You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For billing, daily, and hourly methods, the current models use temperature or HDD/CDD. While unmodified temperature variables are highly correlated with the energy use of HVAC systems, there are a few modifications to the temperature variables that are readily tested using available data and could potentially lead to further reductions in residual variance. In other words, they may be more appropriate for representing the dynamics of outdoor heat penetrating the building shell. In modeling thermostat data, I have seen average out-of-sample CVRMSE reductions of up to 15-20% at the site level from these modifications, so I am curious whether they will have a similar impact on the daily or hourly methods for CalTRACK.
The two modifications I propose are:
Lagged rolling averages for temperature or HDD/CDD: Heat takes time to penetrate the building shell. Especially for hourly calculations (e.g. HDH/CDH in Issue CalTRACK Issue: Calculate Degree Days using Degree Hours when available #120), averaging the previous 2-3 hours can mitigate the impact of spikes and dips on assumed usage, and align the temperature variable with actual consumption. A heat build-up term with a weight to prior temperatures that decays with time difference (e.g. 100% for t, 75% for (t-1), 75%^2 for (t-2), etc.) should also be tried.
Incorporation of relative humidity for a heat index and wind for wind chill: Although much of the work for heat indices (a.k.a. sultriness indices) revolves around their application for humans or animals, HVAC systems do work harder to get rid of heat when it is hot and humid, and a heat index term has consistently beaten dry bulb temperature in my modeling of whole building energy use. I have had some success with using wind chill, although I don't think it would work as well for C&I.
Proposed test methodology
Use a test C&I data set and a test residential data set
I'd like to support the first modification (Lagged rolling averages for temperature or HDD/CDD) but would suggest this could be analyzed as part of the more general issue #124.
I believe that this method would be applied to hourly temperatures, but apply to both hourly and daily methods, and deals more with short term conduction of heat through the building shell. As far as I understand it, #124 is much longer term. Is that not the case?
@jkoliner: True; but #124 targets the same offset between temperature changes and energy changes. If we were to confirm a significant impact from daily variations via that issue, perhaps the same setup & approach & testing might be shared with this issue? Also, we're not aware of anyone getting paid yet based on the hourly methods so perhaps this finer detail is not as urgent (?).
Prerequisites
(Addressed in issues CalTRACK Issue: Improve handling of buildings with weather-energy misalignments #124 and Improve regression results by taking thermal inertia into account #111, which can be rolled into this scan if approved)
Articles #s: 3.1, 3.3, 3.9
Description
For billing, daily, and hourly methods, the current models use temperature or HDD/CDD. While unmodified temperature variables are highly correlated with the energy use of HVAC systems, there are a few modifications to the temperature variables that are readily tested using available data and could potentially lead to further reductions in residual variance. In other words, they may be more appropriate for representing the dynamics of outdoor heat penetrating the building shell. In modeling thermostat data, I have seen average out-of-sample CVRMSE reductions of up to 15-20% at the site level from these modifications, so I am curious whether they will have a similar impact on the daily or hourly methods for CalTRACK.
The two modifications I propose are:
Proposed test methodology
a. 1 to 6 hour lags, inclusive and exclusive of current temperature
b. A couple heat build-up specifications
c. A couple specifications for relative humidity (e.g. from https://journals.ametsoc.org/doi/pdf/10.1175/1520-0450%281979%29018%3C0861%3ATAOSPI%3E2.0.CO%3B2) and wind chill.
Acceptance Criteria
Demonstrated reduction of in-sample and out-of-sample CVRMSE and/or bias (i.e. not harming either while improving at least one)
The text was updated successfully, but these errors were encountered: