Update PV Live multiple times?

[JackKelly]

PV Live national updates many times:

It updates multiple times within the first 3 hrs of the initial estimate. And those updates can be quite significant.

It updates at 2300 and 1100 UTC.

It sometimes updates a few days after the initial estimate.

(See this plot for details of when PV_Live is updated).

Should we grab some of (or all of) these updates? (IIUC, we currently grab the initial estimate in realtime, and then we grab the updated estimate at 1100 UTC)

To keep things relatively simple, I think my suggestion would be something like this:

• As per the new PV_Live README, update our "in-day" estimates every 5 minutes for the first 3 hours after the initial estimate. But keep every one of these "in-day" updates (identified in our DB using PV_Live's new updated_gmt field) so we have the option to train our ML models (in the future) on these different "in-day" estimates
• Ignore the 23:00 UTC update. It's tiny.
• Grab the big update at 11:00 UTC (which Peter implemented already, last night!)

Related

• Need to request the "updated" (day-behind) PV_Live later in the day #219

[peterdudfield]

So I think the steps would be

• Run a new service 3 hours after deliever, maybe lets say 4 for safety
• create new regime, we have "in-day" and "day-after", so maybe its called "in-day-3h"
• Adjust GSP Consumer to get "in-day-3h" which ignores the first 3 hours of data, as they will be the original estimate

[JackKelly]

SGTM!

[peterdudfield]

How important do you think this is? On a scale of 1-3?

[JackKelly]

2