PV Forecast in-accurate after a week of heavy snow

Question

question

jochenku asked Dec 16, '23

PV Forecast in-accurate after a week of heavy snow

Hi!

I live in southern Germany and we had unusualy winterly weather conditions in early December with 50cm of snow during 24 hours. Due to this snow my PV modules were producing almost no power for appr. 5 days.

In the meantime the snow has completely cleared, however the PV forecast still is way off. Although the irradiance has been mostly as forecast, the daily production has been on average 164% (max. 229% / min 122%) compared to the respective forecast during the last 7 days.

As far as I understand, the algorithm consideres degradation of the PV modules or increased shading (eg. due to tree growth).

Is it possible that one week of sudden and total loss of production messes the forecast completely?
How long will it take to recover?
Is it possible to reset the algorithm?

My issue is, that the forecast was pretty much spot on until the snowy period and I use it in a Node Red Flow to set the min. SOC and the inverter Power as well as other parameters. If the forecast is way to low my tools do not work very well (actually do not work at all).

I can gladly provide furhter data if that might help...

Best regards, Jochen

ESS solar forecast

2 |3000

Attachments: Up to 8 attachments (including images) can be used with a maximum of 190.8 MiB each and 286.6 MiB total.

Modifications Space

Node Red Space

Answer 1 · 2023-12-17T07:52:52Z

jochenku answered · Dec 17, 2023 at 07:52 AM

Just to add a bit more information I attach the irradiance data for the last 7 days and the relationship between actual and forecast PV yield.

I have acces to 5 other installations in my region (thus also affected by the snow event) and they all show the same behaviour: PC forecast way too low sind December, 01.

Anybody else having similar issues?

A response from @Guy Stewart (Victron Community Manager) or any other staff member would be much appreciated.

Best regards, Jochen

1702799150578.png (371.6 KiB)

1702799878525.png (10.0 KiB)

2 |3000

Attachments: Up to 8 attachments (including images) can be used with a maximum of 190.8 MiB each and 286.6 MiB total.

Answer 2 · 2023-12-18T03:59:47Z

Guy Stewart (Victron Community Manager) answered · Dec 18, 2023 at 03:59 AM

Hi @JochenKu,

Is it possible that one week of sudden and total loss of production messes the forecast completely?

That is going to have a significant negative effect on the forecast accuracy. I don't think something like that was part of the sample dataset examples.

How long will it take to recover?

In the worst possible case it will take 28 days to 'work through' the system and then that data will no longer be part of the estimation algorithm.

Is it possible to reset the algorithm?

You can manually reset the dataset by resetting your installation location in VRM, and then placing it again. It might take 2 days for the estimate to resume again, so worse off for that window, but then better off again.

2 |3000

Attachments: Up to 8 attachments (including images) can be used with a maximum of 190.8 MiB each and 286.6 MiB total.

Answer 3 · 2023-12-18T05:16:06Z

jochenku answered · Dec 18, 2023 at 05:16 AM

Thank you very much for the quick reply, I will proceed as suggested (reset the VRM location) and come back with the results.

Have a great week, Jochen

1

2 |3000

Attachments: Up to 8 attachments (including images) can be used with a maximum of 190.8 MiB each and 286.6 MiB total.

larsea-dk commented · Jan 22 at 09:04 PM

Any succes? Just nice to have a reply/feedback for other people experiencing similar deviation :-)

0 ·

Answer 4 · 2024-01-23T17:31:37Z

jochenku answered · Jan 23 at 05:31 PM

Unfortunately not really any kind of success:-(

I changed the location of my installation to a city several hundred km away and initally I saw the well known message 'forecast will be available after xx days of data'.

However when I reset the real location the next day, the 'old' and too low forecasts re-appeared again. So no help there except for waiting the a.m. 28 days. And right after this period was over, the next area of strong snow moved in, so here we go again...

I would strongly recommend to make the algorithm more resilient to significant periods of snow. If during the wintertime and after several weeks of a goof match between forecast and actual production the production falls significantly behind and the weather is at or below freezing then the production data of these days should not be interpreted as degraded performance but rather be ignored for refining the forecast...

Best regards, Jochen

1

2 |3000

Attachments: Up to 8 attachments (including images) can be used with a maximum of 190.8 MiB each and 286.6 MiB total.

Guy Stewart (Victron Community Manager) ♦♦ commented · Jan 24 at 06:46 AM

Hi @JochenKu,

Thanks for following up, I'll pass on the feedback.

1 ·

question