I have the following timeseries and unfortunately no other information except time and holidays.

enter image description here

What methods could work for a probabilistic forecast of this? I thought about some regime changing model. How could i detect the regimes?

enter image description here


  • $\begingroup$ You're more likely to get people to look at your question if you use an informative title. $\endgroup$ – mkt - Reinstate Monica Sep 27 '19 at 8:44
  • $\begingroup$ With the frequent zeroes you could use an intermittent model like Croston’s, but there’s claearly some kind of event here as you mention with regime change. How can you not have any data on what drives these pulses? Do you, or someone else, know at least in hindsight the cause? $\endgroup$ – Chris Umphlett Sep 27 '19 at 10:09
  • $\begingroup$ It is the power demand from an industrial facility. If a worker decides to switch on a process that runs for a few hours (sometimes over night) we get these plateaus. The other peaks are from normal daily operation. $\endgroup$ – OhmSweetOhm Dec 13 '19 at 18:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.