Problem: The forecast is way too low so the battery is already full around 11:30 in the morning.
Installation:
3 * Multiplus II 5000kva
Charger mppt 450/100
Inverter SMA 5000LT
Cerbo
DESS
Battery 64KW
Because the battery is full so quickly it starts sending power to the grid at the most expensive time of day.
This does indeed happen; in my case, battery gets full at around 13:00 and then the solar production goes to waste. However, since I don’t have negative prices, I don’t mind that much. Yet.
At your case, of course the system should NEVER sell to the grid when the prices are negative, this is ridiculous. You should disable all “feed-in excess” under ESS → Grid-Feed-in on Remote Console, and if that doesn’t work, you could add specific times for battery discharge to grid through Dynamic ESS settings.
I will show tomorrow’s forecast and it is at least 100% under estimated, then tomorrow I will put in the actual data for comparison.
Erik,
Since when are you using this forecast feature? Note that it uses, learns from historical data. In other words. It will need some time to get more accurate. And in case you’ve recently added a PV input or changed something, that might also have some impact on the forecast.
Your yield being underestimated must have a reproducible cause. Hopefully. Can you check/verify to see if the current estimate matches with the yield of your MPPT or SMA inverter? Perhaps data is missing / skipped from either one of them.
Maybe it’s unrelated, but I’m trying to help so please excuse me for being wrong.
It indeed does learn from historical data, but that applies to consumption, afaik. It takes up to 30 days to get the consumption profile up to date. Solar production should get up to date much sooner, since it mainly depends on maximum production and the weather forecast of the installation’s location.
Here’s the solar forecast for my test DESS system:
Same for solar forecast. It uses blocks of four square kilometer of your location, and it has no clue about information about azimuth, kWp, shading and what not. That is exactly why their AI scripts need to learn this from your data over time.
