Solar Forecast

nbullus

I've starting gathering data for March on the WW solar forecast vs Solcast vs actual generation.

Although it is early doors, the WW forecast does seem to be very conservative. Yesterday it underestimated by 46% and midday today it has already under estimated by 34%. I accept that my forecast is being done at 2am when the battery is scheduled to recharge so will never be 100% accurate due to weather changes after an early morning re-charge, but there seems to be a massive discrepancy with Solcast and I could easily have avoided the battery being re-charged today had the WW forecast been higher.

Is there any forecast correction factor we can apply to make the forecast more optimistic if we find it is consistently under-estimating, or more pessimistic if it is over estimating?

Thanks in advance for advice.

admin

nbullus We do indeed have a multiplier available and will expose this in the advanced settings. Just got some further backend work to do too.

psiloc

There is a "correction factor" setting in the advanced setting page. I have mine set to 0.9.

This won't affect the solar forecast as much as it it affects what SoC target WW bases on it. I'm sure it would achieve the exact same thing in practice.

psiloc

My forecast was massively wrong today too 😅 predicted 5.5kWh and I'm already up to 10.1kWh and it's only 1PM.

It's the first one to be so incorrect that I've noticed. It's inevitable though. I noticed that most weather forecasts thought it was going to be cloudy today

admin

psiloc Yeah it's important to not let one outlier skew the general results. I'd recommend keeping a few weeks worth of actuals vs estimates.

perfrank

admin Thank You, a correction factor to "expected solar" will be helpful.

I also experience a pessimistic WW forecast, typical between 25 and 100 % below realized . I did make a statistic over the values, but gave it up because I could not find a way to automate getting data from the WW activity log. Any ideas?

Some strange results of Smart charging made me wonder: Does WW Smart charging take in account when the days solar energy is expected, relative to the time slot?
Background for the question is, that I, in order to WW getting newest forecast, placed a Smart charge slot in the end of the off grid period, which today resulted in a unexpected 2 kWh charge.

gummy-front

admin

perfrank

I also experience a pessimistic WW forecast, typical between 25 and 100 % below realized . I did make a statistic over the values, but gave it up because I could not find a way to automate getting data from the WW activity log. Any ideas?

Good point. We should perhaps make this available to people as a CSV download.

Ultimately, combined with users actuals, we could run this against our unit tests and identify and correct under and over estimations in some cases, though we would need to be careful and need a fair amount of data.

Does WW Smart charging take in account when the days solar energy is expected, relative to the time slot?

Sort of. Before 11AM in the user's local timezone, we use the same day for solar forecast. After 11AM, we use tomorrow, but there may well be a problem here where some tweaks are required. Maybe it would be better to have the cutoff for tomorrow around 14:00-15:00 local time?

We'll run some unit tests around time cut offs.

We definitely will need more granular calcs when baking this into Agile Octopus slots.

psiloc

For the record, the solar forecast is fixing to be bang on for me today.

perfrank

admin Good point. We should perhaps make this available to people as a CSV download.

Would be nice when making statistics.

Ultimately, combined with users actuals, we could run this against our unit tests and identify and correct under and over estimations in some cases, though we would need to be careful and need a fair amount of data.

Thinking loud: A moving average of realized output over last few weeks, divided by forecast for same period seems to me to be a good starting point for a correction. factor This will adapt correction to changes in whether and seasons. Should/could probably be an option in [Advanced], if realized.

admin Maybe it would be better to have the cutoff for tomorrow around 14:00-15:00 local time?

I think so. Your battery should be fairly large to need next days forecast in calculation before evening. In my case, I have 12.8 kWh, whish more or less correspond to 20-24 hour use.

admin We definitely will need more granular calcs when baking this into Agile Octopus slots.

Sure. Living i DK, I am not on that. It is a very British thing😉

jjb

@admin Not wishing to over engineer a solution…. But it feels like solar forecasting would an ideal candidate for machine learning. If you can get the actual data from the inverter you could set it up to track the solar forecast, the actuals and potentially other data points such as time of year and weather forecast. It could then adjust the correction factor automatically for each individual user. This would bespoke forecast for individual circumstances, for example I have some shade factor to consider and others have multiple different orientations.

No expectation for this to be done as I know you have a load of other things in the works but if you don’t ask…

jeremy

I have had WW setup with Agile Octopus and functioning for only 24 hours.
The solar prediction has been completely wrong.
It is predicting 0kWh whereas by 10.00 am I had accumulated 1.5kWh … and it is still sunny.
Because of this prediction, the battery was charged to 100% overnight, meaning today I am just having to shed the excess.
This is not really how I expected the system to work, could I have set this up wrong?

admin

jeremy We’ll update this to be empty, rather than 0, to avoid confusion.

It’s currently not being used for Agile charge slots.

jeremy

admin Hi, I am not sure I understood your answer.
Are you saying that Solar prediction is not being used for Agile Octopus?
If so, how would I prevent this overcharge situation I face today?

admin

jeremy Correct, it isn’t being used.

I guess you would reduce your slots for charging in the interim, but it’s something that we’ll be introducing to Agile soon.

jeremy

admin I realised already that I did not understand the significance of time and charging slots.
I have one schedule (00:00 to 06:00) with 8 slots in it.

How would I go about adjusting this so that for instance it only charged to 75%?

admin

jeremy Because the slots are only 30 mins, we basically provide you with the conditional 'only charge if SOC less than' option for the charge window. And if the condition is not met, the slot is skipped.

You could lower that and see how you get on, but note that the SOC less than condition gets evaluated at the start of each condensed block of adjacent slots within a window.

jeremy

admin yes, I had already used that setting, using 85% as the value.
The 85% value seems to have been ignored.
I guess I have misunderstood something?

Do I need to use more than one schedule or time slot?
Where do you document this?

admin

jeremy There is currently no target SOC setting for Agile charge windows. The SOC less than condition is for the slot not to get actioned.

We can put a target SOC in, which might make sense. And later on allow a Smart toggle which uses Solar Forecast to calculate target SOC.

johalareewi

admin The SOC less than condition is for the slot not to get actioned.

Has it changed? It used to be the SOC condition was only tested at the start of a block of slots, not at the start of each slot.

nbullus

admin This sounds like a perfect solution having the toggle to use specific SOC or forecast as its as flexible as possible for the user.

admin

johalareewi Typo, slot in this case meaning slot or block of adjacent slots.