Solcast Predictions - Accurate or Complete Waste of Time?

Signed up for solcast a few days ago. Have set it up to pinpoint my exact location and am just using the free api for home users.

Does anyone else use it, and if so... how accurate are their predictions vs reality?

We have a near perfect South facing roof, perhaps 2-3º of Due South, there is not a single bit of shading on the property. No trees within a 100 or more meters and none that are taller than any roofline within visual range. We are also uphill from the rest of the town at possibly one of the highest points in the immediate area.

So we thought we'd be getting decent generation... but for the second time I've noticed that we've had clear skies, nice strong sunshine and very weak generation... Especially compared to what's being predicted. The two images below show the biggest difference, the only other 2 graphs I have have been for very low generating days and they never got above a few hundred watts generation all day.

I know the panels are capable of producing higher amounts, just earlier in the week I saw it hit 3.27Kwh generation on a 4kwh system... I'm just confused how the amount of direct sunlight and their 'predictions' can be so wildly inaccurate.

Any explanations would be appreciated... I'm trying to judge when is the best time to put on the dishwasher to maximise solar generation. 😆 But if it's like this all day, then there is no suitable time to make use of the power and we'll be importing most of it from the grid as usual.

Not even putting more than a couple of % into the battery today either... I've had worse days with cloud cover and got a 50% charge from it.

I have pretty much stopped looking at the solcast predictions as they can be wildly out, sometimes low and sometimes high, sometimes very close.
I think in my case it is because i have an east west split pv and you can not set that up with solcast.

Two parts to my answer:
Is Solcast accurate? Fairly, but it's a statistical model. What you're not seeing due to the simplicity of the GivEnergy front end is that Solcast actually outputs three sets of half-hourly estimates with 90%, 5% and 10% certainty. Looking at my system's Solcast estimates for today, I can expect between 4.6 and 20 kWh of generation with a median of 15.9 kWh. Weighting the confidence factors at about 35% is a safe bet.

Is it a waste of time? As implemented by GE in the app, yes: it comes with little explanation, it's not integrated with any controls and doesn't take advantage of the 3-point estimates. You could just as easily watch the weather forecast or go outside. As part of an integrated control system, on the other hand, it's great. I take the Solcast estimates, combine with consumption data and use this to automatically set an overnight SoC target and it works a dream.

Anomnomnomaly No one can expect solacast to be perfectly accurate. I use it all the time as a general guide to what to expect in my area. However, I need to add to that my experince of the local weather to decide which way to lean from the forecast. If you are anywhere near the coast, mountains, hill ranges, large bodies of water, including major rivers then local knowledge is key to making the most of it. On the whole I find it more accurate than, say the BBC.co.uk forecasts for my village.

As an example of what I mean for local knowledge: I live in a rural area which is very flat but I'm in a loop of the river trent. When I moved here I would see dark clouds and rain moving in and expect it to hit but my neighbour would look up and say 'Nah, it'll move east and leave us alone'. True enough it would head straight towards us and veer off and around. You could even see the rain dumping to the east but it would be dry here. It took me a few years to get anywhere near the same knowledge.

For me, generally, if solcast says no or low generation or lots of generation its right and if its medium generation I can confidently expect generation will be at the upper end of it's forecast.

I find Solcast to be a reasonable estimate. But local weather conditions can render it inaccurate, especially if they are changeable through the day.

Overall, I use it to see if I am going to have a good vs a bad day of solar generation, is all.

Anomnomnomaly I think everyone else has nailed the solcast points.

I just wanted to focus on one thing:

I've noticed that we've had clear skies, nice strong sunshine and very weak generation...

If that's true, then that's really nothing to do with Solcast. If you're getting weak generation in strong sunshine, something is wrong.

It certainly looks odd, can you post a picture of your 'Power Graph' for yesterday. Also a bit more brief detail on the installation may help. e.g. No. of panels, Battery, IBoost (?) etc. and how long installed. and/or any recent changes.

I was finding solcast very unreliable until I had some trees pruned. I'm gobsmacked how much a difference it has made to my generation.

I didn't even realise that they have a limit to the number of times you can refresh the data... Just 10 before it cuts you off.

Considering I'd not even used it today... I just checked and it's already used 2 of them for some reason.

Anomnomnomaly I had my solcast account hooked up to the portal, and that burned through my api requests.

positor1 I have 2 different strings set in solecist and combined in the portal. You definitely can do that with a free account.

I have been logging the daily accuracy of both SolCast and Forecast.Solar (which is built into Home Assistant) for the last 6 months to assess the viability to use the forecast data for a over night SOC target to minimise export.

Generally SolCast (blue) over predicts and Forecast.Solar (orange) under predicts the actual generation (black):

4kWp array SSE facing

naltsta
that's odd. i even read on their site that you cant. can you share a link of some instructions of how to do so?

stevelewis

How have you got it to modify the overnight target? We've got Octopus Go so currently fill the battery to 100% overnight. It would be much better if, like today where we've exported 6.7kwh as the battery is full we'd only take a small charge (if any required) to get us through the night into generation time.

AlunR Set the target battery charge to a percentage. I'm the same as you and I set the battery to charge to 50% in the Go period last night. After dropping to 30% before generation was strong enough to start charging generation then topped that up to 100% by 2:00 this after noon.

AlunR my software runs separately to the app. It takes daily consumption and the Solcast forecast to work out how much overnight charge is needed and writes this value to the inverter using the GivEnergy API.

positor1 I added a second "site" in solcast with the same location but different orientation and slope. Then in the portal you add both Resource ID"s and your one "Solcast API key"

stevelewis As part of an integrated control system, on the other hand, it's great. I take the Solcast estimates, combine with consumption data and use this to automatically set an overnight SoC target and it works a dream

using Home Assistant?

I added Solcast 2 weeks ago, the numbers are all over the place, look at the actuals! Any idea what's going on?
Is this down to a "feature" of having an AC Plant with 2 inverters and 2 batteries or something else?

At least in the daily graphs the actuals v estimates are quite accurate 🙂

MarkMilward don’t worry, I think it’s definitely dual plant.. mine are completely wrong. (Also have dual inverters and batteries). I’ve been told that it’s due to data corruption because of the two inverters. Earlier on I had my system installed as two separate systems in parallel (no master/slave) and data all made sense. Since I’ve had it as master/slave, it’s maybe running better in sync sometimes but data is totally broken. Hopefully EMS (if it ever arrives) should allow for the system to process the data correctly

I don't really use GE portal for anything and all my work is done in Home Assistant - I have see some wild swings in solar forecast from solcast so I now use two sources: solcast.com and forecast.solar and use the average value and that is pretty much spot on (well as forecasts go - yesterday for example average forecast was 16 kWh and I got achieved 15.1 kWh so that is close enough what I need)

Im not convinced by Solcast. At best an estimate. Their periods of sun rarely materialise. Plus they have no idea how shaded my panels are.
These days ill use bbc weather to guide me for tomorrows sun.
The cinic in me says it is just a free way of them collecting the PV data from my location.

Roberto These days ill use bbc weather to guide me for tomorrows sun.

I use this as the most accurate predictor https://www.metoffice.gov.uk/weather/maps-and-charts/cloud-cover-map#

Not sure why, but solcast works just fine 90% of the time for me. So far today it is accurate to within 0.5kWh.

wilfleek

Since I originally posted this, the predictions have got a lot better. It's possible that my generation was out because of other issues that led to the dongle failing and being replaced.

Since then towards the end of Feb... it's been about 80% accurate... in retrospect. I've noticed that the 'predictions' keep changing through the day as they get better weather data I guess. Mine is normally under what they predict... but occasionally spikes much higher than what they predict.

When I look back on the previous days... my generation is normally about 15-20% lower than their prediction.

But after a couple of months use, I'm finding it a reasonable way to see ahead for the next few days.

If I see a couple of days of poor generation coming up... I might decide to do a couple of loads of washing a few days early, and make use of the extra generation I'm getting... Or wait until the mid afternoon to do some because it's going to be better then.

It's not accurate enough to rely upon... but it now gives me a rough idea of what I might generate and if I'll get the battery charged up.

stevelewis
Steve - I have just installed the Solcast tool within the GIvEnergy Templates. I'd like to do what you mention: automatically set an overnight SOC target based upon the future weather forecasts. I assume you have written a piece of code / api for this. Is this something you are willing to share? With notes on how to install and how to use. If not is there somewhere I can get this from.

Thank you.

DavidMoore1 www.github.com/salewis38/palm

This has been mine, I have found that its been pretty reliable unless I get hit by an unexpected pain in the ass cloud lol

I usually go to teh solcast main site to have a look though personally as that doesn't take up any api calls and I have found day to day it has been pretty accurate in determining how much of a nice day I can expect to time washes and whatnot.

EDIT
Just realised how old this post is lol

stevelewis
Thanks