solcast_forecast_today attributes forecast and detailedForecast don't match, for the same period

[BobC76]

Hi, this is a great utility, thanks. I've been trying to reproduce the solcast graph with the 10/90 confidence area using the "detailedForecast" attribute and the ApexCharts card. But I noticed that this doesn't match either the solcast website representation of my home solar site or the "forecast" data in HA for the same sensor and period. This seems odd.

For example from today's data I see in HA:

forecast:

• period_start: '2022-12-16T10:00:00+00:00'
  pv_estimate: 1.4065
  detailedForecast:
• period_start: '2022-12-16T10:00:00+00:00'
  pv_estimate: 0.7285
  pv_estimate10: 0.59025
  pv_estimate90: 0.7384

Any ideas?

[oziee]

from what I can see since every few hours the integration gets the "actual" data of past time, which then does not have a value for estimates, because past time periods are then actual/known

only forecasts for future time can have a prediction/estimate of value

if the forecast estimate was for 10kw but it was cloudy and rainy and only got 2kw, the graph would show 10kw until the actual data was got and would redraw to 2kw

currently the only thing might be to test is there is no pv_estimate10 or pv_estimate90 and then use the pv_estimate value and work out 10% and 90% to fill I the gaps

there is no data for past time estimates (only future shaded regions)

[fboundy]

I think the OP's point is around differences between the forecast attribute and the detailedForecast attributes.

detailedForecast appears to be 50% of the forecast value for the same period. The uom for the entity is kWh rather than kW so I think this is correct.

On your point about history vs forecast: I'm seeing that the detailedForecast 10% and 90% attributes are still there for past periods. I've just done an update at 10:48 and there's still a range of data for this morning:

[oziee]

I'm seeing that the detailedForecast 10% and 90% attributes are still there for past periods.

This is only because the past actual data is only checked every few hours, so until that updates the past 10-90 would still be there

As for the other comments. If you read the forecast vs detailed, you would notice that forecast is hourly and detailed is the whole detail data supplied by Solcast in 30min points, but that isn't even 50% because it might be 3kw for the first 30min of the hour and the next 30min might be cloudy and only 0.5kw

[BobC76]

Thanks. I was coming to a similar conclusion, but hadn't had time to recheck the data coming in as my system is currently offline. If I recall correctly, the HA solcast integration attribute "detailedForecast" is in 30m periods, so need to add the two half hours together to equate to the other 1hrly "forecast", doh!

However the posts above beg the question of how are "estimated actuals/live" and "forecasts", as defined by Solcast represented in the HA solcast integration data?

(I'm interested in seeing the spread of that 10-90 range as it was predicted at say midnight for the day ahead, and how my "actual" compares, going back over several weeks, not just 'today'. By definition, if my model is set up right, and their predictions are any good, it should be within that interval almost all the time, while it will no doubt vary a fair bit from the single point 50% prediction)

[oziee]

attribute "detailedForecast" is in 30m periods, so need to add the two half hours together to equate to the other 1hrly "forecast", doh!

Not exactly.. you have to add half of each together then add together.. (or add then halve) since the values are 30min and not 60min (hourly)