Simulation Calibration

6 posts / 0 new
Last post

I have used equest a number of times to show clients the relative
effects of changing from standard efficiency to higher efficiency
equipment or say from T8 to T5 lighting systems. Generally, pretty basic
models.

I now have a client a (a municipality) that is interested in
"calibrating" the models to get more precise information.

I've heard the term "calibration" as it relates to models, and I think I
understand the concept, but how is this actually done? Is it just a
matter of improving the details of the model, or is there testing of
walls and other systems to determine actual installed values?

Are there any good articles out there on this topic?

All help appreciated.

Thanks,

Chris Mullinax, P.E. LEED AP

Chris Mullinax's picture
Offline
Joined: 2011-09-30
Reputation: 0

To me, "calibration" would require loggers on the lighting, major HVAC
components and electrical panels to record such things as operating
schedules, temperatures and peak kW demand, and matching the model
results to the annual utility data, usually within 5% for annual
electrical consumption, gas use and electrical demand. The model will
likely require several iterations to develop the calibrated baseline,
but then you have the increased confidence level in predicting "X% less
energy use" when you change "Y" in the baseline.

Regards,

William Bishop, EIT, LEED(r) AP

Bishop, Bill's picture
Offline
Joined: 2011-09-30
Reputation: 0

The FEMP M&V Guide has a section (4.6.3) on model calibration.

http://www1.eere.energy.gov/femp/financing/superespcs_measguide.html

Marcus Sheffer

Marcus Sheffer2's picture
Offline
Joined: 2011-10-02
Reputation: 0

Check out "Computerized Building Energy Simulation Handbook" by James Waltz... there's a click-through to Amazon.com on my blog for this fundamental energy modeler's reference manual. An entire chapter is devoted to model calibration. In addition, there are numerous quantitative and anecdotal tips for improving the quality of and reducing the time required to build energy models.

For an example related to model calibration, you may find that only a few or perhaps no data loggers are required -- setting up trend logs, taking instantaneous branch circuit amperage readings (for relatively constant load circuits such as lighting) and accessing existing building schedule information resident in a modern EMCS may be sufficient.

The goal, as Chris stated, is for your model to be within ? 5% of actual energy consumption as documented by the utility bills, month-to-month. Chris is also spot-on about the iterative nature of the process... it's easy to underestimate the actual time commitment required.

Brandon Nichols, PE, LEED AP

Brandon Nichols's picture
Offline
Joined: 2011-10-01
Reputation: 0

A few other comments regarding calibration

Calibrating to annual utility use is better than nothing, but still not
worth much. Calibrating to monthly utility bills can be pretty good, but
that is very dependent upon your goals.

First, note that DOE-2 and derivatives are (still, I believe) hourly
simulations. (EnergyPlus can use shorter time steps.) Utilities usually
bill demand based on a 15-minute period, or less commonly on a 30-minute
period. Perhaps some bill on an hourly period, but I haven?t run into that.
Depending upon when the peak demand occurs, and what is causing it, the
15-minute and hourly peak demands can be VERY different.

Second, it is next-to-impossible to truly calibrate a building in the
absence of zonal energy use information. VAV box operation and reheat can
be particularly problematic, but there are many other areas of concern.
Building simulation programs generally assume proper or at least decent
operation. However, this is seldom the case: economizers don?t work
properly, setpoints are suboptimal, there is simultaneous heating and
cooling due to leaking valves or other problems, etc.

A calibrated model MAY improve savings estimates for many retrofits, but
again, it depends upon whether the retrofit will be impacted by such
problems. One example is a conversion of a system from constant flow to
variable flow. First, most simulations don?t do a great job of simulating
variable speed pumping or VAV fan operation, although this is improving.
Second, most variable flow systems don?t operate optimally, and proper
operation is what we typically assume in simulation.

(For documentation of some of the issues with simulating variable flow
systems, refer to the dictionary for DOE2.2R at

http://doe2.com/download/DOE-22/DOE22Vol2r-DictionaryRefrigeration.pdf

It states, in part, that

?Pressure is not considered in the air-side calculations. VAV fan power is
modeled using a simple curve without considering where the pressure drops
exist in the system, where the static pressure sensor is located, or what
the static setpoint is.?)

Put another way, it is pretty easy to simulate a constant flow system, and
the energy use for that system, since it is relatively unlikely that the
system won?t be operating as expected. However, a variable flow system is
likely to not be operating as expected, is not particularly well-modeled by
DOE-2, and hence the savings estimates from even a calibrated simulation
might not be very good.

This is not to discourage you from calibration?far from it. Your model will
certainly be better after calibration, and you will learn a ton in the
process. However, I encourage you to note the previous resources and
comments. It certainly is VERY ?easy to underestimate the actual time
commitment required:? In the early 1990s I performed multiple
research-level calibrations. For one large building, I calibrated a DOE-2
model to hourly end-use AND zonal subsystem data as part of a research
project. The process was months, not days. That was a very detailed
calibration, but it demonstrated how a model calibrated to monthly data can
still be far from reality.

William E. Koran, P.E.

Bill2's picture
Offline
Joined: 2011-09-30
Reputation: 0

There is an easier way to get a calibrated simulation. Our firm
provides EZSim -- a simplified simulation model developed
specifically to calibrate to monthly utility bills. If one does not
have interval or submetered data, why develop a model in more detail
than you have data to support? EZSim computes one monthly result at
the whole building level and uses actual (not TMY) weather data. See
http://www.ezsim.com/

Model fit can be quite good. The specification of "5% of annual" is
not meaningful in a statistical sense. I prefer to specify in terms
of MSE. Note that the savings estimate is based on comparing two
years (actual and baseline) so the standard errors add in quadrature.
What does this mean? Monthly modeling at MSE of 5% gives an annual
consumption estimate with MSE of 2%. Comparing two years for a
savings estimate results in MSE of 5% of annual applied to the
difference. A 90% Confidence Limit is .96 SE so we have a savings
estimate of +/- 5%. Not too bad -- given that we should be looking at
savings of, say 30%, a 5% confidence band is adequate. If you are
looking at measures with smaller savings, there is not sufficient
resolution. All this can be done very inexpensively using utility
bills as the measurement in an hour or less.

The IMPVP referenced by FEMP and LEED fails to provide an adequate
explanation or discussion if these precision issues. As far as I
know, EZSim is the only modeling tool that reports MSE of the
calibration fit and allows for confidence estimates. The caveat is
that is does not explicitly provide for the myriad of design options
available in a detailed model. However, the model can be tuned to
match any other model. Note that it does so only at the monthly
energy level, not at the hourly level.

David Robison

Dave Robison's picture
Offline
Joined: 2011-10-02
Reputation: 0