WEBVTT 00:00.000 --> 00:11.360 Yes, hello. I am Sol, this is Alex. We're not Peter Durfield. Sorry about that. You 00:11.360 --> 00:16.200 couldn't do it, so we've got, you've got us instead. We'll do our best. And yeah, we're 00:16.200 --> 00:21.640 going to be talking to you today about our open source global solar forecast and dashboard. 00:21.640 --> 00:26.640 I'm going to do it in this order. I'll talk a bit about who we are and why we think we 00:26.640 --> 00:31.360 can do this and then I will hand over to my farmer talented colleague who tell you about 00:31.360 --> 00:39.200 the actual meat of the matter. So, without further ado, who are we? Well, we're two members 00:39.200 --> 00:46.360 of open plan at fix. We're a non-profit open source startup and we work primarily on renewable 00:46.360 --> 00:53.360 energy forecasting. So we do solar forecasts. We do wind forecasts. We do cloud forecasts. 00:53.440 --> 00:57.120 You say cloud forecasts, they're not renewable energy. I know, but we forecast them anyway, 00:57.120 --> 01:02.000 because we love it. We love forecast them so much. It is actually useful, I promise. Yeah, we'll 01:02.000 --> 01:08.800 come to that. And we do this to try and reduce emissions in the UK, see our two emissions. And 01:08.800 --> 01:14.400 how do we reduce emissions? Why do we do solar forecasting? Why does that help? Well, the idea 01:14.400 --> 01:20.320 is solar is growing, right? It's going to be the biggest, the most used form of electricity 01:20.320 --> 01:25.040 generation by 2040. It's huge. We love to see it. But there's a problem with solar. We know 01:25.040 --> 01:30.960 it. It's fickle, right? It's the clouds of there. It doesn't work as well. If there's pollution 01:30.960 --> 01:34.640 in your atmosphere, it doesn't work as well. If your panel is on upside down, it doesn't work at all. 01:35.360 --> 01:39.680 At night time, it goes away, right? It's hard. So, we thought that there was something 01:40.560 --> 01:44.880 something there where we could save a lot of CO2 by basically making it more accurate, 01:44.880 --> 01:49.920 because imagine if you're running a grid and your grid is made up of mostly solar. It's 2040. 01:49.920 --> 01:54.640 We're living in the future. And there's a big error bar between your generation 01:55.840 --> 01:59.360 that you think you have and what you actually have, because the solar is fickle like I said. 01:59.360 --> 02:03.440 It makes it hard to balance your grid, but it's a man's spike that you're trying to match. 02:03.440 --> 02:08.960 It doesn't line up with the generation that you have. You get a blackout. So, that's what 02:08.960 --> 02:14.560 happens here in the UK. We basically have these big gas turbines, and they are always running. 02:14.560 --> 02:20.320 They're always releasing CO2, and they can be turned up when the solar isn't quite what we 02:20.320 --> 02:24.640 thought it would be. And that's bad for the environment. So, we say, okay, if we can provide a really 02:24.640 --> 02:29.600 accurate solar forecast, then we can tell that the grid operations, there's going to be this much 02:29.600 --> 02:35.040 at this time. They can line it all up, and they can turn these off altogether. And that saves CO2. 02:35.600 --> 02:41.760 Which is brilliant. And so, that's what we did. We built a model PVNet. We trained it on 02:41.840 --> 02:47.840 a load of data, and we provide that to the national grid, and we have this looking forecast, 02:47.840 --> 02:52.240 and they use that in the operation room, and it's quite out to tell how well it's working, 02:52.240 --> 02:56.400 because there's a lot of decisions in the tree, but we think we're saving a lot of CO2 already in the UK, 02:56.400 --> 03:01.840 and they seem to like it. So, that all works well. Obviously, this is an open source conference. 03:02.640 --> 03:07.840 We want to make this kind of utility of solar forecast available to as many people as we can. 03:07.840 --> 03:12.480 So, how do we take something like PVNet, which is big, which is kind of run as a service. 03:12.480 --> 03:15.920 It's hard to contribute to. How do we take that and how do you make it more readily available? 03:16.480 --> 03:23.760 Less UK centric, more contributing, contribute to friendly. And that's where we thought the idea 03:23.760 --> 03:27.440 of the global solar forecast, which is what Alex will not tell you about. 03:29.120 --> 03:30.800 Thank you. Thank you. Thank you. 03:31.840 --> 03:37.760 Yes, so something so has already touched upon. We have something that works quite well for the 03:38.720 --> 03:46.800 UK, but it is quite bulky. So, the dream would be to provide accurate solar forecasts for 03:46.800 --> 03:52.800 everybody everywhere, who ever needs it. If you want to forecast for your solar panel at home, 03:52.800 --> 03:58.800 to know when to charge your car, if you want to run an energy grid somewhere else, that is not the UK, 03:58.800 --> 04:05.840 which is a very normal thing to do. And obviously, there's 14 of us. We can't do it alone. 04:06.800 --> 04:13.040 So, the obvious solution was to take an open source and hopefully get people to contribute, 04:13.040 --> 04:19.280 and then we can together build the beautiful solar bunker future that will eventually arrive, hopefully. 04:19.280 --> 04:23.280 I'm not going to go into the whole, why open source is great to build. I know which gone through 04:23.280 --> 04:32.080 it and time. So, it started with us creating open courts, which is the lightweight version of what 04:32.160 --> 04:38.720 we're doing in the UK. It's a smaller ML model. It just runs on three open weather forecasts. 04:38.720 --> 04:44.240 It's wrapped into a Python library. You can download it and run out of your laptop, just tell it, 04:44.240 --> 04:51.200 something about assets. As advertised in full lines of code, you might have seen this picture before, 04:51.200 --> 04:57.760 because we've presented open courts in 2024 at Foster. So, maybe some of you have been here. 04:58.080 --> 05:05.360 And initially, oh, this doesn't read on this projector at all, but this is a map. I promise. 05:08.240 --> 05:13.600 Yeah, eventually it was a replacement for the UK service, so an open UK service. 05:13.600 --> 05:18.480 But people started using it in other countries and telling us about it and we have the tracker 05:18.480 --> 05:23.760 of little flags when people tell us about them using it somewhere else on a GitHub page. 05:24.640 --> 05:30.960 And we decided to put like a wrapper around it to take a global as well. 05:33.200 --> 05:39.760 Yeah, so this is what the global forecast is. It takes centerpoints for the each country. 05:40.640 --> 05:47.360 It gets the capacity for it. It produces forecast 48 hours ahead for solar power in the country. 05:47.840 --> 05:51.760 And there is a beautiful yacht on top of it, which is quite fun to explore, 05:51.760 --> 05:56.640 which we are going to do now. Come along and let's see if we can crash that thing. 05:59.600 --> 06:00.240 Maybe we did. 06:02.400 --> 06:04.400 Yeah, my bad. 06:06.400 --> 06:11.840 Yeah, okay, maybe not a demo today. That's very sad. I should have done that beforehand. 06:12.080 --> 06:26.960 Okay, well, it's waking up. Let me talk about how we expected to be used and how you can get involved. 06:26.960 --> 06:30.880 So again, obviously we're compared by ourselves, and this is very much a work in progress. 06:31.600 --> 06:37.280 It's under the Herded Runs Open Quartz, which has been trained on UK side data. 06:37.280 --> 06:45.920 So basically solar roofs of panels, a generation from that. So it might not perform as well in different 06:45.920 --> 06:51.680 geographies. And we also don't really know how it's doing in different geographies. We have some 06:51.680 --> 07:01.440 benchmarking for the UK, but not for anywhere else. And yeah, so if somebody knows of other 07:01.600 --> 07:07.120 generation data that we can get access to to train a model wider, or wants to train a different 07:07.120 --> 07:12.240 model. You can swap them in and out in the open Quartz. It was to help us point benchmark it. 07:12.240 --> 07:17.040 Just if you use it, please tell us. If you don't, and it's because there is some feature that 07:17.040 --> 07:22.640 you would like us to have. Also please tell us, maybe we'll add it. Maybe it's a bit works now. 07:24.640 --> 07:27.760 Oh boy. I do apologize. 07:28.640 --> 07:46.960 Hello. Hello. Is it the wig? Yay! Okay. So no. It takes a bit to focus for the entire world. Please forgive 07:47.440 --> 07:57.600 this. But I don't think we'll maybe have time to have it a little look, but if you do get time 07:57.600 --> 08:04.560 in your free time, it does look kind of cool. Please take a look. I'm going to buy him this morning. 08:05.920 --> 08:14.400 My mum. Hmm. Yeah. Maybe I can tell you a white cloud forecast thing is useful, 08:14.400 --> 08:23.440 well we're waiting for this to load. We use satellite imagery in our model. It tends to tell you 08:23.440 --> 08:29.520 quite well. Let's tell you a lot about where the cloud's are, and that obviously helps with solar energy. 08:29.520 --> 08:33.840 But unfortunately, you can only have satellite data in the past. So we're starting forecasting 08:33.840 --> 08:39.120 satellite data into the future to see where the cloud is going to be in the future. Okay. 08:39.600 --> 08:51.440 Speedy runs through. So we have this fun chart. Okay. So this is the solar generation for the 08:51.440 --> 08:57.440 entire world. And you can see who's contributing the most at any given point. 08:59.280 --> 09:03.280 Yeah. I was going to do a pop quiz to ask which country contributes this giant police 09:03.280 --> 09:10.080 bikes, but it's obviously China no prices for that one. Yeah. Fun fact about this. This is 09:10.080 --> 09:15.840 15 minute resolution and like at peak there is enough solar power forecast like produced in the 09:15.840 --> 09:24.400 entire world to power the entire world for 15 minutes, which is nice. And then this is what we 09:24.400 --> 09:31.360 had a sneak peak of before. So this is how much it is for gas surge in each country and we can normalize 09:31.520 --> 09:37.360 by the country's capacity if I can hit the button. Yes. So instead of gigawatts, you can see 09:37.360 --> 09:43.920 like the percentage of the assets that are utilized. Yeah, which is a lot easier to see for 09:43.920 --> 09:49.680 different countries on the same scale and you can also see where the sun is at a given point. 09:51.520 --> 09:58.560 Yeah. You can click into a country, have a forecast for that one specifically. 09:59.280 --> 10:07.840 Oh thank God it works. And also see the capacities for the world. 10:09.520 --> 10:16.080 Oh dictionary capacity. Yeah. There is this nice table. We've added maybe 20 or 40 of these. 10:16.080 --> 10:21.360 The arrest was done by the open source contributors and also somebody transformed our 10:21.360 --> 10:25.280 very laying to the map into the globe that you've just seen, which was very cool. 10:26.240 --> 10:30.640 And there is some documentation. I don't know how to work a Mac. Please forgive me. 10:32.000 --> 10:37.200 That you could have read instead of listening to me. But I can drive this as well. 10:38.400 --> 10:44.960 And illustrate a guide. Okay. So hopefully I can go. 10:47.440 --> 10:52.800 Yeah. Okay. Thank you very much for listening. Sorry for some technical difficulties. Please find 10:53.360 --> 10:57.840 this. Please connect with us. Please contribute to us. Thank you for having us. 11:04.160 --> 11:04.800 Questions? 11:06.800 --> 11:14.240 Nice. I'm lucky. It's trying to work with the API. And like he, so book he asks, I'm who to talk to 11:14.240 --> 11:23.920 about the problems talking to the fraud solar API. Like is trying to access the API? 11:24.720 --> 11:31.520 Okay. And has some errors and wants to. Okay. Well I didn't think I can debug this in real 11:31.520 --> 11:38.320 time, but do email us. There is a support. Cool. Yeah. It should be there should be an email in the 11:38.320 --> 11:46.880 docs. Are great operators using this in production? And it's the UK TISO. Right? 11:46.880 --> 11:56.000 Yes. So our main model, Pivenet, is a bit of a beast. So we run a service in real time for the UK 11:56.000 --> 12:00.400 National Grid. They do use it in the control room to have an idea of how much thought is 12:00.400 --> 12:04.400 going to be forecast. What is going to be there in the country? It is definitely used. 12:05.120 --> 12:13.040 Open Quartz. I think we forecast for a couple of people as well with like solar farms. 12:14.960 --> 12:21.680 Yeah. I hope that answers this. Cool. Oh, I have a hand there. 12:22.000 --> 12:27.520 Do you just briefly about cloud? How did you come up with a few people? 12:27.520 --> 12:34.720 Uh, the question was, uh, can we can I tell them, can I tell everyone around about clouds and what 12:34.720 --> 12:40.720 the granularity of the forecast is? I will not remember the granularity, uh, spatially off the top of 12:40.720 --> 12:48.240 my head. Uh, temporally, I think it's five minutes late, maybe 15 minutes late, depending on the 12:48.320 --> 12:52.800 service that is used in the model. I'm sorry. This is not my project. It's based on the 12:52.800 --> 12:58.880 you met sat, uh, mess your sat data. So whatever the granularity that is in terms of spatial resolution. 12:58.880 --> 13:07.600 Yeah. Yeah. Um, is there a follow-up? I don't know if I talked enough. Okay. Please. 13:18.240 --> 13:35.200 Yeah. So the question was, uh, in Australia, there is a system that uses rooftop solar, uh, 13:35.200 --> 13:41.280 generation data in real time to see if there's clouds in the neighbourhoods and then it's 13:41.280 --> 13:46.320 moving towards you if I understand this correctly. Yeah. Uh, and if we're doing something similar 13:46.400 --> 13:52.000 or thinking of doing something similar. So yeah, uh, actually a my journey at OCS started with, 13:52.800 --> 13:58.000 a master's project that did something similar. We tried to, like, get the solar farms, like, 13:58.000 --> 14:05.360 uh, get the data from neighbouring solar sites. Uh, to contribute to the model, the problem is, 14:06.000 --> 14:11.680 uh, in the alive service, you can't really rely on the quality of data, uh, 14:11.760 --> 14:18.320 unfortunately in the UK and, like, the, uh, it got out. No, I have a bit of reports. So, yeah. And in 14:18.320 --> 14:22.560 Pivenet and our main model, we don't actually put in live data because of that reason. There's just 14:22.560 --> 14:29.280 not, um, there isn't a reliable stream. We've thought about sky crammers on and off, but then 14:29.280 --> 14:40.880 never got around to it. Yes, please. Sorry. What was the, uh, 14:41.760 --> 14:49.200 what was coming up? Yeah. So, yeah, we've had people talk to us about how 14:49.200 --> 14:54.960 focus and issue. We're not doing anything, uh, folk related at the moment, as far as I know, 14:56.400 --> 15:01.600 there's always too much to do, but it's kind of on the list somewhere. We take, um, we take 15:01.600 --> 15:05.760 from the end of repeat later, we take a visibility parameter and that has some of the fog. 15:05.760 --> 15:10.640 Yeah. You know, in co-related, but not in terms of satisfying. Yeah. So it goes into the model as, 15:10.640 --> 15:15.200 like, a data stream, hopefully, um, like, a variable that captures it. 15:17.200 --> 15:19.920 Time's up. I'm told. Thank you very much for having us.