3 ms·

At the end of 2016 I left contracting gig for a full time position to work on helping solve intermittency problem with solar PV power generation through forecas

At the end of 2016 I left contracting gig for a full time position to work on helping solve intermittency problem with solar PV power generation through forecasting. With a team of 3 including myself, the only professional software developer, we have launched and run a solar radiation and PV power forecasting/observation API (solcast.com.au) that can provide solar radiation and PV power forecasts world wide that update every 10-30 minutes based on satellite coverage. For the past 10 months this API has been freely accessible whilst we validated our approach and expanded to cover the globe. After great feedback from users, we are now planning a big update to make it even easier to use and to integrate live PV output data into forecasting itself. The change to work on something that contributes a large net positively to society’s around the world (making solar based electricity generation more financially attractive to operators/home owners a like all over the world) has been hugely rewarding and look forward to the growth of solar power generation in 2018.

nairboon 9y ago
What model do you use for power forecasting? Do you use different models for different forecast horizons e.g 1h, 1day, etc. or did you find that (recursive) multi step forecasting with the same model produces better results?
- layoric 9y ago
  We forecast radiation just radiation data as the base for all power forecasting (conversion from solar radiation -> PV power done on the fly) Solar radiation forecasting incorporates a few models for 1-7 days, and for now casting (0-4 hours) is based on NWP wind forecasting combined with our own cloud tracking, ML and blending with NWP models from 4-24 hours. Some of the above might be slightly off as I’m not a meteorologist or study solar radiation modelling. Luckily Solcast founders are and we are also partnered with a project at the Australian National University working with some extremely bright people to get the science right. If you want to get further into the details, feel free to email Nick Engerer or James Luffman (contacts on our website) about the science, or post on our community forums (forums.solcast.com.au).
therealdrag0 9y ago
That's really cool. I'd love to work on something like this.

nairboon 9y ago

What model do you use for power forecasting? Do you use different models for different forecast horizons e.g 1h, 1day, etc. or did you find that (recursive) multi step forecasting with the same model produces better results?

layoric 9y ago
We forecast radiation just radiation data as the base for all power forecasting (conversion from solar radiation -> PV power done on the fly) Solar radiation forecasting incorporates a few models for 1-7 days, and for now casting (0-4 hours) is based on NWP wind forecasting combined with our own cloud tracking, ML and blending with NWP models from 4-24 hours. Some of the above might be slightly off as I’m not a meteorologist or study solar radiation modelling. Luckily Solcast founders are and we are also partnered with a project at the Australian National University working with some extremely bright people to get the science right. If you want to get further into the details, feel free to email Nick Engerer or James Luffman (contacts on our website) about the science, or post on our community forums (forums.solcast.com.au).

therealdrag0 9y ago

That's really cool. I'd love to work on something like this.