development and performance evaluation of a methodology...
TRANSCRIPT
Development and performance evaluation of a methodology, based
on distributed computing, for speeding EnergyPlus simulation
Vishal Garga*
, Kshitij Chandrasena, Jyotirmay Mathur
b, Surekha Tetali
a,
Akshey Jawaa
a Centre for IT in Building Science, International Institute of Information Technology,
Hyderabad, India bCentre for Energy and Environment Malaviya National Institute of Technology,
Jaipur, India
Vishal Garg
Head & Associate Professor,
Centre for IT in Building Science,
IIIT Hyderabad, India.
email: [email protected]
Kshitij Chandrasena
Student, B.Tech Final Year,
IIIT Hyderabad, India.
email: [email protected]
Jyotirmay Mathur
Co-coordinator, Centre for Energy and Environment,
Associate Professor, Mechanical Engineering Department
Malaviya National Institute of Technology
Jaipur, India
email: [email protected]
Surekha Tetali
Student, MS by Research,
Centre for IT in Building Science
IIIT Hyderabad, India.
email: [email protected]
Akshey Jawa
Student, MS by Research,
Centre for IT in Building Science
IIIT Hyderabad, India.
email: aksheyjawa@ research.iiit.ac.in
Corresponding Author: [email protected]
This paper presents an approach for speeding EnergyPlus simulations. The computing
run time of an energy simulation depends on several variables and is directly
proportional to the simulation RunPeriod. In the proposed approach, data parallelization
is achieved by breaking an annual simulation into several segments of smaller
RunPeriod, each handled by a separate computer/processor. The speed gain achieved by
running 12 one-month RunPeriod segments in parallel as compared to single simulation
of twelve months is between 3 to 6 times. Segmentation of simulation has resulted in
minor deviations between the results obtained through segmented simulations and annual
simulations. Methods for reducing these deviations on annual and monthly basis are
presented in this paper using twelve benchmark models each simulated for five cities. On
annual basis, a maximum deviation of 0.06% was observed in cooling, heating, and
lighting consumption. In a month-to-month comparison between the segments and
annual simulation, the maximum deviation was 1.7% for heating and 0.8% for cooling.
Keywords: EnergyPlus, energy simulation, simulation run time, parallel simulation, simulation
speed up
1. Introduction
There has been an increased effort among Architects, HVAC engineers and designers
to implement energy conservation features in buildings. This has resulted in an
increased use of energy simulation software in the design process. Energy simulation
programs can help in achieving energy efficient and cost effective designs.
EnergyPlus [1] is a new-generation building energy simulation program based on
DOE-2 [2] and BLAST [3], with numerous added capabilities. EnergyPlus includes
many innovative simulation capabilities such as time steps of less than an hour,
modular systems and plant integrated with heat balance-based zone simulation,
multizone air flow, thermal comfort, water use, natural ventilation, and photovoltaic
systems. Though EnergyPlus has these innovative capabilities, its major limitation is
that it runs slower than DOE-2. According to a study conducted by Tianzhen H., et al
[4], at a 15-minute time step, EnergyPlus runs much slower than DOE-2.1E by a
factor of 105 for a large office building to 196 for a hospital building. At a 60-minute
time step, EnergyPlus still runs even slower than DOE-2.1E by a factor of 25 for the
large office building to 54 for the hospital building.
According to EnergyPlus run time analysis report [5] prepared by the
Simulation Research Group, EETD, LBNL, simulation settings that can have
significant impacts on EnergyPlus run time include the length of RunPeriod,
Number_of_Timesteps_per_Hour for loads calculations, heat balance solution
algorithm, solar distribution and reflection calculation algorithm, system convergence
limits, shadow calculation interval and the length of the warm up period. This analysis
shows that the longer the run period, longer is the EnergyPlus run time.
Some efforts to use parallel computing for reducing simulation time for a
group of simulations have been reported. Zhang Y [6] developed a Java based tool to
run EnergyPlus on parallel machines specifically for the parametric analysis where
multiple design alternatives have to be analyzed simultaneously. GenOpt [7] is an
optimization program, used to carry out the parametric analysis using multiple
computers / processors. Both these tools help in speeding up the parametric
simulations but do not address speeding individual simulation runs.
In this paper, an approach that uses data parallelization paradigm has been
proposed to increase the simulation speed of single simulation run. In this approach,
annual simulation is segmented into smaller multiple simulations, each of which can
run on a dedicated CPU in a computer cluster or on different cores on the same
computer. This segmentation reduces the simulation run time and increases the speed
of simulation. Segmentation of simulation results in minor deviations between the
results obtained through segmented simulations and annual simulations. To achieve
accurate results, effect of the number of warm up days and shadow calculation days
were analyzed to arrive at different alternatives. This method was tested on 13 models
and 5 cities covering various climatic conditions. Speed gains of 3.2x to 5.8x were
achieved by running 12 one-month RunPeriod segments in parallel as compared to
annual simulations. On an annual basis, a maximum deviation of 0.06% was observed
in cooling, heating, and lighting consumption of buildings. In a month-to-month
comparison between the segmented and annual simulation, this deviation in the worst-
case scenario was 1.7% for heating and 0.8% for cooling energy consumption.
2. Approach
One parameter that significantly affects the runtime of a single simulation run is the
RunPeriod of that simulation. RunPeriod is the object in EnergyPlus which contains
several fields, including information on the begin date and end of the simulation. This
information is assigned by the user in the EnergyPlus Input Data File (.idf). In the
approach presented in this paper the annual simulation was divided into twelve
segments (splitting the single annual .idf into 12 monthly .idf files) with smaller
RunPeriod (monthly) that were then run in parallel. This monthly simulation took
significantly less time in comparison to the annual simulation. These monthly
simulations were independently run in parallel on several computers. Results of all the
parallel runs were then collated. This results in increase in simulation speed. This
approach is explained in Figure 1. In this study, simulations have been performed on a
cluster of computers and the results of these segmented simulations were compared
with the results of the respective months in the annual simulation. It was observed that
segmentation of simulation resulted in minor deviations between the results obtained
through segmented simulations and annual simulations. The causes of these
deviations and the alternatives to reduce them are discussed in the following sections.
Figure 1
2.1 Deviations Observed
When the annual simulation was segmented into twelve monthly simulations it was
observed that there were some deviations in the monthly cooling and heating energy
consumption of the segmented simulations as compared to the corresponding months
of the annual simulation. This was observed in all the 65 cases (13 models x 5 cities)
that were simulated (details given in Section 3- Simulation Runs). Table 1 gives the
percentage deviations in monthly heating and cooling for one of the 65 cases
(“ElemSchool” model for Chicago climate)
Table 1
Besides the deviations observed on the monthly and annual values, large
deviations were also observed in the hourly values of heating and cooling. Figure 2
shows the percentage deviations in the hourly values for cooling and heating energy
consumption obtained in the segmented simulations with the corresponding hourly
values of annual simulation of “ElemSchool” model using New York weather file.
Figure2a
Figure2b
2.2 Causes of Deviations
The deviations discussed in the previous section were due to the following factors:
(1) Mismatch between the day of week for the dates in segmented simulation
and corresponding dates in annual simulation: The field
Day_of_Week_for_Start_DayDay, can be set in the simulation model. The
value given in this field can be used to override the day of the week indicated
in the weather file. Valid days of the week (Sunday, Monday, Tuesday,
Wednesday, Thursday, Friday, and Saturday) can be entered. When weekdays
are used, each subsequent day in the period will be incremented. If this field is
filled with a valid day of the week, all the segmented simulations will take that
day of the week as a static value for the Day_of_Week_for_Start_Day. This
will result in different day of week for the same date in the annual simulation
and the segmented simulation. This difference can further lead to a difference
in the number of total working days in the annual simulation and the sum of
all working days in the segmented simulation, which will therefore affect the
results. For example, if 1st of July is Monday in the annual simulation and the
field Day_of_Week_for_Start_Day is set to Sunday, then the start day for the
seventh segment (July) will be Sunday and will not match with the day of
week in the annual simulation. Further, there will be four Sundays in the
month of July in the annual simulation whereas in the segmented simulation
for July there will be five Sundays. A difference of even one working day can
cause a significant difference in the energy consumption for that month.
(2) Inadequate warm up if the first day of segment is a holiday: EnergyPlus
performs a warm up on the first day of simulation period in order to set the
values of certain variables. Convergence of the simultaneous heat
balance/HVAC solution is reached when the criteria for either the loads or
temperature is satisfied. In case of an annual simulation, the simulation starts
on 1st January, and the warm up calculations are carried out for first day i.e.
1st January. However, for individual segments, the warm up takes place on the
first day of the segment i.e. the first day of each month. In case, the first day of
the month happens to be a holiday, then the warming up of the model for that
segment will be done according to non-operational conditions. This will result
in deviated results for the next few working days of that segment and hence
affect the total monthly consumption.
(3) Dates of shadow calculation in annual simulation not matching with the
dates for segmented simulations: To determine the amount of solar radiation
entering a building and to calculate the amount of cooling or heating load
required to maintain the set point temperature, shadow calculations (sun
position, etc) are performed by EnergyPlus. By default, EnergyPlus performs
these calculations for a frequency of 20 days throughout the RunPeriod. The
shadowing Calculation_Frequency variables in the idf specify the number of
days after which the next shadowing calculation period starts . The dates for
which shadowing calculation is performed depends on this frequency. There
can be deviations in the results of simulations if the shadowing calculation
dates in the segmented simulation do not align with those in the annual
simulation. Therefore, it is important to align the shadowing calculation dates
of the segmented simulations to map with the shadowing calculation dates in
the annual simulation.
2.3 Alternatives Proposed to Decrease the Deviations
The deviations mentioned may be reduced by changing some of the simulation
settings and variables. The solutions proposed are:
(1) Selecting Day_of_Week_for_Start_Day from the weather file: If the Day of
week for Start Day is chosen as UseWeatherFile, the day of week for all the
dates in the annual simulation will be in synchronization with the day of week
for the corresponding dates in the segmented simulation, hence ensuring the
same number of working days between the segmented and annual simulations.
By default, EnergyPlus specifies a particular day (such as Monday) as the Day
of week for Start Day, due to which the start day of the each segmented
simulation will be this day specified. However, this same date in the annual
simulation would be some other day, due to which there would
synchronization problem of the day and dates in annual and segmented
simulations. UseWeatherFile option picks the calendar day for that particular
date from the weather file.
(2) Ensuring first day of the segment is a working day for adequate warming
up: If the first day of the month happens to be a holiday then some days from
the previous month can be added to the segment to ensure adequate number of
working days between the first day of the simulation and the first the of the
month. This increases the RunPeriod of the segment and the first day of
segment would not be the 1st of the month. These extra number of days added
to each segment would be referred to as warmupPlus days in this paper.
(3) Synchronizing shadow calculation dates of the segments with the annual
shadow calculation dates: If the shadowing calculation dates of the annual
simulations and those of each segmented simulation are kept in
synchronization with each other, the effects of unaligned shadow calculations
can be minimized. This is achieved by adding some more days from the
previous month to the segment so that the start date of the shadow calculation
period of the segmented simulation align with the annual shadowing dates.
This extra number of days added before each segment will be referred to as
shadowSync days in this paper.
The following measures were taken to implement these solutions:
(1) The Day_of_Week_for_Start_Day element in the RunPeriod class was
changed to UseWeatherFile. This ensured that the total number of working
days for all the months in the segments were the same as the total number of
working days in the annual simulation.
(2) To every segment 7 warmupPlus days of simulation were added before the
simulation days of the segment. This ensured adequate warm up especially if
the 1st day of the month happened to be a holiday.
(3) Some more days were added in the beginning of the segment to ensure that the
shadow calculation dates in the segment mapped with that of the annual
simulation.
Before arriving at this proposed solution, some more variants were analyzed. In
addition to the 7 warmupPlus days, simulations were done with 0, 14 and 21
warmupPlus days. Two more variants with synchronized shadow calculation days
were also analyzed. This resulted in six different alternatives, which were based on
the combination of warmupPlus days and shadowSync days. These six combinations
were experimented on to find out an alternative that results in the least deviation in
cooling, heating, lighting and equipment consumption when compared with the
annual simulation.
The six alternatives and their names as used in the paper are:
(1) W0: 0 warmupPlus days
(2) W7: 7 warmupPlus days
(3) W14: 14 warmupPlus days
(4) W21: 21 warmupPlus days
(5) W0Sync: shadowSync days with minimum allowed value as 0
(6) W7Sync: shadowSync days with minimum allowed value as 7
For all the six alternatives, the Day_of_Week_for_Start_Day is selected as
UseWeatherFile
3. Performance of the alternatives
As discussed in earlier sections warm up and synchronization of shadow calculation
dates have a significant impact on the simulation runtime and the accuracy of the
results. Simulations were conducted to evaluate the performance of the six
alternatives proposed in the previous section in terms of accuracy and speed up.
“ElemSchool” model has been simulated for the five cities (Chicago, San Francisco,
Tampa, New York, and Houston) with all the six alternatives. Effect of shadow sync
and warm up days on accuracy of results and the speed gain are discussed in Sections
3.1 and 3.2.
3.1 Accuracy of the alternatives
The six alternatives were simulated to analyze the deviations in results of the
segmented simulations in comparison with corresponding results of the annual
simulation. Large deviations in monthly consumption values were observed when
“ElemSchool” model was simulated for Chicago weather data. Table 2 provides the
percentage deviations in Cooling Electricity consumption for various months between
the segmented simulations and annual simulations for all the six alternatives. The
maximum deviation in monthly cooling consumption was 0.098% for W21 alternative
in the month of July. Table 3 shows the percentage deviations in the heating energy
consumption. The maximum deviation in monthly heating consumption was 1.87%
for W0 and W0Sync alternative in month of April. In both the tables, the minimum
deviation values for each month are highlighted in bold.
Table 2
Table 3
Some of the observations from Tables 3 and 4 are:
For cooling, W0Sync and W7Sync show least deviations
For heating, W7Sync shows the least deviations
W14 shows better accuracy for both heating and cooling amongst W0, W7,
W14 and W21.
In the annual results, W7Sync show significantly less deviations compared to
the other methods.
These observations clearly indicate that the alternatives W0Sync and W7Sync are
more accurate and are proved to be better by a significant difference when compared
to the other alternatives. As observed in the Table 2 the deviations in cooling
consumptions are higher in summer. Since the cooling values are of lower orders for a
city like Chicago, a small deviation in the cooling consumption will give a greater
percentage difference. Another interesting observation is that W14 shows better
results than W21, even when W21 has more days of simulation before the first day of
the segmented simulation. The more accurate results of W14 are due to the shadowing
frequency of 15 days in this model. Extra 14 days synchronizes the shadowing periods
better than the extra 21 days taken in W21.
Tables 3 and 4 show the comparison of the six alternatives based on their
performance on monthly and annual basis. W0Syn and W7Sync both give fairly
accurate results for monthly and annual values. Deviations in the hourly values were
then checked to compare the two alternatives. Large deviations in hourly consumption
values were observed when “ElemSchool” model was simulated for New York
weather data.
Figure 3a and 3b show percentage deviations in the hourly cooling and heating
consumption between the segmented simulations and the annual simulations for the
W0Sync alternative. The maximum deviation in hourly value of cooling was 54% and
heating was 18.1%. It can be observed from the graphs that even when W0Sync is
accurate on the monthly and the annual values, there are big deviations on the hourly
values.
Figure 3a
Figure 3b
Figure 4a and 4b show the percentage deviation in the hourly cooling and
heating consumption between the segmented simulations and the annual simulations
when applying the alternative W7Sync.The maximum deviation in hourly value of
cooling was 0.01% and heating was 0.04%. Hence, W7Sync is not only accurate on
the monthly and the annual values; it is also very accurate on hourly values.
Figure 4a
Figure 4b
The graphs clearly indicate that the deviations observed in the hourly data for
W0Sync were reduced in the W7Sync alternative. The deviations in the graph also
strengthen the proposition that the addition of warmupPlus days is important. April,
for instance shows heavy deviations in both cooling and heating. This can be
attributed to the fact that 1st April is start date for one of the shadow calculation
periods. Hence in the W0Sync setup, there are 0 warmupPlus days leading to
deviations. While in the W7Sync setup, there are 15 warmupPlus days (since the
shadowing frequency is 15 days) which results in negligible errors. Since the
shadowing frequency for this simulation set is 15 days, other months are able to align
appropriately and result in fewer deviations. However, if the frequency is some other
number, for example 20, which causes less alignment, then these deviations will
increase. Due to this, the W7Sync algorithm will work efficiently for any value of
shadow frequency and will provide accurate results.
3.2 Speed gain of the alternatives
To evaluate the performance of all the alternatives in terms of simulation run time, the
model “ElemSchool” with “Chicago” weather file was used and the time taken by
each simulation was observed. Table 4 shows the simulation run time (in seconds) and
speed gain of monthly segments, the time taken by the annual simulation (in seconds)
and the effective speed gain for the six alternatives. Speed gain is obtained by
dividing the time of annual simulation by the time taken by segmented simulation.
Variation can be observed in the time taken by simulation by the segments for
different months. This is due to the difference in number of shadow calculations and
the time taken for convergence of various HVAC calculations. It is observed that in
colder months such as January and December the time taken is more due to an
increased number of iterations in the HVAC. The segment that takes the maximum
time governs the effective speed gain for that alternative. Hence, the minimum
speedup value obtained for that set of segments is highlighted in bold. It is clear from
the table that the overall speed gain decreases with the number of warmupPlus days.
W0 is fastest but not very accurate and W7Sync is slower but more accurate. Due to
this accuracy, W7Sync is selected for demonstrating the proposed approach of
speeding up EnergyPlus using parallel computing.
Table 4
3.3 Performance Analysis of W7sync
After selecting W7sync alternative for demonstrating the proposed approach, this
alternative has been further analysed to observe the effect of the following on the
speed gain: Number_of_Timesteps_per_Hour, number of processing units, and
overheads with respect to extra time taken in transferring files over the network and
collating the output.
(1) Number_of_Timesteps_per_Hour: The ElemSchool model for Chicago
climate with varying time steps has been simulated to study the impact of time
steps on speed up. It has been observed that the simulation speed gain
increased with the increase in the value of Number_of_Timesteps_per_Hour.
Speed gain with the value of Number_of_Timesteps_per_Hour as 1 is 2.73
Speed gain with the value of Number_of_Timesteps_per_Hour as 4 is 3.15
Speed gain with the value of Number_of_Timesteps_per_Hour as 20 is
4.12
(2) Number of processing units: To demonstrate the effect of number of
processing units on the speed gain we simulated ElemSchool model for
Chicago on 2, 3, 4, 6 and 12 processing units. It has been observed that the
speed gain decreases with the decrease in number of processing units. Speed
gain with the number of processing units is shown in Table 5
(3) Overheads with respect to extra time taken: This whole process has some
overheads with respect to extra time taken for pre-processing .idf file, sending
the .idf files to processors, collecting the results from processors and collating
the results to form a single file. The overhead depends largely on network
speed, the size of .idf file and the size of the output file which in turn depends
on the number of variables and their reporting frequency. In the simulations
which were performed it was observed that the overhead was close to 5% of
the time taken by the slowest segment. For example, for ElemSchool model,
the overall overhead was about 15 seconds, and the time taken by the slowest
segment was 422 seconds, so the total overhead is 4.02 %.
4. Simulation and results
In order to demonstrate the approach over the selected alternative W7Sync,
simulations were performed using EnergyPlus V 4. Thirteen different building models
were simulated across five different cities. Details of the simulation models, weather
data and the results are provided in the following sections.
4.1 Building Prototypes
The U.S. Department of Energy (DOE) – through three of its national laboratories –
has developed a set of standard benchmark building models for new and existing
buildings [8]. These models are used in the paper to test the proposed approach. The
revised and latest models for EnergyPlus v5 are now referred as „commercial
reference building models for new construction‟ [9]. The changes between the
standard benchmark model used in this paper and the new reference building models
are listed in „summary of changes from v1.2_4.0 to v1.3_5.0‟ document [10].These
models help in providing consistent standardized models for the analysis of the
results. To analyze the performance of the proposed alternative over different models
and climates, the benchmark buildings that come along with the EnergyPlus
installation were used for the simulations. Thirteen different building models- twelve
benchmark buildings for new construction and one example model from EnergyPlus
were used. The example file of EnergyPlus installation is a 5 Zone VAV model with
daylight sensors and is used to check the concept and the accuracy of results for a
building with daylight sensors. Some important characteristic features of these models
are listed in Table 6.
Table 6
4.2 Climate Zones
All the thirteen different models were simulated for five different cities of USA which
belong to four different climate zones as shown in Table 7.
Table 7
4.3 Performance of the proposed alternative- W7Sync
To evaluate the performance of the proposed alternative, 65 cases (13 models x 5
cities) were simulated. Results show that W7Sync results are reasonably accurate for
all the cases, as observed in the earlier section. To show a compact summary of the
results achieved for all the 65 cases, the deviations in annual cooling, heating,
equipment and lighting energy consumption and the speed gains are as listed in the
Table 7.However, analysis has been performed on all the cases and the deviations in
hourly consumptions and the monthly consumptions were noted to be very minor, in
similar limits as discussed in the previous sections.
Table 8
4.4 Results
From the simulations performed over thirteen different models using five different
cities, it was observed that the speed up achieved varied from 3.15x to 5.84x. The
minimum speed gain was in model ElemSchool for Chicago, which was 3.15x. The
maximum gain was achieved for the MidApt and Chicago weather, which was 5.84x.
The average gain observed for all the 65 cases was 4.77x. One interesting observation
made was that the speed gain for a model depends on the climate for which the
simulation is run.
The maximum deviation in cooling consumption was noted in SmallOffice model and
in heating consumption it is noted in MidApt model. The maximum and minimum
deviations annually across all models are as shown in Table 9.
Table 9
Table 10 shows the maximum percentage deviations that occurred in the monthly
consumption of heating, cooling and lighting, out of all the 65 cases that are
simulated. The deviations in case of equipment electricity consumption were found to
be 0 for all the cases.
Table 10
5. Conclusion
The approach of dividing a simulation into segments and running them in parallel
decreases the simulation run time, and is accurate enough to be used practically for
increasing the speed of an EnergyPlus simulation. The proposed approach has been
applied over 13 different models and five different weather files, to check the
accuracy of the results achieved when compared to the annual simulation results. On
annual basis, a maximum deviation of 0.06% was observed in cooling, heating, and
lighting consumption. In a month-to-month comparison between the segments and
annual simulation, the maximum deviation was 1.7% for heating and 0.8% for
cooling. The speed gain in the simulation would range between 3x to 6x. For
performing huge number of simulations during the conceptual design stage and for the
parametric analysis, the proposed approach can be very useful. And, for the final
analysis a single run simulation as per conventional approach can be performed to get
precise results. Further study would include the development of a tool that will use the
described algorithm on a cluster of computers or processors.
References
[1] EnergyPlus Energy Simulation Software, Available from: www.energyplus.gov
[2] DOE-2, Available from: http://gundog.lbl.gov/dirsoft/d2whatis.html
[3] BLAST-Building Load Analysis and System Thermodynamics, Available from:
http://www.cecer.army.mil/facts/sheets/cf-47.pdf
[4] Tianzhen H, Fred B, Philip H, Stephen S, Michael W. 2008. Comparing Computer
Run Time of Building Simulation Programs. Building Simulation , 1 (3), 210-213.
[5] Hong, Tianzhen. (2009). EnergyPlus Run Time Analysis [online]. Lawrence
Berkeley National Laboratory: Lawrence Berkeley National Laboratory. LBNL Paper
LBNL-1311E. Available from: http://escholarship.org/uc/item/36h4m5z0
[6] Yi Z. 2009. “Parallel” EnergyPlus and the development of a parametric analysis
tool. In: Eleventh International IBPSA Conference, 27-30 July 2009 Glasgow,
Scotland: 1382-1388.
[7] GenOpt-Generic Optimization Program, Available from:
http://gundog.lbl.gov/GO/index.html
[8] P. Torcellini, et al. 2008. DOE Commercial Building Benchmark Models. In:
ACEEE Summer Study on Energy Efficiency in Buildings, 17-22 August 2008 Pacific
Grove, California.
[9] Commercial Reference Building models for New Construction, Available from:
http://www1.eere.energy.gov/buildings/commercial_initiative/new_construction.html
[Accessed 4 October 2010]
[10] Summary of Changes from v1.2_4.0 to v1.3_5.0, Available from:
http://apps1.eere.energy.gov/buildings/publications/pdfs/commercial_initiative/refbld
gs_changes_v40tov50.pdf [Accessed 4 October 2010]
Table 1: Percentage deviation in monthly cooling and heating consumption
between segmented and annual run
% Deviation
Months Cooling Heating
Jan 0 -0.0004
Feb 0 -0.1908
March 0.0004 0.0019
April 0 -1.8722
May -0.0071 0.5679
June -0.0017 0.2258
Jul 0.0977 0.0925
Aug 0.0089 -0.5372
Sep -0.0076 0.0948
Oct 0.0555 0.1534
Nov -0.0013 0.0856
Dec 0 -0.0118
Annual 0.0248 -0.1128
Table 2: Deviations in cooling electricity consumption for the six different
alternatives that were simulated for the “ElemSchool” and Chicago climate. The
maximum deviation was 0.098% for W21 alternative in the month of July
Alternatives
Months W0 W7 W14 W21 W0Sync W7Sync
Jan 0 0 0 0 0 0
Feb 0 0 0 0 0 0
March 0.0004 -0.0012 0.0004 -0.0012 0 0
April 0 0.1064 0 -0.0129 0 0
May -0.0071 -0.0794 -0.0025 -0.0362 -0.0071 -0.0025
June -0.0017 -0.0316 -0.0008 -0.0408 0 0
Jul 0.0977 0.097 0.097 0.0984 0.042 0.042
Aug 0.0089 0.0315 0.009 0.0315 -0.0008 -0.0008
Sep -0.0076 -0.0097 -0.008 -0.0087 -0.0068 -0.0068
Oct 0.0555 0.0156 0.056 0.0526 -0.0012 -0.0012
Nov -0.0013 -0.0178 -0.0178 -0.0006 0 0
Dec 0 0 0 0 0 0
Annual 0.0248 0.0188 0.0228 0.018 0.0082 0.0086
Table 3: Deviations in heating consumption for the six different alternatives that
were simulated for the “ElemSchool” and Chicago climate. The maximum
deviation was 1.87% for W0 and W0Sync alternative in month of April.
Alternatives
Month W0 W7 W14 W21 W0Sync W7Sync
Jan -0.0004 -0.0004 -0.0004 -0.0004 -0.0004 -0.0004
Feb -0.1908 -0.0436 0.0095 -0.0387 0.0038 -0.0024
March 0.0019 -0.0579 -0.0236 -0.0395 0 0
April -1.8722 -0.0638 0 -0.0793 -1.8722 0
May 0.5679 -0.0641 0.0002 -0.0288 0.5679 0.0002
June 0.2258 -0.078 0.0181 -0.1015 -0.0145 0
Jul 0.0925 0.0641 0.0666 0.0715 0.0413 0.0376
Aug -0.5372 -0.1311 -0.0869 -0.1311 -0.103 -0.1118
Sep 0.0948 -0.1657 -0.0318 -0.1198 -0.0037 -0.0037
Oct 0.1534 0.1344 0.1456 0.1803 0.0511 0.0511
Nov 0.0856 -0.0967 -0.0967 -0.0944 0.0172 0.0172
Dec -0.0118 0.0174 -0.0193 -0.0411 0 0
Annual -0.1128 -0.0235 -0.0114 -0.0314 -0.0843 0.0021
Table 4: Simulation run time (in seconds) and speed gain of monthly segments,
the time taken by the Annual simulation (in seconds) and the effective speed gain
for the six alternatives.
Month W0 W7 W14 W21 W0Sync W7Sync
Jan 355 3.7 360 3.6 357 3.7 358 3.6 356 3.7 358 3.7
Feb 277 4.8 354 3.7 379 3.5 426 3.1 282 4.7 385 3.5
March 195 6.8 223 5.9 267 4.9 292 4.5 262 5.0 266 5.0
April 179 7.4 192 6.8 209 6.3 222 5.9 179 7.3 212 6.3
May 178 7.4 192 6.8 204 6.5 217 6.0 178 7.4 207 6.4
June 178 7.4 194 6.7 205 6.4 218 6.0 180 7.3 211 6.3
Jul 203 6.5 217 6.0 236 5.6 242 5.4 207 6.4 240 5.6
Aug 179 7.4 195 6.7 210 6.3 222 5.9 183 7.2 215 6.2
Sep 183 7.2 198 6.6 211 6.2 225 5.8 189 7.0 218 6.1
Oct 175 7.5 189 6.9 199 6.6 214 6.1 180 7.3 211 6.3
Nov 228 5.8 242 5.4 243 5.4 264 4.9 235 5.6 261 5.1
Dec 336 3.9 388 3.4 407 3.2 423 3.1 357 3.7 422 3.2
Annual 1316 1308 1314 1302 1316 1331
Effective
speed gain 3.7 3.4 3.2 3.1 3.7 3.2
Table 5: Speed gain achieved for ElemSchool model simulated for Chicago
climate, with varying number of processing units
Number of
processing units
Speed Gain
12 3.2
6 2.6
4 2.3
3 2.1
2 1.8
Table 6: Characteristics of the thirteen building models used for testing the
proposed approach of speeding up EnergyPlus
Model Name
Flo
or
Area
(th
ou
san
d m
2)
Nu
mb
er o
f F
loors
Zone Definition HVAC System HVAC Plant
Inte
r z
on
e S
urfa
ces
Peo
ple
Lig
hts
Win
do
ws
Da
yli
gh
t
Zo
na
l E
qu
ipm
en
t
Cen
tral
Air
Ha
nd
lin
g
Eq
uip
men
t
Sy
stem
Eq
uip
men
t A
uto
size
Co
ils
Pu
mp
s
Boil
ers
Ch
ille
rs
To
wer
s
ElemSchool 6.871 1 x x x x x x x x
Fastfood 0.2323 1 x x x x x x x x
HighSchool 24 2 x x x x x x x x x x x
Hospital 18.697 6 x x x x x x x x x x x
LargeHotel 9.366 6 x x x x x x x x x x x
LargeOff 42.757 12 x x x x x x x x x x x
MedOff 4.952 3 x x x x x
MidApt 3.135 4 x x x x x x x x
Retail 3.882 1 x x x x
SitdownRestrau 0.511 1 x x x x
SmallHotel 1.958 2 x x x x
SmallOffice 0.511 1 x x x x
5ZoneVAV 0.8 1 x x x x x x x x x x x x
Table 7: Climate data used for running the simulation models
City Climate zone Climate type Weather file name
Chicago 5A Cool Humid USA_IL_Chicago-
OHare.Intl.AP.725300_TMY3.epw
San
Francisco
3C Warm-
Marine
USA_CA_San.Francisco.Intl.AP.724940_TMY3
.epw
Tampa 1A Very hot-
humid
USA_FL_Tampa.Intl.AP.722110_TMY3.epw
New York 5A Cool Humid USA_NY_New.York-
J.F.Kennedy.Intl.AP.744860_TMY3.epw
Houston 2A Hot- Humid USA_TX_Houston-
D.W.Hooks.AP.722429_TMY3.epw
Table 8: Percentage annual deviations for Cooling, Heating, Equipment and
Lighting consumption and effective speed up for 13 models and 5 cities
Model Chicago San Francisco Tampa New York Houston
ElemSchool
Cooling 0.0086 0.0062 0.0039 0.0007 0.0052
Heating 0.0021 -0.0011 -0.0053 -0.0010 -0.0025
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 3.15x 3.68x 3.76x 3.33x 3.5x
Fastfood
Cooling 0.0088 0.0451 0.0032 0.0116 0.0047
Heating 0.0002 -0.0004 -0.0021 0.0004 0.0017
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.3x 4.86x 5.05x 4.54x 4.79x
HighSchool
Cooling 0.0020 0.0083 0.0056 0.0051 0.0104
Heating 0.0015 -0.0021 0.0036 0.0020 0.0038
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.92x 5.43x 5.46x 5.37x 5.4x
Hospital
Cooling 0.0002 0.0006 0.0009 0.0005 0.0006
Heating 0.0006 -0.0012 -0.0019 -0.0015 -0.0021
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 5.61x 5.63x 5.7x 5.2x 5.44x
LargeHotel
Cooling 0.0180 0.0295 0.0145 0.0223 0.0170
Heating 0.0004 -0.0002 -0.0020 0.0006 0.0009
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.84x 5.22x 5.77x 5.23x 5.58x
LargeOff
Cooling 0.0245 0.0351 0.0119 0.0176 0.0029
Heating 0.0009 -0.0001 0.0108 0.0049 -0.0013
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 5.23x 5.47x 5.18x 5.31x 5.04x
MedOff
Cooling 0.0233 0.0211 0.0112 0.0263 0.0172
Heating 0.0005 -0.0044 0.0021 0.0001 0.0018
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.16x 4.4x 4.3x 4.31x 4.29x
MidApt
Cooling 0.0205 0.0302 0.0164 0.0231 0.0164
Heating 0.0256 0.0165 0.0185 0.0046 0.0051
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 5.84x 5.42x 5.8x 4.92x 5.62x
Retail
Cooling 0.0078 0.0418 0.0059 0.0100 0.0057
Heating 0.0034 0.0007 0.0015 0.0011 -0.0012
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.36x 4.5x 4.44x 4.53x 4.34x
SitdownRestrau
Cooling 0.0043 0.0348 0.0020 0.0062 0.0031
Heating 0.0001 -0.0001 -0.0016 0.0001 0.0006
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.33x 4.84x 5.07x 4.46x 4.69x
SmallHotel
Cooling 0.0123 0.0153 0.0106 0.0141 0.0116
Heating 0.0002 -0.0014 -0.0016 -0.0001 0.0012
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 3.4x 3.51x 3.51x 3.42x 3.5x
SmallOffice
Cooling 0.0491 0.1599 0.0277 0.0607 0.0334
Heating 0.0001 0.0003 -0.0007 0.0007 0.0011
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0000 0.0000 0.0000 0.0000 0.0000
Speed Gain 4.18x 4.58x 4.58x 4.53x 4.55x
5ZoneVAV
Cooling 0.0207 0.0437 0.0113 0.0207 -0.0337
Heating -0.0009 -0.0103 0.0025 0.0021 -0.0108
Equipment 0.0000 0.0000 0.0000 0.0000 0.0000
Lighting 0.0054 -0.0042 -0.0220 -0.0399 -0.0168
Speed Gain 5.45x 5.47x 5.44x 5.61x 5.43x
Table 9: Maximum and Minimum percentage deviations in annual heating,
cooling, lighting and equipment consumption
Heating Cooling Lighting Equipment
Minimum Deviation 0.0001 0.0002 0 0
Maximum Deviation 0.0256 0.0607 0 0
Table 10: Maximum percentage deviations in monthly heating, cooling and
lighting consumption
Cooling Heating Lighting
Maximum Value -0.8015 1.7407 -0.4307
Model Medium Office Small Office 5ZoneVAV
City New York New York Tampa
Figure1: Flow diagram of the entire process
Figure2a: Percentage deviation in hourly cooling electricity between segmented
simulation and the annual simulation for ElemSchool model with Chicago city.
The maximum hourly deviation is 100%
Figure2b: Percentage deviation in hourly heating consumption between
segmented simulation and the annual simulation for ElemSchool model with
Chicago city. The maximum hourly deviation observed in an hour is as high as
1340 %.( Instead of the peak a general range in which the percentage deviations
exist has been considered to scale the plot)
Figure 3a: Percentage deviation in hourly cooling electricity between segmented
simulation and the annual simulation for ElemSchool model with New York city
weather. The maximum hourly deviation is 54%.
Figure 3b: Percentage deviation in hourly heating consumption between
segmented simulation and the annual simulation for ElemSchool model with
New York city weather. The maximum hourly deviation is 18.1%.
Figure 4a: Percentage deviation in hourly cooling electricity between segmented
simulation and the annual simulation for ElemSchool model with New York city
weather. The maximum hourly deviation is 0.01%.
Figure 4b: Percentage deviation in hourly heating consumption between
segmented simulation and the annual simulation for ElemSchool model with
New York city weather. The maximum hourly deviation is 0.04%.