Season 1 Episode 6: Morgan Pare on the Role of Battery Optimisation for the Green Transition

This week on the Hypercube Podcast, Morgan Pare, Lead Data Scientist Australia at Habitat Energy, joins us. Habitat Energy offers a specialist optimisation and trading platform for grid-scale battery storage and renewables in the UK, the US, and Australia.

LISTEN ON SPOTIFYLISTEN ON APPLE
S1E06 Morgan Pare

In this episode, we covered:

  • How to optimise merchant revenues for battery energy storage systems and renewable assets using data modelling.
  • Why it is crucial to regularly retrain machine learning models to account for data drift and improve forecast accuracy.
  • Why a "human in the loop" approach works to give traders digestible insights while allowing them to take manual control as needed.
  • How battery optimisation can de-risk investment in renewable assets and why it is vital to the green transition.
  • How data modelling and automation are empowering traders operating in Australia's 5-minute energy markets

The weekly Hypercube podcast sits down with leaders in the energy and utilities sectors to explore how data analytics can help businesses make smarter decisions and accelerate business growth.

CHAPTER MARKERS

[0:44] Morgan Pare overviews his work at Habitat and the Australian energy market.

[2:07] Morgan explains the approaches his team is taking to deliver robust energy forecasting and why classical machine learning approaches don’t work.

[04:49] Morgan explains why price-taker strategies are common in the Australian energy market.

[05:48] Morgan and Host Adam discuss the impact of data drift on energy forecasting models and the need for model hygiene.

[7:40] Morgan explains probabilistic forecasting and the benefits of taking this approach.

[08:55] Adam asks what Habitat looks like for the end user, with Morgan explaining that they use a human-in-the-loop approach.

[10:15] Morgan discusses the importance of battery optimisation as part of the green transition and the key considerations, including how they differ when considering non-lithium batteries.

[15:23] Morgan and Adam discuss the critical differences between the UK and Australian energy markets, focusing on the pace of change and the differences in data feeds.

[17:20] Morgan discusses why automation is critical in 5-minute markets to allow traders to capitalise on opportunities.

[20:02] Morgan discusses the importance of data scientists in the energy industry and their role in accelerating the green energy transition.

TRANSCRIPT

Hypercube Podcast Transcript

Title: Morgan Pare on the Role of Battery Optimisation for the Green Transition

Host: Adam Sroka
Guest: Morgan Pare

Intro: Welcome to the Hypercube podcast, where we explore how companies in the energy and utility sector leverage data analytics to make smarter decisions and accelerate business growth. I’m Adam Sroka, founder of Hypercube, a strategic consultancy that supports asset owner-operators, traders, route-to-market providers, and energy services companies to unlock the power of data.

If you’re interested in hearing real-world examples of how data and AI are advancing the energy sector, this is the show for you. 

Adam Sroka: Today, I’m very excited to be joined by Morgan Pare from Habitat Energy. Morgan, great to speak to you. Would you like to introduce yourself for everyone?

Morgan Pare: Hi everyone listening. My name’s Morgan Pare. I lead a team of data scientists for Habitat Energy. Habitat Energy is in the business of optimizing merchant revenues for battery energy storage systems and renewable assets. That involves primarily building optimization models, bidding strategies for bidding large grid-scale batteries into the NEM.

So that’s the National Electricity Market for those that aren’t familiar with the Australian space. And then also building market price forecasts to feed into those models. So that’s what we get up to day to day.

Adam Sroka: First things first here. Do you enjoy the culture over there? Like, what’s the kind of data culture over there compared to over here?

Morgan Pare: I mean, the data culture is really awesome. I think that there may be slightly different, and maybe I wouldn’t say behind the curve, but certainly going through that journey of maturation. I think you either see companies that are kind of very much stuck in their old-school, mature ways, building lots of things on data tools. But a lot of companies as well—Australia seems like a place where you can take risks. So you can actually be, if something doesn’t go quite right in Australia, it’s not the biggest market that you mess up in. So, quite interesting.

I’m seeing a lot of actual kind of data startups here. When I was in Web3, there were a lot of Web3 startups here as well. It’s an exciting place to be, and you get a whole mix of data professionals.

Adam Sroka: So yeah, let’s crack on with what your day-to-day looks like and forecasting in energy systems and some of your expertise around that. What are the best approaches or things that you’re using in your day-to-day to do really good, really solid energy forecasting to get the best performance you can for Habitat?

Morgan Pare: Yeah, I’ve been forecasting quantities of it in the energy sector for quite a while now. I’ve definitely seen some different techniques and approaches. I think the key thing that stands out to me is that you really can’t just throw these classical kind of machine learning approaches at the problem and expect it to be market-leading. I think that you can’t just get by on time series forecasting alone. What you really need is a strong level of expertise on the quantity that you’re trying to forecast. That only really comes through industry experience, sitting down with subject matter experts, getting into the documentation that the grid operator might provide.

I’m sure you love doing that as well, Adam.

Yeah, and then you’ve got to try and work out how to explicitly code factors into the model that allow the model to forecast those prices. For those that aren’t familiar with the imbalance price, it’s essentially a market price that reflects the system imbalances within the national grid within the UK. But in terms of how we forecast the imbalance price, in the past, we spent a long time trying to understand the actual calculation for how to calculate the imbalance price. You need to bring in lots of different bid stacks, lots of different offers, acceptances within the market, and other quantities. When we forecast that price, we build a quasi-calculation around what we think that price should be ahead of time. We have the components that make up that price. We bring them all together and encode those directly into the model. You can do a little bit of time series magic in the background—what was the previous price, what’s the price been looking like in the past amount of time. Then you can also add in some other exogenous features. Things like weather can also drive that price. It’s more difficult to forecast that price directly. Similarly, in the Australian NEM, the National Electricity Market is slightly different, but it’s a price stack, a clear model. It depends massively on how other participants are bidding. The market operator obviously is not going to give you that information straight away. They’re not going to tell you live how participants are bidding, but they will give you windows into that, both in their live data and around the historic data. You’ll get access to those bids the day after, and in the live window, they give you other quantities that you can access. For example, price sensitivities, and you can use those. I would say that the market prices in Australia—the distribution is quite messy. It’s not a very nice distribution to forecast, but some of these other quantities are much nicer to try and forecast. You can use those to forecast those and use those as proxies that allow you to back out the market price as you would expect.

Adam Sroka: So project clear means it’s quite reasonable to just be a price taker strategy there. Is that quite common?

Morgan Pare: It certainly can be quite common. You can bid quite a lot lower than what you expect the price outturns to be. I think if you’re a small asset that doesn’t have the chance to move that price, then it’s a strategy that we see quite often. Of course, if you’re a bigger asset and you do have a chance to move that market, then actually your market price forecasting might want to take that into account as well. You might want to do a bit of co-optimization across those two. But certainly, price-taking strategies are quite common.

Adam Sroka: Do you have a ballpark for big in that context?

Morgan Pare: Big. I mean, yeah, it’s a good question. I would say once you get above 50 megawatts, 100 megawatts.

Adam Sroka: Okay.

Morgan Pare: Yeah, and you actually see quite a lot of sticky prices within the NEM, within the regions of the NEM. That’s where big thermal generation is just, you know, they’re quite often setting the price at their prices that they’re bidding in. That’s another thing that makes it quite difficult to forecast because then you have sticky prices and distributions. It just can be a bit more difficult.

I was also just talking a little bit about another thing that you need, and that’s really excellent model hygiene. It’s a bit more of a data-scientific thing rather than getting into what actually makes these prices outturn where they are.

I think that you need to be able to retrain your models regularly. Quite often, you see data drifts within market prices, especially as new entrants come into the market. They start making bids, changing market prices that way. I mean, we retrain at least every 24 hours and we also have drift detection as well, which could trigger automated retraining. You also got to use the correct window training size as well. You don’t want to go too far back—you’ve got outdated data, but you want it to have a good, well-trained model. The other thing that you need to do is make sure that you focus on model uptime. I think we’ll find quite often that a missed forecast is a lot more detrimental than having a slight increase in forecast accuracy. So yeah, that’s also something that we monitor a lot is model uptime. That’s one of my key KPIs is model uptime. If that starts to drop, then we need to think about fixes and better data quality handling and that kind of thing.

Adam Sroka: Actually, talking about window size and data drifts, did you see much of the impact of the Ukraine war? Because over here, it completely banjaxed a lot of the training—you couldn’t use data because it had unprecedented impacts on the markets and things like that.

Morgan Pare: I mean, it’s a good question. I think Ukraine broke out just before I joined Habitat. So I can’t say from personal experience if that was the case, but I think we’re slightly more insulated down here. I think that Australia has its own thermal generation resources. I think it’s slightly more insulated that way. Yeah, a lot of it is very much coal-fired plants in Australia still, that’s all mined within Australia. So I think it was probably slightly more insulated than the rest of the world.

Adam Sroka: You hinted at probabilistic forecasting earlier being really important. Can you expand on that, like what it means?

Morgan Pare: Yeah, for sure. I mean, I think that it’s really important if you want to have a super competitive edge. I’m not sure many energy companies right now are doing strong probabilistic forecasting. It certainly requires a certain skill set and it’s definitely more prone to some errors than classical point forecasting, b ut I think that it gives you a better idea, especially when you start to think about risk. We’re quite often seeing, for example, in Queensland recently, we’ve had some big thunderstorms come in. When thunderstorms come in, it becomes really difficult to forecast solar outturns and therefore what that does to prices. So it’s kind of unclear whether the solar will be in the system as that ramps off during the evening and that can cause prices to be quite volatile. So yeah, a probabilistic forecast, you can take a look at that. You can say, okay, what’s my P30? So what’s the 30th percentile of my probabilistic forecast?

What’s, you know, Yeah. What’s my P20? What’s my P10? And you can make decisions within the context of that. So you can say, okay, I’m not going to charge if I see that there’s a really high chance that prices are going to be higher than a certain amount, you know, that can feed into how you optimize the asset.

You know, or you might say, actually, there’s a small chance that prices will be really high. So I’ll just discharge through that period, you know, with a, with a kind of more. a riskier profile. 

Adam Sroka: Just a little bit on like the engine of the machinery of that, thinking about the end user. So the end user, you’re making recommendations to traders. How do they consume that? And like, are you running what-if scenarios and giving them like a dashboard with, if this happens, like he’s probably in excess action and things like that?

Morgan Pare:  Habitat’s approach, we call it human in the loop. Um, you probably heard that term throughout the industry as well.

So we give traders what we think is the best option. And then they can choose to steer it. They can either take manual control or they can do kind of more limited intervention steps. But yes, we, you know, it’s certainly something that the traders find valuable is if they change something like, you know, if they set a certain strategy, what does that dispatch profile look like?

Certainly I would say it’s less so on the probabilistic forecast, like, you know, making kind of. Monte CARTO samples from that and showing that to traders and doing all of that kind of good stuff. I think that is potential. But again, that could just overwhelm the traders, so. Mm-Hmm. , you know? Mm-Hmm. . We’d rather just give traders something that’s digestible and they can use and they can steer it how they like.

And you know, I think if you talk to different traders, they have might add different appetites on that and they might just want a thousand screens with all the different outcomes. Yeah. Right now that’s not been super high priority for us

Adam Sroka: The wonderful world of optimization and ethic broadcasting and things like that. Is that something that you’re doing over at Habitat? Are you, have you carried some of that expertise over with you?

Morgan Pare: I mean, optimization is our bread and butter at Habitat. I mean, we, you know, if you look at the independent scores, especially in the UK, we’re, you know, topping those charts when it comes to the battery optimization.

We’re trying to bring that across to Australia, you know, smaller team here trying to, you know, Push that through. And yeah, it’s certainly bringing in some of those previous experience from origami. I think the battery optimization, it’s one of the most important challenges in the energy space for the green transition.

If you think about if you’re an investor and, or, you know, a founder of one of these renewable assets or a big battery, you really want to de risk that business case of investing. So if you can hand that battery off to an optimizer, who’s really skilled at what they do, they’re going to generate, you know, a lot more revenue and that’s going to de risk that business case for you.

But it’s also just taking care of the asset as well. You know, these battery systems are, they can be fallible. They can have issues. I think you had a guest on a previous podcast talking about all the data that these batteries give you about the health of those batteries. And, you know, we also need to consider that as well when doing the kind of optimization as well.

And yeah, and doing that for our, for the, for the customers

Adam Sroka: So, not to give away any secrets or, but if I’m a new team trying to spin up a new grid-scale battery optimization set-up, what should I be thinking about? What’s important to me?

Morgan Pare: You know, maybe we can step back, take a step back and think about it from, from the layman’s terms.

So, you know, if we, if we’re thinking about, okay, I’m, I’m new to the world of optimizing batteries. What does that look like? Well, You might think, okay, we’ve got this duck curve throughout the day, especially so here in Australia, when, you know, prices are slightly higher in the morning, they, the solar comes on in the day, prices get massively compressed.

And then in the evening  peak, when everyone’s switching the kettle on for the footy, it’s on go or AFL. And the evening is when you need the electricity. So you’re looking to charge the battery in the day and then discharge it through the evening. I think that’s quite a naive view of how you do forecasting.

Certainly, there’s elements of that, but it goes a lot, lot deeper than that, as I’ve already hinted to. So, you know, you’ve got to think about it. There’s lots of markets that you can be in, so you might not be just in the energy market. You might be in providing frequency control services. So ancillary services, I think they’re also called in the UK,  you know, that, that’s a key decision point for traders.

Another thing is around cycling and degradation of the battery. So each cycle that you do, it comes with an inherent cost reflected in the degradation that that does to the battery. And if you think about it, you know, we all know lithium-ion. Batteries degrade over time and you can operate your battery in a way that makes the lifetime of that battery be five years or it can be 10 years.

And you know, how do you actually, you know, if you think about large cell replacement costs, what’s the best way to operate that battery to, you know, what’s the perfect amount of time and you know, that kind of thing. So we can essentially turn that into some kind of what we call a hurdle rate. So each cycle has a, you know, if you’re doing a cycle of a battery, you want to X amount of dollars that cycle X amount of pounds for that cycle to be worthwhile.

So you know, we can think about things like the cost of cell replacement in later life and that kind of thing. And you might decide actually you want to be in a frequency control market that allows your battery to stay idle. So it’s not cycling and generating degradation. There’s also lots of other degradation things that happen, things like C rates and keeping state of charge very low, et cetera.

Another interesting piece is when it comes to warranty periods, right? So quite a lot. Often you get. Several hundred, hundred cycles per year, and that’s what the manufacturers give you in warranty periods. If you think, okay, I’ve got 350 cycles this year. Maybe I’ll cycle the battery once a day, not about evens out.

And again, that’s not really the optimal view. There might be periods of the year when it’s better to do more cycles. So for example, in the summer here in Australia, and it also might be, you might think, okay, well I’m at, let’s say June, July, August, and I’ve only used up, let’s say half of my cycles. So you might want to be a bit more aggressive in the later part of that year to use up the full warranty as well.

So, you know, we build all of that into our optimization models as well.

Adam Sroka: Is there a difference with like non lithium based batteries? 

Morgan Pare: Yeah, for sure. So one of them is they don’t have really much degradation at all. The battery life cycle is a lot, lot longer and they’re also a lot longer duration as well.

So it’s kind of interesting to see how that plays out. In terms of, can you, do you still do your kind of classic discharge through the evening or are you doing some kind of different approach given the cycles and the duration of those batteries? Also, you know, you’ve got to switch these pumps on. So it’s an electrolyte solution.

You can’t generate electricity unless the stacks are filled and the pumps are going on the battery. That uses quite a little bit, quite a lot of electricity, you know. And it also changes based on the state of charge of these batteries. The fluids can get more viscous, um, and it comes harder to pump the electrolyte around the actual battery in the stacks.

You’ve got to consider that as well, right? If you want to activate a battery, you’ve got to wait, let’s say five to 10 minutes for the battery to actually get enabled. And then also when you’re on, there’s going to be either a parasitic draw from the actual state of charge of the best, or you’re going to have to bring in power from the grid and patient that power.

So that’s another, like, how do you model that in a non-societal model?

Adam Sroka: So we talked about, we talked about kit, we talked about forecasting, I guess, just generally one of the things that I guess we like to believe over here in the UK.  But is the number of services and the pace of change in the energy market themselves is quite, it’s quite forward thinking. We’re quite advanced in the way our energy systems build. Does that hold true? Is, what’s the, the kind of, are there core differences? Whereabouts is Australia in front of us?

Morgan Pare:  I mean, it’s a great question. It’s certainly something that I’ve learning about new every day. You know, I’ve been in the energy sector here for a year now, and you know, there’s definitely things that come up that still surprised me.

They go, that’s kind of similar to how it was done in the UK or a bit similar. I’m sure you probably have similar thoughts about UK services coming online. I would say that one thing is like the pace of change definitely feels slightly slower here. I remember my time at Origami, you know, we were constantly bringing on new services in Australia.

We’ve just recently had a new. We’re calling it a very fast F CAS response. So that’s the one-second response market. Um, so that’s been new and that’s something we’ve had to build. It’s changing a little bit how we do bidding next year as well. That’s IE double S standards that are coming in. But I would say overall, they feel more manageable.

I think that like you said earlier, you’re not necessarily having to throw out a lot of the work you’ve previous it was designed somewhat differently that we had to kind of hit the drawing board again

Adam Sroka: I didn’t do API. We need to figure it out and get the data from.

Morgan Pare:  Yeah, exactly. But I mean, I think that, you know, in the Australian energy market, I think that the data feeds definitely a lot different to the UK. I think that the UK, you know, it has like the Lexan. services and some of the other ones that are actually quite nice to, to access the data. I mean, you might disagree, but certainly, you know, it’s definitely different. You know, we, we, a lot of it here is built for like a data wrapper. And so getting that data into the system is, is, is different.

Adam Sroka: Different is a very diplomatic word to use there. How does that actually play out in practice? 

Morgan Pare:  Yeah, it was a super interesting conversation when I first, you know, first couple of weeks around the lunch table with the MD here in Australia and, You know, I was like, okay, so it’s a five-minute market. Like you have gate closures. Do you have physical notifications? And yeah, he was like, no, we don’t have any of, any of that. You could, you have five minute settlement periods. You can pretty much change what you want to do for each five minute period up until that period begins, obviously within compliance restrictions.

You know, not only that, but There’s around, you know, across all the markets you have around 500 decision variables about where to allocate your generation and load to into those markets. So when you have this five minute market, it’s quite interesting. I mean, I always say that automation is king here, right? 

Those prices changing every five minutes, there’s really great arbitrage opportunities. But because those prices change so quickly, it’s really hard for traders to actually manage those themselves. So, you need to have an automated system that’s going to help the traders or, you know, either automatically get grab those opportunities or help traders to do so.

Another thing is, because you can change your, you know, again, within reason, because you can change your intentions right up until that period starts. Price forecasting becomes a lot more real-time. So you get the best price forecasts,  a couple of minutes, three or four minutes before that, that interval begins, the traders really don’t have time to react to those price forecasts.

Whereas if you have an automated system, it does. So, like I said, it’s why we kind of in Australia, we automate as a baseline here in Australia. And we have the traders in the loop, and I would say the traders are somewhat less in the loop here in Australia than they would have been in other markets, just because it’s changing so quickly.

And there’s certainly that, you know, there are times when our automated solutions just can’t handle, you know, unforeseen market conditions and traders jump in.  We build an automated system first here.

Adam Sroka: Yeah, that’s really different, isn’t it? Because your traders are acting like course correction, like protection against kind of unmodeled shock, but they can’t be making a decision for every seven month period because it’s just too fast.

Morgan Pare: yeah, exactly. And I think a large premium previously was placed on, you know, you have your strategy set throughout the day and you just kind of let that play out and make minor tweaks as time goes on. Whereas now you’re seeing a big rise in what we call auto bidders within the market. So They’re doing these also bidders. Yeah, it’s kind of interesting. I think the regulators are still trying to figure out what’s the best approach. And certainly we have regulators, regulation front of mind whenever we design new strategies, et cetera. So yeah, that’s just as another layer of interest and complexity to some of these problems.

Adam Sroka: So  before we do go, I just wanted to ask, is there anything you want to plug or promote? Like if people want to learn more about what you’re up to at Habitat, yeah, what’s going on, where should we point them to?

Morgan Pare:  Yeah, for sure. I mean, you know, if, if you want to get in contact with me directly, you can do so via LinkedIn.

So, you know, it’d be great to connect. You can just search me up. I’m sure if you. There’s not too many Morgan Pare’s in the world, so hopefully you’ll find me there. Professionally, I think if anything I said resonated around  price forecasting, around optimization, if you’re looking for, for someone to help optimize any risk grid scale assets you’ve got, then yeah, you can find us at Habitat Energy. Yeah. We’d, we’d love to get in touch. And I think more generally, I’d just like to kind of plug the role of data scientists in energy. I think I talk to a lot of young people nowadays about, you know, Which industries work for them? I often say, you know, try lots of things, but I do have a special place in my heart though for energy. I think that, I think you’d agree, Adam, the problems are super engaging. They’re super chunky. They quite often lend themselves well to ML tooling. Obviously, it shouldn’t be the first. tool that you reach for every time, but there’s certainly value that can be gained from them.  I think it’s crucial that we get as many, you know, as much young talent into this industry as possible to really speed that green energy transition.

Adam Sroka: All right. Well, look, lovely to speak to you as always. Thank you very much and have a good time at Habitat. 

Morgan Pare: Yeah. Likewise, Adam. Thank you so much for having us, on the podcast. It’s great to see what you guys are up to. 

Adam Sroka: Cheers. Thanks very much. 

Outro: And that’s it for this episode of the Hypercube podcast. Thanks for tuning in today.

If you have any questions about the topics we covered, you can reach out to us on LinkedIn or check out our website at  wearehypercube.com. You can also join Beyond Energy, our Slack community of data leaders from the sector. There’s a link to sign up in the episode description. We’re just getting this show off the ground, so if you like today’s episode, please leave us a rating, review, or subscribe wherever you get your podcasts.

It all really helps. See you next time.