NREL / Wattile

Deep Learning-based Forecasting of Building Energy Consumption
BSD 3-Clause "New" or "Revised" License
9 stars 0 forks source link

Wattile model output should either be time zone naive or UTC #293

Closed stephen-frank closed 10 months ago

stephen-frank commented 10 months ago

It appears that Wattile model output timestamps are currently localized as server time (?) when they should be localized as UTC. At this time, the data intake workflow converts to UTC on import, so the model expects UTC and therefore, I presume, the model output also will be UTC. However, when testing on SkySpark, I found the model output timestamps were localized as America/Denver even though they only made sense when interpreted as UTC.

At this time, because of the UTC conversion on initial data import prior to training, the model output should be one of the following:

  1. UTC, or
  2. Time zone naive (no time zone)

For more context, see https://github.com/NREL/nrelWattileExt/issues/47. Interpreting the model output as UTC is a short term fix; I will post a different issue for a longer-term fix (that will involve breaking changes).

stephen-frank commented 10 months ago

This one should be addressed prior to a new Wattile release.

stephen-frank commented 10 months ago

Per discussion 12/11, the output time zone is going to match the time zone of the input data timestamps. So if you feed it UTC it will return UTC. SkySpark just needs to ensure the data is in UTC when fed to the model.