Episode Transcript
[00:00:10] Speaker A: Welcome to the EU Energy Projects Podcast, a podcast series from Enlida and France focusing on the clean energy transition for the European Union and the EU Commission funded energy projects that will help us achieve it.
[00:00:28] Speaker B: Welcome to today's EU Energy Project podcast and today I'm joined by Dr. Sparos Muzakitis who is the coordinator of the Energyguard project and he's going to talk about that project today.
Welcome Sparos.
[00:00:45] Speaker C: Thank you very much for the invitation. I'm very happy to be here.
[00:00:49] Speaker B: Can you please start with a brief introduction to the energyguard project and what its aims are?
[00:00:55] Speaker C: Yes. Energyguard is a three year Horizon Europe project that has been coordinated by the Institute of Communications and Computer Systems. We are currently 60 partners from seven countries who are combining our expertise in the energy sector. Artificial intelligence. Our main goal is to develop and operate an open test and experimentation facility that thanks to, to give the opportunity to developers and data scientists that are currently working on AI solutions for the specifically for the energy sector to validate their products.
And we are offering an experimentation facility that is under real world conditions.
And we are offering this in a cost effective and easy way. And the idea is to bring these AI solutions closer to the needs of the energy sector market. So we're bringing major physical and virtual facilities all across Europe that they're covering different parts of the energy value chain.
And so basically what we're giving is bridging the gap between developing a promising AI solution that hasn't been tested in real conditions and in order to validate, utilizing data from real energy systems. In a nutshell, what then are the
[00:02:24] Speaker B: specific challenges around growing the use of AI in the energy sector?
[00:02:32] Speaker C: There are several interconnected challenges.
First of all, energy is a special sector because we are talking about critical infrastructures that can affect all aspects of our society.
And this is something that we can see, unfortunately, in recent wars, how they pick the energy infrastructure as their primary target.
So the consequences of an AI, of an error in an AI system can go way far beyond poor prediction. So it can affect, for instance, it can affect the grid variability, it can affect the environment, it can affect the safety of infrastructure and even people. And it can have a major risk in society.
So this means that an AI model that it's been created in the energy sector cannot simply be adopted before it performs well in an individual laboratory.
So it needs to be robust and secure and to be tested under realistic conditions.
We see now that we live in an era of AI hype.
You have seen thousands of products that have flooded the market every day. We get news about new products that their reliability and security risks are unknown. So there is a huge need for trust in the energy sector, because if a risk is realized, can have catastrophic cascading effects to the energy systems.
So another challenge is that we have a growing use of AI. But at the same time, especially for the energy sector, there is limited access to good data and infrastructure.
And many AI solutions today, they don't have very reliable results because these unreliable results mainly come from the lack of real ground truth observations, shall we say, meaning observations that are real and they are not synthetic data, for instance.
So in general, the AI tools are as good as their training data.
And the problem is also that you cannot validate a product on a live electrical grid on one hand, but on the other hand, you need very good ground truth observations.
And to make matters worse, the data are usually distributed across many organizations.
Usually they are in different formats and under different levels of quality.
Access might be restricted for commercial or privacy issues.
As we have seen also during our projects, many data may have incomplete or inconsistent records, there might be errors in measurement and other issues. All this affect in the end the accuracy and the fairness of an AI solution.
Another major issue is cyber security, that it will be a growing concern, especially with the advancement of AI.
So as the grids are now becoming smarter, we have interconnected devices and smart meters that become more widely used everywhere.
But this means also the potential attack surface also grows.
It's difficult to train and test an AI system against every unusual event or operational condition it might encounter. So it's very important. A challenge is how you can test this AI model in order to be secure with the new advancements in AI.
For instance, we hear a lot about AI agents, that they bring new opportunities, but they also bring a lot of risks.
And last but not least, we have also many regulatory issues.
The developers must navigate the European AI act, so they need and there are many rapid changes in the AI landscape and the regulatory scene. So it's difficult to navigate and you need to know if you are aligned with the legislation.
This all brings a lot of cost to SMEs and startups in order to develop and use AI for the energy sector.
[00:07:11] Speaker B: I mean, you've already alluded to the importance of testing, but can you go into a little more detail about why testing and experimentation is so important and what the focus of those activities are?
[00:07:28] Speaker C: Yes, the test experimentation are very important because a successful model, as we call a laboratory, so that the data scientist does in, in the university, let's say, or an SME that creates an AI model, it might give him very good accuracy results within the laboratory.
But this doesn't mean that the solution is ready for operation.
The energy systems are very complex and dynamic and conditions can change, for instance, because of the weather, because of the equipment behavior, because of unexpected disturbance in the grid.
An AI model needs to be tested against realistic data and under a wide range of operational scenarios.
And it's many times that data scientists or new AI developers don't even know what to test.
Even knowing what you need to test you your models against it is very important.
And if we are talking about deeper integration of AI, for instance, when we are talking about agentic workflows, you need to check all potential outcomes and interdependencies. You need to ensure you have the guardrails to ensure that there are a lot of safety layers.
So in a nutshell, it's very important because it allows developers to understand how their solution will behave in real condition, like on a real grid, and to identify the weakness and risks before they actually adopt this AI solution to the grids.
For the energy sector, we need to move carefully from the development of an AI solution to a critical infrastructure and we need to test it with realistic conditions.
So this is the perspective of data scientists, but it's also very important from the policymaker's perspective in order to understand what are the risks and legal problems or obstacles from the policies that have or have not yet been applied.
[00:09:50] Speaker B: So, turning to any energy garden, in particular, what facilities does the project bring together and how were these selected to
[00:09:58] Speaker C: cover the full value chain within energyguard, we bring five top European nodes of testing facilities.
First, in Portugal we have RD Nester that provides a digital twin, meaning a virtual representation of the Portuguese transmission network together with real time power system simulation.
So this allows to test AI solutions at the transmission grid scale in order, for instance, to see the grid operation, how your AI solutions performs for a real realistic grid operation. Stability forecasting of the demand and fault analysis.
Second, we have Sedercmat that it's basically a living lab of a smart microgrid that contains wind and photovoltaic generations. It has hydropower, it has battery storage, it has electrical vehicles and many more. So it offers an environment for testing, forecasting of renewable the flexibility, management of the grid, storage optimizations and in general the operation of the microgrid.
The third facility that we offer, basically it's a combination of further facilities for hydrogen generation and this being operated by cea, Cartif and Blue Energy together also with cmat so we have different, basically in these facilities we have different type of electrolyzers.
So these facilities can help AI developers do prognostics and diagnostics of hydrogen generation facilities.
The fourth facility that we have is a high fidelity digital twin of multi apartment residential buildings in Iriga in collaboration with Riga.
And we also have a renewable energy community in Italy, in Adirondoco, where we can test AI for photovoltaic forecasting, battery usage, energy sharing, local optimization and predictive maintenance.
So the different parts of these facilities were not simply selected to have facilities in different countries, but the idea was to have a complete experimentation environment that has the diversity and complexity of the the whole system.
[00:12:43] Speaker B: What criteria then are being put in place to test an experiment against? Because that's obviously the important aspect.
[00:12:53] Speaker C: Yes, and this is a very good question, because things are a little bit more complicated than that. We're not applying one universal pass or fail set of criteria for every AI solution because we are talking about the entire energy value chain and we are talking about energy applications that are very different.
What we're doing is to combine application specific performance indicators. So if we have for instance a forecasting or if we have a predictive maintenance, we need to see each application specifically. So we have indicators for the specific set of different problems. And then on top of that we have a common framework for the trustworthiness of the AI, the cybersecurity and the regulatory alignment.
So at the data science level, let's say at operational phase, for each experiment, we begin with a baseline and a defined measurement and verification plan.
So the relevant criteria depend on the case.
A model for forecasting accuracy has different criteria and thresholds from the resilience, for instance of the microgrid or for the diagnostics and remain life predictions.
So this is about the actual application. A second layer of criteria has to do with the quality of the AI solution where we assess code quality, the model accuracy and the robustness. There are specific metrics for that long range of metrics. And at the same time we have also criteria around explainability, uncertainty, human oversight, fairness, privacy. Basically, we're drawing all the criteria from the assessment list of trustworthy AI solutions that is coming from the EU that has a framework and has all these dimensions.
Lastly, we have cybersecurity assessment that covers the complete deployed application.
And this has to do with number of vulnerabilities, found, software dependencies exposed, secrets, any kind of malicious code identified within.
And finally we also assess it and we give some direction and guidelines to the users about the applicable regulatory and Ethical framework, in particular the aqa.
So we test whether the AI works and whether it will continue to work under difficult conditions and whether it's secure and trustworthy.
[00:15:34] Speaker B: So looking ahead, are there plans to expand the TEF nodes to other countries and sort of maybe longer term. What plans are there for the long term operation of these nodes?
You know, particularly beyond the ending of the project? Is the idea for them to operate on a commercial basis or what's the thinking there?
[00:15:58] Speaker C: Yes, so basically expansion is part of our long term vision. What we are currently building is what we call minimum viable product, the mvp. With the initial features, there is not a fixed list of additional countries or facilities.
The project began with five facilities that cover the entire value chain to cover a large majority of the needs.
And the architecture that we used is modular and extensible, meaning that we can up to the common framework that we have for testing AI pipelines and having an acceptance environment for testing the applications. Against this criteria we can expand it to more nodes.
So we will focus first on nodes that provide for instance a missing technology not covered from the current nodes, an energy vector or a testing capability.
So we mostly plan to do that based on the needs and based on some missing, let's say, or some identified needs and it won't be necessary per country.
About the commercialization, long term operation was considered from the beginning of the project. It's the type of the project innovation action and it needs to be done from the beginning of the project and rather been an afterthought.
So we have already started developing a billing system that will be tied to the resources usage currently being developed.
And we also want to link it with the marketplace of the European AI and demand platform.
So we have already identified the exploitable assets and how we will what we need in terms of cost in order to sustain it.
And we have added also features for monitoring the resources to having subscription plans or usage based subscription. These are under development.
So yes, in general we are also talking with the other energy devs in order to see collaborations in terms of the future operation and commercialization of the project.
[00:18:28] Speaker B: Yes, I was going to ask about how energyguard fits alongside the other TEF projects and what benefits you envisage emerging from cooperation with these projects.
[00:18:43] Speaker C: The European Union has identified the need in order to have test experimentation facilities in many sectors. They quickly identified this need. So there are testing experimentation facilities for health, for agriculture, for smart cities and other sectors.
Energy Guard's role within this TEV project is to bring the energy sector perspective into the wider European goal for having AI test and experimentation facilities.
So we provide a link with the European End Demand platform, which basically is what the Commission plans to have as one stop shop for AI solutions in Europe.
And one of our formal of our primary objectives is to synchronize and collaborate with devs from the other sectors and to see also to exchange what they have been doing, exchange of knowledge tools and also promote each other. And this includes also collaboration with digital innovation hubs, European data spaces and AI factories as well as other European initiatives.
So in a simple term, energyguard is an energy focused contributor to the TEF projects.
[00:20:06] Speaker B: Good. Well on that note, I'd like to thank you Sparos. It's important to hear about projects such as these that are really going to help drive forward the use of AI in Europe. So thank you very much for your time.
We look forward to following the project as it proceeds. Thank you.
[00:20:24] Speaker C: Thank you very much for inviting me and we'll be happy also to invite you also as a user to the platform.
[00:20:30] Speaker B: Thank you.
[00:20:33] Speaker A: You've been listening to the EU Energy Projects Podcast, a podcast brought to you by Enlit and Friends. You can find us on Spotify, Apple and the Enlit World website.
Just hit subscribe and you can access our other episodes too. I'm already Direct Limo. Thank you for joining us.
