I’ve addressed this issue, it will be fixed when I make the next release.
ha-open-meteo-solar-forecast 0.1.13 has been released. This release adds the following new features:
- Re-add cloud cover compensation code with some fixes.
- Make the remaining energy today sensor use 15-minutely watt data instead of hourly data.
NOTE: you might need to readjust your efficiency factor.
3 Likes
ha-open-meteo-solar-forecast 0.1.14 has been released. This release updates to open-meteo-solar-forecast 0.1.16.
I updated to 0.1.13 last night and all forecasts are now unavailable. Will try 0.1.14…
EDIT: Yep that works!
Comparing available forecasts and production “now”
Comparing available forecasts “tomorrow”
Comparing forecasts and production “today”
Comparing “forecasts vs. production” 14 days history
After using the integration for a few weeks, my findings are:
- It almost always overshoots, by 5% or more.
- The forecast is very rarely too low, and that’s usually only for part of the day, not the overall production.
- It doesn’t matter that much if it was cloudy or sunny, the forecast is either way a bit too high.
- The forecast dynamically changes during the day, and it always seem to start very high and then correct downwards. Because of this, checking the past forecast shows much better prediction than what it really was at the start of the specific day.
- There are weird discrepancies between the graph, the total prediction for the day and the remaining. For example it claims it will do 40 kWh today, I can see only 30 kWh were produced, and the estimated remaining production is only 2 kWh.
- Besides all of that, the forecast is still useful and it’s much better than the default HA forecast.
those few points you make are down to the fact that the Integration did not have any cloud cover compensation in it’s calculations…
if you look 5 posts before your own you can see that’s now added back in from v 0.1.13
so while no doubt your findings and observations are what you’ve observed they are somewhat moot now that the Integration’s forecasting approach has changed…hopefully we all see a more “cloud friendly” forecast (or maybe that’s cloud unfriendly) from here on in.
Ah, you are right. I’m on .12, I somehow missed the post mentioning the code reload for the cloud cover compensation. I’ll update the integration and see how much it changes. Thanks.
1 Like
So I updated the integration. I’m not sure if it should update the forecast for today, but if it did, it was in the wrong way. It’s still overshooting today’s production by a lot. It is a sunny day here, but while it’s bright, the sky is not blue, I think there are very high clouds. Also it’s very hot, which reduces the efficiency of the panels. Is this integration taking that into account?
I’m planning to use Open-Meteo’s historical weather API to train a model that could more accurately predict the PV output. I’ve previously used Prophet for time series forecasting but I’m not sure about how well it would work for this application, HOWEVER, I am optimistic that it will work better than the models I’m currently using.
Edit: I just need to find some PV power generation datasets :')
3 Likes
Im not sure if the days are correctly specified or if solcast has it wrong or if this integration just counts days differently?
It counts days differently,
D0 = today
D1 = tomorrow
D2 = after-tomorrow
etc
Is it perhaps possible to add Peak Forecast Today and Peak Forecast Tomorrow?
It already exists, it’s “Solar production forecast Highest power peak time - today” and “Solar production forecast Highest power peak time - tomorrow”?
Im talking about actual Values not times :
1 Like
Right now I’m working on the ML forecasting code, so this won’t be a priority for the time being. At any rate, this is already achievable via a template sensor: {{ states.sensor.energy_production_today.attributes.watts.values() | max }}. There is no need for an additional sensor from the integration.
Thank you. Works
Big underestimation for me on .14:
Actuals are 9.56kWh at 1:20pm, so Solcast is looking about right.
I’m probably going to just get rid of the cloud cover compensation again until I could get the ML model ready. I felt like the current compensation would work because for me it seems that only the high clouds are considered but apparently not.
1 Like
