Advanced battery analytics with Shyam Srinivasan (CEO and Co-Founder of Zitara Technologies)

Hello, everybody, and welcome to the transmission podcast. It's me, Quentin. And this week, we've got Sian, CEO and cofounder of Zittara Technologies Inn. Now Sian comes from a company before this that I really admire.

It's Nest who make the thermostats that we all know and love and, got bought by Google. And now they're doing battery analytics software to figure out how to get the most out of assets and make more money with them. So really enjoyed this conversation. We talk a lot about hardware and software at the edge and what that really means and the impact that this kind of software can have on asset owners and ultimately returns for battery assets. So if you like this one please do hit subscribe and like and put a comment in the comments. It really does help us increase the reach and get to more listeners.

Let's jump in.

Shyam, thank you for joining us on the podcast.

Quentin, good to see you.

So, Zatara, should we talk about that to start with? What is Zatara?

Yeah. Absolutely.

Zatara, we've been around for about six years.

We're a thirty five person company based here in San Francisco, and we're focused on, real time edge monitoring and controls for BESS.

So we try to make your BES more profitable, safer, and last a lot longer.

And and how do you do that then? So what what are the thirty five people do? Are they software engineers, or are they, you know, other types of engineers?

Yeah.

Primarily, we have, two two categories of engineering at Satara, software engineers that build controls, so controls experts.

A lot of our team comes out of kind of the autonomous driving world. So we are really experts at edge controls and cloud edge controls.

But, you know, really, the other half of our r and d team is battery modelers and algorithms experts. So we get really deep into the chemistry of batteries, and that's what allows us to control them better at the edge.

And what's the what's the reason that I mean, we've had a couple of, battery system analyst companies, if you like, or analyst software companies on. And, they've done a really good job of answering this question, so I wanna ask you to you as well. What's the reason that a customer so you're a grid scale actually, I'll do this differently. Hold on. I'm gonna edit this.

And so we spend our our whole life on this podcast talking about grid scale batteries, which is, of course, a massively growing market and lots of different types and chemistries and and, blends of, batteries and battery systems out there.

So for for a owner of energy storage assets, what's the advantage for of of coming to Zatara versus going to the the original equipment manufacturer, say Samsung or Panasonic or CATL?

Yeah. Absolutely.

Well, you know, across the board today, we've seen kind of the dominance of one chemistry, and that's been the last few years of, LFP batteries being by far and away the cheapest option, for grid scale.

They are notoriously difficult to manage, you know, especially as the formats of these sort of, these cells and and the the choices start to get, larger and more consequential, you know, we see different kinds of effects make it harder to tell how much power or energy you have at a given time, even all the way down to the cell or rack level. So you could look at one of these sort of, battery voltage curves, and kinda compare LFP to previous technologies like NMC, or, you know, in the cell phone era, LCO. And you'd see kind of nice linear curves, for previous chemistries that told you, you know, for a particular state of charge, how much voltage you should expect.

And just because systems, you they measure voltage as a proxy for state of charge. So when your when your when your iPhone tells you it's got ninety percent or eighty five percent, the way it's doing that amongst other things is actually measuring voltage and then looking it up on a little chart.

Exactly. Yeah. If you wanna get to looking it up on a little chart, you often have to let that battery idle for a little while.

And so then even more complex is trying to suss out while you're using the battery, while you're doing that discharge, how much more voltage drop has happened because of the discharge. And so if you can back all of that out while the battery is discharging, then you can tell how much power and energy you have.

It turns out your OEM systems are often twenty to thirty percent off during a discharge.

Wow. Twenty to thirty percent. Is that twenty, thirty percent, error, I guess?

Yes. That's right. All the way down at the cell level. And you'll see corrections come out to the telematics that end up getting reported for your market operation or for, you know, for your grid op for your ISO.

So, you know, you'll see smoother signals at the site level. But if you drill down and look at, you know, individual inverters or racks within a bus, you can often see these twenty, thirty percent jumps, frequently.

And so measuring state of charge or the the percentage of energy in a battery to keep it simple, Why does this why does this even matter? Why is this so important?

Well, you know, it depends what you're trying to do with your battery that day. But almost no matter what, more hours of power translates to more revenue, on the on the whole over a long period of time. So, you know, our customers, you know, they may be trading, energy wholesale. So they'll have a a sort of day ahead activity, and then a a, you know, a real time activity on the next day.

And if you, you know, bid the most aggressively you possibly can day ahead, you're not gonna get necessarily awarded all of that energy that you've you've put up, for, for sale.

During the day, you really wanna know, especially kind of towards the end of the day when prices might be peaking, exactly how much you have left in the tank.

And the real time signals that you have to do that today, if they're inaccurate, you're either, you know, incurring sort of opportunity cost by not trading as aggressively as possible, or you're ending up, you know, with a base point deviation in a market like ERCOT, or, you know, other penalties for for not being able to serve what you've

the beginning of this conversation,

you talked about this problem getting worse. So the the change in chemistry over time from let's just do it most recently from NMC to LFP, which is now the the the dominant, battery cell that's being installed around the world for all the good reasons about supply chain, you know, input materials cost.

But if I read between the lines, you're saying that actually measuring state of charge from voltage of LFP cells is more challenging than an MC. Is that correct?

Yeah. That's right. Well well, with any battery, I think once they're old, cold, or under load, you know, you like we like to say that they're they're more difficult in those moments.

But with LFP batteries, we're seeing shocking problems just on day one where, you know, you might expect a few years before you have problems on previous chemistries.

And so how do you guys do this better than the original cell manufacturers? What's the secret sauce?

Yeah. Well, you know, it's it's not really even cell manufacturers that have spell specialized in this to date.

You know, largely, we think about cell manufacturing as this incredibly complex chemical manufacturing, kind of miracle. If you've ever been to a battery factory, it's a very, you know, clean room facility, more complicated than manufacturing silicon.

What a what a thing.

It turns out this is a wholly separate skill set than writing software.

Today, you may see a module maker in the middle of the value chain being tasked with, you know, provisioning a battery management system hardware and usually getting the software just kind of on onboard that BMS by default.

These are often some of the most profit squeezed players in in the battery value chain today. And, you know, they may be focused on just razor thin margins for assembling those cells into a module, but they're really not putting a lot of r and d into the into the equation.

Additionally, they don't really have any LTSA or, you know, contract or warranty associated with the performance of the algorithms on the BMS.

You know, usually an availability contract says my inverters or some percentage of the inverters in my system need to be up and functional, but nowhere does it need, mean that, you know, fifty percent SSC means I have fifty percent of my energy, usually not in the contract.

And when do you guys get involved? So say, I'm an asset owner. I'm gonna build a new well, I've got existing assets operational right now or I'm considering building a new one. When do I say, right. I've gotta call Sean and get his team involved.

Absolutely. Well, today, we're focused on retrofits. So, we'd love to talk to you if your, you know, max power, high sustainable limit readings from your, LFP battery are noticeably wrong. You're running into base point deviations.

You've noticed that the SoC is not accurate, and you care about your your trading clearing, you know, as expected.

We think about it kind of as a market clearing problem because when you submit that day ahead offer, you're including SoC pairs with your with your power and and and price.

And so if your SOC is ten percent off, that means that your bid isn't clearing correctly.

So if you run into those kinds of problems, we'd like to talk to you. We also love to engage more greenfield projects. So when you're at at about your thirty percent, completion mark, we do love to come in and audit your networking diagram. You know, this is just sort of a service that we like to provide. We think that it's very important if you're constructing and planning to own and operate that you have access to the right data and that you have access to the right control. And so thirty percent is right about the time when we can, you know, get ahead of any issues.

But then then we gotta wait, until the until the asset's operational, to really help.

But you have, going back to the first thing you said there, which is that if you if you see if you're an asset owner and you see these kind of errors or you're seeing deviations, then, then pick up the phone. The thing is being able to see those things in the first place. Right? Because you've got measurement challenges compounding on top of each of that.

You think about I'm sure you know this already, but, measuring you know, what what do you measure at the cell level, at the right level? What does the BMS see? And then you've got balance of plan and, transformers where and then you get to the the boundary meter, the settlement meter. There's a lot of losses and noise in between those things, and you get you get paid on the boundary meter.

Right? So, how can folks think about, you know, they own and operate assets, making sure that how how do they measure that? What should they be doing or thinking about to make sure they can find out whether the data's right in the first place? What's your what's your guide?

For sure. Well, you know, this is actually something that we we use the the MOTO, tool for, and we're gonna be getting, onboard with that really soon. We're really excited. We look for places where people are either, you know, generally underperforming the market.

But if we can get down to time series, we're looking for places not just where people had base point deviations, but where they appear to have purchased, you know, energy, at high prices unexpectedly. So particularly, you know, in markets where the customer might have committed to something after a peak event, that's a that's a really, really costly day. What we see happening is, you know, prices start to tick up during the day. The customer starts to discharge their battery, and then their state of charge or state of energy reading, probably goes down faster than anticipated.

The way we can tell is that maybe because they have an ancillary service or something committed in a future hour, they end up purchasing energy at this peak price.

So, you know, putting aside base point deviations, the fact that they ended up purchasing energy at, you know, in in Texas close to the peak sometimes, this can wipe out, you know, weeks or a month of of profits to to just have this one event.

What we tell customers to look for and an analysis that we'll do for free for you, we'll we'll do a complete analysis for for you in terms of the accuracy of your signals. So if you wanna understand your balancing state, your state of charge, your state of energy, your power signals, which ones are accurate or inaccurate, you can send us data and we'll figure it out. But a a key place to look is to say, my system was reporting that I had a hundred megawatt hours at this at this time, then my, you know, revenue grade meter told told me that I put out fifty megawatt hours over the next hour.

If my state of energy says less or more than fifty megawatt hours by quite a lot, that's where we start asking questions, if that makes sense. So we compare for that SOC, does that well reflect actually the percentage of energy in the asset when you compare it to what went out on the revenue grade meter, especially if you can correct for things like auxiliary loads that you mentioned.

Yeah. Everything comes back to that network diagram that you mentioned at the beginning. How about access to data then? Because, I I just think it's absolutely terrific that we have companies like you guys and there's some other European folks doing similar who are saying who are really challenging what is possible with the data versus the OEMs. I think it's, I think it's absolutely brilliant.

And, yeah, by being highly specialized, there is a there is an argument that you can really add a lot of value versus, the OEMs.

But do the OEMs play nice?

And don't you need access to data? Because historically, a lot of our customers who own assets, they're surprised by how, in some cases, some are really sophisticated. But many cases, they're surprised how little access to data about their own assets that they even have.

And you think that that that in this this day and age, if you spend fifty million dollars on a on an asset, then you've got a right to see under the hood, but that's not always the case. So, yeah, are the OEMs playing nice? What do you advise companies put in their procurement and and construction, contracts?

And how do we how do we create an environment where all this data is open for companies like you to come in and really make a a a big difference?

Yeah. Absolutely.

Well, you know, there's there's one notable vendor in the space, that's particularly closed with data, and that's Tesla. Probably not a secret.

We outside of, you know, those kinds of projects where, honestly, we we've steered our customers still looking at evaluating us on on non Tesla sites of which there are many.

You know, we've never had a problem with an OEM not wanting us to have access to particular cell level or battery data.

It it ends up working out to much more of an IT challenge.

If the site hasn't been configured to be get get getting access to the most granular cell level data, there may may be elements that just sort of average or down sample that information along the way out the, out to your historian.

And, you know, it's a matter of wrestling with your sort of integrator or, supplier chain to be able to get access to that data.

What we have seen, though, is that more modern systems and things that are getting constructed today, the data architecture is generally more modern, and and it's easier to get access to whatever data you'd like.

You know, when we look at assets that were constructed in twenty twenty one, twenty twenty two, again, it's not that anybody doesn't want us to have access to the data, but sometimes it's a pain in the ass.

Yeah. And then if you the last thing you wanna do is start demanding that, you know, there's a cost element too, and customers, well, asset owners don't wanna have on-site servers or really high bandwidth connections for cloud storage and all the additional cost there. It becomes, is it even worth it?

So this error can we come back to this error, this state of charge error?

Does it get worse over time?

Yes. Absolutely.

So, generally, what we're looking at is the state of charge of individual cells, within a rack, but there might be four hundred cells adding up to your DC bus voltage within that rack.

And you're limited by the state of energy of, you know, at during discharge, the lowest cell on that stack. Now that could be the most degraded cell in the stack. It could be the cell that had had the most, you know, temperature and therefore had leaked the most internally.

It could be the coldest cell on your stack that is actually gonna hit its voltage limit before the rest.

So at any given time, your actual your algorithm is actually trying to figure out which of these four hundred cells is gonna limit that rack then behind whatever your kind of, AC top topology is, when are different AC blocks in your system gonna turn off. So depending on how you've configured things, you might have thousands of cells, and you're actually trying to guess which is the one that's gonna limit you, at any given time.

So as your cells degrade, you anticipate there being more variability from cell to cell, and then it becomes that much more complex to figure out which one's gonna be the limiting one.

You mentioned, you know, the sort of bandwidth and the data connections. This is definitely exactly where Zatara tries to solve that problem.

You know, in a particular site, there's four hundred cells in Iraq. We might be looking at ten thousand cells or, you know, tens of thousands of cells in a system.

And, you know, the way that we address this in order to be real time is actually exactly the way you said. We put a server into your cabinet, and we live primarily onboard that server.

This means that, you know, we don't have kind of a dashboard or analytics services the way that some of our friends at twice or a cure do.

We're just the controls, and we're replacing the signals that you're using in your existing system.

So, you know, however you are getting data off of your EMS, however you are using those for market operation, and then whatever telematics are coming from your EMS and going into your cloud going into your ISO, we are replacing those directly at the edge.

And that means that we need to be on prem, in order to be real time, in order to be able to soak up all that data quickly enough, but then also to be compliant.

You know, you're not allowed to use cloud solutions in the loop for dispatch. And so, we fit right into your NERC sub case with our, kind of secured server.

So you're physically going on-site with a big metal box and putting it in in a server rack somewhere?

That's exactly right. On a per kilowatt hour basis, this is much cheaper than getting the data off.

So a hot topic at the moment in our world is cell balancing. There's just been so much change and improvement and iteration in how systems are balancing cells. And I know that, you and your team spend a lot of time thinking about this.

Could you just, for our audience, just go straight from the top? What is cell balancing, and why does it matter? And how are you guys thinking about this problem?

Yeah. Absolutely.

Well, you know, cell balancing in a system that's brand new where all of your cells are identical and have all started at exactly the same start of state of charge should be a nonissue.

Unfortunately, right after moment one, this idea that every battery has the same capacity and that every cell has is at the same state of charge goes right out the window.

So, you know, as I mentioned, behind a single rack where you've stacked four hundred cells in series, and then depending on your AC topology behind that, you may be limited by whichever cell in that system has discharged to zero faster.

And that that may be because you can't discharge more of the cells while you're limited by one. You know, the simplest example is in a single series rack, you can only discharge the cells together.

Now this cascades up the system. Right? We have racks that are organized into DC buses that then, kind of might connect in different configurations to a a PCS to form an AC block. And depending on how many of these AC blocks are, in your system, as well as what controls you have behind each one to balance across different racks, you're gonna have different limitations at any given time in your system and different trade offs to take those batteries down and allow them to balance. So, typically, what we're seeing from OEMs today is that in order to trigger kind of a balancing condition, you often need to bring that battery all the way up to a hundred percent or all the way down to zero percent, and let it idle in order to trigger balancing. And this is because with those LFP batteries, the very top and the very bottom, there's a little bit more observability into the state of charge because those voltage curves are a little bit friendlier.

If you choose the wrong subsets of your, asset to take down for this balancing activity, you won't be improving the performance of your battery, and you may be actually, leaving a lot of sort of opportunity cost on the table by picking the wrong element to balance. But because you have to go through this sort of onerous exercise of taking that rack all the way down to zero, you know, it is very critical to pick the correct element of the system to balance.

If you do this systematically and and hit the out of balance racks with this balancing activity consistently, and it can take a a very long time to get back in balance, then you can actually maximize the range over which you have max power in your system.

So if your cells are, you know, twenty percent out of balance, you might find that you can only get your max power between ten and ninety in your system. And we find that ten percent is a very common figure for how much out of balance your system is. But over time, you might actually be adding one to one and a half percent mean imbalance per month, and your present balancing strategy may or may not be addressing that at all. So we'll see, you know, customers that have worked for with their OEM for a long time instituting some of these new approaches to balancing, but the, the main balance is still steadily drifting up at that one and a half percent to a a a month, and they're getting less and less of their actual max power band.

The operational challenges of managing these big batteries are it's just fascinating because you have measurement you have measurement problems like we've just talked about ranging from the granularity of time series data to where it's actually being measured.

And then you have these operational challenges. Once you've measured it, you you're not sure whether it's right or wrong in some cases. You have these operational challenges about, you know, real physics about different cells, having different voltage curves. And then operationally, you have this compounding problem of balancing that every month is getting more and more challenging.

And at some point, you have to kind of rip rip the Band Aid off, go down to zero, and go back to full full charge, which is expensive and, comes with well, it's, an opportunity cost. And, figuring all this out is is just fascinating. And so how how do you guys approach this problem then? If you're thinking about balancing a a data led approach to balancing, what would you advise asset owners to think about?

Yeah. Absolutely.

We think asset owners should think about automation and intelligent automation.

So, you know, the worst case scenarios we see today, we have asset owners that have to take their entire site down in order to perform this balancing activity.

You know, the the better sort of EMS systems that we're seeing out there today enable customers to go PCS by PCS, take a partial outage, and only balance part of the site.

But what we think is really merited is what you get when you install Zatara, which is a prioritized intelligent balancing system. You know, number one, we're giving you real time observability into the state of imbalance behind every one of your AC blocks. So that's not only critical to knowing how much power you have right now. But if you go through this activity today, we're the first time you're getting feedback on how much you've actually unlocked in your system.

And so, you know, ultimately, that's best for setting up automations. That means that you can identify which blocks are costing you the most hours of power, make sure that those are balanced whenever there's an opportunity.

And then we wanna configure things like making sure that we take advantage of time when you're not planning to use that, asset or where the costs are low for taking that one PCS down and automatically just have the asset always optimizing itself because it does take really a long time and persistence to get these things back in balance.

We we just think you need to be taking every possible opportunity.

And, you know, typically, a battery is only really making money for six hours a day. So we wanna use the rest of the possible time.

And so can you do me a favor? Can you for our audience who is primarily folks who own and operate battery systems and, you know, grid scale assets maybe already online or developing and financing these.

At the contracting stage, what should they be adding into the contract? What's the bare minimum for data requirements, that they wanna make sure is in the contract? What's your what's your kind of one zero one guide for that?

You know, we get this request quite a lot, and we're gonna be publishing something here in the next month or so, hopefully.

We think that there's a key SCADA points list that is really important for making sure that you're monitoring your power reliability as well as your safety. And, you know, that's a big topic, this week after what happened at Moss Landing.

We think it's really important to ensure that if you want, you should be able to go in and get access to high granularity cell level voltage data. And if possible, this should be historized, not necessarily taken off of the asset and into cold storage, but you should be able to get, you know, pretty high granularity voltage data from every single cell in your system.

Now not every customer of ours puts a large enough historian on their system to be able to do this. And so at the very least, they're getting the min and max voltages from each rack.

And for a relatively new system, this would be okay.

But, you know, this is perhaps the most important thing in addition to the the current or power ratings from each rack in your system as well as the temperature readings that you're able to get your hands on.

If you have a problem, it's very critical to be able to go back and pull that data that's not that's only at the edge and and use it to be able to debug. You know, why did I have a safety issue? Why did I have that base point deviation?

If you're not doing it, already today, we we highly recommend it.

Yeah. You need to store the boring data when nothing is happening, just for when something does happen. Okay. Now for the last two questions. So firstly, Shyam, is there anything you'd like to plug for our audience or, our listeners? It's generally battery energy storage, grid scale battery energy storage folks from Europe, US, and Australia.

Yeah. I think, just to pay attention to, what we're gonna be launching this month, at at Zatara.

So we are coming out with our best owner operator focused solution.

This is our on prem box, and we'll have a couple different flavors.

Our on prem real time solution for soaking up all of the data that you need, historizing it correctly, and spitting out replacement real time signals that you can use for market operation.

So we think we're the only solution on the market that actually does this on prem real time.

We'll also have a big announcement with our, kind of corporate partner, Emerson, related to Zatar Live being available to customers of the Emerson Ovation DCS, system.

So stay tuned for that.

Very cool. And, also, you're doing hardware, which is great. Hardware is back in vogue finally. As an engineer, I'm so happy. It feels like startups everywhere are getting back into doing real things in the real world.

Absolutely.

Very, very cool.

Not the hardware itself. It's just a hardened, you know, Dell server.

But, ultimately Well, I'll give you the kudos anyway.

I appreciate it.

And, Sean, what's your contrarian view? What's the thing that you believe that not a lot of people do?

Yeah.

We think that battery signals can work at the edge. And I think we see so many different sort of folks out there that provide cloud analytics.

And what we think is really underloved in the world today is things that you can do onboard hardware directly at the edge because this is where I'm at.

What is for, folks who aren't in the in in in this world, what is the edge?

Well, the edge is is a it's basically two things. One, it's the idea that you don't need to transfer data from your site, from your premises across the Internet to get anything done. You know, we can do all of the computation and analysis of that data directly on premises at your site. So the edge is the ability to do that with limited compute. You know, we're not using a supercomputer in the cloud.

But, also, it's the ability to do that in a secure and compliant way that doesn't open you to any kind of, you know, vulnerability or a threat. And so, really, a lot of the reasons that we're doing edge compute, it's really twofold. One, we can get access to so much more data without paying the data transfer costs. And then, two, it's for cybersecurity and resilience.

Often, for your case, if you really wanted to solve this problem, you're not allowed to use a cloud solution. You have to use an on prem one like us.

It's funny. There seems to be there was in the last fifteen, twenty years, there's been a huge shift to cloud. Everything's moved to the cloud, and that's been enabled with, you know, three, four, five g over those that that period.

And it's pretty much in every sector you're seeing a a move now to cheaper compute on chip and not bothering to send stuff. So for example, a lot of you know, you you only have to look at, consumer electronics and the use of large language models. And the the the way that you can squeeze models down now to run locally and all of the security and privacy benefits of that are immense. It it'd be interesting to see whether this decade, we see a switch back away from cloud and back to people doing compute themselves.

AWS has had one hell of a run, you know, growing, like, fifteen, twenty percent year on year forever. So maybe maybe their time has come to an end.

Maybe, Yeah. Yeah. I I I think that it's always been a partnership, and and maybe it's about that privacy element in the end. Right? Maybe you could get the best performance with a cloud model.

But if it's really important that you stay secure or if you don't want your data going up to the cloud, then it's always been really important to have that edge compute.

You know, I was fortunate to be one of the first few hardware engineers at Nest to start my career.

And, you know, we were definitely the a pioneer in the IoT space, making a distributed resource available through a cloud API.

But I think the thing that everybody cared about was making sure that their data about their preferences, how how they moved in and out of their home, that kind of thing, did not go up through the cloud service. And so we had to figure out how to marry what you wanted to do locally in order to be private, with what you could do to to make the the resource available to the grid. And so, you know, I think we're seeing that scale to front of the meter assets now, which is is just wild and and really enjoyable.

And what was it like building let's do this. What was it what was it like building Nest? I absolute I think everyone who has a Nest absolutely loves it. It's just by far the best product. It looks the it looks the part on the wall.

It was so awesome when it first came out. And even now, they are it's kind of it is the leader. Right? So if you're involved at net at NEST in the early days, what was that like?

You know, for a a kid coming right out of college, it was the best possible experience in the energy industry. I think it was this moment of real optimism.

And we saw, you know, utility companies jumping on doing pilots with us, because I think people loved, you know, delighting their consumers. And it's so rare for a utility innovation office to to to be loved, by their consumer, by providing them such a beautiful object.

But I think the trick was, you know, how do we actually start to make firm resources out of distributed networks like this? And I think, you know, edge intelligence became a real part of that. And, you know, I think everybody's kinda surprised the batteries weren't so firm.

So it's surprisingly Yeah.

The idea of using edge compute to to to understand what's happening at the battery, ended up being necessary as well. So it's kinda full circle for me.

Very cool. Well, Shyam, I wanna say a massive thank you for joining us on the podcast. We'll put some links to all the stuff that you guys are doing in the show notes. So if you're listening to this and wanna check them out, please do. And, when I'm over in the West Coast, I come come by for a beer.

Absolutely. Yeah. Everybody should come visit our battery lab in San Francisco.

Awesome. Thank you very much indeed. See you soon.

Quentin, thank you so much.