[Bldg-sim] "Maslow's" hierarchy of calibration
chris.malcolm.yates at gmail.com
chris.malcolm.yates at gmail.com
Tue Jul 23 03:54:05 PDT 2024
Hi Karen,
“Maslow’s” hierarchy of calibration
In answer to my own question, perhaps a good starting point is to target the minimum median, and then apply hourly variation.
Please excuse my ASCII art (it’s meant to be a triangle!).
________^_________
/ \
/ \
/ \
Review NMBE & iterate required \
_____________ ___________________ \
Review CVRMSE & iterate if required \
_ ____________ _______________________\
Apply hourly variations to build \
up energy usage \
Apply holidays \
_________ _______________________________________\
Set a “universal” generic profile to apply to all gains. \
This is set to a diversity (say 20% in the above example) \
And should have enough “headroom” and range to model \
_________ _____________________________________________ \
Create a model with known parameters \
Quantify unknown knowns \
___________________________________________________________
Carpet and box plots
I use similar carpet or colour-area plots. Yours is similar, but I think you are tracing an average day in your example? I like your “synoptic” view of modelled v. actual.
I create the carpet plot using a pivot table hack in Excel and ECAM processed data as the source:
Another handy tool in ECAM is the Boxplot. I will usually target the median as the most likely observed outcome.
Regards
Chris
From: Karen Walkerman <kwalkerman at gmail.com>
Sent: Tuesday, July 23, 2024 10:29 AM
To: Julien Marrec <julien.marrec at gmail.com>
Cc: Chris Yates <chris.malcolm.yates at gmail.com>; bldg-sim at onebuilding.org
Subject: Re: [Bldg-sim] "Maslow's" hierarchy of calibration
Chris - one tool I've found very useful is to create a heat-map that compares modeled data with actual data. I usually do this with hourly data. Then I'll create a grid that shows actual and modeled data and break it down by hour of the day and day of the week. I can further filter it by time of year, or outdoor temperature. Here's an example of total electrical demand for the whole year broken down by day and hour of day:
In this example, we can see that the model is missing some night-time base loads, there is better agreement during periods of time with high occupancy and that the occupancy schedule and building opening / closing schedules need to be adjusted.
I've found this approach can really help me identify what parameters to experiment with, and it is also an excellent tool for communicating with clients.
--
Karen
On Tue, Jul 23, 2024 at 3:41 AM Julien Marrec via Bldg-sim <bldg-sim at lists.onebuilding.org <mailto:bldg-sim at lists.onebuilding.org> > wrote:
Hello Chris,
Apologies if I didn't understand the question correctly, but I understood it as how to automate (part of) the calibration process.
I don't think you should do without first taking a hard look at your initial model, refine it until almost satisfied, before you can start the calibration process itself. One example would be you should probably take a close look at your output:variables to ensure that you are supplying the right amount of outside air to your zones.
I also don't think you should blindly let some tool touch any parameters there are in your simulation file. There are things you can measure or reasonably approximate, and those should be defined in your model. Then there are things that are hard to guess, or hard to guess right, and those should be what you play on.
These caveats aside, there are a number of tools and algorithms that can help you automate the calibration process itself.
One example of such is LBNL's GenOpt <https://simulationresearch.lbl.gov/projects/genopt> which is software agnostic (a front-end for it that I have used in the past with both eQuest and EnergyPlus is ExcaliBEM, from Canada's SIMEB, not sure if still out there or not)
You basically inform the software of the parameters you want to vary and how (for eg, infiltration from 0.3 to 1.0 ACH, R-value +- 30%, heating system efficiency range, etc), and define an objective function, which can be a python script. I've done one that basically tries to minimize the CV(RMSE) and NMBE %, and penalizes heavily if not meeting the thresholds of Guideline 14. The tool will run simulations and try to minimize the objective function.
The OpenStudio Analysis Framework <https://github.com/NREL/OpenStudio-server> is a great candidate if you're working with the OpenStudio ecosystem.
JEPlus+EA <http://www.jeplus.org/wiki/doku.php> can do that for EnergyPlus (and is (was?) bundled in Design Builder if I recall correctly)
There are a variety of similar questions on unmethours.com <http://unmethours.com>
Best,
Julien
--
Julien Marrec, EBCP, BPI MFBA
Owner at EffiBEM <http://www.effibem.com>
T: +33 6 95 14 42 13
LinkedIn (en <https://www.linkedin.com/in/julienmarrec> ) | (fr <https://fr.linkedin.com/in/julienmarrec/fr> ) :
Le mar. 23 juil. 2024 à 09:06, Chris Yates via Bldg-sim <bldg-sim at lists.onebuilding.org <mailto:bldg-sim at lists.onebuilding.org> > a écrit :
Dear bldg-sim’rs,
I’m interested in how people save time and improve accuracy when it comes to calibrated simulation.
Calibrated simulation attempts to model a vast amount of parameters. In comparison to regression methods it’s completely overspecified. We do it in the hope of creating a model (a “digital twin”) that gives us many more what-if? scenarios than regression could on its own.
However, the process is often fraught with dead ends. Of course it is! There are too many details. We can waste a lot of time on getting one parameter right, only to find it has little effect. Even worse, we could overlook something major and compensate with our own erroneous assumptions. Or, we can work through a whole model and find something very late on, causing us to backtrack over previous work. The act of calibration seems beyond detailed management because of the sheer number of building types, servicing approaches, operational parameters, embedded issues (this is a longer list but I’ve ran out of words…).
The exam question is:
* Does anybody follow a simple high level framework (a bit like Maslow) for calibration?
Sorry. I could have asked that at the start of the email 😐
All the best
Chris
_______________________________________________
Bldg-sim mailing list
http://lists.onebuilding.org/listinfo.cgi/bldg-sim-onebuilding.org
To unsubscribe from this mailing list send a blank message to BLDG-SIM-UNSUBSCRIBE at ONEBUILDING.ORG <mailto:BLDG-SIM-UNSUBSCRIBE at ONEBUILDING.ORG>
_______________________________________________
Bldg-sim mailing list
http://lists.onebuilding.org/listinfo.cgi/bldg-sim-onebuilding.org
To unsubscribe from this mailing list send a blank message to BLDG-SIM-UNSUBSCRIBE at ONEBUILDING.ORG <mailto:BLDG-SIM-UNSUBSCRIBE at ONEBUILDING.ORG>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20240723/f4a0a135/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 122269 bytes
Desc: not available
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20240723/f4a0a135/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image004.png
Type: image/png
Size: 29732 bytes
Desc: not available
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20240723/f4a0a135/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image005.png
Type: image/png
Size: 302517 bytes
Desc: not available
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20240723/f4a0a135/attachment-0002.png>
More information about the Bldg-sim
mailing list