I've ALWAYS avoided predictive modeling (telling a client your energy usage
will be below X) as I've been scared. Now I'm pretty deep into a project
performing Green Star energy modeling and my client has popped the question.
We've discussed "why it is hard" and the big three un-knowns: Infiltration,
Occupant behavior!!, and Actual Weather. They still want me to go forward
and are willing to put a bunch of caviates and an agreed margin on the
results.
This is an old 1930's building 200,000 sq ft with original metal window
frames (single pane) that will not be replaced (heritage). We're insulating
the roof and basement ceiling and a completely new HVAC system. Is there a
standard process I can follow or specific suggestions the community could
offer?
So far my idea is to try and guesstimate infiltration, box in comfort
behavior with lease agreements and a survey, and put some margin on weather.
--
Jason
Get utility bills (if you can), or Use Energy Star or CBECs for a reference building (CBECs median may be higher than a 1930 building !); based on these, you can present a range between 'simulated' savings and 'estimated' savings.
Kapil Upadhyaya, LEED AP
I've been getting within 10% without calibration on the last five or six
buildings I've done where I've tracked the bills afterwards. The key is
to be thorough and careful. If you're not getting reasonably close, how
can you justify your recommendations for energy efficiency measures?
Morgan Heater, P.E.
I don't believe there can be any hard rules about the percentage of
error when calibrating a model. The margin of error is proportional
to the amount of information available. The rest is engineering and
experience. Attached is a typical report we use to illustrate how
close the calibration is.
This particular project is a community centre, pool, with an arena
adjacent for heat exchange. A tricky one.
Nice graphs, Chris! Even nicer calibration!
Random comment regarding % deviation for an individual month: Natural gas consumption often goes to near-zero in summer months. This sometimes makes any deviation from the actual usage a very large number, which in reality may be a tiny fraction of total use. There's no simple solution for this; it's not worth worrying about the deviation because it's so small a fraction of overall use, but it looks really bad to non-technical readers. I've been reporting R-squared values calculated from the monthly deviation. Not perfect, but it places the focus on the overall month-by-month match instead of a single month.
Morgan, you apparently work with excellent commissioning agents and very responsible owner / building managers! 10% sounds too good to be true.
James V Dirkes II, PE, BEMP, LEED AP
Morgan - 10% is pretty damn close - well done!
That being said - I would like to respond to the second part of your email. Personally I think it is possible to be slightly inaccurate, but still correct.
In my experience it is when you are in the early stages (when there is a huge margin of error) that the modeling is actually at its most useful. Very often it's the early on normative modeling that leads to good design decisions. For example when modeling a VAV system versus a chilled beam system, I might not have the exact BTU number, but if I can get the usage patterns appropriately represented (cooling, heating, fans & reheat) then the result might not be 'accurate', but correct. Your overall numbers might go up and down with occupancy and weather, but the decision still remains a good one. This is where having the experience and the backing of a good database of comparable buildings is invaluable. CBECS is good for overall usage, but it's even better to be able to look at how the BTUs are spent over the year. Investing the time to collect utility bills for your building type is worth a lot.
Often we worry about the accuracy of our results to the nth degree before making recommendations, but by the time we come up with the magic number its too late - the design has morphed, or been finalized so your numbers are either outdated or redundant (in some cases both).
Vikram Sami, LEED AP BD+C
I don't worry too much about the monthy deviation. It is one more
indicator to use to help understand the differences between the model
and actual bills.
Those graphs are part of an Excel workbook that is used to adjust the
model characteristics, shells out to run the model, then sucks the
hourly results back into the workbook. It is a very fast cycle -
review, adjust run the model and review again. No "hand work" to
extrac the results or to adjust the model.
Hey Morgan:
Does your data include years 3 to 6 after construction? It would be great
to see if usage is in line with predictions after real work use of the
building for a while. What type and size of buildings were they in what
geography?
Thanks,
Andy