historic weather files for model calibration

3 posts / 0 new
Last post

A recent LEED MV plan review comment asked "please indicate the proposed calibration method to account for the local weather conditions during the performance period."

This raises the question for me - are model results significantly impacted by the difference between historic weather data and performance-period data?

When I'm engaged in Measurement and Verification, I can install a weather station that records data for performance period to allow for calibration. (http://www.gsd.harvard.edu/research/gsdsquare/Publications/BuildingSimulation2009.GundHallModel.pdf)

But for an investment-grade energy audit with historic bills, I'm not sure where to turn for all the variables needed to construct a weather file.

Anyone on this list have a recommendation?

(I'm specifically looking for NYC, 2009, weather data for an eQUEST model.)

Aaron Dahlstrom , PE, LEED(r) AP

Dahlstrom, Aaron2's picture
Joined: 2011-09-30
Reputation: 4

Aaron-

The question "are model results significantly impacted by the difference
between historic weather data and performance-period data" is a complex
one, however, here is an approach that may be helpful if you need to
construct your own custom weather file.

You can purchase historical weather data available from NOAA station
data from the NCDC . However,
these data sets do not include solar radiation data and often need some
QC work (remove extraneous observations, fill gaps, etc). In the past,
I have also downloaded weather data from the EnergyPlus

site in IWEC format, and converted to .bin using a combination of the E+
weather data processor and EPW conversion tool available on the DOE-2
Website . However, although
a great deal (a FREE service), in my experience, this data can have gaps
that are too large to fill with the algorithms NREL prescribes, which
then leads you to either piecing this together w/ some other data
source, or, only evaluating the model performance over the periods you
have data for. For many of my projects, I need a complete year or more,
so the E+ method was not ideal.

I recently created some DOE-2 .bin weather files using the following
approach, using data formatting/calculation procedures automated in
MATLAB, and then the "DOEWth.exe" utility (an older, command line
program also available from DOE-2.com) to generate a .bin weather file.
Once automated, this process can be completed in a relatively short
period of time. The following is a brief explanation of the process I
used that may give you some ideas:

1) Downloaded NOAA data (Integrated Surface Data) for closest site.

2) Cleaned data of extraneous points and filled gaps in data using NREL

filling routines. There is usually always one or more observations
recorded for every hour, however, in many cases one or more of the
variables you need for the simulation weather file may not have been
reported.

If you can obtain measured solar data for your location, skip steps 3-5

3) Used the Zhang-Huang solar model (discussed in the E+ engineering
manual) to estimate total horizontal solar radiation with custom model
parameters developed using least-squares regression to TMY3 data. In
this case, I assumed TMY data was more or less "true", however, for one
station data I worked with, I noticed some cloud cover observations did
not appear to be consistent the reported solar radiation data. However,
the custom coefficients yeilded better results than the default
coefficients reported in the E+ literature when I compared the model
output to a short sample of actual measured solar radiation data I had.

4) Used another model to determine the diffuse solar radiation
component from the total global radiation. In my case, I used a custom
model developed for the Pacific NW (published in a thesis) however,
there are many similar models developed from various datasets. Orgill
and Hollands is a popular model, although they are all very similar.

5) Once you have the two solar components above, direct normal solar
radiation can be readily calculated.

6) With all of the necessary data now assembled, format into the TMY2
format for processing into a .bin file using the DOEWth utility. Using
the DOEWth output summary file and a weather file plotting program, like
D-View, you can "inspect" the measured data to make sure it matches what
you expect and aligns with daily or month averages more readily available.

A few notes:
- The DOEWth utilty does have format methods that will calculate solar
data, however, I choose to pre-process the solar data using the solar
radiation models I preferred, which leads you to using the TMY2 format
method.
- I did not calculate illuminance data since my project did not include
daylighting controls. There are models available for calculating
illuminance, and the DOE-2 program may use a model to estimate it from
solar radiation data (need to brush up on this section of the engineers
manual to confirm this).
- Related to the above comment, there are many different models out
there for calculating solar radiation/illuminance data from other
measured parameters. I choose the above because I felt like the models
best captured the variables that I thought were important, and to
lesser degree, I could readily implement them in my programming. Other
than comparing these models to actual TMY data, I have not rigorously
compared these model to others available, so you may want to explore
others.
- For any source of weather data you pursue, I would emphasize reviewing
how data is filled and non-measured variables are calculated (i.e. what
models were used).

I just realized this post may use the word "model" a record number of
times, but hope you find it useful.

David Reddy

David Reddy3's picture
Offline
Joined: 2011-09-30
Reputation: 0

Hi Aaron,

Building your own weather file can be difficult and tedious, especially gathering the appropriate solar data. Here are a couple alternatives that may be appropriate depending on the specific context of your project:

1 - Compare the heating and cooling degree days between your performance period and the TMY weather. If you have multiple years of utility data and associated weather data, the weather variations each year might average out to be very similar to the TMY. If they are moderately different, you could attempt to adjust modeling results based on this difference?
2 - If you create a custom weather file for a short performance period (e.g. one year or less) and tune your model to the utility data from that period, then your results are only applicable to performance during that period. Maybe that's what you want. A different way to look at it would be to generalize the results to a typical year (i.e. the TMY weather). One method is to develop regression fits of measured energy use as a function of outside air temperature. For example, graph hourly hot water use against measured hourly outside air temperature and do a curve fit. The relationship is generally linear but perhaps with a change-point (where the slope changes). With the regressions, you can apply these relationships to the weather data in the TMY file to generalize the measured energy use to energy use during a typical year. Thus, calibrating a TMY-based energy model to the normalized baseline would be an apples to apples comparison and the results would be generalized to typical conditions rather than just the specific conditions during your performance period. There's more to it, but that's the general gist.

Also, there was a great spreadsheet tool that was shared on bldg-sim a couple years ago that used macros to automatically gather and compile actual (hourly or even sub-hourly) weather data from weather stations on www.wunderground.com. But no solar data. I use this tool exclusively now for gathering real weather data.

Good luck,
Hwakong

Date: Mon, 29 Nov 2010 22:09:48 -0800
From: david.j.reddy1 at gmail.com
To: ADahlstrom at in-posse.com; bldg-sim at lists.onebuilding.org
Subject: Re: [Bldg-sim] historic weather files for model calibration

Aaron-

The question "are model results significantly impacted by the difference between historic weather data and performance-period data" is a complex one, however, here is an approach that may be helpful if you need to construct your own custom weather file.

You can purchase historical weather data available from NOAA station data from the NCDC. However, these data sets do not include solar radiation data and often need some QC work (remove extraneous observations, fill gaps, etc). In the past, I have also downloaded weather data from the EnergyPlus site in IWEC format, and converted to .bin using a combination of the E+ weather data processor and EPW conversion tool available on the DOE-2 Website. However, although a great deal (a FREE service), in my experience, this data can have gaps that are too large to fill with the algorithms NREL prescribes, which then leads you to either piecing this together w/ some other data source, or, only evaluating the model performance over the periods you have data for. For many of my projects, I need a complete year or more, so the E+ method was not ideal.

I recently created some DOE-2 .bin weather files using the following approach, using data formatting/calculation procedures automated in MATLAB, and then the "DOEWth.exe" utility (an older, command line program also available from DOE-2.com) to generate a .bin weather file. Once automated, this process can be completed in a relatively short period of time. The following is a brief explanation of the process I used that may give you some ideas:

1) Downloaded NOAA data (Integrated Surface Data) for closest site.

2) Cleaned data of extraneous points and filled gaps in data using NREL filling routines. There is usually always one or more observations recorded for every hour, however, in many cases one or more of the variables you need for the simulation weather file may not have been reported.

If you can obtain measured solar data for your location, skip steps 3-5

3) Used the Zhang-Huang solar model (discussed in the E+ engineering manual) to estimate total horizontal solar radiation with custom model parameters developed using least-squares regression to TMY3 data. In this case, I assumed TMY data was more or less "true", however, for one station data I worked with, I noticed some cloud cover observations did not appear to be consistent the reported solar radiation data. However, the custom coefficients yeilded better results than the default coefficients reported in the E+ literature when I compared the model output to a short sample of actual measured solar radiation data I had.

4) Used another model to determine the diffuse solar radiation component from the total global radiation. In my case, I used a custom model developed for the Pacific NW (published in a thesis) however, there are many similar models developed from various datasets. Orgill and Hollands is a popular model, although they are all very similar.

5) Once you have the two solar components above, direct normal solar radiation can be readily calculated.

6) With all of the necessary data now assembled, format into the TMY2 format for processing into a .bin file using the DOEWth utility. Using the DOEWth output summary file and a weather file plotting program, like D-View, you can "inspect" the measured data to make sure it matches what you expect and aligns with daily or month averages more readily available.

A few notes:
- The DOEWth utilty does have format methods that will calculate solar data, however, I choose to pre-process the solar data using the solar radiation models I preferred, which leads you to using the TMY2 format method.
- I did not calculate illuminance data since my project did not include daylighting controls. There are models available for calculating illuminance, and the DOE-2 program may use a model to estimate it from solar radiation data (need to brush up on this section of the engineers manual to confirm this).
- Related to the above comment, there are many different models out there for calculating solar radiation/illuminance data from other measured parameters. I choose the above because I felt like the models best captured the variables that I thought were important, and to lesser degree, I could readily implement them in my programming. Other than comparing these models to actual TMY data, I have not rigorously compared these model to others available, so you may want to explore others.
- For any source of weather data you pursue, I would emphasize reviewing how data is filled and non-measured variables are calculated (i.e. what models were used).

I just realized this post may use the word "model" a record number of times, but hope you find it useful.

David Reddy

360 Analytics
Building Energy Analysis Consultants
mail: 12354 16th Ave NE, Seattle, WA 98125
office: 206.420.7918
mobile: 206.406.9856
web: www.360-Analytics.com

On 11/29/2010 11:54 AM, Dahlstrom, Aaron wrote:

A recent LEED MV plan review comment asked ?please indicate the proposed calibration method to account for the local weather conditions during the performance period.?

This raises the question for me - are model results significantly impacted by the difference between historic weather data and performance-period data?

When I?m engaged in Measurement and Verification, I can install a weather station that records data for performance period to allow for calibration. (http://www.gsd.harvard.edu/research/gsdsquare/Publications/BuildingSimulation2009.GundHallModel.pdf)

But for an investment-grade energy audit with historic bills, I?m not sure where to turn for all the variables needed to construct a weather file.

Anyone on this list have a recommendation?

(I?m specifically looking for NYC, 2009, weather data for an eQUEST model.)

Aaron Dahlstrom , PE, LEED? AP

Hwakong Cheng's picture
Offline
Joined: 2011-09-30
Reputation: 0