Off Grid production and tuning


I have a 5kw array at my off-grid house that I installed it a couple months ago.

I recently discovered this site and set up my site as a rooftop solar. My long term plan is to use it combined with data logs of my consumption to predict my battery charge for the next 3 days.
I can then use this to automatically run my generator for my bulk charge when I know solar will not be sufficient in the next few days. Ideally this will allow me to avoid running the generator when I know there is a good solar day coming tomorrow. Nothing worse then burning fuel to charge when there is sun coming.

I’ve been manually comparing my production to the forcast data and I’ve determined tuning would be beneficial. I just set it up to push real measurements on a 10 minute interval but only when in bulk charge mode. I think for off-grid limiting to readings during bulk is critical, in absorb I could be only charging at 1kw when the sun is actually capable of 3kw because my batteries can’t take the full power. This would mess with the turning of my array.

Couple questions based on my setup.

Does anyone else use Solcast for an Off-grid array?

If I only provide measuremnts when in bulik mode will tuning even be possible? Basically I’ll provide measurements from sun up until I hit absorb, some days this is by 10 am, other days I don’t hit it at all. I also do not provide measurements once the sun goes down and they are all 0s.

1 Like

Hi Erik,

Sounds like a good use of our system. We don’t have full visibility, especially for our free-tier users, for the exact use case for the data. It ranges from global solar var races to assisting the operation of country wide electricity networks and lots in between.

We do know about some opensource efforts that use our systems to assist with battery management, for example ->

So this might help with some ideas. There are also research projects I believe using our service around the idea of battery management but not sure if they are specifically looking at off-grid scenarios or not.

Tuning will still work if you are providing them in bulk lots. Even if it is once a week updates, sending daily data it will still work the same. Our automated tuning system checks for new measurements about once/twice a day to then look back on all historical measurements compared to satellite based observations to capture system characteristics to tune your specific system. For specific yearly shading, these won’t get captured until we can see measurements exhibiting these situations and learn from them, so you might see additional delay but this will improve overtime.

Consumption prediction will be a challenge so anything you are willing to share here I’m sure will be useful to others. We created the free-tier to support hobbyists and researchers to encourage sharing of these kinds of use cases so others can learn and optimize use of the data. We like hearing about these use cases too :+1:.

Hope this has helped, and let us know if you have any other questions. :+1:


Thanks Darren

I think I’ll leave it pushing my 10 minute average when in bulk mode for now. At this point I’m only tinkering so no rush to have my array tuned better. Unfortunately I don’t have any saved historical data to push.

For consumption data I was planning on recording the average wattage of what I consume in 10 or 15 minutes windows to my own SQL database. Once I have a large enough sample of data I can start running reports based on time of day, day of week, if the house is occupied or not, outdoor temp etc and try to come up a model that can predict my expected consumption. I can use the same predict and validate with real data to tweak my models. This is probably at least a year or 2 away.

I’m excited to see what I can accomplish with this.


I’m doing something similar to this but off peak charging a Powerwall 2.

How long does the tuning process take and is it best to send data 24 hours a day, or just when the system is actively producing?


Hi @scienceintheshed,

The tuning process will only succeed once there is enough high quality data it can use. Generally this is about 6-12 weeks worth of data. Sending all measurements is better provided the 0 production is due to lack of solar radiation. 5-15 minute data resolution.

More information in our, docs, let us know if you have any questions and please share results here as it can help others.

1 Like