Inside Flow-Based Market Coupling | Time2Talk #206: Paul Pontenagel & Frank Boerman

Hello, welcome to Time2Talk. Time2Markets, energy trading podcasts where in breakdown the latest developments for trading experts and industry leaders. Welcome to Time2Talk. Today we have another guest today from Tenet in the Netherlands, which of course lies close to my heart me being Dutch. Today we have a rank Boerman, specialist with internet on the operational side of the flow based market coupling. Welcome Frank. Thank you. Can you tell us a bit about who you are, what do you do at Tenet, what are your interests, I mean your title says it partially, but please introduce us. Yes, I'm Frank and I've been working at Tenet for four years now and I mainly work as a flow based capacity calculation expert in the operation side. So I do a lot of operational monitoring analysis to these stuff like that to support the operation, I'm not an operator myself. And can you walk us a bit through the capacity calculation for the market participants, very important to know of course on the flow based market coupling, the PTDF, the power transmission distribution factors, all very important. What is your precise role in defining the capacity calculation and let's zoom in a bit into that. So my role as I said is more of a more support. So I look a lot at the results, the final domain results and then what the market does with the capacity that we provide to see if we can further optimize the process and to explain a bit more what it is. So in day at market coupling and these days also an interday there are two steps. So you have the step that calculates the capacities that we can safely release to the market and you have the market step. And most people focused on the second one but the first is just as important. So we as I always say we set the playing field. So we define the space within which the market can move. And these days that is in flow based capacity calculation and it is slightly more advanced than what we used to do let's say 10 years before. So 10 years before the situation was simpler we defined one number per hour per border direction. So let's say one and a half gigawatts, Netherlands to Germany and then something else Germany to Netherlands and then we did that for all borders in the in the coupled area. And that is that is fine that works but that is not necessarily the most efficient way because you determine a value in isolation per border while in truth if there is a lot of import for example from Belgium to the Netherlands it is very possible that in the grid that enables more expert export from the Netherlands to Germany and vice versa. But if you're in an ATC as it is called available thrice capacity which is still used on some borders. If you're in that world than as a TSO is 10th you calculate through different scenarios and you basically pick the lowest value to be safe in all possible scenarios. While now in flow based we can actually define a set of possibilities is no longer one number it is a linear algebra constraint for the mathematically inclined constraint matrix and then you basically give a quite large set of possibilities to the market and then the market algorithm you via will find the optimal point given all market orders. So and that is that is a very important step. It's also highly transparent but it's also complicated. So that's the thing we provide a lot of data and a lot of things you can look at but you need to know what you what you look at and this is currently live in the in the core region so that's basically most of continental Europe so that's the Benelux from uns Germany Austria and then half recent Europe but not for example Italy, Bucharia, Greece, Spain those have their own ATC system still Scandinavia is their own region but they have recently been moved to the flow based domestic calculation methodology but yeah we are separate regions so we calculate flow based domain separately and then the market algorithm stitches it together. Thank you for the introduction there and yeah zooming in of course into flow based that's a more exciting topic of today I think you said that you make a lot of information public to the market I think the most valuable piece of information there is of course your ptdf matrix that tells you basically on every piece of the grid how much capacity is available to dive a bit more technical you have your parameters and your RAM value your remaining available margin data if you do some derivations you can basically derive what are your maximum important export values for every country in every hour. Okay can you walk a bit more through how is this ptdf formed and I'm not meaning like conceptually but already diving a bit more into detail from from tenet sides you have the knacks every individual element of this matrix how are they defined what kind of fundamentals determine how the RAM will be how these coefficients will be and how is this variation on all these individual knacks. Yeah so the funny thing with flow based is that a lot of things are are done centralized right so that is that is much more efficient so from our individual tenet site what we put in is obviously our individual grip model for the for the Netherlands and every tso does that for their respective area so tenet germany does it for for their part the Netherlands for the Netherlands so that we deliver the other thing we deliver is a starting point of the knack is so it's it's important to know that the knack list is not just an element is actually an n-1 situation so it is a full name is a critical network element and contingency so it is an element a to b in the situation that element c to d has an outage unexpectedly and there is of course a base case so just element a to b when everything is all that's the base case and we make a list of all possible combinations that we as experts from from from from the tso site deemed possible relevant and in an ellipse we have made the fairly simple choice to in in terms of elements include almost the entirety of the 380 kv top level grid and then we have made combinations with elements basically nearest to it as so the the the german interconnector made a dealer we have made outage situations with two aims have to swallowing the south etc it's also possible to define am i's one situations for elements abroad obviously because you know the degree doesn't understand borders so there are also knacks for that we have for a line in nellands in the case of an outage abroad so so that's an input so you have the net model you have the knack list and what we also define is the gresk the generation load shift key and what that means is that the p3ds we don't calculate that happens in the centralised system and what it does broadly speaking is it it tries to move in that position so the sum of all export and import for all for yeah we call hubs not countries because you can also have virtual areas etc so the the generic term is hub the price zones right yes yes but not in all cases so there are also virtual hubs in terms of the a lego DC cable but that is a modeling detail that that's not super important but anyway for every hub what the system does is it pumps up the net position up and down and then with the gresk factors is basically translation from a net position to individual notes in your network so it maps for example if for every 100 megawatt that the nellands exports more some aims half for example gas plant will move up let's say 10% of that so that is then your factor and that we deliver and then the system starts pumping it all up and down and then it calculates for all of these m-1 situations it checks if I do this if I pump up the nellands what is the effect on the flow on my element and that is how p3ds are formed ram is formed by checking what is left after we subtract various flows that we know is there so a starting port you take the maximum flow that can be over element then you have a certain safety margin for all the assumptions that we do and we have some reference flows or flows that we assume is it's there regardless of market interaction but there are some loop flows that there are various flows in this and then the part that is left simply that we call the ram and is different per knack per hour right so the ram is always never the same and the ram is then really the flow that is left for the market this is not always physical it's also something called virtual capacity due to regulatory requirements. So we have in Europe the so-called oh my cat is going to join the party. We have the so-called 70% rule which means that in principle we have to release 70% capacity on all on all our elements. However physically that might not exist. So then there is a part there's a part which is virtual. And this part we then later validate if it's really possible later in the process. We validate if this virtual capacity really can be safeguarded by remedial actions. So that means really specced concentrating. And if there is not a case and if an only if we can prove that in the validation phase then we can actually lower the target a bit and then the ram also goes down. And is that basically one of the steps that is between the early PTDF and the final PTDF? Yes. So we publish from the top of my heart that my head three stages of flow base calculation which is almost every single one that we also have an operations. In operations we actually have four. But the last one doesn't matter that much which is the pre-final one and the only difference is long term nominations which is only active on a couple of borders so that the change is not that big. So we publish indeed an initial intermediate and final domain which is final main is the one that we use for market coupling. And indeed the reported changes in between is stuff like virtual capacity, validation steps, derogations that we can lower the target for various legal reasons. So this is in between in between all of these steps. And there's also filtering of knack steps. So to come back to your question how do we define the knack list? Well we as experts define a static one. So we think of things that might be relevant. There is an expert assessment. We also use some simulation tool for that. But in principle there's an expert assessment. And then what happens is that the system in the initial computation already comes up with the PTDF. And then there is the so-called 5% threshold filtering that if a knack on no hubs reaches a 5% threshold then we basically deem the constraint to be irrelevant. And then it is removed from all four of the steps. And that helps a lot because we have a gigantic amount of knacks in in whole of Europe because again, the past five years in total. Yeah I think over there is the statistic that I think it's slightly less now. But it used to be that we had over 24 hours and a half a million for every day. Before filtering on the 5% rule. Yeah, yeah. So that's really everything. And those are also in the final domain but some then are disabled. We filter them out and then you can see branch daters is out. And so that is also the step. So we made the knack list and the knack list as a filtering in the centralized process. And then it moves on to minra production, validation phases, and some long term nomination flows. But for Nelland that is pretty irrelevant. Yeah, that's I always forget. I think the checks of lacking in border or something. So many Eastern Europe that they still have long term nominations. Yeah, you see you can basically do it everywhere but it's very little used, typically of course. And that's why. No, no, no, no, you're not allowed to do it anywhere, anywhere anymore. So long term nominations, like physical nominations is is actually in only couple of borders left. Most borders, including the Netherlands, have financial rights. So it's not a physical nominated flow. It is a certain form of hedge product with them with the market, with the TSO as the counterparty. There's actually some discussion on this now. Yeah, implicit capacity on all the borders basically. Yeah, yeah, so jao capacity. Yeah, and there is indeed that that's auction of on jao and there is now some discussion if you want to reform that but there's no real clear answer to that yet. But that brings me to the other topic as so that you that we also talked about before, which is the LTA domain. Because flow base is only one part of it. We also have this LTA domain. So that's long term capacity that we would like to release on the border regardless of what is going on. In order to safeguard our financial hedge products that we base the sell to market parties on a yearly monthly basis. And what we have chosen as a European system is that we basically put that in. Let's really, really think we cannot handle it but they're pretty minor values. So LTA Cortilo, it doesn't have that much anymore. But you basically then have two domains. So you have an LTA domain, which is a value per border because in the auction, it is actually just good old fashioned A to C per border. And you have the flow base domain. And we both have those. The LTA domain, of course, we don't have to calculate it's just a value per border. And we have the flow base domain, which we calculate. And then we send this to Euphemia, European Markth coupling algorithm. And then that stitches together. Just to jump in shortly. Basically, at FMC domain is a very complex polyhedron. And let's say 13 dimensions if you look at the flow base core. LTA is more of a rectangular domain, basically a square domain in 3 cube and 3 dimensions, but the high dimensions still square because these LTA values are fixed basically for every hour. Perpendical. Yeah. Yeah. So if you throw that out in terms of a y-axis and position, x-axis and position, if you have a value, which is independent of what the other borders do, then indeed you get a square. While in flow base, it is dependent on each other. And then you call something that we always lovingly call the potato because you have this kind of convex shape. Just to jump in here. So basically, also for the audience to understand that even though flow based is basically the most wealth-eller generating system that we have compared towards the old school ATC or towards the LTA, the day-at-market settles based on the domain, which is basically a dual domain, FMC and LTA. So your physical, static LTA domain, which is basically a backup solution, is still of very much relevance in the total domain calculation for the day-at-market. For now, yes. But there are talks to remove it and to have an LTA product to be purely financial and to have a financial risk also on the TSO side. But it's unclear how that will work in the future. So there is no, there is some discussion on it. There are, I think also, and so we probably something on this. But there's no decision taking on that yet. But it's just a matter of how you decide how you want to shape it. But because of the large amount of virtual capacity that in the European regulation, we now release, it usually vastly outstrips the LTA domain. So the LTA domain is also losing in relevance because the flow based domain keeps growing, also due to grid expansions. And then this minimum value from LTA is not that relevant anymore. But there are situations where your LTA domain is slightly larger than your FMC domain in some directions. And then if you want to backtest your day-at-market, you need to optimize an LPA, basically. And you might settle in the LTA domains. If you only model flow based on some specific days, you will have a very wrong out forecast on the day-at-market because you miss the domain. So that is absolutely true. However, it's becoming less true because the more virtual capacity under the European Action Plan that we release, the less relevant to LTA is because you have to imagine if we do a validation step. And we could back on virtual capacities that includes LTA. So if we do, if we say we cannot save practice part, it doesn't matter if there's LTA or not, we will curtail the only difference being that parties that have bought the product will be compensated. Wow, obviously if you have no LTA rights, you also will not be compensated if we think the domain is only. When we look at the forecasting of the day-at-market, in the end, a large part of our audience are speculative traders. They love forecasts. They love trading. They love spreads. Of course, within the whole flow based system or LTA, you get spreads when your LTA domain or your flow based domain is very small. You have a very small run, so you very easily hit a shadow price that you settle on the surface of your polyhedron. So a very small run will affect, make the very small domain, making very large spreads between all these countries. No, that is possible. It also depends on the grid. So they are parts of Europe which simply have a weaker grid. And especially the connection between Western East and Europe, which now that we have flow based core becomes more and more visible, especially in more extreme weather patterns, because we are more and more getting a weather-dominated system. So what we saw both in the summer, as well as a couple of weeks ago, is that in Eastern Europe, so that I'm talking about the Balkans, I had a very high load, either in the summer due to cooling or now due to heating. Which makes the prices explode while it not necessarily makes a huge difference in the West. And the reason for this is then that if we want to move power between Western and Eastern Europe, you basically have to go through the old historic Austrian Hungarian pipeline. So then have you have Germany, Austria, Hungary, the rest because Hungary is the borders most of the Balkan. And especially Austria has a weaker grid. So this pipeline is simply less developed. And in defence of Austrian colleagues, this is historical artifact, which is being worked upon a lot, which then you can really see divergence. And then you can also see that it pulls up the rest of the region with it. So recently the Greek government has complained about the fact that Greek prices go up for reasons that they do not understand. And that can mainly be explained by the fact that Euphemia re-routes a lot of power to the areas where it's really tight, which is the Balkan is in Europe. So then we talk about Hungary, Croatia, Romania, which all three are by the way in the core region, while Bosnia and Syria are not. So that really, then you really see this bit indeed. And then because then there are simple simply not enough capacities. And then an LTA domain will not help anyway. So you can do whatever you want, but if you in terms of virtual things with capacities, not there, you cannot move the power. So in terms of forecasts, what is always very important is to understand where these larger breaklines lay and really check the weather and load the load patterns. This also happens in Western Europe. So there's always this rule of thumb in France, that in France, due to a lot of cheap nuclear, historically speaking, they all have a lot of heating with electricity, which is usually older types of heating that we're not talking about a heat pump, then we're just talking about the resistor, which we put electricity through. So that was the answer to heating, right? Yeah, yeah, yeah, yeah, of course. Yeah, I was thinking. No, I have so. And then you get this rule of thumb that for every degree of freezing, you get one gig, what more of load? I don't know how accurate that is anymore, but it used to be like the rule of thumb. And that can really, you know, jolt up, which we've seen in in 2021-22 when there was not enough nuclear fleet availability. So those things are very even even separate from plastic calculation. Those things are always there and are very important for forecasts to realize. And then indeed, if you go to a more detailed, more nuanced out-to-our difference, then you can really look at the plastic calculation. But even then, that is forecastable. So I often get a lot of, especially now, the Nordics that they have moved to the flowway system. I get a lot of questions or complaints, like, oh, it's very complicated, which is part of the truth. It is complicated. But it doesn't mean it's a black box. So a lot of knacks behave in a similar pattern. So you can actually forecast the domain even with simple time series as a start. It's absolutely possible. Because a lot of knacks have a certain pattern which moves with the wind and solar in feet, which is quite predictable. Can I hold you here for a second? Yes, this is getting really interesting because when I was trading, people always said in the market, like, oh, you can forget about forecasting this PTDF because it's so volatile and without massive fundamental changes, you can get a totally different PTDF due to variation on RAM or extra knacks or critical branches as they were called back then. But you basically, yeah, from TSO perspective, now state the opposite. Can you emphasize a bit more where lies the value then for the speculators, the forecars, the trading houses, what is typically forecastable on these knacks? So the PTDFL is themselves. If you look at time series, there are definitely patterns in that. So even if you do not even take a very fundamental approach, you can look at the historical domains, see which knacks are usually limiting or limiting on the borders of your mathematical constraints matrix, and then do a forecast with weather inputs. That is, of course, never 100% correct. It is true that sometimes it can flip. But most things are weather pattern driven and/or availability driven. So if there are large capacity line outages, which is published by TSOs, or there are large generational units outage, thus influences your GDSK matrix that I talked before. That can have a significant impact. Those things are forecastable. Of course, not everything is forecastable. So the validation phase of TSOs, for example, is very hard to forecast because it is based on our own internal forecast of remedial action potential. So read the specch and counter trading may mainly really specch potential, which from an outside perspective, it is very hard to do because there is a lot of competition sensitive information in that, availability, so generators, etc. So that is difficult. But the validation phase doesn't always trigger very hard, especially not in Nellis in Germany. So if you ignore that for a bit, you can still do quite a fundamental thing. And you can also go deeper, especially now the Nordics, I've seen multiple companies, quasi-companies, which you can hire, by the way, you don't have to use yourself, which try to emulate on a fundamental basis the grid itself checks where these knacks are and do their own grid calculations. And while there will never be 100% emulating what we are doing because we might take another assumption. There is no universal truth on a D-2 because you have to remember we calculate on two days ahead. A lot of things can change. So these assumptions are never perfect. And so you will never be exact the same. Even if we run, only if we use the exact same inputs, we will get the exact same output. But if you change minor inputs or you do a forecast again, you can get different results even for us. But you can still get the direction quite well. And that is, of course, the important part here. You don't need every exact thing. But if you know the general direction, if you know that due to some fundamental reason, the maximum expo-projective of France, of course, which is now dominating the European power system due to its nuclear fleet, if you can see that that lowers, which actually happened in May this year and also at the end of the summer, then you know that that is a fundamental impact. And you can use in your forecast, well, we have to tune down the net position in France and then other generators which are next in the merit order list might go up again. So that's a good point you make there. And that's also, of course, what forecasting the PTDF and the Diat Mark, when you want to simulate a Femia, that's where the value lies to see where on what position coordinate to do a settle within your domain and how is your domain different from tomorrow. And if you're very close to the edge of your polyhedral and your Mark's exports reduce because of reducing RAM, your spreads will increase and that's where you can make your money obviously. No indeed. And there are also special cases which you know that it can be very sensitive to these maximum boundaries. Because of course, again, they are a boundary. So I get a question a lot like there is some price spread, but the Netherlands is not until it's maximum net position that that is published. And that makes sense because the maximum position means that you optimize all other hubs for this single maximum. You really push it to the maximum. So this for example, possible to reach the maximum of Netherlands export, Germany would need to be importing 15 gigs. And that's not always a reasonable choice. So for the market coupling. However, if you know that that the max drops quite a lot, of course, that still has an effect. We will almost never reach 100 percent, but especially when the maximum drops, we can reach higher utilization levels. And there are also situations in which countries become very sensitive to that. So a good example of that is solar and Nellens. Right. So we have an absolutely mind boggling amount of solar PV in feet of the Netherlands. We are worth leading in that, especially in my computer. And so what you can see is on very sunny days, if our, if our max net position drops a little bit, it can immediately have a yeah, max export. It can immediately have a very large impact because solar in feet cannot be curtailed because we have the the the solar, the net metering, but also because most of our solar is on a regular cruise. And I don't know about you, but I don't pull the plug of my residential roof. And I mean, the office most don't most don't. So that is stuff that has to feed in. And then the prices crash. So if you know that there is less capacity there, if the Nellens is hugely sensitive to that. And then that can spread like an oil, an oil spill. So then we go very low, but also other countries go very low because the market coupling tries to export of the stuff in any way that is. What you typically see there is if you plot all the the price curves, after my one to 24 over the whole core market, you get this beautiful webs of spreads basically that it's never one market that crashes. But often it will be then one market that spikes as well where you really see this being pulled apart by these. Yes, then it's not uncommon to really move to the minus 3, 4, 500 euro per megawatt hour region, which we have seen. However, this is an extreme case, and this is only a couple of hours a year, but it does exemplify that the nanos is hugely sensitive to this. So in the past, it has happened that Jouk de Veris reasons, once there was an tooling failure at an individual TSO, not us, but a neighbor, but that impacts our capacities as well. And then the domain quaint became quite limited and then the markets react in a violent way because you cannot get loose of all this solar. The same goes for in a lesser extent. Large wind in feet, although wind plants can easily more easily curtail than residential solar. But even then, there is a certain minimum level that you want to output even a negative price, because otherwise you might damage your wind turbines. So if there's a nice storm, then a lot of infeed is there, and then you can also get these collapses and very sensitive situations to possible capacity reductions. Yes. Thank you, Frank. One of the last questions that I would like to ask, and this is maybe not directly related to your role, but when looking at this whole flow-based market coupling, a welfare optimization, basically simulating a Femia. If you want to forecast your day at market, you want to forecast the net position and you need your, you know, you need your bid enough occurs to simulate that. We have Nordic market now where we have flow-based market coupling in the Nordics. If you want to be able to simulate a Femia and make a proper day at forecast, assuming the flow-based market coupling in the Nordics, you want to forecast the price per price on. You need to have bid enough occurs per price on. In the Nordics, these are not available per price on. They're only available on an aggregated level per country. That's an issue to forecast these markets in a very precise manner. Do you know what is the driver behind this? What is the politics behind this? And will there be any change on one of these way of publishing? As far as I know, this is not only for Nordics. They had curves as published by the Nemos. The electricity exchanges are always the aggregated version. And this is mainly due to, I think this is mainly due to competition, to protect sensitive data. So this is a choice made on a European system level. Personally, I do not foresee that to change anytime soon. I know that I think Nemos do provide some details about anonymized bits, etc, which you can buy of them. But in terms of really settled curves, and it's also difficult because the curves that you see, the aggregated curves are already shifted by Euphemia. So the curve per hub is a very difficult question, because how do you define that? Do you take out all import export? Then that also can have some other effects. Do you simply shift them back? Do you adjust your merit order? So that is, and also then becomes the question, what is available to what? Because in principle, there are multiple solutions to this. There are multiple ways to pull the curve outside of each other. So I think that is difficult, and I do not foresee that to change for now. But it is good to state this is not only the Nordics. So in day ahead, curves that you can buy from either Nemos directly or various data aggregators is the same for the whole Estak region. The Nordics are more sensitive due to the nature of the. Yeah, they have smaller zones. So they have smaller zones. And then to model that, it's nicer to have a lot of curves. So that is completely understandable. In Europe, we have bigger zones, which is also something I'm in discussion right now. So then in terms of forecasting, it's slightly slightly different. But it's a difficult topic to fix. It is. Frank, I think we can nicely round up our conversation of today. I think lots of available inside interesting topics. And I think we could go on for a few more hours, but that might be for another day maybe. I would like to mostly thank you for today for all your expertise for being available to us and the audience. Super insightful. I hope to meet you again soon. And I do wish you all the best of your own. Yes, thank you very much. Let's say enjoy what you do. [BLANK_AUDIO] [BLANK_AUDIO] [BLANK_AUDIO] [BLANK_AUDIO] [BLANK_AUDIO]