Forecasting tomorrow's sunshine (to better use the solar)

SD_DRACULA

From what I can see most days when Forecast.Solar predicts under 20kwh it's a bad day with under 5kwh real production.

irishchris

Forecast.Solar for me is always a waste of time and never very accurate for my location. Solcast on the other hand has been nearly always fairly accurate. Today as example had 46.4 predicted with 47.1 generated. But agree the lower predicted days seem to be the less accurate ones. 15kwh predicted tomorrow but forecast not looking great on met.ie

reklamos

@connesha I have pulled your code from repo and there is a lot of going on in there. Before I start disecting the code, is there a simple way for me to just get hourly numbers of tomorrows forcast?

connesha

https://www.boards.ie/discussion/comment/118930722#Comment_118930722

The solar forecast is based on history, so requires a DB table, etc.

It was written for specific needs I had, and for figuring stuff out, so it may not currently be easily consumable into someone else's codebase. Over time I'll refactor pieces into something more manageable. I hadn't originally planned or sharing it this early.

At its simplest, it takes your generation and radiation for the past number of days, and calculates the ratio between them. Then the solar forecast for the future days is ratio * radiation. There's a bunch off stuff for selecting which of the past days to use for calculating the ratio, based on choosing days similar to the day you're forecasting, and how far to go back in time based on the time of the year (sun angle)

For pulling the radiation from met.ie see WeatherForecast

The db table that stores the daily stats where solar forecasting is done from is solar_forecasts. DbForecastDaily reads and writes to that.

ForecastAction uses WeatherForecast and DbForecastDaily to make future daily solar forecasts.

In the code version I posted, Hourly Radiation (DbRadiationHourly) is only for graphs, so you don't need to take that. I'm doing a bit more with that in the current version I'm working on

So, if you want to get the minimal running, I'd start with:

• Create the solar_forecasts DB table
• Just focus on getting ForecastAction working (you can remove the other actions to start with)
• Remove references to DbRadiationHourly from ForecastAction
• The only service you really need to get running is DbForecastDaily (refresh() it before using). You can remove the other Services.
• If you don't want to use SolarConfig as I have it (its backed by a db), then create a simple config class with the few values DbForecastDaily needs from SolarConfig (you can see their values in the README
• Use harness_forecast.py as your entry point to start with. Forecasts are written to the solar_forecasts DB table, so check that after

connesha

Following the discussion about the poor forecast last Wednesday on all the forecasting tools, and how solar forecasts are only as good as the weather forecasts they are based on, the following graph taken this afternoon shows it pretty well...

Looking at today on the bottom graph: the solid red line shows the hourly radiation from met.ie that was locked in at 1:45AM (this solid line never changes after 1:45am)

The dotted red line is the "live" forecast. It continuously retrieves the data from met.ie every 10 mins throughout the day, and "locks" them on an hour by hour basis. I.e. the "live" value at 11:00am is what met.ie returned just a few minutes before 11:00am, and so on hour by hour

The difference in the red lines shows how the weather forecast changed between 1:45am and during the day up to 17:31

At 1:45am, the total day radiation was 5076, whereas the sum of the live ones is 3925.

The solar forecast for 5076 radiation was 37kwh, whereas the live radiation of 3925 would have been 28.7kwh.

The vertical dotted red line is the "current time" when the graph was taken. The actual generation for the day up to that point in time was 27.6kwh

While this "live" data doesn't change anything for deciding whether to charge a battery, as you need to make a decision on charging in the middle of the night, it is useful to see during the day, and perhaps could make some automation decisions based on it...

bullit_dodger

Just said i'd swing back on this topic and update people with how the algorithm is getting along. Pretty good is the verdict I think.

Not really sure what happened on the 5th of April (looks like I don't have logs for that day), which in turn meant that the next couple of days were out until the algorithm "self-adjusted" to include the dodgy value. I guess I should filter out zero values when computing the averages, but I didn't think of it when I was writing it.

SD_DRACULA

Is that an excel sheet or something you can easily share for others to put our details into?

Looks interesting 😀

JayBee66

I'm amazed at how well SolCast is at forecasting solar radiance for the day. All the more, considering that I live down a hole in North KK, have an East/West configuration so I configured with the perpendicular to that axis and there are trees resulting in sunset being a little earlier due to shadowing.

Forecast Solar is way off. Seems to think that I live on planet Mercury and is always over optimistic. I have disabled it. I'd prefer to delete it.

As machine learning was my expertise before retirement, I was looking forward to a project using Met Éireann data but with SolCast doing so well, I'll settle with that and better use my time elsewhere.

silver_sky

I'm still using the met eireann forecast that was shared in one of the other threads. It's generally working out OK. Here you can see the past several days. I just put the figures into excel for the graph, it normally just does it's thing in homeassistant.

championc

https://www.boards.ie/discussion/comment/120445557#Comment_120445557

But is this taken from data at midnight, the night before ?

bullit_dodger

met.ie updates every 4 hrs (or so), so you can pull the data from whenever. In my case I pull at 5am, so just before the sun comes up. If you pull the data at 8am then you've potentially missed a few hours production from day 6am to 8am.

I added 3 panels to my own system last week, so the the recent estimated in the last week were off. Should settle down in a few days. I'll make this app available to people in a few weeks.

Aside: Tomorrow could be the best day of the year so far .... at least for Dublin. Haven't looked elsewhere in the country

Delboy5

https://www.boards.ie/discussion/comment/120447068#Comment_120447068

Thanks bullit_dodger i would defo appreciate the app as Im sure some others would. Currently doing a bit of manual work with Excel to see my Predictions, today and tomorrow both giving me over 20kwh...

DC999

https://www.boards.ie/discussion/comment/120447068#Comment_120447068

Class, you extended again. 3rd times a charm :) Another DIY micro inverter?

bullit_dodger

Nahh, I added 3x panels via my original installer to the main roof installation, so they are on the sting inverter. I originally wanted to have these on there back in the day, but SEAI guidelines necessitated that you have 50cm from the edge etc and that meant that I couldn't get them on there and still get the grant.

Grant obtained 6 months ago. It's my dime and my roof now, so on go the panels that I originally wanted. Still plenty of room to the edge and apex (like 32cm or something), but technically not the 50cm that they want. So main roof up to 6.5Kwp and the shed chipping in with 2.3Kwp brining me to 8.8Kwp in total.

I had a look on Google maps, and have yet to find anyone in south city Dublin with as many panels (I've 23). LOL. Maybe there's one out there, but I've not seen them. 18 is the highest I've found.

silver_sky

https://www.boards.ie/discussion/comment/120446418#Comment_120446418

On my setup I run it at 12:01 and then again at 1:30am. Battery charging if required takes place from 2-6.

The forecast later in the morning is probably more accurate but not much use for decision making at that point.

bullit_dodger

One thing I can tell you silver as I was running it on the hour every hour that it (the forecast) doesn't change from 12:01 and 1:30am. They update at 11pm (or 00:00 in the summer), and then they update again at 4am usually. I've mine set to charge from 5am -> 9am (summer time) I left it as late as I possibly could as if for example there was a few watts coming in at 7:30am, might as well grab it also.

It charges at 2.6kw so with a 8.5kw battery it can take a little bit over 3 hrs if empty. If the forecast significantly improves during that charge window, say for example i decided to charge to 80% and then all of a sudden it's blue skies forecasted for the day, then I leave what I have and set a new charge level like 20% or so.

JayBee66

HA forecasted 28.6KWh for today but production was only 25.1KWh. Does God do compo?

I disabled Forecast.Solar so I assume it's SolCast doing the forecast on the Energy dashboard.

Although the sky was mostly blue by midday there were some big clouds loitering over the house and that's something a prediction model can't account for.

I'll use SolCast to tell me roughly when to shower or run the oven so that the PV rather than the grid assists the battery. I'll be happy with that.

silver_sky

https://www.boards.ie/discussion/comment/120448055#Comment_120448055

The two times is more a doublecheck in case there are any issues as I frequently run HA updates after midnight. I've mine done in such a way that it works out the lowest possible charge rate for the time period. I figured this might be better than ramming as much in as quick as possible, just in terms of battery longevity.