Today at 6:35 it looks OK again.
even though the DESS forecast curve still does not fit to the standard energy forecast
Thanks @HansDampf for pointing me to this thread. I started one yesterday (Dynamic ESS solar numbers/forecast is going crazy) with clearly the same issue.
This morning it has sort of settled, that means no crazy solar forecasts anymore of 600+kWh but around 50-60kWh which is more or less correct. However, the hourly Forecasted Solar to Battery numbers are still about 50% too low.
Now that the energy prices for tomorrow are known, still reasonable hourly Energy forecast numbers, also for tomorrow. Although still too low hourly Forecasted Solar to Battery numbers.
I assume and hope that after a few days DESS will adjust and get the forecasted numbers right again? Very curious to find out what has gone wrong on the Victron server.
1 Like
For me CURRENTLY the forecast looks like
this morning still everything was fine. Absolutely no idea why now some hours are okish and others are going crazy.
in DESS part it looks better
For tomorrow:
DESS OK - but every update it might look totally different.
@dognose maybe you can forward this issue again internally?
My solar âTotal forecastâ today is much better than yesterday (48-59kWh). I hadnât checked the forecast graphs, but the bars in the graph today at 11:00 and 12:00 are way too high.
For tomorrow it looks better but still about 30% too low:
And itâs going crazy againâŚ!
Solar Forecast this evening:
Solar Forecast tomorrow:
Energy Forecast tomorrow:
@dognose, whatâs going on here?! Itâs very frustrating. Is there anything I can reset? Is there anything Victron can reset? Where do these crazy solar forecast number come from?
Unfortunately we most likely cannot do anything then wait for victron
We currently have bad weather, so forecast should not be higher than 50kWh a day for me. Maximum per hour should be ~10kWh (if sunny & no clouds) and per day ~80-95kWh.
Iâve upgraded Victronâs Dynamic ESS (DESS) system by integrating a more accurate solar production forecast using Node-RED and the PVGIS API. The default forecast built into DESS often turns out to be overly optimistic or unreliable â especially in systems with oversized PV arrays and multiple panel orientations.
Hereâs what I implemented:
- Installed Node-RED directly on the Cerbo GX (running Venus OS Large)
- Retrieved hourly solar forecasts from PVGIS for three orientations: North, East, and West
- Calculated the combined expected production per hour
- Sent the resulting forecast data to DESS via MQTT/dbus at
/Settings/ESS/Forecast
thanks @ikke050 this could be a workaround thanks for the suggestion. Did you describe this somewhere step by step? (posted your node red flow)
But the issue here seems not to be related on the PV forecast as such (the irradiance forecast in the advanced tab is everytime okish).
The issue here is that the forecast jumps between correct (forget about a few percentages) and absolute crazy values in the energy forecast - where forecast is doubled, tripled upto 5 times more than realistic. Jumps means that now it is OK one hour later crazy then few hours are OK again then the day looks fine and few hours later again absolute crazy values. Somehow random behaviour switchting between realistic and factors off.
That it can work âperfectlyâ I have seen the last few week after Victron fixed something - but now the fix seems lost again.
While I was writing the previous post the forecast corrected it self for today
⌠also next 7 days looks fine
and that is exactly what I meant before. It is not about a bit of inaccuracy but massive (factors) of miscalculation at random? times.
I had the same problems and some more what i have done:
Accurate hourly PV forecast using PVGIS API
Seasonal correction per month
Live cloud cover adjustment using Open-Meteo
Dynamic Min SOC control based on forecasted PV
extra:
Smart switching of in my case Fronius & SmartSolar
Everything runs directly on the Cerbo GX.
Thanks for that, much appreciated! But of course, Victron should fix this and I donât see why this canât be fixed quickly. It worked fine for the last months. Until it didnât.
For what itâs worth and I canât imagine itâs related but when I switched on the manual charge (âKeep batteries chargedâ) at 10:00, the forecast changed from:
to:
Still not correct but more workable figures I guess⌠Weâll see when it goes crazy again.
I had the same problems but even before that it wasnât accurate. tried a few times to reset forecast.. (settings/general/reset forcast) didnât solve the problem.. I have my solar panels on east/west.
My panels are 1/3 east, 1/3 west and 1/3 south, but the forecast was actually pretty accurate.
It never worked and instead off improving it becomes worser and worser.
I donât really mind as I made my own forecasting based on solcast aswel and some machine learing proces that makes a correction between the sollcast forecast and my installation. I have panels in different directions.
Very frequently there is an assumption that the issues might be related to panels in different directions.
AFAIK the algo uses the irradiance forecast and then checks the last x weeks for similar per hour.
If you build a kind of average (which I do not know if victron does) the forecast for the next day should be pretty accurate.
Doing this: no panel size, directions etc. is needed. Of course it will never be as accurate as with all data but that is also not required. I would not mind if it is even 20% of track.
My system is mainly north and some south, and a bit of west. But the forecast worked for a while after I reported the issue and it was âfixedâ. Since Roger also reports it was working fine and and some random point in time changed behaviour this is a strong indication that directions should not matter. At least I dont think this thread issue should be related to directions.
@ikke050 thanks for the github link, if I have some time I will try at the weekend. But having a quick look I already struggle to find where the PV forecast and Tibber prices are retrieved. I do not see any http requests to do so in my imported flow or is there a second flow collecting data?
I wonder if it may be related to the number of different directions / orientations in the system and to whether these different directions are being monitored and forecast separately.
If the forecast looks only look at the total sum of PV production - either because all inverters are aggregated on their side first or because this happens already on customer side because he has an inverter that covers multiple orientations at the same time so that the sum is being calculated without knowledge of Victron, then itâs going to be hard to fit that to a single simple model.
I have an East-West system and the VRM forecast is pretty accurate. But I have two inverters - one for East and one for West - and they are integrated separately into dbus and VRM. Both systems also have nearly the same peak power, but itâs not 100% symmetrical and the house is casting a shadow on the West system half of the day and on the East system for the other half so that only in summer there is an overlap were both systems get sun for a short amount of time. So the total sum of both systems does not fit into a simple model.
Could be that they do as you explained - but would make no sense.
The simplest model I explained above.
- Just take the irradiance (W per squaremeter) forecasted per hour (for this you need the geolocation).
- Look for similar cases in the past (for each hour of the day).
- Take the energy production of those hours (measured values of your system) and
- build an average.
This works independent from shadows directions etc. To be able to do this you needs some weeks of data (otherwise no match will be found). If the duration just considers ~4 weeks you have also a monthly compensation build in avoiding winter results (sun angle) influencing summer calculation, or shadows or trees with leafs only in summer etc.
But even if they would do it different the point within this thread is EXTREM values AND switching between somehow correct and the extreme randomly. Which for me is a clear indication that it has nothing to do with such direction related inaccurancies.
More sense would make a wrong DB query delivering wrong values (other user system etc) which then impacts the results. Why this is happening ârandomlyâ and not alway only Victron can answer.
Hi
My Victron system has the same problem and my panels are all facing east-south
