WEBVTT 00:00:00.880 --> 00:00:04.400 Hi everyone, this is Solomon Freer from PV Lighthouse. 00:00:04.480 --> 00:00:06.871 In this video, I'm going to give you a tutorial ... 00:00:06.871 --> 00:00:10.079 on our latest new feature for SunSolve Power, the Optimiser. 00:00:10.400 --> 00:00:13.524 You may already be familiar with our tool for sweeping cell ... 00:00:13.524 --> 00:00:14.799 and module parameters. 00:00:15.360 --> 00:00:18.398 The optimisation update now allows users to harness an ... 00:00:18.398 --> 00:00:21.898 inbuilt evolutionary algorithm to find optimal combinations ... 00:00:21.898 --> 00:00:24.870 of parameters by exploring the ranges you set for them ... 00:00:24.870 --> 00:00:28.238 and finding the best result based on your predefined metric. 00:00:29.040 --> 00:00:32.145 The benefit over sweeping is that the optimiser doesn't ... 00:00:32.145 --> 00:00:34.853 need to check the full resolution of your desired ... 00:00:34.853 --> 00:00:36.835 ranges, instead learning from each ... 00:00:36.835 --> 00:00:39.808 generation of results and producing new generations ... 00:00:39.808 --> 00:00:42.318 based on the best results of the previous one. 00:00:42.880 --> 00:00:46.299 I won't be going into the detail of the evolutionary algorithm ... 00:00:46.299 --> 00:00:48.731 we use in this video, but instead I'll give an ... 00:00:48.731 --> 00:00:51.492 overview of how to use the optimiser tool to get the ... 00:00:51.492 --> 00:00:52.478 results you need. 00:00:53.040 --> 00:00:55.565 However, if if you go to the about tab ... 00:00:55.565 --> 00:00:58.763 and select Solver, then you can scroll down to ... 00:00:58.763 --> 00:01:01.961 Optimisation routine and options to get more ... 00:01:01.961 --> 00:01:05.748 information about how our algorithm works and how you ... 00:01:05.748 --> 00:01:10.798 can change the user options to fine tune the optimization to your needs. 00:01:14.320 --> 00:01:15.154 So, to start, 00:01:15.154 --> 00:01:18.872 what you'll want to do is load a simulation file that you're ... 00:01:18.872 --> 00:01:20.238 looking to optimise. 00:01:21.200 --> 00:01:24.093 If it's already been solved, then you'll want to duplicate it ... 00:01:24.093 --> 00:01:26.079 so that you can set it up for optimisation. 00:01:26.480 --> 00:01:28.453 And if you don't have a simulation file, 00:01:28.453 --> 00:01:31.383 you can start with one of our provided templates and edit it ... 00:01:31.383 --> 00:01:32.878 to match your target problem. 00:01:35.200 --> 00:01:38.749 So then once you've got your simulation file ready, 00:01:38.749 --> 00:01:42.050 you'll head over to the Solver tab under Inputs, 00:01:42.050 --> 00:01:45.599 and you'll select your solve type as optimisation. 00:01:47.760 --> 00:01:51.744 So from here you can see that in order to start having inputs ... 00:01:51.744 --> 00:01:54.915 for optimisation, you'll need to activate the ... 00:01:54.915 --> 00:01:58.655 wand to enable the adding and removing of optimisation ... 00:01:58.655 --> 00:02:02.639 inputs so you can go over to the tab that's relevant for you. 00:02:03.280 --> 00:02:06.117 So in this case, I've loaded a tandem module ... 00:02:06.117 --> 00:02:08.559 with perovskite and silicon layers. 00:02:08.880 --> 00:02:13.291 So first I'll choose the thickness of the silicon layer ... 00:02:13.291 --> 00:02:16.359 as a sweep input for the optimisation. 00:02:16.760 --> 00:02:19.571 And then secondly, I'll go over and find the ... 00:02:19.571 --> 00:02:23.319 perovskite layer and select that for optimisation also. 00:02:24.360 --> 00:02:27.264 Then on the solver tab, you can now see that there's ... 00:02:27.264 --> 00:02:30.030 a number two here, indicating I've now selected ... 00:02:30.030 --> 00:02:32.381 two different inputs for optimisation, 00:02:32.381 --> 00:02:35.078 and they appear here under Inputs to optimise. 00:02:36.120 --> 00:02:40.150 So the next step is to choose the ranges that you're looking ... 00:02:40.150 --> 00:02:43.522 for to optimise over, so the ranges that you think ... 00:02:43.522 --> 00:02:46.318 you're likely to Find the right answers. 00:02:46.320 --> 00:02:48.551 And you can also choose your 'best guess' 00:02:48.551 --> 00:02:51.438 to help the optimisation algorithm look in the right ... 00:02:51.438 --> 00:02:54.719 regions in those early generations to speed up the process. 00:02:55.840 --> 00:02:58.853 So, for the 'slab 3a thickness', 00:02:58.853 --> 00:03:04.879 which is the silicon layer, I'm going to set a best guess of 180 µm. 00:03:05.360 --> 00:03:06.960 And you can see the units over here. 00:03:07.760 --> 00:03:12.846 And I'm going to allow it to sweep from 100 to 300 with a ... 00:03:12.846 --> 00:03:17.123 resolution of 5 µm, keeping in mind that when ... 00:03:17.123 --> 00:03:22.325 you've got a wider range with a very small resolution, 00:03:22.325 --> 00:03:28.798 it will likely take longer and use up more rays to get to the optimum. 00:03:28.800 --> 00:03:33.158 So if you're not sure, then you can often see it as ... 00:03:33.158 --> 00:03:38.497 best to use a larger resolution at first and get a rough idea ... 00:03:38.497 --> 00:03:41.439 before you really try to hone in. 00:03:42.240 --> 00:03:45.154 So, for the perovskite layer, 00:03:45.154 --> 00:03:49.467 I'm going to allow it to sweep from 20 up to 400 ... 00:03:49.467 --> 00:03:52.731 nanometers, and I'll do this in 20 ... 00:03:52.731 --> 00:03:54.479 nanometer steps. 00:03:54.560 --> 00:03:56.492 Sorry, 20 nanometer steps, 00:03:56.492 --> 00:03:57.920 with a best guess... 00:03:57.920 --> 00:04:00.240 ...let's put it at 100 to start. 00:04:01.600 --> 00:04:04.000 Now, you'll now want to choose your goal. 00:04:04.080 --> 00:04:05.576 So, for this situation, 00:04:05.576 --> 00:04:09.839 I'm going to try to maximise the electrical power of the tandem cell. 00:04:09.840 --> 00:04:12.855 You can also choose the absorbed photon current in a ... 00:04:12.855 --> 00:04:15.519 particular layer or the electrical current. 00:04:17.360 --> 00:04:20.627 So now you get to the optimisation options, 00:04:20.627 --> 00:04:24.348 and you can choose the maximum number of runs that ... 00:04:24.348 --> 00:04:26.798 you'd like the optimiser to use. 00:04:26.800 --> 00:04:28.720 And if it reaches that maximum, 00:04:28.720 --> 00:04:33.079 it will stop and give you the best results so far before it's converged. 00:04:33.640 --> 00:04:36.806 If it's at the point where it's likely to go over your ... 00:04:36.806 --> 00:04:39.456 maximum run number, you can also change the ... 00:04:39.456 --> 00:04:41.959 number of rays that you're using per run. 00:04:42.120 --> 00:04:44.409 But be aware that if you reduce this number ... 00:04:44.409 --> 00:04:46.894 significantly, you're going to experience a ... 00:04:46.894 --> 00:04:48.921 lot more uncertainty in the results, 00:04:48.921 --> 00:04:51.472 which will affect the convergence and also the ... 00:04:51.472 --> 00:04:53.957 validity of the actual optimisation result. 00:04:54.200 --> 00:04:56.040 So don't change it unnecessarily. 00:04:57.160 --> 00:04:59.480 And you can also set a time limit for the optimisation. 00:04:59.880 --> 00:05:02.857 The maximum total raise will be calculated based on the ... 00:05:02.857 --> 00:05:05.639 maximum runs and the raise you've selected per run. 00:05:06.520 --> 00:05:09.361 Then there's the algorithm options for the genetic ... 00:05:09.361 --> 00:05:11.079 algorithm that we're running. 00:05:11.640 --> 00:05:14.139 You can read more about these options, 00:05:14.139 --> 00:05:16.638 as I said in the about tab under solver, 00:05:16.638 --> 00:05:20.386 but for this demonstration, I'm going to stick with these ... 00:05:20.386 --> 00:05:21.479 default values. 00:05:22.120 --> 00:05:26.140 One option that I will draw your attention to is the ... 00:05:26.140 --> 00:05:30.734 convergence tolerance, which basically says you'll see ... 00:05:30.734 --> 00:05:33.318 more explained in the about tab, 00:05:33.318 --> 00:05:37.721 but it tells the optimiser how how sure we want to be that ... 00:05:37.721 --> 00:05:41.741 the best results that we've seen are falling into a ... 00:05:41.741 --> 00:05:46.335 narrower enough range that we're confident that it's not ... 00:05:46.335 --> 00:05:50.355 currently still improving with each generation. 00:05:50.680 --> 00:05:53.765 So it's saying if you set this number to a large number, 00:05:53.765 --> 00:05:55.959 then you're going to converge faster. 00:05:56.120 --> 00:05:59.530 You're willing to accept that there might still be some ... 00:05:59.530 --> 00:06:01.976 uncertainty and you're not currently, 00:06:01.976 --> 00:06:05.163 you've not currently hit a plateau in your results, 00:06:05.163 --> 00:06:08.721 whereas if you set it smaller, then it'll be more strict in ... 00:06:08.721 --> 00:06:12.427 terms of the convergence and it will wait longer before it's ... 00:06:12.427 --> 00:06:15.318 certain that you're not going to do any better. 00:06:16.280 --> 00:06:18.765 But keep in mind that as you set this smaller, 00:06:18.765 --> 00:06:21.317 you're going to go through more generations, 00:06:21.317 --> 00:06:23.399 more runs, you might hit your maximum ... 00:06:23.399 --> 00:06:25.279 run limit, you will use more rays. 00:06:25.440 --> 00:06:27.200 So these are all things to be aware of. 00:06:28.240 --> 00:06:32.365 But so once you're ready, you can just hit play and then ... 00:06:32.365 --> 00:06:37.223 you'll see it submitted the job and it will get started and start ... 00:06:37.223 --> 00:06:41.439 going through some generations once the job has begun. 00:06:44.400 --> 00:06:47.060 In the meantime, while this is running, 00:06:47.060 --> 00:06:50.849 you can already see that it started on generation one and ... 00:06:50.849 --> 00:06:54.799 this download optimisation results button has appeared. 00:06:55.280 --> 00:06:58.966 If you hit download before the optimization is finished, 00:06:58.966 --> 00:07:02.498 you'll be able to download a CSV (file) with the current ... 00:07:02.498 --> 00:07:05.954 runs that have been simulated and you'll get an idea of ... 00:07:05.954 --> 00:07:08.718 where the optimisation is at at the moment. 00:07:08.880 --> 00:07:11.406 But it's not a very long process generally, 00:07:11.406 --> 00:07:14.564 so it's best to just wait until the end to look at the CSV ... 00:07:14.564 --> 00:07:16.318 (file) with the full results. 00:07:17.480 --> 00:07:19.704 In the meantime, while this is running, 00:07:19.704 --> 00:07:22.737 I can look closer at the about tab and step through some ... 00:07:22.737 --> 00:07:24.759 more detail about the convergence. 00:07:26.520 --> 00:07:29.425 So here's this convergence tolerance number, 00:07:29.425 --> 00:07:31.362 the 'k' that I was talking about ... 00:07:31.362 --> 00:07:33.448 before, and you can see from these ... 00:07:33.448 --> 00:07:36.279 images how the convergence tolerance works. 00:07:37.800 --> 00:07:40.519 So as we've got more and more runs, 00:07:40.519 --> 00:07:45.471 if the best runs have still quite a bit of variance about them, 00:07:45.471 --> 00:07:49.937 then this will mean that you're less likely to pass this ... 00:07:49.937 --> 00:07:53.918 convergence test depending on your value of 'k'. 00:07:54.080 --> 00:07:58.185 But as your best runs start to form a tighter distribution, 00:07:58.185 --> 00:08:02.374 then you can be more certain that the confidence interval of ... 00:08:02.374 --> 00:08:06.144 what the best value is has gotten tighter and compared ... 00:08:06.144 --> 00:08:09.998 to the confidence interval of your best results so far. 00:08:10.880 --> 00:08:13.510 Then depending on the value of 'k', 00:08:13.510 --> 00:08:17.411 you can meet convergence based on how tightly these ... 00:08:17.411 --> 00:08:19.679 best values are distributed. 00:08:21.040 --> 00:08:23.777 So now if we go back to the inputs tab, 00:08:23.777 --> 00:08:26.879 we can see it's now up to generation five. 00:08:29.280 --> 00:08:33.717 So I'll put a cut in the video here to skip forward to when ... 00:08:33.717 --> 00:08:35.839 this is finished solving. 00:08:39.400 --> 00:08:42.152 So now that the optimisation is complete, 00:08:42.152 --> 00:08:46.084 you'll see that there's these result values highlighted in ... 00:08:46.084 --> 00:08:48.758 green under the inputs to optimise area. 00:08:49.800 --> 00:08:53.023 In this case, we can see that it found the ... 00:08:53.023 --> 00:08:57.809 best value for the silicon slab thickness was 230 µm and for ... 00:08:57.809 --> 00:08:59.958 the perovskite was 400 nm. 00:09:00.280 --> 00:09:03.570 Unfortunately, this is a case where it the ... 00:09:03.570 --> 00:09:07.519 boundary of my range ended up being the best value. 00:09:08.160 --> 00:09:11.987 So this tells you that you need to run the optimisation again ... 00:09:11.987 --> 00:09:15.279 with a broader range in order to find a true optimum. 00:09:15.280 --> 00:09:19.403 Because it's telling you that it hasn't tested values further ... 00:09:19.403 --> 00:09:23.050 than this boundary and the boundary value gave the best ... 00:09:23.050 --> 00:09:24.239 result possible. 00:09:24.800 --> 00:09:27.545 But looking closer, we can see what the power ... 00:09:27.545 --> 00:09:29.919 that it found for this best result was. 00:09:29.920 --> 00:09:32.916 If you go to the outputs tab and and you check in this ... 00:09:32.916 --> 00:09:35.912 case the 'Module JV'. If you were optimising other ... 00:09:35.912 --> 00:09:38.123 parameters, you could look at some of ... 00:09:38.123 --> 00:09:41.261 these other tabs to find the relevant area to find the ... 00:09:41.261 --> 00:09:42.759 optimum that you wanted. 00:09:43.000 --> 00:09:45.569 But of course, all of this information is also ... 00:09:45.569 --> 00:09:49.559 available in the CSV (file) if you download the optimisation results. 00:09:52.520 --> 00:09:54.527 But so... to prepare for a new ... 00:09:54.527 --> 00:09:58.206 optimisation where we can include a broader range so ... 00:09:58.206 --> 00:10:01.718 that we can test the full optimisation beyond just ... 00:10:01.718 --> 00:10:05.314 getting stuck at this boundary, what we want to do is ... 00:10:05.314 --> 00:10:09.077 duplicate the simulation and then edit these ranges. 00:10:11.240 --> 00:10:13.400 So because this value got stuck, 00:10:13.400 --> 00:10:15.320 we'll increase it to 1000 nm. 00:10:15.720 --> 00:10:18.942 And then we'll also increase this lower value to 300 (nm) ... 00:10:18.942 --> 00:10:21.959 because we don't want to recheck the lower range that ... 00:10:21.959 --> 00:10:23.399 we've already ruled out. 00:10:24.440 --> 00:10:27.193 Then we'll also want to increase this best guess to ... 00:10:27.193 --> 00:10:29.159 make sure it's within our new range. 00:10:30.680 --> 00:10:33.240 But I'll show you in a minute what happens if you don't do this. 00:10:33.640 --> 00:10:36.997 So let's also update the 'slab 3a thickness' 00:10:36.997 --> 00:10:39.719 to look between 200 and 300 microns. 00:10:39.720 --> 00:10:44.120 But let's leave the best guess outside of this range and hit play. 00:10:44.680 --> 00:10:46.609 And you'll see 'Validation Failed' 00:10:46.609 --> 00:10:49.759 'Simulation not started, validations of inputs failed.' 00:10:49.760 --> 00:10:51.960 You can see the Info tab for more details. 00:10:52.360 --> 00:10:55.080 'Best guess too low' on the Info tab. 00:10:55.560 --> 00:10:59.171 So it'll warn you if there's an issue with the way that you've ... 00:10:59.171 --> 00:11:00.759 set up your optimisation. 00:11:01.080 --> 00:11:03.080 And you'll just need to edit this. 00:11:03.800 --> 00:11:08.167 So then we can then change the best guess to a valid value ... 00:11:08.167 --> 00:11:12.439 and hit play and then it'll start and submit properly. 00:11:12.680 --> 00:11:16.598 But what I'll do is I'll stop that one and I'll skip to the finished ... 00:11:16.598 --> 00:11:18.919 version, which I've already run here. 00:11:19.880 --> 00:11:24.044 So in this finished version, it's now found an optimum ... 00:11:24.044 --> 00:11:27.190 silicon Thickness in this case of 270 µm, 00:11:27.190 --> 00:11:29.873 which is nicely within our bounds, 00:11:29.873 --> 00:11:32.834 and for the perovskite layer of 460 nm, 00:11:32.834 --> 00:11:35.517 which is also within these bounds. 00:11:35.600 --> 00:11:39.316 So now we can be a lot happier with the result and we ... 00:11:39.316 --> 00:11:42.035 can go to the outputs tab 'Module JV' 00:11:42.035 --> 00:11:45.660 in this case and look at the maximum power for the ... 00:11:45.660 --> 00:11:50.010 module and see 703.4 watts, which is higher than the value ... 00:11:50.010 --> 00:11:52.638 we saw in the previous simulation. 00:11:52.880 --> 00:11:56.500 So we know that it has improved and we can be ... 00:11:56.500 --> 00:11:58.879 happy with this new result. 00:11:59.680 --> 00:12:04.870 But if we want to look closer, we can then download the ... 00:12:04.870 --> 00:12:08.645 optimisation results as a CSV (file), 00:12:08.645 --> 00:12:13.599 pull them up and then in this CSV file we now have the ... 00:12:13.599 --> 00:12:17.374 columns with the 'Generation number' 00:12:17.374 --> 00:12:21.738 and we can see that it took five generations, 00:12:21.738 --> 00:12:26.220 the 'Run number' taking something around 100 ... 00:12:26.220 --> 00:12:29.877 runs and then the number of rays used. 00:12:30.680 --> 00:12:34.404 So this is a cumulative column and sometimes you'll see a ... 00:12:34.404 --> 00:12:36.543 '-1' as the run number and you'll ... 00:12:36.543 --> 00:12:38.999 notice that the rays don't increase. 00:12:39.640 --> 00:12:44.188 This is because basically a repeat combination of ... 00:12:44.188 --> 00:12:48.519 parameters has appeared in the new generation. 00:12:48.840 --> 00:12:53.354 And rather than rerunning this same parameter combination ... 00:12:53.354 --> 00:12:56.875 that's appeared, we just use the old values and ... 00:12:56.875 --> 00:13:01.118 still pass on those genes within the genetic algorithm. 00:13:01.200 --> 00:13:04.240 But we don't use rays on running this same ... 00:13:04.240 --> 00:13:05.760 combination again. 00:13:06.320 --> 00:13:10.355 So that will possibly confuse you if you haven't watched ... 00:13:10.355 --> 00:13:12.759 this video and you're doing this. 00:13:12.760 --> 00:13:16.411 But hopefully now you'll understand what those -1's ... 00:13:16.411 --> 00:13:18.319 mean and why they're there. 00:13:18.800 --> 00:13:22.049 There's also some notes here telling you which generation ... 00:13:22.049 --> 00:13:25.099 that they were similar to in this notes column at the end ... 00:13:25.099 --> 00:13:26.159 of this CSV (file). 00:13:27.840 --> 00:13:32.195 The most important columns for us though are the 'Output ... 00:13:32.195 --> 00:13:34.697 value', which in this case is the ... 00:13:34.697 --> 00:13:37.662 module power and the 'Output conf95', 00:13:37.662 --> 00:13:41.832 which is telling you the confidence that SunSolve has ... 00:13:41.832 --> 00:13:46.558 in this value based on the ray tracing the number of rays used. 00:13:46.800 --> 00:13:49.840 And so it gives you an idea of the uncertainty in this value. 00:13:52.800 --> 00:13:57.248 So going back to the SunSolve tab and we can see ... 00:13:57.248 --> 00:14:03.920 that in this case we've got a happy result compared to the CSV (file). 00:14:04.400 --> 00:14:09.468 We can see that the value that we got of 703 is looking ... 00:14:09.468 --> 00:14:13.239 good compared to the rest of the range. 00:14:13.800 --> 00:14:17.426 And we can tell that we've basically found a good ... 00:14:17.426 --> 00:14:21.596 optimum within all of the different combinations that ... 00:14:21.596 --> 00:14:25.313 we've checked here and achieving the goal of this ... 00:14:25.313 --> 00:14:27.398 optimisation in this case. 00:14:29.880 --> 00:14:31.552 So that is, in summary, 00:14:31.552 --> 00:14:33.400 how to use the optimiser. 00:14:33.800 --> 00:14:38.518 In future we can do some deeper dive videos subject to ... 00:14:38.518 --> 00:14:41.199 anyone's further questions. 00:14:41.200 --> 00:14:45.053 So if you have anything else that you'd like us to cover for ... 00:14:45.053 --> 00:14:47.862 a future video relating to the Optimiser, 00:14:47.862 --> 00:14:50.993 please reach out and we can see how we can go with ... 00:14:50.993 --> 00:14:52.518 creating such a video.