1
$\begingroup$

I have hourly data and I want to extract the non-thermosensitive part of an electric consumption of a building. The external temperature is available.

Given that the electric consumption is highly volatile, I first thought about using dummy variables for each hour, day (weekday or weekend), and month of the year.

Then I "separated" the temperature into 6 categories: Since the thermosensitive part I would like to extract is the result of heating and cooling (mostly), I separated temperatures below 15°C, between 15° and 25°, and above 25°. I also separated the temperatures when the building was occupied or not, since the presence of people can influence the temperature inside.

My question is: Is it possible to remove the thermosensitive part of the series? And if it is, what method should I use?

Improve this question
$\endgroup$
5
  • $\begingroup$ I'm having trouble following this. What is the "thermosensible part" of the data? Can you give some simple example numbers that illustrate what you want to do? $\endgroup$
    – gung - Reinstate Monica
    Commented Nov 14, 2016 at 17:01
  • $\begingroup$ @gung In Romance languages "sensible" means sensitive. Thus we should understand this to be a question about removing the temperature-related effects of electricity consumption from what appears to be a very fine-grained time series. $\endgroup$
    – whuber ♦
    Commented Nov 14, 2016 at 17:04
  • 1
    $\begingroup$ So is the idea here that you have energy consumption for a building over time, & you want to estimate how much energy is used to, say, keep the lights on, vs how much is used to run the air conditioning? $\endgroup$
    – gung - Reinstate Monica
    Commented Nov 14, 2016 at 17:07
  • $\begingroup$ @gung : Yes, this is what I meant to say. $\endgroup$
    – SdaliM
    Commented Nov 14, 2016 at 17:38
  • $\begingroup$ Since the data appear to be (at least) hourly, heating and cooling degree-days will not be sufficiently granular to do much good. Ideally, you are trying to estimate (and remove) the response of the building HVAC machinery to outside temperatures (as well as to other factors like the number of occupants). That response will be nonlinear and even non-monotonic. Although many techniques exist to handle such general responses, you might make a good start by choosing functions that are compatible with physical theory, as described at stats.stackexchange.com/a/148166. $\endgroup$
    – whuber ♦
    Commented Nov 14, 2016 at 19:39

1 Answer 1

Reset to default
0
$\begingroup$

Your thoughts on incorporating deterministic effects like hour-of-the-day ;day-of-the-week ; holidays (pre and post) etc are in the correct direction and is precisely what is needed . I would not separate the temperatures but rather use heating and cooling degree days as two predictors . One way I have seen this work is to approach it with a parent/child relationship i.e. daily/hourly. I just posted a response What is the frequency in my series and does it matter for my forecasting models (SARIMA)? that you might want to review as it focuses on multi-frequency analysis of time series data which is what you have.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I don't quite understand what you are proposing when you say "I would not separate the temperatures but rather use heating and cooling degree days as two predictors." Do you mean adding a different dummy variable? $\endgroup$
    – SdaliM
    Commented Nov 15, 2016 at 15:23
  • $\begingroup$ Yes .. take a look at the url that whuber recommended . Additionally we have used two variables to model temperature , var1 = 0 if temperature >32 and 32-temp otherwise ; var2 =0 and temp-75 otherwise. These two variable capture the "bathtub response" of ten seen when temperature is important. $\endgroup$
    – IrishStat
    Commented Nov 15, 2016 at 16:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.