‍Renewable energies are in high demand, but the European power grid is not designed to handle the fluctuating amounts of electricity. Marcel Wasserer, founder of the Viennese start-up enliteAI, wants to remedy the situation with an AI solution and relieve the strain on the grids by optimally distributing electricity production and demand.

Clean electricity is mostly discussed in terms of its generation, but its feed-in to outdated grids is much less present in discussions. Yet this is often a problem, as Europe's power grids are repeatedly pushed to their limits by the switch to renewable energies. Through so-called feed-in management measures, production from renewable energies is reduced in output or shut down completely when there is excess production, while in other regions electricity from gas-fired power plants has to be fed in at the same time to meet local demand. This costs money and does not make the best use of available resources.

On the one hand, this happens because wind and sunlight in particular are difficult to plan for, and too much of them causes overloads in parts of the grid; on the other hand, because the grid is not adequately designed to transport large amounts of electricity over long distances: When the infrastructure was built decades ago, most of the electricity came from nuclear, coal, and gas plants that could be ramped up or down as needed. Today, renewables play a much bigger role: in Austria, around 37% of total energy consumption in 2020 came from renewables, and for top-ranked Sweden, the figure was over 60%.

A Viennese company wants to tackle this problem with the help of artificial intelligence: enliteAI GmbH, which was founded in 2017 by Marcel Wasserer together with his former schoolmate Clemens Wasner and management consultant Johannes Stumtner. The start-up offers various services and products that help companies optimize profit or logistical processes. In the past years, enliteAI was able to implement more than 45 individual AI solutions in various companies before the strategy change towards own products. "Currently, we are active in the areas of logistics and production, energy and public infrastructure," Wasserer said.

The AI products from enliteAI work with reinforcement learning - unlike the previously successful neuronal networks, which learn to recognize already fixed results based on training data. With reinforcement learning, on the other hand, the AI acts with its environment - like in a chess or computer game. Each time the agent (as the AI is also called) sets an action, the further possible actions change. Training with simple data sets is therefore not possible; instead, simulations are used. In most cases, the agent only receives feedback at the end of the runtime as to whether the decisions made were correct in the long term.

In order to be able to use AI for real decision-making problems such as supply chains, suitable training environments are needed. For a project on supply chain optimization, enliteAI, as a pioneer in this field, had to develop one itself. "After working on this for about two years and being so deep into this topic, we started looking around to see where else we could use this," Wasserer says.

They came across the Learning to Run a Power Network competition, which has been hosted by French network operator RTE since 2016. "RTE realized that the conventional way of controlling power grids is simply not enough as complexity continues to increase," Wasserer says. AI solutions can help turn the many adjusting screws and find new combinations that relieve the strain on the grids. The system should serve as an advisor and not take over the control itself. enliteAI saw the opportunity to use the new training environment, which is now published as MazeRL on Github, for this problem - with success: 2022 was able to win first place in said competition.

In parallel with the project development, the start-up was already exchanging ideas with network operators and developing its own product in order to be able to bring the technical approach to optimization into operational planning as soon as possible. In addition, the product "detekt" was launched in 2022, which uses image material to detect and localize road damage and road markings. For CTO Wasserer, that's the biggest achievement: "We've managed to develop two extremely sophisticated products as a small company with our own resources." It is also important for him to be deeply involved in the technical details of the projects, because otherwise he finds it difficult to make the right decisions. In addition to the development process, he is also responsible for technical innovation in the company - and for "bringing the right people together."

This already began with the founding: Wasner and Stumtner come from management consulting backgrounds, while Wasserer provided the technical expertise. "We realized that together we were ideally positioned to bring AI into companies," says Wasserer, who has always been fascinated by AI. As a child, he was also particularly interested in computer games: "I didn't want to play them, though - I reprogrammed them instead." That led to a career as a game developer after studying computational intelligence and technical mathematics. Wasserer built a 15-person team that programmed "Ski Challenge" among other successful games. "To pursue my AI interest, I had to wait another ten years - until computers were powerful enough," Wasserer describes.

With enliteAI he could now realize his passion. The team, which now has 15 members, has had some successes in recent years (for example, annual growth of 189%) and has built up a prestigious clientele of a total of twelve ATX and DAX companies. Then, last fall, came the invitation to a workshop at the most important AI conference, the "NeurIPS", in which application fields for reinforcement learning were presented. For Wasserer, it was a milestone to be able to present his project together with the pioneers of the AI industry - Deepmind, Open AI, Amazon and Google.

In the future, Wasserer hopes to identify even more fields where AI can make an important contribution and show what is possible through AI in the field of electricity. "We know from our studies that we could save up to 60% CO2. We want to bring that to the grid now," he says. Among operators, he says, the pressure is on to secure grids against congestion, but because the topic is still very new, the appropriate data must first be collected and simulations created. Wasserer also makes it clear that AI cannot be the only solution for making Europe's power grid more secure - storage options and grid expansion are just as important. Says Wasserer, "But you also have to take advantage of the internal flexibility of the grids that our systems can create."

‍

Marcel Wasserer, Clemens Wasner und Johannes Stumtner gründeten 2017 die enliteAI GmbH. Das Unternehmen wurde für seine Stromnetz-KI 2022 mit dem Sonderpreis „Verena“ des österreichischen Innovations-Staatspreises ausgezeichnet.

Fotos: David Visnjic

‍