Regression being a sum of two regressions

#1
Hi everyone,
I hope someone can help me with this.
I'm analysing energy data vs ambient temperature. Using the total energy vs the temperature is not giving me the needed results to accept the model, but checking more closely the data I realized that during the day the energy consumption is higher than the night (when the place is open). Taking that in consideration I tried to model the energy consumption at night and then during the day. Both models work under this so the final model, the total energy, would be the sum of the day model and the night model.
In my understanding, it could be correct, but I would like to check other points of view.
Thanks!!!
Tomas
 

staassis

Active Member
#2
Such problems are typically addressed by building one time series model using, say, hourly data. Such model would contain daily seasonality, capturing different demand during different times of day. What you have done seems to ignore the early morning and early evening dynamics, providing insights on how nightly demand turns into daily demand and back.
 

noetsi

Fortran must die
#3
it seems like you can simply build a dummy predictor variable with 1=day and 0 = night (or the other way around of course).

Or the much better approach stasis suggests....:p