So I've taken the plunge and got PredBat up and running on HA (on my Raspberry Pi).
I've got 4KW of solar and a 9.4KWh battery, with a hybrid invertor. I'm on Octopus Agile Import and Export.
Given that (as I understand it), PredBat needs historical data to start working well, should I leave it in Monitor mode for a few days before giving it full control, or let it loose sooner, accepting that it's not going to get things right until it has more data?
(By the way, some of Trefor's videos show a 'read only' switch, which seems to be missing now.) Just found it on the raw config card.
J
#2James L
Well, I've switched to 'Control charge & discharge' mode, and have a couple of questions.
The plan seems to be changing quite a lot - is this because PredBat hasn't got enough data yet to act sensibly (see my main Q above)?
Is the 'limit' column in the Plan the maximum SOC it'll charge to? If so, seems low to me. Where can I change this in the config?
If you have a brand new install of GivTCP then yes Predbat will usually need more history to come up with an optimal plan. All depends what you set days_previous to though, if you set it to say:
days_previous:
1
7
And this history isn't in givtcp/home assistant then clearly predbat can't plan based upon it.
If you set your history to just 1 day ago then you might be OK by now,
You might find it useful to turn on 'html debug' which shows the import and export prices after conversion losses. The 42 is indeed the target SOC that predbat is charging to. The following column (mainly green) is the load that predbat anticpates the house will require in each 30 minute slot.
You can't change these target figures that predbat calculates, it calculates what you will need based upon your load history and solar generation. There's no point in charging unnecessarily if your house load doesn't need it. Having said that your half hour history figures are absolutely tiny so maybe you do need to let history build a bit more before turning it on to charge & discharge mode.
J
#4James L
geoffreycoan Thanks Geoffrey - your advice confirms what I suspected, with some useful pointers.
As more history is being gathered, I think the plan is becoming more sensible. I've already got days_previous set to 1, so it wouldn't be modelling that badly even with little data.
I'll increase days_previous in a few days.
G
#5geoffreycoan
James L As more history is being gathered, I think the plan is becoming more sensible. I've already got days_previous set to 1, so it wouldn't be modelling that badly even with little data.
No problem James, sounds good.
I don't know for sure but in one conversation with the author of Predbat, Trefor, he said that days_previous of 1 includes today and advised me not to include it. For that reason my days_previous is set to:
days_previous:
- 2
- 3
- 4
- 5
- 6
- 7
- 8
So I average over the whole of last week. I also have switch.predbat_load_filter_modal set to true so the lowest day's consumption is discarded from the averaging.