D. Turner, A.
Chalifoux, Baselining methodology for facility-level monthly
Acknowledgment energy use—part 1: theoretical aspects, ASHRAE Trans. 103
(1997) 2.
[6] M.F. Fels, K.M. Keating, Measurement of energy savings from
This study is supported by Center For Total Building
demand-side management programs in U.S. electric utilities, Annu.
Performance, Department of Building, National University Rev. Energy Environ. 18 (1993) 57–88.
of Singapore. [7] S. Katipamula, D.E. Claridge, Use of simplified system models
to measure retrofit energy savings, J. Solar Energy Eng. 115
(1993).
References [8] N. Draper, H. Smith, Applied Regression Analysis, 2nd edition, John
Wiley and Sons, New York, 1981.
[1] M. Fels, Special Issue Devoted to Measuring Energy Savings, [9] W. Tan, Practical Research Methods, Prentice Hall, Singapore.
The Princeton Scorekeeping Method (PRISM), Energy and Buildings [10] B.L. Bowerman, R.T. O’Connell, Linear Statistical Models: An
9(1–2) (1986). Applied Approach, 2nd edition, Duxbury Press, Belmont, California,
[2] A. Dhar, T.A. Reddy, D.E. Claridge, A. Fourier, Series model to 1990.
predict hourly heating and cooling energy use in commercial buildings [11] ASME, ANSI/ASME Standard PTC 19.1-1985, Measurement uncer-
with outdoor temperature as the only weather variable, J. Solar Energy tainty: Instruments and apparatus. New York American Society of
Eng. 121 (1999) 47–53. Mechanical Engineers, 1990.
[3] J.K. Kissock, A methodology to measure retrofit energy savings in [12] ASHRAE, ASHRAE GUIDELINE 14: Measurement of Energy and
commercial buildings, Doctoral Dissertation (1993), Department of Demand Savings, American Society of Heating, Refrigerating and
Mechanical Engineering, Texas A&M University. Air-Conditioning Engineers, Inc., Atlanta, 2002.