0
$\begingroup$

I'm creating a model to predict energy consumption in one food production facility. From business, I know that Downtime due to power failure, machine failure and maintenance, etc. is one of the major drivers. The business has given me two possible features to be created for this.

  1. Total Downtime
  2. Downtime as a percentage (calculated as Downtime/Total Planned Runtime )

What should be my considerations while selecting any of these as my feature. Which one is the most helpful feature to interpret and derive action items for business? I appreciate any help you can provide.

$\endgroup$

1 Answer 1

1
$\begingroup$

Energy consumption naturally needs to be measured in a certain period of time. While "downtime" may affect energy consumption, you need to make sure that "downtime" is measured against the same period of time as energy consumption.

Thus "total downtime" (per period) may be a good choice, provided that the period length is comparable. You should make sure that the "total" is related to the same underlying scale. E.g. "total" 6 out of 12 obviously is different from total 6 out of 24. So in this case you would need to bring both figures on the same scale to make them comparable.

This leads to "downtime as percentage" which may also be used. But same thing here. "Planned runtime" should be compareable across periods/cases/observations in order to habe downtime on the same scale.

  • Check if "energy consumption" ($y$) and "downtime" ($x$) relate to the same period to avoid ambiguous measurment
  • Make sure "downtime" is comparable (on the same scale) across periods to have a unique identification of this effect
  • Finally, try different representations of "downtime" in the predictive model, to find the best representation
$\endgroup$
2
  • $\begingroup$ Energy consumption is a daily data. So I will record daily downtime. so both are at the same period. $\endgroup$
    – NAS_2339
    Mar 23, 2022 at 10:16
  • $\begingroup$ "Make sure "downtime" is comparable (on the same scale) across periods to have a unique identification of this effect". Could you please elaborate on this? Thanks! $\endgroup$
    – NAS_2339
    Mar 23, 2022 at 10:17

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.