Recording and presentations: Auto-Bidding: Optimizing Battery performance through energy management systems
17 June 2020
Automated transcription (it may contain errors)
Belén Gallego 1:02
Good morning, ladies and gentlemen. perhapses. Good evening. I just wanted to let you know that we’re going to begin in one minute, okay, but you’re in the right place. We’re talking today about optimizing battery performance through energy management systems. And I would like to invite you all in this one minute that we have until we start to use the chat on the right hand side, and to introduce yourselves your company where you’re joining from. My name is William Godwin, I’m doing in from Madrid today and from the other Insights Team. And with me, we have people also from the Silla team, and they’re gonna introduce themselves in a minute, but please, just literally 30 seconds actually, please just take to the chat and introduce yourselves tag all attendees, please and we’ll begin very shortly
Well, good morning and definitely good evening, wherever you are, because it seems that we have people from everywhere today as well, obviously a lot from Australia but also from Chile, from Mexico and from the US. We will hear about this is crazy tense for you guys, but welcome. Anyhow, so I’d like to ask my two experts to please introduce themselves. Luke, please, if you could open your microphone and introduce yourself.
Unknown Speaker 2:39
Yeah, thank you, Berlin. Good afternoon. Good morning and good evening, everyone. My name is Luke Shama. I’m a business development manager for energy storage about Silla based based in Sydney and been working with Foxy laugh for almost nine years. Now in in different position from from proposal and business development for gas speaking plant, solar PV and an energy storage for the last three years. And I’m very happy to be to be part of this webinar and very eager to keep up some good discussions with with all the attendees.
Belén Gallego 3:23
Thank you very much like okay like to ask Luke as well to introduce yourself.
Unknown Speaker 3:27
Yeah. Hi, my name is Luke Whitmer. I’m General Manager of data science in the energy storage and optimization business unit in Ward Scylla. I’ve been with the company now for five years and we’ve been really busy for a couple years building some neat technologies. And I really look forward to sharing some of that with you guys today.
Belén Gallego 3:56
Thank you very much look, and I can see from the chat I mean, there are people from everywhere. This A lot of Australians by all means, but I mean I’m just gobsmacked by how many Americans Colombians, Mexicans, and then these are Singaporeans Indians. So welcome everyone. We’re going to learn here today about what software systems like your first one. So can I ask you to prepare your screen and mute yourself. And in the meanwhile, I will just was like, prepares. I just like to remind you all how this session is gonna work. You probably been in other sessions that will hold. However, I’ll repeat it again for those who are new and Welcome, everyone. We’re going to have two presentations, one from Luke and one from Luke and we will take questions after they’re done. Okay. For the questions you can use the q&a box that is at the bottom of your screen, the one in the in the bar at the bottom. Don’t use the chat for that the chat is just for us to have chatting but in the in the question and answer in the q&a box is really easy for us to follow the questions so please send them through that and we will get to them and also we have recorded This it will be available in a view in a few days and we will also send along the materials so you can watch them in detail and you will have also the contact details for both we can look so look you can go for it we can see your screen perfectly but yeah here you go
Unknown Speaker 5:20
on good thing Thank you Berlin for the introduction and and welcome everyone again to this to this webinar. Um, introductions already done so let’s dive into into the topic. First of all, I’m going to introduce vasila overall quickly see who we are what we do, which which references we have before we take a deeper look at the ideal Australian market, why it’s very specific and what can we bring here as solutions. And then look, we’ll go a bit more specifically into the optimization through the through our gems, which is wer e m s and we’ll end that with a with a short demo of our m s. So vasila who we are about Eliza is a Finnish company founded in 1834. So, now the next next milestone is not not less than 180 years, 90 years sorry anniversary and hopefully will be also there for the next 190 years. The other intake is, is roughly about 5 billion euro good year BAD YEAR split into two divisions, the first one being the marine industry, so providing solutions and equipment for large vessels, LNG carriers, container ships
Unknown Speaker 6:58
Unknown Speaker 7:00
The other part of the business the other half is about energy solutions. So, which is the focus of today’s today’s webinar
Unknown Speaker 7:13
and for energy solutions,
Unknown Speaker 7:18
we, when we do have a vision which we share across the whole bartylla organization, about the future with hundred percent renewable energies and everything we do all the solutions we are providing here as well in the energy business in the energy space is around enabling these transition to hundred percent renewable energy as fast as possible. So, the first part of our offering is is around first starting power plant based on reciprocating engines. Whether it’s fueled by by by gas or Liquid fueled by biofuel and even we released one pilot with ammonia recently. And the second part of it is actually the what’s really the main interest for today’s webinar is the energy storage and optimization capabilities. So, we are battery storage integrator and also since bartylla acquired the company Greensmith US based company Grizzlies three three years ago also ms provider So, energy management system which is delivered with with our battery storage systems and for all of these assets whether it’s a flexible power plant based on reciprocating engines or an energy storage plant, we we do offer also Lifecycle Services meaning performance Maintenance management over the lifetime of the of the power plant. So, vasila we we are an energy storage integrator meaning we we purchase battery modules we purchase inverters and integrate them we made the design work and wrap up with the guarantees. So, what you can see here is our very standout containerized solution which is then fitted with battery modules inside and inverters on the side. One thing which is which is very important for us is that we are we are technology agnostic So, in terms of battery modules and inverter branding, we are pretty flexible or very flexible even enable to use any type of of modules Whether it’s we’re talking about different chemistries in lithium ion, we’re talking about different inverter brands. I think that’s very, that’s very important in context of the of the Australian market, reasonably is that we are all familiar with the quite lengthy GPS process, which is often the the longest, let’s say lead time for for any energy storage and renewable development. And sometimes, asset owners need also to to decide on an inverter platform in order to kick off those those GPS, GPS and the GPS preparation and submission. So if if you have some auto frame agreements with some of the asset owners and project developers have some frame agreements with inverter manufacturers directly We are very happy to accommodate whatever hardware here and make it work with with our solution. So, then we let’s see where where we have deployed actually those those systems. And so, this is the reference list for energy storage only not not not the global vasila database are installed based, which is more than 70 gigawatts worldwide. And you see here we have referenced or contracted deployed or power plants, energy storage, power plant in operation all over the world with a slightly stronger focus on on the US when it comes to the amount of site.
Unknown Speaker 11:48
The latest reference here is, which we have announced in the UK, also very, very dynamic market with two times 50 megawatts. 440 volt power and we are covering with those references, very large usage very large applications. Well as large as energy storage can procure whether it’s greed deferral, spinning reserve replacement, might very complex microgreen management or frequency regulation. And what’s interesting here is also the references in in the markets which which share similarities with the Australian market. So among others, air quotes or PGM in the US.
Unknown Speaker 12:42
These markets are also
Unknown Speaker 12:45
there are some similarities, some some differences, of course. And when it comes to the topic today, the webinar is optimizing energy storage and, and it’s optimizing also For the merchant markets here, what is similar between casual the UK grid and the grid here? The name is that the beads are cleared on five minute intervals, that there are several rounds of beatings ahead of dispatch, and that there are some there’s market for one energy market for procurement of energy and one or one or several frequency regulations market which which are cleared in terms of with with our not not energy and also of course, they are lots of lots of difference here the type of settlement the amount of of FPS markets, where we have here age in India, Australia, which is which is quite a lot the the type of the noughties incentives market cap or spot pricing cap. So, that makes it very very interesting to compare those markets and to share a previous experience we had in order similar merchant markets with what we can have here in Australia. So as I said energy market or a mood emo market is about one energy on one side and it’s gas on on the other side. And so, if gas again split into more regulation and contingency where contingency is there to be procured to cover larger frequency frequency deviations. What we can see here and you have a view on we have aggregated the average energy price for each Each market here whether it’s energy regulation or contingency, from 2017 to 2019. And what you can see here is is very specific to the Australian market. That’s one of the differences. It’s highly dynamic. It’s very dynamic, it’s moving a lot. So, last year’s we we saw that most of the, when someone having a battery storage asset connected into in the name would have earned most of the revenue through the reg raise market. And there was a quite an increase compared to 2019 while contingency raise was a steep steep decrease compared to two years ago. And that would be actually interesting to to hear in the audience. If you guys have some, some battery storage connected if you also share this the same the same view here. So, it’s a highly dynamic market and that’s exactly the reason why why you need some optimization of it, where basically the optimization can also adjust dynamically. Without that you have to manually you know, set up set up your your ATMs and your bidding system. What is also very specific here is, is that there are lots of rule changes in in Australia at the moment or actually rules which were supposed to change like the five minutes settlement and which have been postponed, but also changes into the primary frequency response. So the mandatory part of it, which will which have been released, like two weeks ago, couple of weeks ago for the final rule, and which will then kick in
Unknown Speaker 17:00
As such it it will not impact the operation of the battery, but it will impact a lot or it may impact a lot how the other assets are behaving and and thus it will also It may also impact how emo how much a mo is procuring of F gas REG and if gas contingency. So, there are lots of moving pieces here, and the only advice we can have when when it’s so, to ensure that basically whether you are an asset owner finance or equity investor or OR operator that you have any MS and a bidding system which is flexible enough to cover for all those regulation changes. And also which is not like a black box where you don’t see what’s inside and you have to live with it because if the market is changing you also Be able to understand how to how to change what’s how your your plant is, is beating. And that’s one one of the solution or the solution we have is called gems. So it’s the stands for green sneeze energy management system. It’s a it’s a like software suite with different layers has been which has been deployed in in all of our energy storage power plants which have been constructed so far.
Unknown Speaker 18:36
And he has different layers from
Unknown Speaker 18:40
the control of the of the battery system themselves. So communication with the BMS, ensuring that you have the correct warranty conditions that you maintain you don’t go out of the of the warranty frame. But also if you have a hybrid system, let’s say when prosper storage on our solar plus storage, also a layer which enable enable us to optimize the full hybrid system. And, and the dispatch of every every single asset within within this hybrid system. And then the third layer is actually where the market interface the beading part, which is actually one of the one of the most critical part reason is you can have a nice battery storage assets with you know, best technology etc. If If you don’t know how to beat or if you beat it from, you’re gonna just exhaust the battery, deplete them and and basically not getting the full the full value for eight meaning while in Australia on the merchant market meaning full revenue for it. So it’s the EMF part and building party, certainly here One of the even most important than vendor hardware, and then the hardware itself, because it’s a smaller part of the project, but actually driving the larger value of over over the lifetime of the asset. And yeah, and Luke will drive to drive you through the details of these bidding system.
Unknown Speaker 20:23
Unknown Speaker 20:27
Let me get my screen shared here.
Belén Gallego 20:35
And once look gets the screen up. Just a reminder q&a box at the bottom send your questions and another reminder, we are recording we will send you the materials. You ready look.
Unknown Speaker 20:47
So do you see my screen?
Unknown Speaker 20:48
Yes, we do access to options. And then
Unknown Speaker 20:52
perfect. Yeah, this is just the next slide from where as Louie ck was going. So Luke, thank you so much for that. That kickoff, I’m gonna dive right into some of the really just a candid conversation about auto bidding in general how these types of computing technologies can be applied. And we know that there are options in this space. There’s traditional legacy trading houses that are engaging in this space. And there’s other intelligent, automated bidding solutions out there on the market as well. This is an area where, regardless of which technology you pick, we wanted to try to talk about our view on the right way to do things just as an educational approach to Helping each of our potential customers and just the general public who’s who’s
Unknown Speaker 22:07
aware of these types of systems
Unknown Speaker 22:12
has has a broader understanding of what it is that is going on in these types of platforms. And so with that,
Unknown Speaker 22:23
there’s a lot of analogies that people
Unknown Speaker 22:26
Unknown Speaker 22:29
If we go as far back to, you know, when the stock exchange first started getting traded by computers, there are some analogies there. It’s a little bit different because of all of the regulation and market structure associated with energy markets, where pure speed was a really important piece of the financial trading industry here where we are constrained to submitting bids every five minutes. fractions of seconds don’t really matter. But the timing does still play a part because as you optimize, the longer your algorithm can spend optimizing, and the more data you can ingest, that takes actually computationally longer, the smarter your algorithm can be. And so the rules associated with the timing of bid placement and gate closure and how you have to optimize as quickly as possible all have to be comprehended inside such a platform. And so, a little bit of background on where the gems platform has come from. These screenshots on the right are really associated with our island grid, full grid control, economic dispatch platform where we are controlling multiple hybrid assets. Providing a forecasted view of the plan of where the battery state of charge is going to go of where the Which engines are going to turn on and off at what times based on the load profile forecasted and those renewable forecasts. And in an auto bidding scenario, these types of configurations with multiple different power plants, whether it’s hybridized, renewables with batteries, or even larger portfolios that include a retail load curve, or a, some thermal assets as well. All of that can be optimized together. And so all of that is is the beat is plugged into a single algorithm that can determine what’s the best way to dispatch my plants. And so one key difference here when we’re doing our island control, we’re more like the now we’re more like the the system operator where we’re focused on levelized cost of energy. Whereas the objective function the problem that you’re solving when you’re auto bidding is maximum Summarizing your profits with your portfolio, and it’s the market effect that actually makes the overall market cost less for consumers, as each power plant owner is trying to make the most money possible, but it’s that competition that drives those costs down. So, things like incorporating the price sensitivity, the pre dispatch forecast sensitivity into an algorithm is really important, because if you can have an algorithm that is able to anticipate how much price slippage a certain bid or a certain amount of cleared power will actually have on the grid, then your algorithm can be able to understand whether you should be constraining yourself actually in your bids so that you don’t undercut the whole market too much, but you actually are able to sustain something that is still profitable for you. As an owner, but it does still actually reduce the price to all the consumers as well. And so
Unknown Speaker 26:08
most of what
Unknown Speaker 26:11
these types of economic dispatch
Unknown Speaker 26:15
problems are are linear quadratic programming, objective functions and so for automating when there’s a lot of good reliable forecast information. This is an approach that works. And this is something that we offer as part of our our gems platform is advanced algorithms, including things that are directly tied to very traditional linear programming module. So here, this is a formal optimization. It requires a very detailed model. So you have to actually capture all of the different constraints and costs and create This solution space where the problem is always looking for a minimum. And you have to leverage solver techniques that don’t get stuck in local minima, but are actually able to find the global minima within the appropriate amount of time. This type of approach enables more control and human interactions. There are a lot more knobs and intuitive ways for people to study the inputs, adjust the inputs and have an impact on what the auto bidding solutions are in real time requiring less manual bid intervention. Because you can actually manually adjust some of the inputs separately while the auto bidding process continues. And it can provide a very detailed forecasted view. A second approach is a more traditional machine learning approach, very common for this type of problem. Problem of decision making, do I place a bid at this time of day and at what power level should I place it? So reinforcement learning is a, an algorithm that is trained on historic data. So you take multiple years of historic data, you can even adjust those, those training data to include expected future prices, if you want it. And so that approach goes through a process called exploration where the algorithm explores all of the solution space and says, Well, if I place this kind of bid, when the forecasted prices are x, y, and z, and when my my system state my battery state of charge is is within this range, what’s the outcome over the next couple of hours and it repeatedly runs simulations of millions of these combinations? And explores the whole solution space and determines which actions are the most lucrative so there’s a reward anytime the right type of actions are taken. In the case of this type of wholesale market bidding, it’s the revenue that you earn during that time would be the reward in that problem. And then once there’s a specific clear path, then the exploitation phase can take place where the model actually runs and exploits that that approach that statistical best choice in each of these different situations. So that as the battery state of charge is changing, and as the market price forecasts are changing, the algorithm dynamically can pick which types of bids it should place in which markets and at what levels to be able to exploit And actually make money to ensure that the the battery state of charts gets positioned at a high state of charge before the market has high price spikes where discharging would be required. So this type of approach is a little bit more like a black box.
Unknown Speaker 30:18
There’s there’s not
Unknown Speaker 30:21
there’s not always a clear forecasted view on exactly what’s going to happen because this is more statistical in nature, it’s based on
Unknown Speaker 30:31
Unknown Speaker 30:33
experience of what was a good thing to happen. And so,
Unknown Speaker 30:39
this this type of algorithm
Unknown Speaker 30:42
is something that that can be constantly learning constantly being retrained with different processes to ensure that as those market dynamics change, that that solution space is always accounted for. So these two different options we as a company, have been researching exploring different methods in these areas. Now for several years, as we’ve been preparing for the different opportunities for batteries, it’s been very clear that that a battery is a more expensive asset in a portfolio is central to the dispatch of multiple assets within a portfolio. And so that’s really what has led us to the point where these types of automated modules are available within the platform to be configured for your specific situation, whatever that might look like.
Unknown Speaker 31:46
So this is this is a chart that is really useful just to describe what some of the control a human trader person, whether it’s it’s someone at an actual trading desk 24 by seven managing Multiple assets. Or if you’re a smaller IPP, maybe you just have one person who’s your trading, market focused person, and they’re not available 24 seven. But during the day, they monitor the site review simulations and would be able to change settings and let the site just run
Unknown Speaker 32:20
in auto mode, most of the time, that here I’m
Unknown Speaker 32:28
just looking at these two charts, let’s study the top one first. So blue is the battery state of charge. And Orange is the energy spot price. And so here, a very common knob that we talk about in these auto bidding scenarios is the energy throughput cost. So this is a parameter. There’s lots of other parameters within the model depends on all the different configuration for each site. But if it’s more than expensive to use the battery than the automated logic won’t use the battery, it will try to wait for higher price deltas, it will leverage other assets it might have in its portfolio to accomplish a certain objective or requirement. So here looking at the blue, comparing them between the two, the top chart has a lower throughput cost and the bottom one has a higher one of $30. And so you can see that just for example, in this section, the battery state of charge has a lot more activity in the top chart than here, there’s a lot of little arbitrage during these, this volatility here that’s happening because the battery can exploit more more activity here than it does down here. And so you can see that because the the price of using the battery is higher, the battery simply isn’t used as much and in the context of battery warranties. Extending battery life.
Unknown Speaker 34:04
In real time,
Unknown Speaker 34:07
a trader or any type of algorithm that is configured to make the most money will drive the battery into the ground very quickly. If you don’t have some type of statistical view on how much throughput is your allowance, are you trying to use x number of megawatt hours of throughput this month or this year or this week, the longer the time range, the better lets you have more flexibility. There’s an opportunity cost factor here. So you want to use your battery while the market prices are good. And so if there is opportunity, you should capture that as quickly as possible. As more batteries come onto the grid, then some of these this price volatility will be stabilized and there’s an opposing force that has more renewables come online than the price Ability gets worse. And so these the growth of renewables is driving this growth in battery systems that are dynamically participating in these markets for this very reason. And having control over what level of sensitivity you want your system to react to the grid width is really important as an asset owner to ensure that you’re able to maximize your returns not only today, but through the life of the asset.
Unknown Speaker 35:33
Another really important piece that is
Unknown Speaker 35:39
often talked about among
Unknown Speaker 35:43
our different engagements, and even just various articles, studying the existing batteries with the publicly available information that’s out there is related to forecast accuracy and this is not just something in Australia but In all of these types of battery optimized CES, battery centric optimized systems, the accuracy of your forecasts is really central to achieving good results. So here are one of the main forecast inputs in any type of Australia automated bidding scenario is the pre dispatch forecasts that come from IMO. So, as those data are coming in,
Unknown Speaker 36:35
it’s rare that a system
Unknown Speaker 36:38
would be relying solely on that information. There’s a lot of forecast inaccuracy in those raw data’s, but at the same time, those data’s are really the most.
Unknown Speaker 36:52
Most knowledge packed
Unknown Speaker 36:55
are anybody’s forecasting system is not going to Nestle certainly no more information than what the dispatch operator knows about different plants having maintenance schedules about potential outages in other places of the grid, the congestion that’s happening because of the whole grid network and all the interconnections. And how power can flow from point A to point B, with all of the renewable forecasts across the whole region. And so those pre dispatch forecasts contain the most information but at the same time, people are dynamically reacting to these forecasts all the time. And so here’s just an example where blue is the pre dispatch forecast and orange is what actually happened. And this is very common to see in the data in Australia, where in the pre dispatch forecast, there’s some type of big price spike that actually it’s showing up here as something like a two hour spike from our 15 to 17. This blue is saying it’s going to be a really high priced couple of hours. But in reality, yes, it did actually spike up that high, but not the whole time. There’s a lot of intermittency throughout there. And even even having this type of high orange spike sustained as long as this one did is really rare. That’s not very typical, it’s more common to have a small spike or something like that. But the the key piece of the puzzle here is having algorithms and a system that is trained in a way to be aware that these previous batch forecasts have error to track that error to provide visibility to that error to the users and to be positioned to to be flexible and not have such dependence on not not expect these to actually happen fully. So Just an architectural diagram. And this is a generic one that fits all of the different system operator interactions that we have engaged with globally. So some of the US is owes particularly, but also it’s the same in in Australia, where within the system operator, within IMO, there’s multiple systems. It’s not it’s easy to think of them as one entity. They’re the system operator dispatch entity. But in reality, there’s multiple systems that have to be integrated with and so having a one stop shop where these orange lines that connect our software platform are already established off the shelf is really important to enable Low Risk project and so here the power plant controller communicates with the battery technology or the renewable plants, whatever those types of interfaces are. We have standard API’s for communicating with those devices. It’s very common for our customers to already have existing renewable assets or justing Thermal assets. And so there’s some select data that we need to be connected with on each of those power plants to ensure that that the system has the data it needs to operate intelligently. There’s our connection between hips. Back, there’s our connection between the power plant controller and the cloud. And then our standard API for third party access and inputs. Meanwhile, with the system operator, our cloud is where we’re retrieving All the data that’s available from their price data and pre dispatch forecasts. It’s where we connect to actually would place bids. And in certain cases, there’s a third party between us and the operator the through bid placement entities that place the bid directly for us, specifically, if it’s common for our customers to already be using some type of trading partner for settlement services and things like that. And that partner may have a technology, a platform already in place for bid placement. And so sometimes there’s a connection through a third party on that side for bid placement and there’s cases where we’ve connected directly for bid placement as well. And then after we get cleared, those bids get consumed and
Unknown Speaker 41:55
cleared by the system operator than any dispatch instructions and for entertainment. markets or ancillary service markets get get sent to the powerplant controller directly by the system operator where we would clear any types of F cast contingency types of markets, we’re clearly detecting our own frequency deviations. And doing frequency response dynamically ourselves according to that cleared power in those markets.
Unknown Speaker 42:27
while providing feedback, in terms of
Unknown Speaker 42:31
reporting back to the system operator, what’s our power limit of availability and things like that?
Unknown Speaker 42:43
A really important note is
Unknown Speaker 42:46
on the fleet director side by offloading all of the site data to the cloud, providing visibility
Unknown Speaker 42:55
there, there’s a lot of good
Unknown Speaker 42:59
secure You don’t want analysts and market traders logging into your PowerPoint controller. That’s not that’s not appropriate in modern cybersecurity practices. And so that’s, that’s a really important piece. So being cognizant of time, I’m going to switch to the demo for a few minutes here, and then we’ll have at least 10 minutes for questions.
Unknown Speaker 43:27
Unknown Speaker 43:33
I’m gonna log into a demo of a 10 megawatt battery in Australia with the auto bill bidding logic running and we’ll be able to poke around and see the clearing
Unknown Speaker 43:44
of this simulated environment
Unknown Speaker 43:49
and get an idea for our platform. This is end to end encryption with two factor authentication so that you can get in safely so got to come out with My phone and click that guy.
Unknown Speaker 44:03
Belén Gallego 44:06
Whatever you want is getting really happy now, you know,
Unknown Speaker 44:09
stuck with me this long. I wasn’t following the questions. I hope people aren’t yelling at me start the demo. I’m so just clicking right in here. If you had multiple sites, you do have a nice map view and you can have different virtual power plant operations and logics running between them
Belén Gallego 44:31
yesterday, you know, we’re only seeing a screen that says demo. Okay, so you need only sharing. Yeah, sharing because you’re sharing the document, you’re doing very well and share your Chrome. We didn’t want to see the buzzword Anyway, you know. There we go. Now
Unknown Speaker 44:49
we can see it like I should log back out those you can see that
Unknown Speaker 44:54
the login screens nice, I’ll show it at the end if people care. Cool. Um, Right, so you get a nice map view. If you had multiple sites, you can see them all in one place get alerting from all of them. There’s just some some random examples here on the left, we’re mostly focused on on this site today. If you have virtual power plant logic, you want to abstract things across multiple sites that can also be managed here. But just a diving into one of these sites. This is a very standard landing dashboard layout for us. We have a highly customizable interface where any data within our system you can plot it doesn’t look like the battery’s doing anything right now, which is unfortunate. It was 20 minutes ago.
Unknown Speaker 45:48
Unknown Speaker 45:50
So here, we’re just showing effectively what’s the site doing this is just a battery, no renewables co located We’re tracking all the different prices and each of the markets. The battery state of charge is here. There’s some summary statistics on across the top in terms of overall what’s the power plant capable of doing right now. So it’s a 10 megawatt plant. If one of the the units gets disconnected or is down for maintenance, then this number would update. The automated system passes that information along to the auditor so that it would immediately adjust those bids due to a site on availability. As we go down at the bottom of this main page, there’s a bunch of cards that represent different groups of devices. So just diving into this one briefly. We take object oriented programming very seriously the relationship between these devices is abstracted for that reason, so you can see We’ve sort of navigated down through the system. So if you wanted to have a specific view on what is available within the gems platform, all this data gets streamed off of the site so that you can have very granular view remotely. And if we just sort of look at the last five minutes, I mean, this is this is updating in real time. All the data that’s that’s available from the different devices is visible here.
Unknown Speaker 47:29
And so some simple things.
Unknown Speaker 47:34
One of the one of the, we only have a few high level parameters visible today, in this public demo we are, there’s a lot of proprietary work that we can’t share in this public setting. And so we’re only sharing some very select things here. But simple stuff, just to wrap your head around some of the high level things that you You can do, just as an example the in the slides, I talked about the battery throughput cost. So if we have been running with a certain battery throughput cost, and you’ve run some simulations with our gems analyzer platform, which is a separate interface, and you’ve decided on some new set of parameters that you want to apply, then you can come in here and you can actually apply those and say, Okay, I actually want to run with a higher throughput cost of my battery throughput has been going up, we made a lot of money, but I’m on track to actually use more of my battery throughput than I want to use at this time of the year, or whatever. And so you can change that. And you can say, well, the other thing I noticed in my simulations was that the energy market is causing me to have a lot of throughput. And I’m not actually making as much money in that market as some of the other ones. So I want to limit my participation level. I want to really have the algorithm focus on keeping power in the reg market and the F Cass market. And so you can just come in here and say, you know, I only want I only want two megawatts to ever be bid in the energy market for charging. And I’ll keep my my discharge here in case there’s a big spike. So you can have a symmetric participation in the different markets. But we’re going to force the algorithm to try to charge using the reg market more than anything else. So if the SOC is getting low, it’s going to have to do that in a way that actually makes some money. You can save those changes. And when the algorithm gets to the next point in time where it’s it’s able to update those bids, then then those parameters will take effect. And so, last couple minutes before we go to q&a, I’ll just share in this report section,
Unknown Speaker 49:53
like last 24 hours.
Unknown Speaker 49:57
We’re pulling in all the different This one is To set up for New South Wales, actually, so we’re pulling in all of the pre dispatch, forecasts, price information, live and tracking all the different nine markets here. You can see there was a price spike up to $300. Earlier today, all right, guess what? Today? I guess that’s yesterday. Yeah, because it’s noon in Australia. So this was yesterday evening. And the interesting thing is, is when we track different error metrics, root mean squared error and the bias on these different forecasts, we can actually see that
Unknown Speaker 50:44
an hour before
Unknown Speaker 50:47
Unknown Speaker 50:49
spike actually happened. There was some error in that forecast. So it was $136 off it. It was not accurately forecasted an hour before it actually showed up. But 10 minutes before it showed up, it was it was only $65 of error. And so you can actually have visibility into how good the raw forecasts are coming in. And our algorithm has access to all these different error metrics as well, so that we’re tracking that performance as an input also to the model to ensure that our bids that are being generated, have that learned knowledge within them as well? So
Unknown Speaker 51:43
I don’t think
Unknown Speaker 51:46
there’s there’s lots more we can show in the demo. But again, in the in the context of this public setting, we want to just keep it a little bit short and as time is wrapping up, I think we’re open for questions now.
Belén Gallego 52:01
Thank you very much look very nifty tool that you have going there. I assume that it learns more. So it’s kind of like machine learning style, or are you guys changing algorithms at the back?
Unknown Speaker 52:12
or How is it? Yeah, great question. So there are aspects of it that that can be set up for self learning. A lot of this stuff is pretty new. So while we’ve been setting up the back end to support that most of our stuff in production today is pretty hands on just because of the nature of these markets and how we don’t want these million dollar assets doing too much on their own yet we were keeping pretty close track of them. An important thing to keep in mind as the world does move more towards more autonomous retraining and self learning as time goes on. It is really just tracking performance and having boundaries and alerting and alarm setup which our platform Form supports fully out of the box in terms of being able to track different performance metrics and getting text alerts and email alerts however you configure it to alert you in different thresholds get crossed so that you can say like hmm i didn’t i we used to be performing better than that I’d better go look at it that way things don’t go on for too long without people paying attention
Belén Gallego 53:24
I can imagine like you can go crazy if you’re trying to understand you know the like innocent out of every single market that has battery storage in the world because they’re also different and you know, the stacking of values still not clear in different markets so like to come a lot of work for you guys agree field
Unknown Speaker 53:42
or we all read some questions to myself a week flagged some for answer. Yeah, we can go to town on that. Well, I think about what I want to answer next.
Belén Gallego 53:51
Okay, excellent. So, like, you you’ve answered a lot, actually, but one of the ones that you have live is when did Plants owned by different operators in the same network all are operating on a gems platform and potentially fighting each other as they’re operating on the same building algorithms.
Unknown Speaker 54:11
Yeah, this one is a good one. First of all, we’re very eager to reach this point where several assets are beating with the gyms and and it is true that these assets would actually, if owned by by different owners would compete compete with each other as if they’re bidding let’s say, two separate storage assets being on the same merchant market and would have like the same algorithm, then what what would be different is of course, that would be an issue if, for example, vasila would have incentives on on revenue sharing, for example, or incentives to sum or to optimize more one asset versus the other. That’s a very good one. I think there there’s also to provide some confidence here. There’s also a kind of manual setup which is possible to do which was have an influence on on the bidding behavior as Luke explained about the throughput cost. So the systems would not be like hundred percent similar first. And I think it’s also the components we can provide is is by providing some transparency here on you know, how the plant or how the building system is operating. And maybe Luke, you you have some more and more comments on that which is, which is a very, very good and be tricky question.
Unknown Speaker 55:40
Yeah. I mean, um,
Unknown Speaker 55:45
our approach is that our algorithm is not ours. We are a we’re a, an optimization machine learning platform. Our technology is We’re pretty open to our customers once we get inside of an NDA. And there’s, you know, an active project, we’re working on something contracted. We work very closely with our customers to do what they want to do. And so, yes, in the sense of the fact that the markets are set up for competition, yes, there’s risk of batteries driving each other to the bottom. But in the end, each, each asset owner is going to have to determine their own thresholds of what they’re comfortable with, in terms of the operating profiles of their systems, and it’s your asset. So it’s not up to the automated logic to fight with each other. It’s up to you to think what’s what should my strategy be and what kind of things do I want to do? In our platform, it gets configured that way, and so he’s short we have recommendations, we have lots of simulations and we have a sort of a standard way of doing things. But in the end, the algorithm is going to be different enough on each project and with the different geographic reality of the electric grid. It’s never going to be to a point where we would get, you know, dispatched in such a way as to you know, really collapsed the whole market unless there was so many batteries on the grid to get there. And at that point, I think that the grid will be set up in a way to actually manage that better.
Belén Gallego 57:31
So I’m gonna have to change. Thank you very much. Okay, another one for Luke that you had marked. Just wanted to know if gems can be appropriately configured to take weather forecast data for renewable energy assets and energy market. They had real time live historical data to predict the demand and basically automatically suggest bid values to the project owners, which again,
Unknown Speaker 57:53
was Yeah, that was a pretty long question. So I thought better to take it live. And we do have a system Especially micro grids, where we integrate lots of forecasts into our our dispatching algorithm, whether it’s renewable forecast, so, day ahead or short term forecast for wind, for for solar, we can integrate as well, but also load forecast. So that also all these forecasts will be incorporated to the algorithm to better to have a better state of charge management. And so short short answer is yes. And I think it’s it’s also when we talk about hybrid power plants or renewable power storage. It’s one of the of the key of optimizing here the whole system not only each part individually, but the whole hybrid system and the dispatch of of the whole unit.
Belén Gallego 58:55
Perfect and I’m gonna do one more for you because you marked it and that’s Luke, you can take it on from If you want to answer a few, okay, can you use gems on one side other battery management systems, and what level of interface will be required with existing battery controllers power generation and customer loads?
Unknown Speaker 59:13
Yes, that’s a good one about the interface. And it’s relating to the kind of first layer of control of gems on how to control the battery modules themselves, the battery rack so on the top of its of each battery rack, we have a BMS, which is supplied by by the module manufacturer, so we don’t manufacture our own BMS M. And here we ensure that there is a proper communication between the our energy management system, the overall architecture, and the BMS on the top of each rack, and how the interface like the technical details on the interface. I leave that to Luke about if we’re talking about what type of you know signals or communication protocols?
Unknown Speaker 1:00:06
Sorry, I was reading questions. Was that a question to me?
Unknown Speaker 1:00:09
Yeah, more more about the communication about the EMF and the BMS, what type of communication protocols are because yeah. From from, you know from battery modules manufacturing. Yeah, I mean, most
Unknown Speaker 1:00:24
most the mssp canvas or mod bus, we have our own fully integrated turnkey battery systems, which is what this platform comes with. But so you if you have some type of separate power plant controller, you have to have a battery interface with that that’s, that’s pre integrated, right. So it’s quite separate from the market bidding side of things.
Belén Gallego 1:00:55
Go ahead with your questions. I know you have a few heard Yeah,
Unknown Speaker 1:00:58
so I was just reading a few here in general. Let me try to bucket a couple of them as we wrap up
Unknown Speaker 1:01:04
the question about renewable energy,
Unknown Speaker 1:01:08
coal, you know, being able to co optimize between that and storage storage if there’s like a solar plus storage PPA, how does that work? All of that really comes down to the commercial side of how you should Who’s your off taker? And what type of flexibility Do you have inside of that PPA to participate in open markets? Is your interconnection big enough to support? Actually, when you register with the grid? In Australia, your renewable plant is registered separately from your battery anyway, but you might have some type of transformer or interconnection constraint that you’ll have to manage. So yes gems, manages all of those complexities for a specific site
Unknown Speaker 1:01:57
and then some other questions here?
Unknown Speaker 1:02:07
specific to some of the stuff we’re showing, Adrian’s got some questions here, related to the throughput cost, having it as a function of something else like SOC, so that it’s dynamically getting it adjusted. It’s a very interesting idea. It turns some things into nonlinear. Non, there’s some nonlinearities. When you start doing that, and also you you would end up in some situations potentially where the algorithm would would not do the things you want it to do. I think it’s an interesting idea that I haven’t thought about too much. And there may be some reason to apply that throughput as a function of something other than SOC, directly, because once you get a low SOC, I mean, you’re going to want to charge back up. And it’s you’ve already done that discharge. Just to have it increasing as you go down, it’s a little bit too late to change those those bids. You really want to have accurate forecasts and a plan for the day that’s going to maximize your revenue because for the most part, you’re a price taker, you do have some impact on the prices, especially if you’re a big battery, or if you’re in you know, South Australia, then you’re a little bit more of a price setter. But in general, you’re you want to plan out what you’re going to do that day and stick to it. And so the the throughput cost is a bit more of a longer term thing than something you’d want to change throughout that algorithm. Question is the demo working on actual live data from from them? Yes. It’s not and then we are self clearing ourselves. So we are not, not the nem not not the the AMD dispatch engine No. So we’re not actually integrated to that dispatch engine, which is being asked so we’re simulating the clearing ourselves in this demo, but in the context of a specific project, then that type of simulation can can be performed.
Unknown Speaker 1:04:29
Any other questions stood out to you guys,
Unknown Speaker 1:04:32
maybe one or four energy storage can be used to be used for energy storage sizing optimization, not out operation, which is a very good one actually, because, well, it’s the same, the same background algorithm which is running not for the same purpose. And we do have a function if you when you when you’re having a second look at the slides in the gems presentation, there is one one moment Dual code analyzer is what we use also in order to design a system and especially a hybrid system. So let’s say if you have you want to have 50 megawatt firm capacity of renewable, what type of what size of solar? What size of storage? Would you be able to install? It’s it’s an analysis which which we do internally is kind of complex. It also takes time so we don’t generally do that widely open. But on a project by project basis, this is also something which we can have a look at.
Unknown Speaker 1:05:40
Yeah, I’ll just there’s another question kind of related here. Can your tool perform this is from do a no can your tool perform a forecast of IRR for a potential storage project so as Luke said we do have capability in house of simulating and helping size and maximize IRR and Specific combinations. It’s not something that, you know, if you buy the tool or buy a license to the tool that you can do yourself. It’s something that, you know, our data science team will support. Very few selected key projects that are really on track to closing, you know, we’re not in the business of helping you figure out your business case, really. But I think in terms of narrowing in on things, and finalizing system design, that’s really what we leverage that tool for.
Belén Gallego 1:06:32
Okay, thank you very much leaking. Look. Thank you very much for sharing with us today and show that mobile I think that’s what everyone likes to see. And thank you very much for our audience. Don’t forget, you can get in touch with both of them, you know, in the materials, that information is there and you will receive it by the end of the week. So thank you very much again, everyone. Thank you very much, Luke, and Luke, and thank you, everyone in the audience and see you next time.
Unknown Speaker 1:06:58
Yeah, thank you. See you And Tony’s day to contact us for all the answered question.
Belén Gallego 1:07:04
Okay, thanks. Thank you.
Transcribed by https://otter.ai