1

Topic: Question regarding tuning the Metrix model...

I have a question regarding tuning the Metrix model. I am currently running 4.0 version.

The regression equation for calculating the baseline consumption is of the form:

kWh = x1*(# of days)+x2*(# of CDD)+Offset

In tuning the parameters of the model (x1, x2), we make sure we get a good fit with highest R2 and least bias, CVRMSE. If the model R2 is 85%, so what it means is 85% of the variation in the data can be explained by our model. The rest 15% can be caused by variables other than( # of days) and (# of CDD).

So my question is
What is bill match? If we have a reasonably good fit do we still need to do a bill match by using multipliers?

What exactly is a multiplier and when do you use it?

2

Re: Question regarding tuning the Metrix model...

Tuning your meters is something like pitching in baseball. As long as you get the batter to strike out you win. There are many types of pitches you can throw and many theories of how is the best way to pitch.

When tuning your meters to weather, the object is to best represent your building's energy usage pattern. Again, there are many different theories of how this can be done.

Of course, just tuning the meter works fine. This is what I do. As long as the statistical indicators tell me that the fit is good enough, then I see no point in doing anything in addition to the tuning.

Some energy analysts believe that you should tune the meter, and then billmatch on top of the tuning. Although I don't do this, I don't think I can make enough of a case about it to say that it is wrong. It isn't. It is just another way.

When you tune a meter, AND DO NOT billmatch on top of the tuning, you are making the assumption that the fit line you have produced is a reasonable representation of how the facility responds to changes in weather or any other independant variables you may have. If a bill is a little above or below he fit line, we assume that this randomness, or noise, is unexplainable, and we don't know what caused it, and we further assume that we don't expect it to happen again in the future. There is noise, of course, but over time it cancels out, as half of the noise is greater than the fitline and half below the fit line, so we ignore it. (Of course, this is assuming your statistical indicators are acceptable.)

When you tune a meter AND billmatch on top of the tuning, you are assuming that the randomness, or noise, you find in the base year will continue year after year. For example, if January in the base year was higher than the fit line by 5%, then by billmatching on top of the tuning, you are assuming that January's bills will always be 5% higher than the fit line.

The baseline equation is:

Baseline kWh = (const * #Days + coeff * #CDD) * multiplier + offset

You have seen the (const * #Days + coeff * #CDD) before. That is the fit line.

But the multiplier and offsets are baseline modifications, and when you tune AND billmatch you are either adding the multiplier (percentage offset) or the offset (absolute offset) to your tuning equation.

Take a look at the Tuning Contract (which Metrix 3 and Metrix 4 produces), and you will see how these are calculated.

3

Re: Question regarding tuning the Metrix model...

I personally use the bill match to create a 0 deviation. I have found that customers get a bit anxious when the baseline totals are more (Not so much when they are less.) than the actual consumption, even if it is only a couple of %. By bill matching the actual data and the baseline data will generate a 0 deviation.

Hope this helps!