I explained to you why this is related to your north-facing system and the specific weather conditions of the last few days. Give the system a few days. One more thing… the system changes its forecasts every 15 minutes. Mine has also gone up and down by a factor of 2-3 over the last few days. That was due to the regional weather. I’m very dependent on the weather for my job and have several weather forecasts, including professional ones. None of them have been regionally accurate over the last two days. Sometimes a cloud band moving a few kilometers differently or high fog that doesn’t clear up can have serious consequences for the PV.
Na I do not believe the learning theory I do believe that I/we got maybe an update (or more likely accidentely correct):
Now I see the forecast as a range the first time.
and yes the energy forecast did go up and down by factors but the irridiation forecast not (here at least). If the irridiation is pretty stable the energy forecast should be as well - that’s why I believe it is a bug they maybe are working on.
But lets move from believing to facts again:
Energy forecast for tomorrow 135-165:
Irridiation forecast:
So roughly 2.x (realistic) to 4 (very optimistic) times of todays actuals. Means we should end up at ~55-80kWh tomorrow - VRM states 135-165kWh. Lets see who is right my super simple way or VRMs algo.
update next morning VRM has slightly reduced the forecast for today 27.03
Similar Irridiation
but
to 128-156kWh
Result 43,8kWh
Irradiantion result
So “I” was to optimistic by 25% but VRM by factor 3. Irradiation has changed but not dramatically.
Last day then a pause for min a week.
Today it should be sunny
But that sunny? 207kWh min?
Irradiation:
I would assume ~40% more than yesterday so I am guestimating a solar engery output of ~61kWh today (most likely bit less).
Update with the result: 52.2kWh
“I” was 15% to optimistic. VRM by factor 4 wrong. I really would like to understand what Victron is calculating.
So several days later - the forecast for today
taken yesterday evening - super optimistic 96kWh
and from this morning - more on the pessimistic side 41kWh
Comsumption forecast has not changed but solar has changed a lot over night - no weather forecast is stable showing blue sky!
My forecast would be 55-65kWh (733 yesterday today 744W/sqm peak and yesterday we had 62kWh but it will be warmer so less efficient)
The forecast in the early morning seems now to be much better then the evening of the day before which looks no longer that bad (factor 3-10) instead 1.5-2 wrong. Why there is such a huge difference I cannot understand - looks like completely different algos.
Real result will be added this evening.
Here it is 58,6kWh:
Wait a few days, and your consumption forecast will adjust to your actual consumption. This works very reliably.
Consumption forecast works as expected even though it is far more difficult to forecast the consumption than the solar.
Solar for today forecasted yesterday evening 143kwh
this morning 32kWh
My forecast (simply taking the irridiation forecast and rule of three) is ~59kWh as irradiation (weather) today should be similar as yesterday. Picture taken yesterday evening
Why are both values (evening of yesterday and morning today) sooo different. It also looks like the better the weather the bigger the difference. Still my conclusion they are using different algorithm for both and both are not working as they should and it seems no one wants/can transparently explain how it is done.
Result 56,9kWh:
Wie still get the unrealisitic astronomic high forecast in the early evening
Lukily in the morning the forecast is now better so DESS can work somehow as expected for the day. For me it looks like switching between two completely difference approaches.
today we had 54kWh - irradiance forecast shows less so why should we get 167-204kWh as forecasted. Pretty sure tomorrow morning this should somehow fit.
My solar forecasts have been way off the past few days as well.
If I’m not mistaken, end of last week forecasted solar yield was close to zero and this week they’re unrealistically high.
For today, forecast was 20 to 45kWh each hour, out of a 3kWp array 
Live update: forecast just updated to slightly more realistic values.
Live update2: system just switched back to bonkers forecast for today.
For tomorrow it’s looking more realistic so far, so it seems the system is somehow recovering.
Maybe there’s a glitch in the data source used to predict solar yield ?
I also thought it is getting better - but it seems to jump between somehow OK in the morning and completely off in the evening since weeks. Where the morning version is improving.
I honestly still don’t understand how the system calculates and noone explains the parameters considered.
Yes, I understand it can get frustrating when a system suddenly starts behaving erratically and you’re left without an answer or explanation.
Double so since Victron has a high reputation of actively engaging with their customer base.
On the other side, Victron doesn’t really have an obligation to explain how their algorithms work.
It would be very much appreciated if they do of course, but I understand it if they choose not to disclose that information - even if it’s frustrating when it’s clear that errors are being made.
From my external point of view, a LOT is currently buzzing at Victron R&D department.
There’s new hardware expected, VRM seems to be undergoing major updates and VenusOS v3.60 is probably splitting some brains as well at the Victron HQ.
Add all that up and I can understand that, from a customer’s point of view, community engagement seems to be low.
Somehow I’m fairly confident that Victron staff is taking notes of the ongoing hickups and either is already working to fix them or will soon get to fix them.
My crystal ball is, unfortunately, kind of foggy at the moment but I do sense that in a few months (weeks ?) something big is going to drop that will (hopefully) relieve the Victron engineers from a lot of stress so they’ll have time to pick up the less critical notifications again.
I completely agree with you regarding future developments and current problems. I, too, get the feeling from many of the comments you’ve made that there’s a lot going on in the Victron world. But is the PV forecast one of these areas that needs work? I actually have the feeling, and can confirm from my personal experience, that the PV forecast is going relatively well. Not perfect, but there are no complete failures either. Integrating this forecast for further periods into the future is a completely different discussion. Bear in mind that the person who started the post has a very, very specific situation that’s far removed from many other situations. Could that be the problem?
1 Like
For me pretty strange that you are constantly pointing towards me/my system being the potential problem. I don’t think a solar system pointing to north is that super special - unusual maybe but the forecast algo should be able to handle this as well. It should not even matter if you would have several systems pointing to random directions.
I was showing that a pretty accurate forecast is easy to achieve with a simple rule considering past days energy production and irradiance forecast. There is absolutely no need for complex AI learning or other buzz stuff to do a simple one day ahead forecast.
I am reporting the facts that I see and asked for an explaination/help to improve. That should be the normal way to improve right (even if it would be a corner case)? If something is wrong with my system also happy to correct.
From my point of view the factors wrong forecast looks like a bug. Why else should it jump between somehow realistic and completely of values. I am offering help to debug and improve but seems no one is willing to look into the issue.
It might be low prio for VRM developers - taken and accepted. Then this might be still valuable for others seeing potential misbehaviours and motivate them to add further data to nail down the issue.
Note:
I like the victron system and the idea of DESS. I had many years e.g. SMA and other stuff, but the open system of victron convinced me building a new system being able to control everything and integrate into home automation. Works great overall. Minor stuff like this solar forecast to be improved to be even better.
For me, however, your almost exclusively north-facing location is the only plausible explanation. As I’ve reported several times, it works for me and many others without any major issues. There have been some discussions about percentage deviations, but not to the extent that you’re experiencing. What else could be the cause of your problem? Or is it a specific geographical location with unusual meteorological conditions? How else is it possible that the system works well for hundreds of users but is so off-kilter for you? What is your explanation for this phenomenon?
But this is no reason why this should not be a bug. Similar reason I hear often from developers - “works on my machine” - can I then ignore others where it is not working? I found some (not many there I agree) similar cases - unfortunately no thread came to a conclusion and threads were closed due to inactivity.
I have shown that no matter what system you have the one day ahead forecast is simple using existing data. No reason at all to be factors off. That is what I expect from a professional system development as well.
Or a releasenote: “Be careful north systems will not work properly.”
That’s exactly why I think it is a bug.
As said it meanwhile switches between somehow correct and completely off values within an hour and then back again. In the morning it is somehow OK now and correcting the charging behaviour. That’s why no longer high prio for me.
We have ruled out all known other factors, like wrong setting of geolocation etc. If there is a part in the algo triggering this misbehaviour it would be interesting to all of us which one it is so we can check this in future.
That’s exactly what I ask Victron - only they can explain what they are calculating. The most simple answer is bug in algorithm.
Lets close this bug or no bug discussion and focus on the facts and data hoping they will fix this as well.
Since the thread is relatively long already let me illustrate the issue once more as today seems to be a good example:
Yesterday we had a solar production of 46,3kwh - sunny morning then bit worse then improving:
the irradiance forecast yester looked like:
which fits pretty good to the curve of the result.
Today irradiance forecast is:
which means slightly higher peak but far more constant while earlier afternoon. Looking at the curve (area of the curve which equals energy) it might be 20-30% more. So I would assume 9-14kWh more production today, which would mean ~55kWh.
VRMs solar forecast of today looks currently like 184-225kWh. The peak hour at nine with more than 30kWh production in one hour. How should this be possible with a 21kWp system? Expectation, as irradiance until lunch is similar, should be similar as the day before means 4-5kWh between 8 and 9am..
As consequence this means the battery can be easily charged with solar only. Which is not the case.
Instead DESS should decide to charge the battery when the price at ~2pm is low.
Pretty sure the system will correct itself while the morning but the initial forecast is again completely off. Typically it changes the solar forecast within a few hours from >>100kWh to more realistic ~50kWh without any changed irradiance/weather forecast. Since the numbers are sonextremly different there must be a paramter in Victron algorithm influencing the result in a wrong way. Would be interesting to know what is part of the formular to correct it.
Again: VRM’s PV forecast is calculated by Solcast, not Victron, so it’s not fully under Victron’s control I’m afraid.
All you can do is to subscribe to Solcast yourself and manage to get accurate forecasts out of it by tweaking certain parameters.
1 Like
Are you sure that the resulting PV energy forecast is handled by Solcast? How should they do this as there is no data except the location about the system. Typically Solcast needs kWp and angles of each single system (next to location).
Note: I have even a solcast account
Result looks OK ~10kW peak power. Even for yesterday after lunch the cloud dip is visible. So Solcast does a ok job.
Is it not more like that the sun irradiance is used from solcast? The irradiance is always matching and looks OK.
And honestly it does not matter - as end user I look at the result provided by the E2E system.
Result of today:
So DESS corrected itself and charged battery from grid and PV energy resulted close to what I calculated by irradiance (visible in VRM portal) and energy production of yesterday.
And for most people, it seems to work relatively well. It can always be better! At least not by a multiple. This then raises the question: Why doesn’t it work for you? This is called error analysis. And that leads me to the explanation for the northern system. But then you accuse me of making an assumption and demand that Victron disclose their algorithm. Is that the way forward?
This is the wrong question. Question is “why does the system produce this wrong results with the input given?”. Since direction north /east whatever is no input parameter it should not matter at all.
If direction would matter east west should also have issues.
I did not ask to disclose the algorithm itself but to give at least an indication what can influence such results.
It is obviously wrong forecast produced by a victron system not my intepretation or similar. Just facts provided for an analysis by someone who has more insights - so please stop pointing back to me as you obviously cannot or not willing to help.
I’m not defending or accusing anyone. What I’m wondering is why something isn’t working for you that isn’t a problem for others. There must be a reason. As for algorithms, I think that’s one of Victron’s core assets, and they certainly won’t disclose them. Why would they?